A power station parameter matrix analysis method

Abstract

The utility model provides a kind of power station parameter matrix analysis method, belongs to power station fault diagnosis technical field, solve how to design a kind of power station parameter matrix analysis method, solve the numerous power station parameters, data effective information density is low, lead to the problem that fault cause analysis is difficult, and the massive data stream generated occupies a large amount of storage space, by power station parameter data matrixization representation, constructs parameter data matrix, by matrix operation, the rank of matrix is as quantization index, effectively distinguishes the information density of parameter, the correlation between different parameters can be analyzed by the multi-parameter matrix constructed, not limited by parameter number and time sequence length, and by matrix similarity operation, judge the similarity of unit state, provide quick judgment basis for similar fault in later period;Dimensionality reduction is carried out to operation data matrix using matrix singular value decomposition method, retain the part of large singular value proportion, can extract fault feature, realize compressed storage, save storage space.

Description

Technical Field

[0001] This invention belongs to the field of power plant fault diagnosis technology and relates to a method for power plant parameter matrix analysis. Background Technology

[0002] Power plant equipment malfunctions often manifest as abnormalities in parameters such as temperature, pressure, and flow rate. Operators need to quickly determine the cause of the malfunction based on existing indicators to make operational adjustments or perform equipment repairs. Therefore, power plants deploy numerous monitoring points for real-time monitoring. However, the large number of parameters and the low density of effective data information make malfunction analysis difficult, and the resulting massive data streams occupy a significant amount of storage space.

[0003] To extract the feature information contained in fault data, effective signal analysis methods are needed. Meanwhile, to save storage space, power plant data storage systems perform periodic data cleanup on historical data, diluting it. This approach may result in the loss of valuable data.

[0004] In existing technologies, the literature "Matrix Analysis Method for Thermal Power Plant Thermal Systems" (Xing Deshan, Journal of North China Electric Power University, July 2006) and "Matrix Analysis Method for Thermoeconomics of Thermal Power Plant Thermal Systems" (Guo Jianglong, Proceedings of the CSEE, January 2004) aim to derive a complete and universal heat balance or A system of equilibrium equations. This involves combining several known formulas with fixed parameters to form a system of equations. Because the system of equations can be simplified using determinants, the aforementioned literature utilizes the simplification capabilities of matrices to derive a general calculation formula, which can then be substituted into subsequent calculations using specific parameter values. The literature "Optimization Analysis of Substation Monitoring Information Point Tables Based on Improved Neural Networks" (Zhang Yu) uses a matrix that is the weight matrix in the algorithm, with a fixed form and calculation method.

[0005] Many signal analysis methods are effective at processing single-parameter time series data. However, power plant parameters have clear physical meanings and strong correlations between them, a crucial characteristic often overlooked by these methods. While big data mining and analysis methods can uncover correlation rules between parameters and thus determine their correlations, these methods have limited ability to mine correlation rules between multiple parameters and struggle to analyze the correlations between unit states.

[0006] Currently, the power plant data storage system uses an equal-interval data cleaning method. This method does not take into account the inherent characteristics of different types and parameters of data, resulting in the loss of effective data. Summary of the Invention

[0007] The purpose of this invention is to design a power plant parameter matrix analysis method to solve the problems of numerous power plant parameters, low effective information density of data, difficulty in analyzing the causes of failures, and the massive data streams that occupy a large amount of storage space.

[0008] The present invention solves the above-mentioned technical problems through the following technical solutions:

[0009] A method for matrix-based analysis of power plant parameters, which represents power plant parameter data in a matrix format:

[0010] A single-parameter matrix of power plant parameters is constructed. The single-parameter matrix is ​​used to analyze the changes of specific parameters in different periods. The rank of the single-parameter matrix is ​​calculated as a quantitative index of the information density of the specific parameters. Small fluctuations in the time series of specific parameters are eliminated by superimposing several random white noise matrices. Effective information is extracted by performing singular value decomposition on the single-parameter matrix, thereby achieving compression of specific parameter data.

[0011] A multi-parameter matrix of power plant parameters is constructed. This multi-parameter matrix is ​​used to analyze the changes in different parameter values ​​within the same time period. The correlation between multiple parameters is analyzed through matrix similarity operations to determine the similarity of unit states, thereby identifying similar fault states between units.

[0012] The technical solution of this invention represents power plant parameter data in a matrix form, constructs a parameter data matrix, and uses matrix operations to use the rank of the matrix as a quantitative indicator, effectively distinguishing the information density of parameters. The constructed multi-parameter matrix can analyze the correlation between different parameters, without being limited by the number of parameters or the length of time sequence. Furthermore, through matrix similarity operations, it can determine the similarity of unit states, providing a rapid basis for judgment when similar faults occur later. The method of singular value decomposition is used to reduce the dimensionality of the operating data matrix and retain the part with a large proportion of singular values, which can extract fault features, achieve compressed storage, and save storage space.

[0013] As a further improvement to the technical solution of this invention, the method for constructing a single-parameter matrix of power plant parameters is as follows: For a data sequence of a single specific parameter, it is truncated according to a time period T, with different periods serving as different row vectors of the matrix; simultaneously, the data within a period is sampled at equal intervals according to the characteristics of the parameter, diluting the data sequence, and the matrix is ​​constructed as follows:

[0014]

[0015] In the above formula, the subscripts 1…n are the period labels, and different subscripts represent parameter values ​​in different periods; the superscripts 1…t are the sequence numbers of the data in a period, and different superscripts represent the parameter values ​​arranged in chronological order in a certain period.

[0016] As a further improvement to the technical solution of this invention, the method of using the rank of a single-parameter matrix as a quantitative indicator of the information density of a specific parameter is as follows: When the parameter is stable, the different row vectors in the matrix are all equal or in the same cluster, and the rank is very small; when the equipment or unit is in an abnormal state, the parameter changes significantly, and the time series changes are different from the normal sequence, which is reflected in the increase of the rank of the matrix; when the parameter time series of each period is different, the matrix is ​​a full-rank matrix; a full-rank matrix means that the parameter changes more frequently during this period, and the equipment or unit is unstable.

[0017] As a further improvement to the technical solution of the present invention, the method of eliminating small fluctuations in the timing of specific parameters by superimposing several random white noise matrices is as follows: For parameters that still fluctuate within a small range when stable, this manifests as glitches in the digital signal. For glitches, the fluctuation range is relatively small compared to the true value. Therefore, a random white noise matrix is ​​generated based on the fluctuation range and superimposed on the original data signal. Taking advantage of the characteristic that the mean of white noise is 0, the noise of the signal itself is masked by introducing uniformly distributed white noise into the original data multiple times.

[0018] As a further improvement to the technical solution of this invention, the method for compressing specific parameter data by extracting effective information through singular value decomposition of a single-parameter matrix is ​​as follows: The parameter matrix is ​​decomposed into several low-rank matrices, as expressed by the formula:

[0019]

[0020] Larger singular values ​​correspond to parts containing more information, while smaller singular values ​​correspond to parts containing less information. The singular values ​​are sorted in descending order. The sum of a few terms accounts for more than 70% of the total singular value sum, so the matrices corresponding to these terms are retained. Here, σ1…σ r The singular values ​​are arranged in descending order, and their indices u and v are the left and right singular vectors corresponding to that singular value, respectively.

[0021] As a further improvement to the technical solution of this invention, the method for constructing a multi-parameter matrix of power plant parameters is as follows: Multiple different parameter sequences are truncated according to the same time period, and the time series are sampled at equal time intervals to obtain the parameter values ​​of different parameters arranged in chronological order within the same time period, constructing a matrix in the following form:

[0022]

[0023] In the above formula, different row vectors represent the time point sequence of different parameters within the same time period; the subscripts 1…t in the above formula represent the sequence number of the data of the parameter within a specific period.

[0024] As a further improvement to the technical solution of this invention, the method for analyzing the correlation between multiple parameters through matrix similarity operations is as follows: Different parameter time series within the same time range are used as different rows of a matrix to generate a matrix, where each row's letter represents a parameter. Through matrix operations, a multi-parameter rate of change matrix is ​​constructed as follows:

[0025]

[0026] Different letters in this matrix represent different parameters. Each row of vectors represents the trend of the parameter's change within the same time period. When the trends of two parameters are consistent, they can be represented by a set of vector bases. When there is a positive correlation, the vector base is multiplied by a positive number, and when there is a negative correlation, the vector base is multiplied by a negative number. Therefore, when the multi-parameter matrix is ​​not a full-rank matrix, it indicates that the parameters are correlated.

[0027] The advantages of this invention are as follows: The technical solution of this invention represents the power plant parameter data in a matrix form, constructs a parameter data matrix, and uses matrix operations to use the rank of the matrix as a quantitative indicator, effectively distinguishing the information density of the parameters. The constructed multi-parameter matrix can analyze the correlation between different parameters, without being limited by the number of parameters or the length of the time sequence. Furthermore, through matrix similarity operations, the similarity of the unit states can be judged, providing a rapid basis for judgment when similar faults occur later. The method of singular value decomposition is used to reduce the dimensionality of the operating data matrix and retain the part with a large proportion of singular values, which can extract fault features, achieve compressed storage, and save storage space. Attached Figure Description

[0028] Figure 1 This is a graph recording the unit's output parameters. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0031] The power plant parameter data is collected in time-point sequence. To enable matrix analysis, a transformation method needs to be designed: For the data sequence of a single power plant parameter, it is truncated according to the time period T, with different periods serving as different row vectors of the matrix; simultaneously, the data within a period (one row) is sampled at equal intervals according to the characteristics of the parameter, diluting the data sequence, and constructing the matrix as follows:

[0032]

[0033] In the above formula, the subscripts 1…n are period indices, with different subscripts representing parameter values ​​within different periods; the superscripts 1…t are the sequence numbers of the data within a period, with different superscripts representing the parameter values ​​arranged sequentially in time within a certain period. Constructing such a matrix allows for analysis of the parameter's variation across different periods, analysis of the generality of data trends, and facilitates the identification of outlier parameter values.

[0034] For multiple parameters, their correlations can be analyzed by constructing a single matrix. Different parameter sequences are truncated to the same time period, and the time series are sampled at equal time intervals to obtain the parameter values ​​arranged chronologically within the same time period. The matrix structure is as follows:

[0035]

[0036] In the above formula, different row vectors represent the time point sequence of different parameters within the same time period. The subscripts 1…t in the formula indicate the sequence number of the parameter's data within a specific period. By analyzing different parameter values ​​within the same time period, the correlation between them can be explored, thus providing a basis for subsequent fault diagnosis based on parameter anomalies.

[0037] For some parameters, even when stable, they still fluctuate within a small range, which manifests as glitches in the digital signal. (See...) Figure 1 As shown in the figure, the normal startup process on June 15th showed a relatively stable parameter timing sequence, but glitches still existed in the unit's output parameters. During the subsequent two startups, while glitches remained in the parameter sequence, the parameter values ​​also fluctuated significantly.

[0038] Therefore, a glitch elimination method needs to be designed before performing parameter matrix analysis. (Observation) Figure 1 As can be seen, the fluctuation range accounts for a small proportion of the true value. Therefore, a random white noise matrix can be generated based on the fluctuation range and superimposed on the original data signal. Taking advantage of the fact that the mean of white noise is 0, uniformly distributed white noise is introduced into the original data multiple times to mask the noise of the signal itself through multiple artificially added noises.

[0039] For power plant parameters, most data have a range of variation. When the unit or equipment is in a stable state, parameter values ​​remain stable or change smoothly; when equipment fails, parameters containing fault information may change drastically in a short period. For a parameter matrix, when the equipment is in a stable state, the element values ​​within the matrix are the same or similar. Different row vectors in the matrix are all equal or in the same cluster, resulting in a very small rank. In special cases, all element values ​​in the matrix are constants, resulting in a rank of 1. Therefore, when a parameter matrix can be represented by its own submatrices, it indicates that there is similarity within the matrix, reflecting the stationarity of the parameters and the stability of the equipment state, i.e., a very small rank.

[0040] For a parameter matrix A (m×n matrix), the elements at the intersection of any α rows and β columns constitute a k-order submatrix of A. The determinant of this submatrix is ​​called a k-order minor of A. When the equipment or unit is stable, the parameter matrix A exhibits internal similarity and can be represented by this k-order minor after computation. Therefore, the rank of this k-order minor is also the rank of the parameter matrix A, reflecting the information density of matrix A. When the equipment or unit is in an abnormal state, the parameters change significantly, the time-series point changes are irregular, and different row vectors lack similarity. Matrix A approaches a full-rank matrix, with a very large rank. From the above description, it can be seen that the rank of a parameter matrix can reflect the information density of the data.

[0041] The correlation between power plant parameters is reflected in their changing trends; that is, a change in one parameter causes a positive or negative correlation with another parameter. Based on this, a multi-parameter rate of change matrix is ​​constructed to achieve correlation analysis of the parameters. Different parameter time series within the same time range are treated as different rows of a matrix, generating a new matrix. Each row's letter represents a parameter, and matrix operations are used to construct the following matrix.

[0042]

[0043] Different letters in this matrix represent different parameters, and each row of vectors represents the trend of that parameter's change over the same time period. When two parameters have the same trend, they can be represented by a set of vector bases. A positive correlation results in the vector base being multiplied by a positive number, and a negative correlation results in the vector base being multiplied by a negative number. Therefore, when the multi-parameter rate of change matrix is ​​not a full-rank matrix, it indicates that the parameters are correlated.

[0044] By combining multiple parameters monitoring the status of a device or unit and constructing them according to the above method, a matrix A reflecting the unit's status is obtained. When the unit is running stably under the same operating conditions, the parameter values ​​are basically consistent, and the matrix values ​​of the two status points are almost identical, indicating similarity. The unit may also have similar states under different operating conditions, as shown in Table 1 below:

[0045] Table 1: Unit Status under Different Typical Operating Conditions

[0046]

[0047]

[0048] As can be seen from the table above, when the unit is under typical operating conditions, such as 75% load and 50% load, some parameter values ​​may differ, but they exhibit a proportional relationship, and the trends of change between the parameters are consistent. The unit state matrix A itself describes the relationship between parameters under a specific state. If the relationship between parameters is consistent in two states, the unit states are similar.

[0049] For a given state matrix A1, it describes a dynamic change (dynamic law between parameters) in the coordinate system (current unit state). If there exists another state matrix A2 in a different coordinate system (another unit state) that describes the same dynamic change (dynamic law between parameters), then matrices A1 and A2 are similar, i.e., A1~A2.

[0050] Power plants prioritize fault analysis, but the low density of effective information in parameter data makes analysis difficult. Power plants typically store data for entire time periods, consuming significant storage space and making it difficult to detect anomalies. By constructing a parameter matrix A using the matrix method described above, we obtain a parameter matrix A. Matrix A describes the overall dynamic changes of parameters over a period of time. Multiple patterns of change may exist in matrix A, with the direction and magnitude of change corresponding to different eigenvectors and eigenvalues. These patterns vary in strength; stronger patterns are more valuable for analyzing fault information contained in anomalies. Therefore, it is crucial to retain eigenvectors with larger eigenvalues ​​(larger dynamic changes).

[0051] As we know from the construction method of the parametric matrix, matrix A is not necessarily a square matrix, and therefore matrix A does not necessarily have eigenvalues ​​and eigenvectors. For any parametric matrix A, if we want to preserve some strong regularities (certain directions with large amplitude of change), we should preserve the singular vectors corresponding to the larger singular values. The largest singular value corresponds to the direction with the greatest stretching effect, and the direction with the greatest stretching is the direction with the largest data variance distribution, containing the most information. Therefore, to obtain the singular values ​​of the parametric matrix A:

[0052]

[0053] Here σ1…σ r The singular values ​​are arranged in descending order, with u and v corresponding to their indices as the left and right singular vectors, respectively. Larger singular values ​​contain more information, while smaller singular values ​​contain less. If the sum of several terms accounts for more than 70% of the total singular value sum, the matrices corresponding to these terms are retained, achieving dimensionality reduction storage.

[0054] The goal of this invention is to construct a holistic method based on matrix analysis to analyze various issues such as power plant data processing, parameter correlation, system state identification, and power plant parameter feature mining and storage. It emphasizes utilizing matrix properties, principles, and theorems to analyze the inherent patterns in power plant data through custom-designed special structures, achieving in-depth data mining. The invention also analyzes and demonstrates why matrix knowledge can be used to analyze such problems, addressing the theoretical support for pattern analysis. Since power plant parameters are time series, this invention first proposes a matrix construction method for single-parameter and multi-parameter time series, and this constructed matrix is ​​not limited to specific parameters. For the common spike phenomenon in power plant data, a method for eliminating spikes is proposed. The purpose of constructing parameter matrices is to analyze the inherent properties of parameters and the correlation between parameters: 1) The information density attribute of a parameter determines the degree of dilution in the power plant's storage of that parameter data. This invention proposes using the rank of the matrix as an information density index and analyzes its feasibility. This index is not limited to specific parameters and has universality. 2) The physical meaning of power plant parameters is clear. This invention proposes a multi-parameter matrix construction form and provides a correlation analysis method for calculating the rank of the matrix, and analyzes its feasibility. No single parameter can fully reflect the status of equipment or units. This invention proposes using a multi-parameter matrix as the status points of equipment or units and analyzes the feasibility of using matrix similarity for equipment status identification. Power plant data is massive and is currently typically compressed for storage. However, how to achieve compression while preserving effective data information is a challenge. This invention proposes using singular value decomposition to compress storage while mining information features.

[0055] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for matrix analysis of power plant parameters, characterized in that, Represent the power plant parameter data in a matrix format: A single-parameter matrix of power plant parameters is constructed. The single-parameter matrix is ​​used to analyze the changes of specific parameters in different periods. The rank of the single-parameter matrix is ​​calculated as a quantitative index of the information density of the specific parameters. Small fluctuations in the time series of specific parameters are eliminated by superimposing several random white noise matrices. Effective information is extracted by performing singular value decomposition on the single-parameter matrix, thereby achieving compression of specific parameter data. The method for using the rank of a single-parameter matrix as a quantitative indicator of the information density of a specific parameter is as follows: When the parameter is stable, the different row vectors in the matrix are all equal or in the same cluster, and the rank is very small; when the equipment or unit is in an abnormal state, the parameter changes significantly, and the time series changes are different from the normal sequence, which is reflected in the increase of the matrix rank; when the parameter time series of each period is different, the matrix is ​​a full-rank matrix; a full-rank matrix means that the parameter changes frequently during this period, and the equipment or unit is unstable. A multi-parameter matrix of power plant parameters is constructed. The multi-parameter matrix is ​​used to analyze the changes of different parameter values ​​within the same time period. The correlation between multiple parameters is analyzed through matrix similarity operation to determine the similarity of unit states, thereby determining the similar fault states between units. The method for analyzing the correlation between multiple parameters through matrix similarity operations is as follows: Different parameter time series within the same time range are used as different rows of a matrix to generate a new matrix. Each row's letter represents a parameter. Through matrix operations, a multi-parameter rate of change matrix is ​​constructed. Different letters in the multi-parameter rate of change matrix represent different parameters, and each row vector represents the parameter's change trend within the same time period. When two parameters have the same change trend, they are represented by a set of vector bases. For positive correlation, the vector base is multiplied by a positive number; for negative correlation, the vector base is multiplied by a negative number. Therefore, when the multi-parameter matrix is ​​not a full-rank matrix, it indicates that the parameters are correlated.

2. The power plant parameter matrix analysis method according to claim 1, characterized in that, The method for constructing a single-parameter matrix of power plant parameters is as follows: For a data sequence of a single specific parameter, it is truncated according to a time period T, with different periods serving as different row vectors of the matrix; simultaneously, the data within a period is sampled at equal intervals according to the characteristics of the parameter, diluting the data sequence, and the matrix is ​​constructed as follows: In the above formula, the subscript 1… n The subscripts are used for period designations; different subscripts represent parameter values ​​within different periods; superscripts 1… t It is the sequence number of the data within a period. Different superscripts indicate the parameter values ​​arranged in chronological order within a certain period.

3. The power plant parameter matrix analysis method according to claim 1, characterized in that, The method for eliminating small fluctuations in the timing of specific parameters by superimposing several random white noise matrices is as follows: For parameters that still fluctuate within a small range when stable, this manifests as glitches in the digital signal. For glitches, the fluctuation range is relatively small compared to the true value. Therefore, a random white noise matrix is ​​generated based on the fluctuation range and superimposed on the original data signal. Taking advantage of the characteristic that the mean of white noise is 0, uniformly distributed white noise is introduced into the original data multiple times to mask the noise of the signal itself through multiple artificially added noises.

4. The power plant parameter matrix analysis method according to claim 1, characterized in that, The method for compressing specific parameter data by extracting effective information through singular value decomposition of a single-parameter matrix is ​​as follows: The parameter matrix is ​​decomposed into several low-rank matrices, as expressed by the formula: Larger singular values ​​correspond to parts containing more information, while smaller singular values ​​correspond to parts containing less information. The singular values ​​are sorted in descending order. The sum of a few terms accounts for more than 70% of the total singular value sum; the matrices corresponding to these terms are retained. σ 1… σ r The singular values ​​are sorted in descending order, and their corresponding indices are... u and v These are the left and right singular vectors corresponding to the singular value, respectively.

5. The power plant parameter matrix analysis method according to claim 1, characterized in that, The method for constructing a multi-parameter matrix of power plant parameters is as follows: Multiple different parameter sequences are truncated according to the same time period, and the time series are sampled at equal time intervals to obtain the parameter values ​​of different parameters arranged in chronological order within the same time period. The matrix form is as follows: In the above formula, different row vectors represent the time sequence of different parameters within the same time period; the subscript 1... t This indicates the sequence number of the data within a specific period for this parameter.

6. The power plant parameter matrix analysis method according to claim 1, characterized in that, The multi-parameter rate of change matrix is ​​as follows: Different letters in this matrix represent different parameters. Each row of vectors represents the trend of the parameter's change within the same time period. When the trends of two parameters are consistent, they can be represented by a set of vector bases. When there is a positive correlation, the vector base is multiplied by a positive number, and when there is a negative correlation, the vector base is multiplied by a negative number. Therefore, when the multi-parameter matrix is ​​not a full-rank matrix, it indicates that the parameters are correlated.

Images