A fault warning method for power generation equipment based on N-order deep multivariate state estimation

The MSET algorithm is optimized by the N-order depth multivariate state estimation method, which solves the problem of inaccurate early warning of power generation equipment fault warning methods in the prior art, and realizes more accurate equipment status prediction and early warning.

CN117743742BActive Publication Date: 2024-12-03BEIJING DATANG GOHIGH DATA NETWORKS TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202311862497.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-12-29
Publication Date
2024-12-03
Estimated Expiration
2043-12-29

AI Technical Summary

Technical Problem

In the existing power generation equipment fault warning method, when the MSET algorithm is used, the normal value of the current working condition cannot be accurately predicted when the equipment is operating in abnormal conditions, resulting in slow warning or false alarm.

Method used

Using the N-order depth multivariate state estimation method, the observation values ​​and historical memory matrix of multiple dimension parameters of the device are obtained, and the residual values ​​of the parameters of each dimension are calculated, and the historical memory matrix is ​​updated according to the target matrix and observation value, and the MSET is optimized. Algorithm to improve early warning accuracy.

Benefits of technology

It realizes more accurate equipment operating conditions prediction, reduces false alarms and false alarms, and improves the timeliness and accuracy of early warnings.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117743742B_ABST
    Figure CN117743742B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for early warning of power generation equipment failure, and relates to the technical field of early warning of power generation equipment failure. The method comprises: obtaining the observed values ​​of parameters of multiple dimensions of the equipment; obtaining the first residual value corresponding to the parameter of each dimension according to the observed values ​​of the parameter of each dimension and the historical memory matrix, wherein the historical memory matrix is ​​generated according to the historical parameter values ​​of the parameters of multiple dimensions of the equipment; obtaining the target matrix according to the historical memory matrix and the parameters of the first dimension when the first residual value of the parameter of the first dimension is greater than the first preset value; obtaining the final estimated value of the parameter of each dimension according to the historical memory matrix, the target matrix and the observed values ​​of the parameters of the second dimension, wherein the parameters of the second dimension are the parameters of the dimensions of the multiple dimensions except the parameters of the first dimension. The estimated value obtained by the present invention more accurately reflects the operating condition prediction of the equipment, so as to provide timely and accurate early warning.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of power generation equipment fault early warning, and in particular to a fault intelligent early warning method and system for power generation equipment in a power plant. Background Art

[0002] In order to promote the consumption of new energy power, deep peak load regulation of thermal power units has become a normal operating condition. The units are in variable operating conditions for a long time, which puts higher requirements on equipment operation monitoring. Although the distributed control system (DCS) of thermal power units has single-point over-limit alarm and related protection functions, it focuses on emergency treatment or post-accident analysis after accidents, and cannot achieve continuous monitoring and early warning of equipment variable operating parameters deviating from normal values, let alone timely giving early warning signals in the early stage of equipment failure.

[0003] In recent years, with the application of big data and artificial intelligence technology in the construction of smart power plants, various power generation groups and power colleges have carried out in-depth research on model-based deviation warning, that is, to establish a prediction model for equipment monitoring parameters, calculate the predicted value of the parameter in real time, and warn the equipment when the real-time monitoring parameter deviates from the model predicted value, and build software systems such as intelligent warning and intelligent monitoring. The algorithms used mainly include neural network algorithm, multivariate state estimation technique (MSET) and time series prediction algorithm, among which the MSET algorithm is widely used because it can obtain the expected values ​​of multiple parameters of the equipment at the same time and has good real-time performance.

[0004] However, in the existing power generation equipment fault warning method and system using the MSET algorithm, when the equipment is operating abnormally and one (or more) state parameters rise or fall abnormally, the predicted value obtained by the MSET algorithm follows the real-time value of the parameter and cannot accurately predict the normal value of the equipment state parameter under the current operating condition, resulting in delayed warning or missed warning. At the same time, the predicted value of the parameter that is highly correlated with the abnormal parameter will deviate from the normal value, resulting in false warning. Summary of the invention

[0005] The object of the present invention is to provide a power generation equipment fault early warning method based on N-order deep multivariate state estimation, so as to solve the problem of inaccurate early warning of the existing power generation equipment fault early warning method.

[0006] In a first aspect, in order to achieve the above-mentioned purpose, an embodiment of the present invention provides a power generation equipment fault early warning method based on N-order deep multivariate state estimation, comprising:

[0007] Obtain observation values ​​of parameters of multiple dimensions of the device;

[0008] Obtaining a first residual value corresponding to the parameter of each dimension according to an observed value of the parameter of each dimension and a historical memory matrix, wherein the historical memory matrix is ​​generated according to historical parameter values ​​of the parameters of multiple dimensions of the device;

[0009] When a first residual value of a parameter of a first dimension is greater than a first preset value, obtaining a target matrix according to the historical memory matrix and the parameter of the first dimension, where the parameter of the first dimension is a parameter of at least one dimension among the parameters of the multiple dimensions;

[0010] The final estimated value of the parameter of each dimension is obtained according to the historical memory matrix, the target matrix and the observed values ​​of the parameters of the second dimension, where the parameters of the second dimension are the parameters of the dimension other than the parameters of the first dimension among the parameters of the multiple dimensions.

[0011] Optionally, obtaining a first residual value corresponding to the parameter of each dimension according to the observed value of the parameter of each dimension and the historical memory matrix includes:

[0012] Using the multivariate state estimation technology MSET algorithm, according to the historical memory matrix and the observed value of the parameter of each dimension, a first estimated value of the parameter of each dimension is obtained;

[0013] According to the observed value of the parameter of each dimension and the first estimated value of the parameter of the corresponding dimension, the first residual value of the parameter of each dimension is obtained.

[0014] Optionally, obtaining a target matrix according to the historical memory matrix and the parameter of the first dimension includes:

[0015] Obtaining a first difference value according to the historical memory matrix and the parameter of the first dimension;

[0016] When the first difference is less than or equal to a second preset value, a first matrix is ​​generated according to the historical memory matrix and the parameter values ​​of the parameters of the first dimension, and the parameter values ​​of the parameters of the first dimension are the parameter values ​​of the parameters of the corresponding dimension in the historical memory matrix;

[0017] Obtaining a second residual value corresponding to the parameter of each dimension according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension;

[0018] When the second residual value corresponding to the parameter of each dimension is less than or equal to the first preset value, using the first matrix as the target matrix;

[0019] When a second residual value corresponding to a parameter of a third dimension is greater than the first preset value, a second difference value is obtained according to the historical memory matrix, the first matrix and the parameter of the third dimension, where the parameter of the third dimension is a parameter of at least one dimension among the parameters of the multiple dimensions;

[0020] When the second difference is less than or equal to the second preset value, a second matrix is ​​generated according to the historical memory matrix and the parameter values ​​of the parameters of the third dimension, and the parameter values ​​of the parameters of the third dimension are the parameter values ​​of the parameters of the corresponding dimension in the historical memory matrix;

[0021] The second matrix is ​​updated to the first matrix, the parameters of the third dimension are updated to the parameters of the first dimension, and the observation values ​​of the parameters of the historical memory matrix, the first matrix and the second dimension are returned to obtain the second residual value corresponding to the parameters of each dimension, until the second residual value corresponding to the parameters of each dimension is less than or equal to the first preset value, and the second matrix is ​​used as the target matrix.

[0022] Optionally, obtaining a second residual value corresponding to the parameter of each dimension according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension includes:

[0023] Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension, a second estimated value of the parameter of each dimension is obtained;

[0024] According to the observed value of the parameter of each dimension and the second estimated value of the parameter of the corresponding dimension, a second residual value of the parameter of each dimension is obtained.

[0025] Optionally, obtaining a first difference value according to the historical memory matrix and the parameter of the first dimension includes:

[0026] Using the MSET algorithm, according to the historical memory matrix and the observed value of the parameter of each dimension, a first estimated value of the parameter of each dimension is obtained;

[0027] Obtaining a first average relative error value according to the observed value of the parameter of each dimension and the corresponding first estimated value;

[0028] Removing parameter values ​​of parameters of the first dimension from the historical memory matrix to obtain the first matrix;

[0029] Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension, a second estimated value of the parameter of each dimension is obtained;

[0030] Obtaining a second average relative error value according to the observed value of the parameter of each dimension and the corresponding second estimated value;

[0031] The first difference is obtained according to the first average relative error value and the second average relative error value.

[0032] Optionally, obtaining a second difference value according to the historical memory matrix, the first matrix and the parameter of the third dimension includes:

[0033] Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the third dimension, a third estimated value of the parameter of each dimension is obtained;

[0034] Obtaining a third average relative error value according to the observed value of the parameter of each dimension and the corresponding third estimated value;

[0035] Removing parameter values ​​of the parameters of the third dimension from the first matrix to obtain the second matrix;

[0036] Using the MSET algorithm, a fourth estimated value of the parameter of each dimension is obtained according to the observed values ​​of the historical memory matrix, the second matrix and the parameter of the fourth dimension, where the parameter of the fourth dimension is a parameter of the dimension other than the parameter of the third dimension among the parameters of the multiple dimensions;

[0037] Obtaining a fourth average relative error value according to the observed value of the parameter of each dimension and the corresponding fourth estimated value;

[0038] The second difference is obtained according to the third average relative error value and the fourth average relative error value.

[0039] Optionally, the method further comprises:

[0040] When the first difference is greater than the second preset value, the MSET algorithm is used to obtain the final estimated value of the parameter of each dimension based on the observed values ​​of the historical memory matrix, the target matrix and the parameters of the second dimension.

[0041] Optionally, the method further comprises:

[0042] When the second difference is greater than the second preset value, the second matrix is ​​used as the target matrix.

[0043] Optionally, the method further comprises:

[0044] When the residual value of each dimension is less than or equal to the first preset value, the MSET algorithm is used to obtain the final estimated value of the parameter of each dimension based on the historical memory matrix and the observed value of the parameter of each dimension.

[0045] Optionally, the method further comprises:

[0046] Obtaining historical parameter values ​​of parameters of multiple dimensions of the device;

[0047] The historical parameter values ​​are processed using at least one of the following methods to obtain the historical memory matrix:

[0048] Expert rule algorithm, density clustering algorithm and equal-interval sampling algorithm.

[0049] In a second aspect, in order to achieve the above-mentioned purpose, an embodiment of the present invention provides a power generation equipment fault warning device using an N-order deep multivariate state estimation technology, comprising:

[0050] A first acquisition module is used to acquire observation values ​​of parameters of multiple dimensions of a device;

[0051] A first processing module, configured to obtain a first residual value corresponding to the parameter of each dimension according to an observed value of the parameter of each dimension and a historical memory matrix, wherein the historical memory matrix is ​​generated according to historical parameter values ​​of the parameters of multiple dimensions of the device;

[0052] a second processing module, configured to obtain a target matrix according to the historical memory matrix and the parameters of the first dimension when a first residual value of the parameters of the first dimension is greater than a first preset value, wherein the parameters of the first dimension are parameters of at least one dimension among the parameters of the multiple dimensions;

[0053] The third processing module is used to obtain the final estimated value of the parameter of each dimension based on the historical memory matrix, the target matrix and the observed value of the parameter of the second dimension, where the parameter of the second dimension is the parameter of the dimension other than the parameter of the first dimension among the parameters of the multiple dimensions.

[0054] In the third aspect, in order to achieve the above-mentioned purpose, an embodiment of the present invention provides a power generation equipment fault warning device based on N-order deep multivariate state estimation technology, including a transceiver, a memory, a processor, and a computer program stored on the memory and running on the processor, and when the processor executes the computer program, it implements the power generation equipment fault warning method based on N-order deep multivariate state estimation as described in any one of the first aspects.

[0055] In a fourth aspect, in order to achieve the above-mentioned purpose, an embodiment of the present invention provides a readable storage medium on which a program or instruction is stored, and when the program or instruction is executed by a processor, a power generation equipment fault warning method of N-order deep multivariate state estimation as described in any one of the first aspects is implemented.

[0056] The above technical solution of the present invention has at least the following beneficial effects:

[0057] The N-order deep multivariate state estimation method for power generation equipment fault warning of the embodiment of the present invention obtains parameter values ​​of parameters of multiple dimensions of the equipment, obtains a first residual value corresponding to the parameter of each dimension according to the parameter value of the parameter of each dimension and the historical memory matrix, and when the first residual value of the parameter of the first dimension is greater than a first preset value, obtains a target matrix according to the historical memory matrix and the parameters of the first dimension, and the target matrix is ​​the optimal characteristic subspace corresponding to the normal parameters in the historical memory matrix. The final estimated value of the parameter of each dimension is obtained by using the historical memory matrix, the target matrix and the parameter values ​​of the parameters of each dimension. The estimated value can more accurately reflect the operating condition prediction of the equipment, so as to provide timely and accurate warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A schematic diagram of temperature variation of a temperature measuring point 1 of a front bearing of an induced draft fan provided by the present invention;

[0059] Figure 2 A flow chart of a power generation equipment fault early warning method based on N-order deep multivariate state estimation provided by an embodiment of the present invention;

[0060] Figure 3 A specific flow chart of a method for early warning of power generation equipment failures using N-order deep multivariate state estimation provided by an embodiment of the present invention;

[0061] Figure 4 A schematic diagram showing comparison results of the predicted value (or estimated value) of the induced draft fan front bearing temperature 1 provided by the MEST algorithm, the predicted value of the MD-MEST algorithm, and the observed value;

[0062] Figure 5 A schematic diagram showing the comparison results of the predicted value (or estimated value) of the induced draft fan current provided by the MEST algorithm, the predicted value of the MD-MEST algorithm, and the observed value;

[0063] Figure 6 A schematic diagram of the structure of a power generation equipment fault warning device using an N-order deep multivariate state estimation technology provided by an embodiment of the present invention;

[0064] Figure 7 A schematic diagram of the structure of a power generation equipment fault warning device using an N-order deep multivariate state estimation technology provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail in conjunction with the accompanying drawings and specific embodiments. In the following description, specific details such as specific configurations and components are provided only to help fully understand the embodiments of the present invention. Therefore, it should be clear to those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. In addition, for clarity and brevity, the description of known functions and structures is omitted.

[0066] It should be understood that the references to "one embodiment" or "an embodiment" throughout the specification mean that the specific features, structures, or characteristics associated with the embodiment are included in at least one embodiment of the present invention. Therefore, the references to "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0067] In various embodiments of the present invention, it should be understood that the size of the serial numbers of the following processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0068] In the embodiments provided by the present invention, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.

[0069] When describing the embodiments of the present invention, some concepts used in the following description are first explained.

[0070] The principle of multivariate state estimation technology (MSET):

[0071] Multivariate state estimation technology (MSET) is a technology used for anomaly detection and prediction of equipment, sensors, and operating parameters in commercial nuclear power plants. The technology was developed to provide an efficient and accurate method to estimate the state of the system and help detect potential problems early for proper maintenance and treatment. MSET mainly performs state estimation by comparing the similarity between the new state and the normal working state. Because the algorithm has good real-time, fast and accurate performance, it has been widely used in real-time state monitoring and early warning of power generation equipment. The principle of MSET is as follows:

[0072] For a device, if there are n interrelated monitoring variables (one variable corresponds to one parameter), then at time t i When monitoring, these n variables can be combined into an observation vector, denoted as X, which is expressed as follows:

[0073] X(t i )=[x1(t i ),x2(t i ),......,x n (t i )] T Formula 1

[0074] By analyzing the historical data of equipment operation and extracting its features, a historical memory matrix of equipment operation is constructed. The historical memory matrix of n measurement points (one measurement point corresponds to one parameter of the equipment) and m states (one state can be understood as the parameter values ​​of all parameters at a certain moment) is denoted as D and is expressed as follows:

[0075]

[0076] Based on the memory matrix D, the model is constructed using the MSET algorithm. The observation vector X of the real-time operation of the device obs (which can also be understood as the real-time parameter values ​​of multiple parameters) is used as the input parameter of the MSET model, and the observation vector X obs It is compared with the normal operating state stored in the memory matrix D. By using similarity theory and linearly weighting each state in the memory matrix, an estimate of the observation vector X under normal operating conditions is obtained. est The estimated value can be used as a reference for normal operation and to determine whether the current observation vector is similar to the normal operation state or has an anomaly. In this way, the state estimation and anomaly detection of the equipment can be achieved.

[0077] Let the observation vector be X obs , the estimated vector obtained by the MSET algorithm is X est , the formula is as follows:

[0078]

[0079] The vector W represents the weight vector corresponding to each state in the observation vector. It reflects the similarity between the current state of the observation vector and the memory matrix. The calculation of the weight vector W is based on evaluating the similarity between the observation vector and each state in the memory matrix, and assigning different weights to each state according to the similarity.

[0080] Based on the optimal estimation principle, the weight vector W can be determined by minimizing the residual, and the formula is as follows:

[0081] W=(D T D) -1 (D T X obs ) Formula 4

[0082] Therefore, the estimated value Xest The calculation formula is as follows:

[0083]

[0084] The correlation between the process memory matrix data may cause the matrix to be irreversible, which may lead to the inability to calculate formula 5 normally. For this reason, a nonlinear operator is used. Instead of matrix multiplication, it can be expressed as follows:

[0085]

[0086] The Euclidean distance between two vectors is selected nonlinearly. The calculation formula is as follows:

[0087]

[0088] When performing a nonlinear operation, if the two vectors are very similar or close, the result of the nonlinear operation will be close to zero. On the contrary, if there is a large difference between the two vectors, the result of the nonlinear operation will increase accordingly. The result of this nonlinear operation can be used to measure the similarity or difference between the two vectors.

[0089] MEST limitation analysis:

[0090] Since the principle of MSET algorithm prediction is to calculate the observation vector X obs (also called parameter value) and the distance similarity between each state in the historical memory matrix to obtain the weight system W, and then estimate the predicted value X est (or called estimated value or estimated vector). Therefore, when the device status is abnormal and the parameter exceeds the range in the historical memory matrix, it will cause the predicted value X est Deviating from the normal value of the equipment under the same historical operating conditions, this results in missed warnings and false alarms when the MSET algorithm is used to develop an intelligent early warning software system for continuous parameter prediction and early warning. Specifically, when the equipment is in an abnormal operating state, during the abnormal increase or decrease of one (or more) state parameters, the predicted value obtained by the MSET algorithm follows the real-time value of the parameter and cannot accurately predict the normal value of the equipment state parameter under the current operating condition, resulting in delayed early warning or missed warnings. At the same time, the predicted value of the parameter that is strongly correlated with the abnormal parameter will deviate from the normal value, resulting in false warnings.

[0091] Specifically, the MSET prediction process is as follows:

[0092] To simplify the calculation process, assume that the historical memory matrix D of a device's operating parameters is a 2×2 matrix:

[0093]

[0094] Observation vector X obs :

[0095] X obs =[x p11 x p21 ] T

[0096] Then we have:

[0097] Memory matrix D transpose D T :

[0098]

[0099] D and D T The Euclidean distance

[0100]

[0101] The inverse matrix

[0102]

[0103] The memory matrix D and Multiplication

[0104] D T and X obs The Euclidean distance is:

[0105]

[0106] Observation vector X obs The estimated vector of the estimated vector is X est :

[0107]

[0108] In the case of abnormal operation of the device, suppose only x p11 Deviation from normal operating value, x p11 >x 11 , x p11 >x 12 , and the deviation is getting bigger and bigger. From the above deduction, we can know that x p11 The predicted value of x' p11 will also gradually increase, deviating from the normal value under this condition, which will cause the warning to lag and also cause x p21 The predicted value x' p21 Increase, causing the parameter x p21 False alarm.

[0109] The following are some examples:

[0110] Example 1: Abnormal bearing temperature of induced draft fan in thermal power plant:

[0111] The monitoring points (also called parameters) of the induced draft fan of a 600MW thermal power unit include 54 points, including the position feedback of the electric actuator of the moving blade, the current of the induced draft fan, the front bearing temperature 1, the front bearing temperature 2, and the front bearing temperature 3 of the induced draft fan. In order to establish the fault warning model of the induced draft fan, the historical data (historical parameter values) of the induced draft fan for one year were selected from the power plant safety instrument system (Safety Instrumented System, SIS), with a data interval of 1 minute, totaling 527040 data. Using density clustering and equal-interval sampling, 1550 historical data of normal working conditions were obtained, and the historical memory matrix of the constructed MSET model was a 1550×54-dimensional matrix.

[0112] like Figure 1 As shown in the figure, the temperature of the front bearing of the induced draft fan at point 1 suddenly failed, and the temperature rose rapidly from 48°C to 102°C. During the process, the other 53 monitoring parameters did not change. The predicted value of the temperature obtained by the MSET algorithm rose from 48°C to 54°C. It can be seen that when the parameter is abnormal, the predicted value obtained by the MSET algorithm changes with the real-time value, and cannot accurately predict the normal value under the current working conditions.

[0113] Example 2: Abnormal bearing temperature of the condensate pump in the fault warning system of a thermal power plant:

[0114] A 330MW thermal power plant built a power generation equipment fault warning system using MSET as the core algorithm. At a certain time, a bearing on the condensate pump motor suddenly failed, causing the bearing temperature to rise rapidly from 36°C to 58°C and continue to fluctuate. During the abnormal process of this measuring point, the predicted value of the motor front bearing temperature also rose and fluctuated rapidly following the real-time value.

[0115] To sum up, in the power generation equipment fault warning method and system using the MSET algorithm, when the equipment operating state is abnormal, and one (or more) state parameters rise or fall abnormally, the predicted value obtained by the MSET algorithm follows the real-time value of the parameter and cannot accurately predict the normal value of the equipment state parameter in the current operating condition; at the same time, it affects the predicted values ​​of other parameters.

[0116] In order to solve the problem of inaccurate warning in existing equipment failure warning methods, an embodiment of the present invention provides a power generation equipment failure warning method based on N-order deep multivariate state estimation.

[0117] like Figure 2 The embodiment of the present invention provides a power generation equipment fault early warning method based on N-order deep multivariate state estimation, comprising:

[0118] Step 201: Obtain the observed values ​​of the parameters of multiple dimensions of the device. The observed value (or observation vector) can be understood as the real-time measurement value of the parameter at a certain time point. The observed values ​​of a parameter at different time points within a period of time are stored to obtain the historical parameter value of the parameter. It can also be understood that the historical parameter value of the parameter includes the measured values ​​of the parameter at multiple time points.

[0119] Parameters of different dimensions can be understood as different types of parameters or as different monitoring points, such as position feedback of the electric actuator of the movable blade, current of the induced draft fan, front bearing temperature 1, front bearing temperature 2 of the induced draft fan, front bearing temperature 3 of the induced draft fan and other parameters (or monitoring points).

[0120] The device may be a power generation device, including thermal power plant equipment, wind power equipment, hydropower equipment, nuclear power equipment, photovoltaic equipment, etc. Taking the device as a thermal power plant device as an example, after obtaining the observation values ​​of parameters in multiple dimensions of the device, the observation values ​​may be stored in the SIS or distributed control system (DCS) of the thermal power plant.

[0121] Step 202: Obtain a first residual value corresponding to the parameter of each dimension based on the observed value of the parameter of each dimension and a historical memory matrix, wherein the historical memory matrix is ​​generated based on the historical parameter values ​​of the parameters of multiple dimensions of the device.

[0122] It should be noted that the historical memory matrix is ​​generated based on the historical parameter values ​​of the parameters of multiple dimensions of the device. After the historical memory matrix is ​​generated, the historical memory matrix includes the parameter values ​​of the parameters of each dimension.

[0123] In this step, based on the observed values ​​of the parameters of each dimension and the historical memory matrix, the estimated values ​​(or estimated vectors) of the parameters of each dimension can be obtained, and the residual values ​​can be obtained based on the estimated values ​​and observed values ​​of the corresponding parameters.

[0124] In an optional embodiment, the process of generating the history memory matrix includes, that is, the method further includes:

[0125] Obtain historical parameter values ​​of parameters of multiple dimensions of the device, specifically,

[0126] Collect historical data of equipment operation for one year from the SIS or DCS system of the thermal power plant, including all status parameters related to the equipment, that is, collect historical parameter values ​​of parameters in multiple dimensions.

[0127] The historical parameter values ​​are processed using at least one of the following methods to obtain the historical memory matrix:

[0128] Expert rule algorithm, density clustering algorithm and equal-interval sampling algorithm.

[0129] In an optional embodiment, obtaining a first residual value corresponding to the parameter of each dimension according to the observed value of the parameter of each dimension and the historical memory matrix includes:

[0130] The multivariate state estimation technique MSET algorithm is used to obtain the first estimated value (or first estimated vector or first predicted value) of the parameter of each dimension according to the historical memory matrix and the observed value of the parameter of each dimension. Specifically, the observation vector X obs (Observed value) Combined with the historical memory matrix, the first estimated vector obtained by the MSET algorithm using formula 8 is X est :

[0131]

[0132] Among them, D represents the historical memory matrix and W represents the weight vector.

[0133] According to the observed value of the parameter of each dimension and the first estimated value of the parameter of the corresponding dimension, the first residual value of the parameter of each dimension is obtained. Specifically, the difference between the observed value and the corresponding first estimated value is taken as the first residual value.

[0134] Step 203: When a first residual value of a parameter of a first dimension is greater than a first preset value, a target matrix is ​​obtained according to the historical memory matrix and the parameter of the first dimension, where the parameter of the first dimension is a parameter of at least one dimension among the parameters of the multiple dimensions.

[0135] In this step, when the first residual value of the parameter of the first dimension is greater than the first preset value, the parameter of the first dimension can be regarded as a parameter of abnormal equipment operation, and processed according to the historical memory matrix and the parameter of the first dimension, or understood as optimizing the historical memory matrix using the parameter of the first dimension to obtain the target matrix.

[0136] It should be noted that the first preset value can be selected according to demand. Preferably, the first preset value is greater than the critical difference value for the device to issue an early warning.

[0137] In an optional embodiment, the method further includes:

[0138] When the residual value of each dimension is less than or equal to the first preset value, the MSET algorithm is used to obtain the final estimated value of the parameter of each dimension based on the historical memory matrix and the observed value of the parameter of each dimension.

[0139] That is, when the parameters of the equipment are all operating normally, the estimated value obtained according to the above formula eight is the final estimated value (or final predicted value) of the parameter of each dimension, and the historical memory matrix continues to be used for prediction.

[0140] Step 204: Obtain final estimated values ​​of the parameters of each dimension based on the historical memory matrix, the target matrix and the observed values ​​of the parameters of the second dimension, where the parameters of the second dimension are the parameters of the multiple dimensions excluding the parameters of the first dimension.

[0141] In this step, the parameters of the second dimension can be understood as the parameters of the normal operation of the equipment. After determining the normal operation parameters of the equipment and the target matrix, the following formula 9 is used to obtain the final estimated value X of the parameters of each dimension: est :

[0142]

[0143] Where D represents the historical memory matrix, W n represents the weight vector, D n represents the target matrix, X nobs Represents the observed values ​​of the parameter of the second dimension.

[0144] In an optional embodiment, obtaining a target matrix according to the historical memory matrix and the parameters of the first dimension comprises the following steps:

[0145] Step (1): obtaining a first difference value according to the historical memory matrix and the parameter of the first dimension;

[0146] Optionally, obtaining a first difference value according to the historical memory matrix and the parameter of the first dimension includes:

[0147] Using the MSET algorithm, according to the observed value of the parameter of each dimension and the historical memory matrix, the first estimated value of the parameter of each dimension is obtained. The calculation process is shown in Formula 8 above. According to the observed value of the parameter of each dimension and the corresponding first estimated value, the first mean relative error value is obtained. The mean relative error (Mean Relative Error, MRE) is the average value of the absolute value of the relative error (relative error refers to the ratio of the error to the true value). The first mean relative error value is obtained by first obtaining the difference between the observed value of each dimension and the corresponding first estimated value, and then calculating the absolute value of the ratio of the difference to the observed value, summing the absolute values ​​of the ratios corresponding to each dimension, and then dividing by the number of dimensions.

[0148] The first matrix is ​​obtained by removing the parameter values ​​of the parameters of the first dimension from the historical memory matrix, that is, based on the backward stepwise regression idea, traversing and eliminating the variables in the historical memory matrix D (the variables are also the parameter values ​​of the parameters of the first dimension) in sequence to obtain the first matrix;

[0149] Using the MSET algorithm, according to the observed values ​​of the parameters of the historical memory matrix, the first matrix and the second dimension, the second estimated value of the parameter of each dimension is obtained. The specific calculation process is shown in Formula 9, where n is equal to 1, X est That is the second estimated value, and D1 represents the first matrix.

[0150] According to the observed value of the parameter of each dimension and the corresponding second estimated value, a second average relative error value is obtained. The calculation process of the second average relative error value is consistent with the above-mentioned first average relative error value, and will not be repeated here. According to the first average relative error value and the second average relative error value, the first difference is obtained, and the first difference is the difference between the first average relative error value and the second average relative error value.

[0151] That is, the first difference is the difference between the average relative error of the observed value and the estimated value of the parameter of each dimension before and after the parameter value of the parameter of the first dimension is eliminated in the historical memory matrix.

[0152] Step (2): when the first difference is less than or equal to the second preset value, a first matrix is ​​generated according to the historical memory matrix and the parameter values ​​of the parameters of the first dimension, the parameter values ​​of the parameters of the first dimension are the parameter values ​​of the parameters of the corresponding dimension in the historical memory matrix, that is, the parameter values ​​of the parameters of the first dimension are removed from the historical memory matrix to obtain the first matrix, or it can be understood that the parameter values ​​of the parameters of the second dimension (the parameters of the second dimension are the parameters of the multiple dimensions of the device except the parameters of the first dimension) are added to the new matrix to form the first matrix D1. After the traversal is completed, the first matrix D1 composed of the variables that meet the requirements (that is, the parameter values ​​of the parameters of the second dimension) is the optimal feature subspace corresponding to the normal operation parameters of the device in the historical memory matrix D;

[0153] After determining the first matrix, the weight matrix W1 corresponding to the parameters of the device at each moment can be solved, please refer to the above formula 9, where n is equal to 1.

[0154] Furthermore, when the first difference is greater than the second preset value, the MSET algorithm is used to obtain the final estimated value of the parameter of each dimension based on the observed values ​​of the historical memory matrix, the target matrix and the parameters of the second dimension. The specific calculation process is shown in the above formula nine.

[0155] Step (3): obtaining a second residual value corresponding to the parameter of each dimension according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension;

[0156] Optionally, obtaining a second residual value corresponding to the parameter of each dimension according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension includes:

[0157] Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension, a second estimated value of the parameter of each dimension is obtained;

[0158] According to the observed value of the parameter of each dimension and the second estimated value of the parameter of the corresponding dimension, a second residual value of the parameter of each dimension is obtained.

[0159] It should be noted that the process of determining the second residual value is basically the same as the process of determining the first residual value described above, and will not be described in detail here.

[0160] That is, in this step, the first matrix is ​​used to continue analyzing the residual value (that is, the second residual value) corresponding to each parameter.

[0161] Step (4): when the second residual value corresponding to the parameter of each dimension is less than or equal to the first preset value, the first matrix is ​​used as the target matrix, that is, the second residual value does not exceed the second preset value, and the first matrix is ​​continued to be used for prediction;

[0162] Step (5): when the second residual value corresponding to the parameter of the third dimension is greater than the first preset value, a second difference value is obtained according to the historical memory matrix, the first matrix and the parameter of the third dimension, where the parameter of the third dimension is a parameter of at least one dimension among the parameters of the multiple dimensions;

[0163] Optionally, obtaining a second difference value according to the historical memory matrix, the first matrix and the parameter of the third dimension includes:

[0164] Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the third dimension, a third estimated value of the parameter of each dimension is obtained;

[0165] Obtaining a third average relative error value according to the observed value of the parameter of each dimension and the corresponding third estimated value;

[0166] Removing parameter values ​​of the parameters of the third dimension from the first matrix to obtain the second matrix;

[0167] Using the MSET algorithm, a fourth estimated value of the parameter of each dimension is obtained according to the observed values ​​of the historical memory matrix, the second matrix and the parameter of the fourth dimension, where the parameter of the fourth dimension is a parameter of the dimension other than the parameter of the third dimension among the parameters of the multiple dimensions;

[0168] Obtaining a fourth average relative error value according to the observed value of the parameter of each dimension and the corresponding fourth estimated value;

[0169] The second difference is obtained according to the third average relative error value and the fourth average relative error value.

[0170] It should be noted that the calculation process of calculating the second difference is basically the same as the calculation process of calculating the first difference, which will not be repeated here.

[0171] Step (6): when the second difference is less than or equal to the second preset value, generating a second matrix according to the historical memory matrix and the parameter values ​​of the parameters of the third dimension, wherein the parameter values ​​of the parameters of the third dimension are the parameter values ​​of the parameters of the corresponding dimension in the historical memory matrix;

[0172] That is, in this step, the parameters of the third dimension are taken as the parameters of abnormal operation of the equipment, and the parameter values ​​of the parameters of the third dimension are removed from the historical memory matrix to obtain the second matrix, or it can be understood that the parameter values ​​of the parameters of the fourth dimension (the parameters of the fourth dimension are the parameters of the multiple dimensions of the equipment except the parameters of the third dimension) are added to the new matrix to form the second matrix D2. After the traversal is completed, the first matrix D2 composed of the variables that meet the requirements (that is, the parameter values ​​of the parameters of the fourth dimension) is the optimal feature subspace corresponding to the parameters of normal operation of the equipment in the historical memory matrix D;

[0173] Furthermore, when the second difference is greater than the second preset value, the second matrix is ​​used as the target matrix.

[0174] Step (7): Update the second matrix to the first matrix, update the parameters of the third dimension to the parameters of the first dimension, update the parameters of the fourth dimension to the parameters of the second dimension, return to the step of obtaining the second residual value corresponding to the parameters of each dimension based on the observed values ​​of the historical memory matrix, the first matrix and the parameters of the second dimension, continue to perform optimization iterations until the second residual value corresponding to the parameters of each dimension is less than or equal to the first preset value, and use the second matrix as the target matrix.

[0175] Combine the following Figure 3 , describing the specific process of the power generation equipment fault early warning method of N-order deep multivariate state estimation provided by an embodiment of the present invention:

[0176] Collect historical parameter values ​​of equipment parameters for one year from the SIS or DCS of the thermal power plant, including all status data related to the equipment;

[0177] According to the historical parameter values, the historical memory matrix is ​​constructed using the house rule algorithm, density clustering algorithm and equal spacing sampling algorithm;

[0178] Input the current observation value X of the equipment from the SIS or DCS system obs , combined with the historical memory matrix, the first estimate obtained by the MSET algorithm is X est ;

[0179] A first residual value of the parameter is obtained according to the observed value and the first estimated value. If the first residual value does not exceed the first preset value, the historical memory matrix is ​​used for subsequent prediction to end the optimization process. If the first residual value of the parameter of the first dimension exceeds the first preset value, the parameter of the first dimension is used as the parameter of abnormal equipment operation, and the parameter of the second dimension is used as the parameter of normal equipment operation. The first matrix is ​​generated according to the parameter values ​​of the historical memory matrix and the parameters of the first dimension. The specific process is: based on the backward stepwise regression idea, the variables in the historical memory matrix D (that is, the parameter values ​​of the parameters of the first dimension) are traversed and eliminated in sequence, and the first difference between the average relative error of the observed value and the estimated value before and after the variable is eliminated is calculated, and the second preset value is set. If the first difference does not exceed the second preset value, the variable (that is, the parameter value of the parameter of the second dimension) is added to the new matrix to form the first matrix D1. After the traversal is completed, the first matrix D1 composed of the variables that meet the requirements is the optimal characteristic subspace corresponding to the normal equipment operation parameters in the historical memory matrix D. The weight W1 corresponding to the current operating condition of the equipment is solved according to the optimal characteristic subspace.

[0180] Continue to analyze the second residual value of the parameters of each dimension according to the first matrix. If the second residual value does not exceed the first preset value, continue to make subsequent predictions. If the second residual value exceeds the first preset value, the parameters of the third dimension are used as parameters of abnormal equipment operation, and the parameters of the fourth dimension are used as parameters of normal equipment operation. Continue to generate the second matrix according to the historical memory matrix and the parameter values ​​of the parameters of the third dimension as the optimal characteristic subspace D2 corresponding to the normal equipment operation parameters according to the above process, and solve the weight W2 corresponding to the current operating condition of the equipment in the optimal characteristic subspace, that is, update the second matrix to the first matrix, update the parameters of the third dimension to the parameters of the first dimension, and update the parameters of the fourth dimension to the parameters of the second dimension. Repeat the above process until the residual value of the parameters of each dimension does not exceed the first preset value.

[0181] The N-order deep multivariate state estimation method for power generation equipment fault warning provided by the embodiment of the present invention is an optimization technology of the MSET algorithm, which can be called the N-order deep multivariate state estimation technology (MD-MSET) algorithm. The historical memory matrix and observation vector of the MSET algorithm contain parameter information of multiple dimensions of the equipment, and the dimensional information is redundant and affects each other. From the analysis of the principle of equipment operation, some dimensional state parameters can determine the current operating point of the equipment. Based on this, the MD-MSET technology is proposed. Its core principle is to analyze the model residuals, optimize the optimal characteristic subspace corresponding to the normal operating parameters of the equipment in the historical memory matrix, and solve the weight corresponding to the current operating condition of the equipment in the subspace, and use this weight to estimate the expected value of each parameter in the original space, and achieve more accurate estimation by automatically controlling the iteration depth.

[0182] The following is a comparison of the power generation equipment fault warning method provided by the N-order deep multivariate state estimation technology (MD-MSET algorithm) and the estimated value predicted by the MSET algorithm using several specific examples:

[0183] The monitoring points of the induced draft fan of a 600MW thermal power unit include 54 points, including the position feedback of the electric actuator of the moving blade, the current of the induced draft fan, the front bearing temperature 1, the front bearing temperature 2, and the front bearing temperature 3 of the induced draft fan. In order to establish the fault warning model of the induced draft fan, the historical parameter values ​​of the induced draft fan for one year were selected from the SIS system of the power plant, with a data interval of 1 minute, totaling 527,040 data. Using density clustering and equal-interval sampling, 1,550 parameter values ​​of normal working conditions were obtained, and the constructed MSET model memory matrix was a 1,550×54-dimensional historical memory matrix.

[0184] Example 3: Comparison results of the predicted value (or estimated value) of the MEST algorithm, the predicted value of the MD-MEST algorithm, and the observed value of the front bearing temperature 1 of the induced draft fan are shown in the figure below: Figure 4 As shown in the figure, the temperature of the induced draft fan bearing 1 point suddenly failed, and the temperature rose rapidly from 48°C to 102°C. During the process, the other 53 monitoring parameters did not change, and the residual threshold of the MD-MSET algorithm was set to 5°C. In the case of abnormal parameters, the predicted value (or estimated value) of the MD-MSET algorithm is closer to the normal value of the current working condition than the predicted value of the MSET algorithm, and is not affected by abnormal parameter changes.

[0185] Example 4: Comparison of the predicted value (or estimated value) of the induced draft fan current using the MEST algorithm, the predicted value using the MD-MEST algorithm, and the observed value Figure 5As shown in the figure, the current of the induced draft fan abnormally increased from 220A to 322A due to internal friction fault, but did not exceed the maximum value in the historical memory matrix, and the other 53 monitoring parameters did not change during the process. Using the MSET algorithm to predict, the abnormal increase in current caused the predicted values ​​of the moving blade opening and inlet pressure, which were strongly correlated with it, to deviate seriously from the actual values, which would cause false alarms of the moving blade opening and inlet pressure; the predicted value using the MD-MSET algorithm was closer to the normal value of the current working condition than MSET, and was not affected by abnormal changes in parameters.

[0186] The embodiment of the present invention optimizes the MSET algorithm in view of the shortcomings of the MSET algorithm. When the abnormal operation state of the equipment is monitored, the optimal characteristic subspace corresponding to the normal operation parameters of the equipment in the historical memory matrix is ​​optimized by analyzing the model residuals, and the weight corresponding to the current working condition of the equipment is solved in the subspace. The method of estimating the expected value of each parameter in the original space and the method of automatically controlling the iteration depth are used to estimate the expected value of each parameter by using this weight. Compared with the MSET algorithm, the present invention can accurately estimate the normal value of the current working state parameter of the equipment when the equipment operation state is abnormal, optimizes the shortcomings of the MSET algorithm, reduces the probability of missed alarms and false alarms of the power generation equipment fault warning system constructed with MSET as the core algorithm, improves the accuracy of the warning, has a large engineering practical value, and is suitable for system software such as equipment fault warning, intelligent monitoring panel and other industry equipment fault warning.

[0187] like Figure 6 As shown, the embodiment of the present invention also provides a power generation equipment fault warning device using N-order deep multivariate state estimation technology, including:

[0188] A first acquisition module 601 is used to acquire observation values ​​of parameters of multiple dimensions of a device;

[0189] A first processing module 602, configured to obtain a first residual value corresponding to the parameter of each dimension according to the observed value of the parameter of each dimension and the historical memory matrix, wherein the historical memory matrix is ​​generated according to the historical parameter values ​​of the parameters of multiple dimensions of the device;

[0190] A second processing module 603 is used to obtain a target matrix according to the historical memory matrix and the parameters of the first dimension when a first residual value of the parameters of the first dimension is greater than a first preset value, where the parameters of the first dimension are parameters of at least one dimension among the parameters of the multiple dimensions;

[0191] The third processing module 604 is used to obtain the final estimated value of the parameter of each dimension based on the historical memory matrix, the target matrix and the observed value of the parameter of the second dimension, where the parameter of the second dimension is the parameter of the dimension other than the parameter of the first dimension among the parameters of the multiple dimensions.

[0192] Optionally, the first processing module 602 includes:

[0193] A first processing unit is used to obtain a first estimated value of the parameter of each dimension according to the historical memory matrix and the observed value of the parameter of each dimension by using a multivariate state estimation technology MSET algorithm;

[0194] The second processing unit is used to obtain a first residual value of the parameter of each dimension according to the observed value of the parameter of each dimension and the first estimated value of the parameter of the corresponding dimension.

[0195] Optionally, the second processing module 603 includes:

[0196] A third processing unit, configured to obtain a first difference value according to the historical memory matrix and the parameter of the first dimension;

[0197] a fourth processing unit, configured to generate a first matrix according to the historical memory matrix and parameter values ​​of the parameters of the first dimension when the first difference is less than or equal to a second preset value, wherein the parameter values ​​of the parameters of the first dimension are parameter values ​​of the parameters of the corresponding dimension in the historical memory matrix;

[0198] a fifth processing unit, configured to obtain a second residual value corresponding to the parameter of each dimension according to the historical memory matrix, the first matrix and the observed value of the parameter of the second dimension;

[0199] a sixth processing unit, configured to use the first matrix as the target matrix when the second residual value corresponding to the parameter of each dimension is less than or equal to the first preset value;

[0200] a seventh processing unit, configured to obtain a second difference value according to the historical memory matrix, the first matrix and the parameter of the third dimension when a second residual value corresponding to the parameter of the third dimension is greater than the first preset value, wherein the parameter of the third dimension is a parameter of at least one dimension among the parameters of the multiple dimensions;

[0201] an eighth processing unit, configured to generate a second matrix according to the historical memory matrix and the parameter values ​​of the parameters of the third dimension when the second difference is less than or equal to the second preset value, wherein the parameter values ​​of the parameters of the third dimension are the parameter values ​​of the parameters of the corresponding dimension in the historical memory matrix;

[0202] The ninth processing unit is used to update the second matrix to the first matrix, update the parameters of the third dimension to the parameters of the first dimension, return to the step of obtaining the second residual value corresponding to the parameters of each dimension based on the observed values ​​of the historical memory matrix, the first matrix and the parameters of the second dimension, until the second residual value corresponding to the parameters of each dimension is less than or equal to the first preset value, and use the second matrix as the target matrix.

[0203] Optionally, the fifth processing unit is specifically configured to:

[0204] Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension, a second estimated value of the parameter of each dimension is obtained;

[0205] According to the observed value of the parameter of each dimension and the second estimated value of the parameter of the corresponding dimension, a second residual value of the parameter of each dimension is obtained.

[0206] Optionally, the third processing unit is specifically configured to:

[0207] Using the MSET algorithm, according to the historical memory matrix and the observed value of the parameter of each dimension, a first estimated value of the parameter of each dimension is obtained;

[0208] Obtaining a first average relative error value according to the observed value of the parameter of each dimension and the corresponding first estimated value;

[0209] Removing parameter values ​​of parameters of the first dimension from the historical memory matrix to obtain the first matrix;

[0210] Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension, a second estimated value of the parameter of each dimension is obtained;

[0211] Obtaining a second average relative error value according to the observed value of the parameter of each dimension and the corresponding second estimated value;

[0212] The first difference is obtained according to the first average relative error value and the second average relative error value.

[0213] Optionally, the seventh processing unit is specifically configured to:

[0214] Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the third dimension, a third estimated value of the parameter of each dimension is obtained;

[0215] Obtaining a third average relative error value according to the observed value of the parameter of each dimension and the corresponding third estimated value;

[0216] Removing parameter values ​​of the parameters of the third dimension from the first matrix to obtain the second matrix;

[0217] Using the MSET algorithm, a fourth estimated value of the parameter of each dimension is obtained according to the observed values ​​of the historical memory matrix, the second matrix and the parameter of the fourth dimension, where the parameter of the fourth dimension is a parameter of the dimension other than the parameter of the third dimension among the parameters of the multiple dimensions;

[0218] Obtaining a fourth average relative error value according to the observed value of the parameter of each dimension and the corresponding fourth estimated value;

[0219] The second difference is obtained according to the third average relative error value and the fourth average relative error value.

[0220] Optionally, the first processing module 602 further includes:

[0221] The tenth processing unit is used to obtain the final estimated value of the parameter of each dimension based on the observed values ​​of the parameters of the historical memory matrix, the target matrix and the second dimension by using the MSET algorithm when the first difference is greater than the second preset value.

[0222] Optionally, the first processing module 602 further includes:

[0223] An eleventh processing unit is configured to use the second matrix as the target matrix when the second difference is greater than the second preset value.

[0224] Optionally, the device further comprises:

[0225] The fourth processing module is used to obtain the final estimated value of the parameter of each dimension based on the historical memory matrix and the observed value of the parameter of each dimension using the MSET algorithm when the residual value of each dimension is less than or equal to the first preset value.

[0226] Optionally, the device further comprises:

[0227] A second acquisition module, used to acquire historical parameter values ​​of parameters of multiple dimensions of the device;

[0228] The fifth processing module is used to process the historical parameter value by using at least one of the following methods to obtain the historical memory matrix:

[0229] Expert rule algorithm, density clustering algorithm and equal-interval sampling algorithm.

[0230] It should be noted that the power generation equipment fault warning device using the N-order deep multivariate state estimation technology provided in the embodiment of the present invention is a device capable of executing the above-mentioned power generation equipment fault warning method using the N-order deep multivariate state estimation. All embodiments of the above-mentioned power generation equipment fault warning method using the N-order deep multivariate state estimation are applicable to the device and can achieve the same or similar technical effects.

[0231] like Figure 7 As shown, an embodiment of the present invention also provides a power generation equipment fault warning device with N-order deep multivariate state estimation technology, including: a processor 701; and a memory 403 connected to the processor 701 through a bus interface 702, the memory 703 is used to store the programs and data used by the processor 701 when performing operations, and the processor 701 calls and executes the programs and data stored in the memory 703.

[0232] The transceiver 704 is connected to the bus interface 702 and is used to receive and send data under the control of the processor 701. Specifically, the processor 701 is used to read the program in the memory 703. The processor 701 performs the following process:

[0233] Obtain observation values ​​of parameters of multiple dimensions of the device;

[0234] Obtaining a first residual value corresponding to the parameter of each dimension according to an observed value of the parameter of each dimension and a historical memory matrix, wherein the historical memory matrix is ​​generated according to historical parameter values ​​of the parameters of multiple dimensions of the device;

[0235] When a first residual value of a parameter of a first dimension is greater than a first preset value, obtaining a target matrix according to the historical memory matrix and the parameter of the first dimension, where the parameter of the first dimension is a parameter of at least one dimension among the parameters of the multiple dimensions;

[0236] The final estimated value of the parameter of each dimension is obtained according to the historical memory matrix, the target matrix and the observed values ​​of the parameters of the second dimension, where the parameters of the second dimension are the parameters of the dimension other than the parameters of the first dimension among the parameters of the multiple dimensions.

[0237] Optionally, the processor 701 is configured to:

[0238] Using the multivariate state estimation technology MSET algorithm, according to the historical memory matrix and the observed value of the parameter of each dimension, a first estimated value of the parameter of each dimension is obtained;

[0239] According to the observed value of the parameter of each dimension and the first estimated value of the parameter of the corresponding dimension, the first residual value of the parameter of each dimension is obtained.

[0240] Optionally, the processor 701 is configured to:

[0241] Obtaining a first difference value according to the historical memory matrix and the parameter of the first dimension;

[0242] When the first difference is less than or equal to a second preset value, a first matrix is ​​generated according to the historical memory matrix and the parameter values ​​of the parameters of the first dimension, and the parameter values ​​of the parameters of the first dimension are the parameter values ​​of the parameters of the corresponding dimension in the historical memory matrix;

[0243] Obtaining a second residual value corresponding to the parameter of each dimension according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension;

[0244] When the second residual value corresponding to the parameter of each dimension is less than or equal to the first preset value, using the first matrix as the target matrix;

[0245] When a second residual value corresponding to a parameter of a third dimension is greater than the first preset value, a second difference value is obtained according to the historical memory matrix, the first matrix and the parameter of the third dimension, where the parameter of the third dimension is a parameter of at least one dimension among the parameters of the multiple dimensions;

[0246] When the second difference is less than or equal to the second preset value, a second matrix is ​​generated according to the historical memory matrix and the parameter values ​​of the parameters of the third dimension, and the parameter values ​​of the parameters of the third dimension are the parameter values ​​of the parameters of the corresponding dimension in the historical memory matrix;

[0247] The second matrix is ​​updated to the first matrix, the parameters of the third dimension are updated to the parameters of the first dimension, and the observation values ​​of the parameters of the historical memory matrix, the first matrix and the second dimension are returned to obtain the second residual value corresponding to the parameters of each dimension, until the second residual value corresponding to the parameters of each dimension is less than or equal to the first preset value, and the second matrix is ​​used as the target matrix.

[0248] Optionally, the processor 701 is specifically configured to:

[0249] Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension, a second estimated value of the parameter of each dimension is obtained;

[0250] According to the observed value of the parameter of each dimension and the second estimated value of the parameter of the corresponding dimension, a second residual value of the parameter of each dimension is obtained.

[0251] Optionally, the processor 701 is specifically configured to:

[0252] Using the MSET algorithm, according to the historical memory matrix and the observed value of the parameter of each dimension, a first estimated value of the parameter of each dimension is obtained;

[0253] Obtaining a first average relative error value according to the observed value of the parameter of each dimension and the corresponding first estimated value;

[0254] Removing parameter values ​​of parameters of the first dimension from the historical memory matrix to obtain the first matrix;

[0255] Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension, a second estimated value of the parameter of each dimension is obtained;

[0256] Obtaining a second average relative error value according to the observed value of the parameter of each dimension and the corresponding second estimated value;

[0257] The first difference is obtained according to the first average relative error value and the second average relative error value.

[0258] Optionally, the processor 701 is specifically configured to:

[0259] Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the third dimension, a third estimated value of the parameter of each dimension is obtained;

[0260] Obtaining a third average relative error value according to the observed value of the parameter of each dimension and the corresponding third estimated value;

[0261] Removing parameter values ​​of the parameters of the third dimension from the first matrix to obtain the second matrix;

[0262] Using the MSET algorithm, a fourth estimated value of the parameter of each dimension is obtained according to the observed values ​​of the historical memory matrix, the second matrix and the parameter of the fourth dimension, where the parameter of the fourth dimension is a parameter of the dimension other than the parameter of the third dimension among the parameters of the multiple dimensions;

[0263] Obtaining a fourth average relative error value according to the observed value of the parameter of each dimension and the corresponding fourth estimated value;

[0264] The second difference is obtained according to the third average relative error value and the fourth average relative error value.

[0265] Optionally, the processor 701 is further configured to:

[0266] When the first difference is greater than the second preset value, the MSET algorithm is used to obtain the final estimated value of the parameter of each dimension based on the observed values ​​of the historical memory matrix, the target matrix and the parameters of the second dimension.

[0267] Optionally, the processor 701 is further configured to:

[0268] When the second difference is greater than the second preset value, the second matrix is ​​used as the target matrix.

[0269] Optionally, the processor 701 is further configured to:

[0270] When the residual value of each dimension is less than or equal to the first preset value, the MSET algorithm is used to obtain the final estimated value of the parameter of each dimension based on the historical memory matrix and the observed value of the parameter of each dimension.

[0271] Optionally, the processor 701 is further configured to:

[0272] Obtaining historical parameter values ​​of parameters of multiple dimensions of the device;

[0273] The historical parameter values ​​are processed using at least one of the following methods to obtain the historical memory matrix:

[0274] Expert rule algorithm, density clustering algorithm and equal-interval sampling algorithm.

[0275] Among them, Figure 7 In the embodiment, the bus architecture may include any number of interconnected buses and bridges, specifically linking various circuits of one or more processors represented by processor 701 and memory represented by memory 703. The bus architecture may also link various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides a user interface 705. The transceiver 704 may be a plurality of components, i.e., including a transmitter and a receiver, providing a unit for communicating with various other devices on a transmission medium. The processor 701 is responsible for managing the bus architecture and general processing, and the memory 703 may store data used by the processor 701 when performing operations.

[0276] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or may be accomplished by instructing relevant hardware through a computer program, wherein the computer program includes instructions for executing part or all of the steps of the above method; and the computer program may be stored in a readable storage medium, and the storage medium may be any form of storage medium.

[0277] In addition, an embodiment of the present invention further provides a computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, each process of the embodiment of the power generation equipment fault warning method for N-order deep multivariate state estimation as described above is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. Among them, the computer-readable storage medium, such as a read-only memory (Read-Only Memory, referred to as ROM), a random access memory (RandomAccess Memory, referred to as RAM), a disk or an optical disk, etc.

[0278] In addition, it should be noted that in the apparatus and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. Moreover, the steps of performing the above-mentioned series of processing can naturally be performed in the order of description or in chronological order, but it is not necessary to perform them in chronological order, and some steps can be performed in parallel or independently of each other. For those of ordinary skill in the art, it is understood that all or any steps or components of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or a network of computing devices in hardware, firmware, software or a combination thereof, which can be achieved by those of ordinary skill in the art using their basic programming skills after reading the description of the present invention.

[0279] Therefore, the purpose of the present invention can also be achieved by running a program or a group of programs on any computing device. The computing device can be a well-known general-purpose device. Therefore, the purpose of the present invention can also be achieved by simply providing a program product containing a program code that implements the method or device. In other words, such a program product also constitutes the present invention, and a storage medium storing such a program product can also constitute the present invention. Obviously, the storage medium can be any well-known storage medium or any storage medium developed in the future.

[0280] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0281] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A power generation equipment fault early warning method based on N-order deep multivariate state estimation, characterized in that: include: Obtain observation values ​​of parameters of multiple dimensions of the device; Obtaining a first residual value corresponding to the parameter of each dimension according to an observed value of the parameter of each dimension and a historical memory matrix, wherein the historical memory matrix is ​​generated according to historical parameter values ​​of the parameters of multiple dimensions of the device; When a first residual value of a parameter of a first dimension is greater than a first preset value, obtaining a target matrix according to the historical memory matrix and the parameter of the first dimension, where the parameter of the first dimension is a parameter of at least one dimension among the parameters of the multiple dimensions; Obtaining a final estimated value of the parameter of each dimension according to the historical memory matrix, the target matrix and the observed values ​​of the parameters of the second dimension, wherein the parameters of the second dimension are the parameters of the dimensions other than the parameters of the first dimension among the parameters of the multiple dimensions; Wherein, obtaining a target matrix according to the historical memory matrix and the parameters of the first dimension includes: Obtaining a first difference value according to the historical memory matrix and the parameter of the first dimension; When the first difference is less than or equal to a second preset value, a first matrix is ​​generated according to the historical memory matrix and the parameter values ​​of the parameters of the first dimension, and the parameter values ​​of the parameters of the first dimension are the parameter values ​​of the parameters of the corresponding dimension in the historical memory matrix; Obtaining a second residual value corresponding to the parameter of each dimension according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension; When the second residual value corresponding to the parameter of each dimension is less than or equal to the first preset value, using the first matrix as the target matrix; When a second residual value corresponding to a parameter of a third dimension is greater than the first preset value, a second difference value is obtained according to the historical memory matrix, the first matrix and the parameter of the third dimension, where the parameter of the third dimension is a parameter of at least one dimension among the parameters of the multiple dimensions; When the second difference is less than or equal to the second preset value, a second matrix is ​​generated according to the historical memory matrix and the parameter values ​​of the parameters of the third dimension, and the parameter values ​​of the parameters of the third dimension are the parameter values ​​of the parameters of the corresponding dimension in the historical memory matrix; The second matrix is ​​updated to the first matrix, the parameters of the third dimension are updated to the parameters of the first dimension, and the observation values ​​of the parameters of the historical memory matrix, the first matrix and the second dimension are returned to obtain the second residual value corresponding to the parameters of each dimension, until the second residual value corresponding to the parameters of each dimension is less than or equal to the first preset value, and the second matrix is ​​used as the target matrix.

2. The method according to claim 1, characterized in that According to the observed values ​​of the parameters of each dimension and the historical memory matrix, the first residual value corresponding to the parameters of each dimension is obtained, including: Using the multivariate state estimation technology MSET algorithm, according to the historical memory matrix and the observed value of the parameter of each dimension, a first estimated value of the parameter of each dimension is obtained; According to the observed value of the parameter of each dimension and the first estimated value of the parameter of the corresponding dimension, the first residual value of the parameter of each dimension is obtained.

3. The method according to claim 1, characterized in that Obtaining a second residual value corresponding to the parameter of each dimension according to the historical memory matrix, the first matrix, and the observed values ​​of the parameters of the second dimension includes: Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension, a second estimated value of the parameter of each dimension is obtained; According to the observed value of the parameter of each dimension and the second estimated value of the parameter of the corresponding dimension, a second residual value of the parameter of each dimension is obtained.

4. The method according to claim 1, characterized in that: Obtaining a first difference value according to the historical memory matrix and the parameter of the first dimension includes: Using the MSET algorithm, according to the historical memory matrix and the observed value of the parameter of each dimension, a first estimated value of the parameter of each dimension is obtained; Obtaining a first average relative error value according to the observed value of the parameter of each dimension and the corresponding first estimated value; Removing parameter values ​​of parameters of the first dimension from the historical memory matrix to obtain the first matrix; Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the second dimension, a second estimated value of the parameter of each dimension is obtained; Obtaining a second average relative error value according to the observed value of the parameter of each dimension and the corresponding second estimated value; The first difference is obtained according to the first average relative error value and the second average relative error value.

5. The method according to claim 1, characterized in that Obtaining a second difference value according to the historical memory matrix, the first matrix and the parameter of the third dimension includes: Using the MSET algorithm, according to the historical memory matrix, the first matrix and the observed values ​​of the parameters of the third dimension, a third estimated value of the parameter of each dimension is obtained; Obtaining a third average relative error value according to the observed value of the parameter of each dimension and the corresponding third estimated value; Removing parameter values ​​of the parameters of the third dimension from the first matrix to obtain the second matrix; Using the MSET algorithm, a fourth estimated value of the parameter of each dimension is obtained according to the observed values ​​of the historical memory matrix, the second matrix and the parameter of the fourth dimension, where the parameter of the fourth dimension is a parameter of the dimension other than the parameter of the third dimension among the parameters of the multiple dimensions; Obtaining a fourth average relative error value according to the observed value of the parameter of each dimension and the corresponding fourth estimated value; The second difference is obtained according to the third average relative error value and the fourth average relative error value.

6. The method according to claim 1, characterized in that The method further comprises: When the first difference is greater than the second preset value, the MSET algorithm is used to obtain the final estimated value of the parameter of each dimension based on the observed values ​​of the historical memory matrix, the target matrix and the parameters of the second dimension.

7. The method according to claim 1, characterized in that The method further comprises: When the second difference is greater than the second preset value, the second matrix is ​​used as the target matrix.

8. The method according to claim 1, characterized in that The method further comprises: When the residual value of each dimension is less than or equal to the first preset value, the MSET algorithm is used to obtain the final estimated value of the parameter of each dimension based on the historical memory matrix and the observed value of the parameter of each dimension.

9. The method according to claim 1, characterized in that: The method further comprises: Obtaining historical parameter values ​​of parameters of multiple dimensions of the device; The historical parameter values ​​are processed using at least one of the following methods to obtain the historical memory matrix: Expert rule algorithm, density clustering algorithm and equal-interval sampling algorithm.

10. A power generation equipment fault warning device using N-order deep multivariate state estimation technology, characterized in that: include: A first acquisition module is used to acquire observation values ​​of parameters of multiple dimensions of a device; A first processing module, configured to obtain a first residual value corresponding to the parameter of each dimension according to an observed value of the parameter of each dimension and a historical memory matrix, wherein the historical memory matrix is ​​generated according to historical parameter values ​​of the parameters of multiple dimensions of the device; a second processing module, configured to obtain a target matrix according to the historical memory matrix and the parameters of the first dimension when a first residual value of the parameters of the first dimension is greater than a first preset value, wherein the parameters of the first dimension are parameters of at least one dimension among the parameters of the multiple dimensions; a third processing module, configured to obtain a final estimated value of the parameter of each dimension according to the historical memory matrix, the target matrix and the observed value of the parameter of the second dimension, wherein the parameter of the second dimension is a parameter of a dimension other than the parameter of the first dimension among the parameters of the multiple dimensions; Wherein, the second processing module includes: A third processing unit, configured to obtain a first difference value according to the historical memory matrix and the parameter of the first dimension; a fourth processing unit, configured to generate a first matrix according to the historical memory matrix and parameter values ​​of the parameters of the first dimension when the first difference is less than or equal to a second preset value, wherein the parameter values ​​of the parameters of the first dimension are parameter values ​​of the parameters of the corresponding dimension in the historical memory matrix; a fifth processing unit, configured to obtain a second residual value corresponding to the parameter of each dimension according to the historical memory matrix, the first matrix and the observed value of the parameter of the second dimension; a sixth processing unit, configured to use the first matrix as the target matrix when the second residual value corresponding to the parameter of each dimension is less than or equal to the first preset value; a seventh processing unit, configured to obtain a second difference value according to the historical memory matrix, the first matrix and the parameter of the third dimension when a second residual value corresponding to the parameter of the third dimension is greater than the first preset value, wherein the parameter of the third dimension is a parameter of at least one dimension among the parameters of the multiple dimensions; an eighth processing unit, configured to generate a second matrix according to the historical memory matrix and the parameter values ​​of the parameters of the third dimension when the second difference is less than or equal to the second preset value, wherein the parameter values ​​of the parameters of the third dimension are the parameter values ​​of the parameters of the corresponding dimension in the historical memory matrix; The ninth processing unit is used to update the second matrix to the first matrix, update the parameters of the third dimension to the parameters of the first dimension, return to the step of obtaining the second residual value corresponding to the parameters of each dimension based on the observed values ​​of the historical memory matrix, the first matrix and the parameters of the second dimension, until the second residual value corresponding to the parameters of each dimension is less than or equal to the first preset value, and use the second matrix as the target matrix.

11. A power generation equipment fault warning device using N-order deep multivariate state estimation technology, comprising a transceiver, a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the power generation equipment fault warning method of N-order deep multivariate state estimation according to any one of claims 1 to 9 is implemented.

12. A readable storage medium having a program or instruction stored thereon, characterized in that: When the program or instruction is executed by the processor, the power generation equipment fault warning method of N-order deep multivariate state estimation as described in any one of claims 1 to 9 is implemented.

Citation Information

Patent Citations

  • Virtualized network element fault analysis method and system based on multi-observation-dimension HMM

    CN111565118A

  • Power plant blower fault early warning method based on multivariate state estimation

    CN112067335A