A self-diagnosis early warning function abnormal state recognition system

Abstract

This invention provides a self-diagnostic early warning function abnormal state identification system, relating to the field of thermal power plant safety technology. It includes a data acquisition module for real-time acquisition of raw data; a data processing module for preprocessing the raw data and fault datasets extracted from historical data; a model building module for obtaining an abnormal self-diagnosis prediction model from the union of binary classifiers trained using the processed data; and a state identification module for inputting data into the abnormal self-diagnosis prediction model, identifying fault data, and sending corresponding alarm signals to the operator station based on the output first fault category. By obtaining the abnormal diagnosis prediction model from the union of binary classifiers trained using preprocessed fault data, and determining the corresponding alarm signal based on the fault category output from the real-time data input to the abnormal diagnosis prediction model, the system ensures safe operation and provides a basis for condition-based maintenance.

Description

Technical Field

[0001] This invention relates to the field of safety technology in thermal power plants, and in particular to an abnormal state identification system with self-diagnostic early warning function. Background Technology

[0002] The ash conveying system is an important component of the dust removal system in thermal power plants. Its operating efficiency directly affects the dust removal efficiency and operating status of the dust removal system. As environmental protection issues receive more and more attention, the requirements for dust removal in thermal power plants are also getting higher and higher, and the work of the ash conveying system has become extremely critical.

[0003] Currently, the control of ash conveying systems still relies mainly on preset parameters for automatic control. When problems such as pipe blockage, conveying timeout, pressurization leakage, valve jamming, and abnormal material level occur during the ash conveying process, the system cannot detect or accurately diagnose the fault in a timely manner, which leads to a decrease in the conveying efficiency of the ash conveying system. In severe cases, it may even lead to damage to the ash conveying system equipment.

[0004] Therefore, the present invention provides a self-diagnostic early warning function abnormal state identification system. Summary of the Invention

[0005] This invention provides a self-diagnostic early warning function abnormal state identification system, which obtains an abnormal diagnosis prediction model by using the union of binary classifiers trained with preprocessed historical fault data; based on the fault category output by the abnormal diagnosis prediction model with real-time data input, the corresponding alarm signal is determined and fed back to the operator station, ensuring the safe operation of the system while also providing a basis for condition-based maintenance.

[0006] This invention provides a self-diagnostic early warning function abnormal state identification system, comprising:

[0007] Data acquisition module: used to collect raw data in real time using a preset device installed at a preset location, and to store the data;

[0008] Data processing module: used to preprocess the raw data and the fault dataset extracted from historical data respectively to obtain the first data and the first target data;

[0009] Model building module: used to obtain an anomaly self-diagnosis prediction model by combining the binary classifiers trained based on the first target data;

[0010] Status recognition module: It is used to input the first data into the abnormal self-diagnosis prediction model, identify the fault data and output the corresponding first fault category, and then send the corresponding alarm signal to the operator station according to the first fault category.

[0011] Preferably, the data acquisition module includes:

[0012] The device installation unit is used to install a first pressure sensor at each ash conveying pipeline in the ash conveying system of a thermal power plant, a second pressure sensor and a position feedback detection switch at the discharge valve of the conveying silo pump group and the pneumatic control valve respectively, and a passive nuclear level gauge at the lower position of the ash hopper.

[0013] Data acquisition unit: used to acquire real-time data from the first pressure sensor, the second pressure sensor, the position feedback detection switch and the passive nuclear level gauge, and to combine the corresponding time data and position data to obtain raw data, which is then transmitted to the data storage unit for storage.

[0014] Preferably, the data processing module is used to perform redundancy processing on the original data and the fault dataset respectively, and then standardize the data to obtain the first data and the first target data.

[0015] Preferably, the model building module includes:

[0016] Sample determination unit: used to arbitrarily divide the first target data into m groups, and determine any m-1 groups of the first target data as the training sample set Y = {y1, y2, y3, ... y n}, where n represents the total number of samples;

[0017] Nearest Neighbor Construction Unit: Used to utilize the training sample set Y = {y1, y2, y3, ... y n}, obtain principal component information Z = {z1, z2, z3, ... z} n};

[0018] Based on Euclidean distance, construct the nearest neighbor Z of low-dimensional spatial sample points. i ={z i1 , z i2 , z i3 , ...z ik+1}(i=1,2,3,…,n) and the nearest neighbor Y of the data space sample points i ={y i1 y i2 y i3 , ...y ik}(i = 1, 2, 3, ..., n);

[0019] Reconstructed sample unit: used to obtain the affine matrix and obtain the reconstruction error based on the inverse mapping of the local error of each low-dimensional sample point to its neighborhood center;

[0020] Align each nearest neighbor Y in the original space i center The reconstruction error is used to obtain the reconstructed sample points y. ri Finally, the reconstructed sample set Y is obtained. r ={yr1 y r2 y r3 , ...y rn};

[0021] Model building unit: It is used to build a binary classifier based on the reconstructed sample set, and then retrain the optimal binary classifier based on the parameters of the binary classifier obtained by the cross-validation algorithm, and finally combine them to obtain the anomaly self-diagnosis prediction model.

[0022] Preferably, the model building unit includes:

[0023] Partitioning blocks: used to divide the reconstructed sample set Y into N category labels based on the reconstructed sample set. r ={y r1 y r2 y r3 , ...y rn The sample with all pairs of different labels in the array is divided to construct the... A set of two classes of samples, where [] represents the integer symbol;

[0024] Training block: Used to train a binary classifier for each binary sample set, and then input the first target data of the remaining set as the test data to judge the classification result.

[0025] If there is only one result from a binary classifier, the binary classifier is labeled as the first positive class, and the final classification result is the category label corresponding to the first positive class.

[0026] If there is more than one binary classifier result, the category label corresponding to the binary classifier with the highest reliability is selected as the final classification result;

[0027] Building blocks: used to retrain the optimal binary classifier based on the cross-validation algorithm, and then combine them to obtain the anomaly self-diagnosis prediction model.

[0028] Preferably, the building block includes:

[0029] Parameter calculation unit: used to obtain the first evaluation parameter of the binary classifier based on the judgment of the classification result of the test data and the actual classification result;

[0030] The formula for calculating the first evaluation parameter is as follows:

[0031]

[0032] Among them, P j Let y be the first evaluation parameter of the j-th binary classifier; jT Let y represent the number of samples whose true classifier correctly identifies the data type in the test data by the j-th binary classifier; N represents the total number of test data; yjF This represents the number of samples in which the j-th binary classifier incorrectly classifies the true category of the test data. Let the precision of the j-th binary classifier be denoted as . This is expressed as the weighting coefficient of the impact of precision on the first evaluation parameter; Let the recall rate of the j-th binary classifier be denoted as . ε represents the weighting coefficient of the influence of recall rate on the first evaluation parameter; ε represents the calculation loss factor for obtaining the first evaluation parameter.

[0033] If the first evaluation parameter of a binary classifier is less than the preset evaluation threshold, then each time, any one group of data from the first target data divided into m groups will be used as the test data, and the remaining m-1 groups of data will be used as the training sample set to train and obtain several groups of undetermined binary classifiers.

[0034] Obtain the first evaluation parameters of each group of undetermined binary classifiers, and combine them with the preset evaluation threshold to select the optimal binary classifier group with the best classification performance;

[0035] By combining all binary classifiers in the optimal binary classifier group, we obtain the anomaly self-diagnosis prediction model.

[0036] Preferred options also include:

[0037] Calculation module: Calculates the reliability of the binary classifier using the formula shown below:

[0038]

[0039] Among them, K s Let represent the reliability of the s-th binary classifier; g(·) represents an increasing mapping function based on the nearest neighbor distance variable within a class, with a range of [0, 1]; l s (C a |x) represents the s-th binary classifier identifying the input test data x as category C. a The decision output is the intra-class nearest neighbor distance; x represents the test data to be classified using the s-th binary classifier; C a Let d represent the category obtained by the s-th binary classifier after classifying the test data x, where a∈{1,2,…,r}; s (C a |x) represents the posterior probability criterion; h(·) represents the mapping function based on the posterior probability;

[0040] When there is more than one binary classifier result, the first reliability corresponding to each binary classifier result is calculated using the formula.

[0041] The category label corresponding to the binary classifier with the highest reliability is determined as the final classification result.

[0042] Preferably, the state recognition module includes:

[0043] Identification unit: used to input the first data into the anomaly diagnosis prediction model and output the first result, and based on the first result, determine the output sample set as fault data and the output fault category as the first fault category;

[0044] Feedback unit: used to determine the corresponding alarm signal based on the first fault category and send it to the operator station.

[0045] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0046] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0047] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0048] Figure 1 This is a structural diagram of a self-diagnostic early warning function abnormal state recognition system according to an embodiment of the present invention. Detailed Implementation

[0049] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0050] This invention provides a self-diagnostic early warning function abnormal state identification system, such as... Figure 1 As shown, it includes:

[0051] Data acquisition module: used to collect raw data in real time using a preset device installed at a preset location, and to store the data;

[0052] Data processing module: used to preprocess the raw data and the fault dataset extracted from historical data respectively to obtain the first data and the first target data;

[0053] Model building module: used to obtain an anomaly self-diagnosis prediction model by combining the binary classifiers trained based on the first target data;

[0054] Status recognition module: It is used to input the first data into the abnormal self-diagnosis prediction model, identify the fault data and output the corresponding first fault category, and then send the corresponding alarm signal to the operator station according to the first fault category.

[0055] In this embodiment, the preset position refers to each ash conveying pipe in the ash conveying system of the thermal power plant, the discharge valve of the conveying silo pump group, the pneumatic control valve, and the lower position of the ash hopper; the preset device includes a first pressure sensor, a second pressure sensor, a position feedback detection switch, and a passive nuclear level gauge.

[0056] In this embodiment, the raw data includes pressure data collected in real time by the first pressure sensor and the second pressure sensor, position feedback signal collected in real time by the position feedback detection switch, level data collected in real time by the passive nuclear level gauge, current time data of the real-time collected data, and position data of all preset devices; the fault dataset consists of a certain preset amount of data with anomalies extracted from historical data.

[0057] In this embodiment, the first data and the first target data are obtained by performing redundancy processing on the original data and the fault dataset respectively, followed by standardization processing. The purpose of redundancy processing is to reduce useless data and avoid resource waste; the purpose of standardization processing is to eliminate the influence of different dimensions of data in the dataset on the data and reduce distribution differences.

[0058] In this embodiment, the binary classifier is a model trained on the first target data to classify two categories; the anomaly self-diagnosis prediction model is obtained by combining the parameters of the binary classifier and all binary classifiers, and is used to identify fault data and output the corresponding fault category.

[0059] In this embodiment, fault data refers to the data output by inputting the raw data collected in real time and the processed data into the abnormal self-diagnosis prediction model; the first fault category refers to the category output by inputting the raw data collected in real time and the processed data into the abnormal self-diagnosis prediction model, such as pipe blockage, conveying timeout, or abnormal material level; the operator station is used to make an effective response based on the acquired alarm signal.

[0060] The beneficial effects of the above technical solution are as follows: by using the union of binary classifiers trained with preprocessed historical fault data, an anomaly diagnosis prediction model is obtained; based on the fault category output by the anomaly diagnosis prediction model with real-time data input, the corresponding alarm signal is determined and fed back to the operator station, which ensures the safe operation of the system and provides a basis for condition-based maintenance.

[0061] This invention provides a self-diagnostic early warning function abnormal state identification system, wherein the data acquisition module includes:

[0062] The device installation unit is used to install a first pressure sensor at each ash conveying pipeline in the ash conveying system of a thermal power plant, a second pressure sensor and a position feedback detection switch at the discharge valve of the conveying silo pump group and the pneumatic control valve respectively, and a passive nuclear level gauge at the lower position of the ash hopper.

[0063] Data acquisition unit: used to acquire real-time data from the first pressure sensor, the second pressure sensor, the position feedback detection switch and the passive nuclear level gauge, and to combine the corresponding time data and position data to obtain raw data, which is then transmitted to the data storage unit for storage.

[0064] In this embodiment, the ash conveying system of the thermal power plant mainly includes a silo conveying pump, an ash conveying pipeline, an air source, an ash hopper, and a control unit; a first pressure sensor is used to measure the pressure data of the conveying pipeline in real time, thereby determining whether there is a conveying timeout or pipe blockage; a second pressure sensor is used to measure the pressure value of the ash conveying silo pump group in real time, corresponding to the conveying sealing performance in the pressurization fluidization stage, thereby determining whether there is a pressurization leakage; a passive nuclear level gauge is used to monitor and control the material level in the silo pump to determine whether there is an abnormal material level; a position feedback detection switch is used to collect position feedback signals in real time, thereby determining whether there is a valve jamming fault.

[0065] In this embodiment, the raw data consists of data collected in real time by all preset devices, corresponding time data, and location data. The time data refers to the current moment data collected in real time by the preset devices, and the location data refers to the specific location where the preset devices are installed.

[0066] The beneficial effects of the above technical solution are: by using pressure sensors, position feedback switch devices and passive nuclear level gauges installed at preset positions to collect real-time data, and then transmitting it to the data storage unit for storage, data support is provided for subsequent fault identification.

[0067] This invention provides a self-diagnostic early warning function abnormal state identification system, wherein the model building module includes:

[0068] Sample determination unit: used to arbitrarily divide the first target data into m groups, and determine any m-1 groups of the first target data as the training sample set Y = {y1, y2, y3, ... y n}, where n represents the total number of samples;

[0069] Nearest Neighbor Construction Unit: Used to utilize the training sample set Y = {y1, y2, y3, ... y n}, obtain principal component information Z = {z1, z2, z3, ... z} n};

[0070] Based on Euclidean distance, construct the nearest neighbor Z of low-dimensional spatial sample points.i ={z i1 , z i2 , z i3 , ...z ik+1}(i=1,2,3,…,n) and the nearest neighbor Y of the data space sample points i ={y i1 y i2 y i3 , ...y ik}(i = 1, 2, 3, ..., n);

[0071] Reconstructed sample unit: used to obtain the affine matrix and obtain the reconstruction error based on the inverse mapping of the local error of each low-dimensional sample point to its neighborhood center;

[0072] Align each nearest neighbor Y in the original space i center The reconstruction error is used to obtain the reconstructed sample points y. ri Finally, the reconstructed sample set Y is obtained. r ={y r1 y r2 y r3 , ...y rn};

[0073] Model building unit: It is used to build a binary classifier based on the reconstructed sample set, and then retrain the optimal binary classifier based on the parameters of the binary classifier obtained by the cross-validation algorithm, and finally combine them to obtain the anomaly self-diagnosis prediction model.

[0074] In this embodiment, the first target data is the data obtained by redundancy processing and standardization of the fault dataset; the training sample set consists of any m-1 sets of the first target data.

[0075] In this embodiment, the purpose of obtaining the principal component information of the training sample set is to extract global feature information of the data, reduce the complexity of data features while ensuring the effectiveness of data information, and improve the subsequent training effect of the model; the purpose of the affine matrix is ​​to inversely map the local structural error to the high-dimensional space, where the local error refers to the difference between each low-dimensional sample point and its neighborhood center; the purpose of reconstructing the error of each nearest neighborhood center in its original space is to obtain the lost local structural information and improve the data representation ability of the sample, where the reconstruction error is obtained by inverse mapping of the local error.

[0076] In this embodiment, the binary classifier is a model for classifying two categories; the cross-validation algorithm is mainly used to obtain the parameters of the binary classifier trained by inputting different groups of test data into different training sample sets, so as to relatively objectively judge the classification effect of these parameters corresponding to the binary classifier; the anomaly self-diagnosis prediction model is obtained by the union of the finally trained binary classifiers, and is used to identify fault data and fault categories in real time.

[0077] The beneficial effects of the above technical solution are as follows: by using most of the first target data as training samples, principal component information is obtained, then a nearest neighbor is constructed for each coordinate in the low-dimensional space, and the local error between each coordinate and its neighborhood center is obtained. Then, the error is reconstructed back to the high-dimensional space by inverse mapping, and the reconstructed coordinates are used as the reconstructed sample set to train a classifier, thereby constructing an anomaly self-diagnosis prediction model. This improves the expressive power of the data while ensuring the classification effect of the classifier and the accurate identification function of the anomaly self-diagnosis prediction model.

[0078] This invention provides a self-diagnostic early warning function abnormal state identification system, wherein the model building unit includes:

[0079] Partitioning blocks: used to divide the reconstructed sample set Y into N category labels based on the reconstructed sample set. r ={y r1 y r2 y r3 , ...y rn The sample with all pairs of different labels in the array is divided to construct the... A set of two classes of samples, where [] represents the integer symbol;

[0080] Training block: Used to train a binary classifier for each binary sample set, and then input the first target data of the remaining set as the test data to judge the classification result.

[0081] If there is only one result from a binary classifier, the binary classifier is labeled as the first positive class, and the final classification result is the category label corresponding to the first positive class.

[0082] If there is more than one binary classifier result, the category label corresponding to the binary classifier with the highest reliability is selected as the final classification result;

[0083] Building blocks: used to retrain the optimal binary classifier based on the cross-validation algorithm, and then combine them to obtain the anomaly self-diagnosis prediction model.

[0084] In this embodiment, the category label refers to the data classification name, such as the pressure value of the ash conveying silo pump group or the pressure value of the conveying pipeline; the second-class sample set consists of two different categories of samples.

[0085] In this embodiment, for example, there is a first target data divided into 10 groups, of which any 9 groups of data are used as training sample sets, and the remaining group of first target data is used as test data.

[0086] In this embodiment, for example, there is data D1 to be tested, which is classified to obtain two results: classifier C1 and classifier C2. At this time, the reliability of classifier C1 and classifier C2 is obtained respectively, and the category label corresponding to the classifier with the highest reliability is selected as the final classification result.

[0087] In this embodiment, the cross-validation algorithm is mainly used to obtain the parameters of the binary classifiers trained by inputting different groups of test data into different training sample sets; the optimal binary classifier is obtained by retraining with suitable training data based on the cross-validation algorithm; and the optimal binary classifier is obtained by combining the anomaly self-diagnosis prediction model.

[0088] The beneficial effects of the above technical solution are: by using the reconstructed sample set to train a binary classifier, and obtaining the corresponding classification result based on the test data; based on the effect analysis of the classification result, the optimal binary classifier is obtained by retraining in combination with the cross-validation algorithm, and then the anomaly self-diagnosis prediction model is established by merging the results, thus ensuring the accurate identification function of the anomaly self-diagnosis prediction model.

[0089] This invention provides a self-diagnostic early warning function abnormal state identification system, the building blocks of which include:

[0090] Parameter calculation unit: used to obtain the first evaluation parameter of the binary classifier based on the judgment of the classification result of the test data and the actual classification result;

[0091] The formula for calculating the first evaluation parameter is as follows:

[0092]

[0093] Among them, P j Let y be the first evaluation parameter of the j-th binary classifier; jT Let y represent the number of samples whose true classifier correctly identifies the data type in the test data by the j-th binary classifier; N represents the total number of test data; y jF This represents the number of samples in which the j-th binary classifier incorrectly classifies the true category of the test data. Let the precision of the j-th binary classifier be denoted as . This is expressed as the weighting coefficient of the impact of precision on the first evaluation parameter; Let the recall rate of the j-th binary classifier be denoted as . ε represents the weighting coefficient of the influence of recall rate on the first evaluation parameter; ε represents the calculation loss factor for obtaining the first evaluation parameter.

[0094] If the first evaluation parameter of a binary classifier is less than the preset evaluation threshold, then each time, any one group of data from the first target data divided into m groups will be used as the test data, and the remaining m-1 groups of data will be used as the training sample set to train and obtain several groups of undetermined binary classifiers.

[0095] Obtain the first evaluation parameters of each group of undetermined binary classifiers, and combine them with the preset evaluation threshold to select the optimal binary classifier group with the best classification performance;

[0096] By combining all binary classifiers in the optimal binary classifier group, we obtain the anomaly self-diagnosis prediction model.

[0097] In this embodiment, the preset evaluation threshold is set in advance, generally 0.85; the undetermined binary classifier is a classifier obtained by training any one group of data in the first target data divided into m groups as the test data and the remaining m-1 groups of data as the training sample set; the optimal binary classifier group is composed of the classifier with the best classification effect, that is, the classifier with the best first evaluation parameter, and is used to obtain the anomaly self-diagnosis prediction model by union.

[0098] The beneficial effects of the above technical solution are: by obtaining the first evaluation parameters of the binary classifier, its classification effect is judged; by combining the cross-validation algorithm, the optimal binary classifier is retrained using the most suitable training data, and the optimal anomaly self-diagnosis prediction model with the best recognition function is obtained by combining the results, thereby improving the fault identification efficiency, ensuring the safe operation of the system, and providing an accurate basis for condition-based maintenance.

[0099] This invention provides a self-diagnostic early warning function abnormal state identification system, which further includes:

[0100] Calculation module: Calculates the reliability of the binary classifier using the formula shown below:

[0101]

[0102] Among them, K s Let represent the reliability of the s-th binary classifier; g(·) represents an increasing mapping function based on the nearest neighbor distance variable within a class, with a range of [0, 1]; l s (C a |x) represents the s-th binary classifier identifying the input test data x as category C. a The decision output is the intra-class nearest neighbor distance; x represents the test data to be classified using the s-th binary classifier; C a Let d represent the category obtained by the s-th binary classifier after classifying the test data x, where a∈{1,2,…,r}; s (C a|x) represents the posterior probability criterion; h(·) represents the mapping function based on the posterior probability;

[0103] When there is more than one binary classifier result, the first reliability corresponding to each binary classifier result is calculated using the formula.

[0104] The category label corresponding to the binary classifier with the highest reliability is determined as the final classification result.

[0105] In this embodiment, for example, there are binary classifier results c1, c2, and c3, with corresponding reliability K1 > K2 = K3; in this case, the category label corresponding to the binary classifier result c1 is used as the final classification result.

[0106] The beneficial effects of the above technical solution are: by determining the class label corresponding to the most reliable binary classifier when there is more than one binary classifier result, the accuracy of providing a basis for condition-based maintenance is guaranteed, thereby indirectly ensuring the safe operation of the system.

[0107] This invention provides a self-diagnostic early warning function abnormal state identification system, wherein the state identification module includes:

[0108] Identification unit: used to input the first data into the anomaly diagnosis prediction model and output the first result, and based on the first result, determine the output sample set as fault data and the output fault category as the first fault category;

[0109] Feedback unit: used to determine the corresponding alarm signal based on the first fault category and send it to the operator station.

[0110] In this embodiment, the first data is obtained by redundancy processing of the original data and then standardization processing; the first result includes the output sample set and the fault category; the alarm signal is a signal emitted by a computer display device, sound device or indicator light, used to remind the ash conveying system of an emergency fault; the operator station can quickly know the location and cause of the fault based on the different alarm signals obtained, and thus take corresponding measures.

[0111] The beneficial effects of the above technical solution are: by inputting the preprocessed real-time collected raw data into the anomaly diagnosis and prediction model, fault data and corresponding fault categories can be quickly obtained. Then, according to the fault category, the corresponding alarm signal is sent to the operator console, which avoids the error rate of manual judgment, improves the timeliness of fault handling, and ensures the safe operation of the system.

[0112] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A self-diagnostic early warning function abnormal state identification system, characterized in that, include: Data acquisition module: used to collect raw data in real time using a preset device installed at a preset location, and to store the data; Data processing module: used to preprocess the raw data and the fault dataset extracted from historical data respectively to obtain the first data and the first target data; Model building module: includes a sample determination unit: used to arbitrarily divide the first target data into m groups, and determine any m-1 groups of the first target data as the training sample set Y = {y1, y2, y3, ... y n }, where n represents the total number of samples; Nearest Neighbor Construction Unit: used to utilize the training sample set Y = {y1, y2, y3, ... y n }, obtain principal component information Z = {z1, z2, z3, ... z} n Based on Euclidean distance, construct the nearest neighbor Z region for low-dimensional sample points. i ={z i1 ,z i2 ,z i3 ,…z ik+1 }(i=1,2,3,…,n) and the nearest neighbor Y of the data space sample points i ={y i1 ,y i2 ,y i3 ,…y ik }(i=1,2,3,…,n); Reconstructed sample unit: used to obtain the affine matrix and obtain the reconstruction error based on the inverse mapping of the local error between each low-dimensional space sample point and its neighborhood center; Align each nearest neighbor Y in the original space i center The reconstruction error is used to obtain the reconstructed sample points y. ri Finally, the reconstructed sample set Y is obtained. r ={y r1 ,y r2 ,y r3 ,…y rn Model building unit: includes partitioning blocks: used to partition the reconstructed sample set Y according to the N category labels of the reconstructed sample set. r ={y r1 ,y r2 ,y r3 ,…y rn The sample with all pairs of different labels in the array is divided to construct the... There are two binary sample sets, where [] represents the rounding symbol; Training block: used to train a binary classifier for each binary sample set, and then input the first target data of the remaining set as the test data to judge the classification result; if there is only one binary classifier result, the binary classifier is marked as the first positive class, and the final classification result is the category label corresponding to the first positive class; if there is more than one binary classifier result, the category label corresponding to the binary classifier with the highest reliability is selected as the final classification result; Building block: used to retrain based on the cross-validation algorithm to obtain the optimal binary classifier, and then combine them to obtain the anomaly self-diagnosis prediction model; Status recognition module: It is used to input the first data into the abnormal self-diagnosis prediction model, identify the fault data and output the corresponding first fault category, and then send the corresponding alarm signal to the operator station according to the first fault category.

2. A self-diagnostic early warning function abnormal state identification system according to claim 1, characterized in that, The data acquisition module includes: The device installation unit is used to install a first pressure sensor at each ash conveying pipeline in the ash conveying system of a thermal power plant, a second pressure sensor and a position feedback detection switch at the discharge valve of the conveying silo pump group and the pneumatic control valve respectively, and a passive nuclear level gauge at the lower position of the ash hopper. Data acquisition unit: used to acquire real-time data from the first pressure sensor, the second pressure sensor, the position feedback detection switch and the passive nuclear level gauge, and to combine the corresponding time data and position data to obtain raw data, which is then transmitted to the data storage unit for storage.

3. A self-diagnostic early warning function abnormal state identification system according to claim 1, characterized in that, The data processing module is used to perform redundancy processing on the original data and the fault dataset, and then standardize them to obtain the first data and the first target data.

4. A self-diagnostic early warning function abnormal state identification system according to claim 1, characterized in that, The building blocks include: Parameter calculation unit: used to obtain the first evaluation parameter of the binary classifier based on the judgment of the classification result of the test data and the actual classification result; The formula for calculating the first evaluation parameter is as follows: Among them, P j Let y be the first evaluation parameter of the j-th binary classifier; jT Let y represent the number of samples whose true classifier correctly identifies the data type in the test data by the j-th binary classifier; N represents the total number of test data; y jF This represents the number of samples in which the j-th binary classifier incorrectly classifies the true category of the test data. Let the precision of the j-th binary classifier be denoted as . This is expressed as the weighting coefficient of the impact of precision on the first evaluation parameter; Let the recall rate of the j-th binary classifier be denoted as . ε represents the weighting coefficient of the influence of recall rate on the first evaluation parameter; ε represents the calculation loss factor for obtaining the first evaluation parameter. If the first evaluation parameter of a binary classifier is less than the preset evaluation threshold, then each time, any one group of data from the first target data divided into m groups will be used as the test data, and the remaining m-1 groups of data will be used as the training sample set to train and obtain several groups of undetermined binary classifiers. Obtain the first evaluation parameters of each group of undetermined binary classifiers, and combine them with the preset evaluation threshold to select the optimal binary classifier group with the best classification performance; By combining all binary classifiers in the optimal binary classifier group, we obtain the anomaly self-diagnosis prediction model.

5. A self-diagnostic early warning function abnormal state identification system according to claim 1, characterized in that, Also includes: Calculation module: Calculates the reliability of the binary classifier using the formula shown below: Among them, K s Let represent the reliability of the s-th binary classifier; g(·) represents an increasing mapping function based on the nearest neighbor distance variable within a class, with a range of [0, 1]; l s (C a |x) represents the s-th binary classifier identifying the input test data x as category C. a The decision output is the intra-class nearest neighbor distance; x represents the test data to be classified using the s-th binary classifier; C a Let d represent the category obtained by the s-th binary classifier after classifying the test data x, where a∈{1,2,…,r}; s (C a |x) represents the posterior probability criterion; h(·) represents the mapping function based on the posterior probability; When there is more than one binary classifier result, the first reliability corresponding to each binary classifier result is calculated using the formula. The category label corresponding to the binary classifier with the highest reliability is determined as the final classification result.

6. A self-diagnostic early warning function abnormal state identification system according to claim 1, characterized in that, The status recognition module includes: Identification unit: used to input the first data into the anomaly diagnosis prediction model and output the first result, and based on the first result, determine the output sample set as fault data and the output fault category as the first fault category; Feedback unit: used to determine the corresponding alarm signal based on the first fault category and send it to the operator station.

Images