Sensor fault positioning method, device and equipment and storage medium

Abstract

The invention provides a sensor fault positioning method, device and equipment and a storage medium, in each round of fault positioning process, k sensors are excluded from N sensors each time to obtain a plurality of first data sets containing original data of N-k sensors, and the initial value of k is 1; for each first data group, based on redundant information among the sensors, reconstructing data of corresponding N-k sensors through original data of part of the sensors; calculating a reconstruction error according to the original data and the reconstruction data of the corresponding N-k sensors; if a first data group of which the reconstruction error does not exceed an error threshold exists in the current round of fault positioning, determining k sensors excluded from the first data group as fault sensors; otherwise, adding 1 to k and executing the next round of fault positioning, so that the accuracy of sensor fault positioning can be improved. In addition, because the method does not need to depend on high-frequency time sequence analysis, the method can be suitable for a fault positioning scene of a nuclear-level sensor with a low sampling rate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sensor anomaly detection, and particularly relates to a sensor fault positioning method and device, equipment and a storage medium. BACKGROUND

[0002] In the construction of a nuclear power plant safety monitoring system, a multi-sensor redundant configuration scheme is generally adopted for key equipment to realize state perception. These sensors collect multiple physical parameters such as pressure, temperature and vibration synchronously to provide basic data support for equipment health assessment and system parameter reconstruction. However, when a specific sensor deviates significantly from the actual working condition due to noise interference, zero drift or hardware failure, its abnormal data participating in fusion operation will cause misjudgment risk of the system state.

[0003] Although a fault-tolerant mechanism is included in the current redundant system design, its effectiveness depends on the reliable implementation of two key technical links: first, the real-time identification ability of the abnormal sensor, and second, the accurate positioning accuracy of the corresponding fault node. Ideally, timely exclusion of fault node data can enable the redundant system to continue to maintain a high credibility of state inversion function.

[0004] However, there are two technical bottlenecks in actual application: first, due to the inherent low sampling characteristics (usually ≤1 Hz) of nuclear-level sensors, the frequency domain analysis method and dynamic signal processing technology currently adopted are difficult to implement effectively; second, with the expansion of the monitoring system scale, the complex nonlinear feature space caused by the coupling effect between sensors makes the detection accuracy of the threshold-based judgment mechanism or traditional machine learning classification model show an exponential decline trend. SUMMARY

[0005] Therefore, the present application provides a sensor fault positioning method, device, equipment and storage medium to solve the deficiencies in the related art.

[0006] In a first aspect of the present application, a sensor fault positioning method is provided, applied to a system comprising N sensors, the N sensors having redundant information through regular arrangement, and the method comprising:

[0007] In a case where a fault sensor exists in the N sensors, in each round of fault positioning process, k different sensors are excluded from the N sensors each time to obtain a plurality of first data groups of original data of N-k sensors, wherein the initial value of k is 1;

[0008] For each first data group, one different sensor is excluded from the corresponding N-k sensors each time to obtain N-k second data groups containing N-(k+1) original data of sensors, and data of the corresponding N-k sensors is reconstructed according to each second data group respectively to obtain first reconstruction data;

[0009] According to the obtained N-k groups of first reconstruction data and the original data of the corresponding N-k sensors, first reconstruction errors corresponding to the first data groups are obtained;

[0010] If there is a first data group corresponding to a first reconstruction error that does not exceed an error threshold, the k sensors excluded in the first data group are determined as faulty sensors; otherwise, k is increased by 1 and the next round of fault positioning is performed until the faulty sensors are located or k reaches a set threshold.

[0011] According to an embodiment of the present application, the method further comprises:

[0012] N third data groups containing N-1 original data of sensors are obtained by excluding one different sensor from the N sensors each time;

[0013] Second reconstruction data is obtained by reconstructing data of the N sensors according to each third data group respectively;

[0014] Second reconstruction errors are obtained according to the obtained N groups of second reconstruction data and the original data of the N sensors;

[0015] If the second reconstruction error exceeds the error threshold, it is determined that there is a faulty sensor in the N sensors.

[0016] According to an embodiment of the present application, the data of the corresponding N-k sensors is reconstructed according to each second data group respectively to obtain first reconstruction data, comprising:

[0017] The N-k second data groups are input into a pre-trained detection model, so that the detection model reconstructs data of the corresponding N-k sensors according to each second data group respectively to generate first reconstruction data corresponding to each second data group.

[0018] According to an embodiment of the present application, the detection model is optimized by minimizing a loss function including reconstruction errors and regression errors in a training stage, and the detection model is specifically used for:

[0019] For each second data group, the second data group is preprocessed;

[0020] Feature extraction is performed on the preprocessed data by a feature encoder;

[0021] According to the extracted features, a regression value corresponding to the second data set is predicted by a regressor, and data of the corresponding N-k sensors is reconstructed by a feature decoder to generate first reconstructed data corresponding to the second data set.

[0022] According to an embodiment of the present application, the method further comprises:

[0023] Obtaining a plurality of sets of sample data, wherein each set of sample data contains normal data of N sensors;

[0024] Using the detection model, third reconstructed data corresponding to the i-th set of sample data after excluding the j-th sensor is generated;

[0025] The mean and standard deviation of all third reconstructed data are calculated, and the error threshold is determined according to the mean and the standard deviation.

[0026] According to an embodiment of the present application, the first reconstructed error corresponding to the first data set is obtained according to the obtained N-k sets of first reconstructed data and the original data of the corresponding N-k sensors, comprising:

[0027] The reconstruction error between each first reconstructed data and the original data of the corresponding N-k sensors is calculated;

[0028] From all the calculated reconstruction errors, the maximum value is selected as the first reconstructed error.

[0029] In a second aspect of the present application, a sensor fault locating device is provided, which is applied to a system containing N sensors, and the N sensors have redundant information through regular arrangement, and the device comprises:

[0030] An excluding unit is configured to, in a case where it is determined that there is a faulty sensor in the N sensors, exclude k different sensors from the N sensors each time in each round of fault locating process to obtain a plurality of first data sets containing original data of N-k sensors, wherein the initial value of k is 1;

[0031] A reconstructing unit is configured to, for each first data set, exclude one different sensor from the corresponding N-k sensors each time to obtain N-k second data sets containing original data of N-(k+1) sensors, and reconstruct data of the corresponding N-k sensors according to each second data set to obtain first reconstructed data;

[0032] An error calculating unit is configured to obtain a first reconstructed error corresponding to the first data set according to the obtained N-k sets of first reconstructed data and the original data of the corresponding N-k sensors;

[0033] a fault locating unit, configured to: if there is a first data group corresponding to the first reconstruction error not exceeding the error threshold, determine the k sensors excluded in the first data group as the fault sensors; otherwise, add 1 to k and perform the next round of fault locating until the fault sensors are located or k reaches a set threshold.

[0034] According to one embodiment of the present application, the apparatus further comprises an anomaly detecting unit, specifically configured to:

[0035] exclude one different sensor from the N sensors each time to obtain N third data groups containing original data of N-1 sensors;

[0036] reconstruct data of the N sensors according to each third data group respectively to obtain second reconstruction data;

[0037] obtain second reconstruction errors according to the obtained N groups of second reconstruction data and the original data of the N sensors;

[0038] if the second reconstruction errors exceed the error threshold, determine that there is a fault sensor in the N sensors.

[0039] In a third aspect of the present application, an electronic device is provided, comprising a processor and a memory, the memory storing machine executable instructions capable of being executed by the processor, and the processor is configured to execute the machine executable instructions to implement steps of the method proposed in the above embodiments.

[0040] In a fourth aspect of the present application, a machine readable storage medium is provided, the machine readable storage medium storing machine executable instructions, and the machine executable instructions are executed by a processor to implement steps of the method proposed in the above embodiments.

[0041] As can be seen from the above technical solutions, in each round of fault locating process, k different sensors are excluded from the N sensors each time to obtain a plurality of first data groups containing original data of N-k sensors, and the initial value of k is 1; for each first data group, based on the redundancy information between the sensors, the data of the corresponding N-k sensors is reconstructed by the original data of part of the sensors; according to the original data and the reconstructed data of the corresponding N-k sensors, the first reconstruction error corresponding to the first data group is calculated; if there is a first data group corresponding to the first reconstruction error not exceeding the error threshold in the current round of fault locating process, the k sensors excluded in the first data group are determined as the fault sensors; otherwise, add 1 to k and perform the next round of fault locating until the fault sensors are located or k reaches a set threshold, so as to improve the accuracy of sensor fault locating. In addition, since the present application does not need to rely on high frequency time series analysis, it can be applied to the fault locating scene of low sampling rate nuclear level sensors.

[0042] It should be understood that the general description and detailed description below are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 is a flow diagram of a sensor fault positioning method provided by an embodiment of the present application;

[0044] Figure 2 is a flow diagram of a sensor anomaly detection provided by an embodiment of the present application;

[0045] Figure 3 is a flow diagram of a sensor fault positioning provided by an embodiment of the present application;

[0046] Figure 4 is a structural diagram of a sensor fault positioning device provided by an embodiment of the present application;

[0047] Figure 5 is a hardware structure diagram of an electronic device shown by an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0048] The exemplary embodiments will be described in detail herein below with reference to the drawings. The following description relates to the drawings, in which the same numbers denote the same or similar elements throughout the several drawings. The embodiments described in the following exemplary embodiments do not represent all the implementations consistent with the present application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the present application as detailed in the appended claims.

[0049] The terms used in the present application are merely for the purpose of describing particular embodiments and are not intended to limit the present application. The singular forms "a," "an," and "the" used in the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0050] In order to make the technical solutions provided by the embodiments of the present application better understood by those skilled in the art, and to make the above-mentioned purposes, features and advantages of the embodiments of the present application more apparent and easy to understand, the technical solutions in the embodiments of the present application are further described in detail below with reference to the drawings.

[0051] In the construction of nuclear power plant safety monitoring system, key equipment generally adopts multi-sensor redundant configuration scheme to realize state perception. These sensors collect multiple physical parameters such as pressure, temperature and vibration through synchronous acquisition, and provide basic data support for equipment health evaluation and system parameter reconstruction. However, when a specific sensor deviates significantly from the actual working condition due to noise interference, zero drift or hardware failure, the abnormal data of the sensor will cause misjudgment risk of system state when participating in fusion operation.

[0052] Although the current redundant system design includes fault-tolerant mechanism, its effectiveness depends on the reliable implementation of two key technical links: first, the real-time identification ability of abnormal sensors, and second, the accurate positioning accuracy of corresponding fault nodes. In an ideal case, timely exclusion of fault node data can enable the redundant system to continue to maintain high reliability of state inversion function.

[0053] However, there are two technical bottlenecks in actual application: first, due to the inherent low sampling characteristics (usually ≤1Hz) of nuclear-level sensors, the frequency domain analysis method and dynamic signal processing technology currently used are difficult to effectively implement; second, with the expansion of the scale of the monitoring system, the complex nonlinear feature space caused by the coupling effect between sensors makes the detection accuracy of the threshold-based judgment mechanism or traditional machine learning classification model decrease exponentially.

[0054] Therefore, the embodiments of the present application disclose a sensor fault positioning method to solve the above technical problems.

[0055] As shown in the figure, Figure 1 Figure 1 is a flowchart of a sensor fault positioning method provided by the embodiments of the present application. The sensor fault positioning method is applied to a system comprising N sensors, and the N sensors have redundant information through regular arrangement.

[0056] Redundant information refers to the fact that multiple sensors in the N sensors observe the same physical quantity (or associated physical quantity), so that part of the data has overlapping or complementary relationship. The implementation of such redundancy includes but is not limited to spatial redundancy (multiple sensors covering the same monitoring area), functional redundancy (multiple sensors measuring different characteristics of the same target), etc.

[0057] In the embodiments of the present application, the N sensors have redundant information through regular arrangement. For example, the N sensors are arranged in a predetermined space, so that the data collected by the N sensors have redundancy, for example, the sensing areas of multiple sensors partially overlap, ensuring that the same target is detected by different sensors.

[0058] The sensor fault positioning method can include the following steps:

[0059] ​S101: In a case where it is determined that there is a faulty sensor in the N sensors, in each round of fault locating process, k different sensors are excluded from the N sensors each time, to obtain a plurality of first data groups of raw data of N-k sensors.

[0060] In a case where it is determined that there is a faulty sensor in the N sensors, one or more rounds of fault locating can be performed to locate the specific faulty sensor from the N sensors. In each round of fault locating process, k different sensors are excluded from the N sensors each time, to obtain a plurality of first data groups of raw data of N-k sensors.

[0061] wherein the initial value of k is 1, and the value of k can be updated in the manner described in S104 (the updating manner of k is not described here, see below for details).

[0062] For example, in a system comprising sensors 1, 2, …, 9, in a case where it is determined that there is a faulty sensor in the 9 sensors, a multi-round progressive fault locating strategy can be adopted. That is, in the first round of fault locating process, 1 different sensor is excluded from the 9 sensors each time, to obtain C(9, 8) = 9 first data groups of raw data of 8 sensors, respectively (2, 3, 4, 5, 6, 7, 8, 9), (1, 3, 4, 5, 6, 7, 8, 9), …, (1, 2, 3, 4, 5, 6, 7, 8). If the specific faulty sensor cannot be located according to the manner described in S102 to S104, k is increased by 1 and the next round of fault locating is performed, that is, in the second round of fault locating process, 2 different sensors are excluded from the 9 sensors each time, to obtain C(9, 7) = 36 first data groups of raw data of 7 sensors, respectively (3, 4, 5, 6, 7, 8, 9), (2, 4, 5, 6, 7, 8, 9), …, (1, 2, 3, 4, 5, 6, 7). This is repeated until the faulty sensor is located or k reaches the set threshold.

[0063] S102: For each first data group, one different sensor is excluded from the corresponding N-k sensors each time, to obtain N-k second data groups of raw data of N-(k+1) sensors, and the data of the corresponding N-k sensors is reconstructed according to each second data group respectively, to obtain first reconstructed data.

[0064] In each round of fault locating process, for each first data group, one different sensor is excluded from the corresponding N-k sensors of the first data group each time, to obtain N-k second data groups of raw data of N-(k+1) sensors.

[0065] For example, for the first data set (2, 3, 4, 5, 6, 7, 8, 9), one different sensor is excluded from the corresponding 8 sensors each time, so that C(8, 7) = 8 first data sets containing 7 sensors of original data can be obtained, which are (3, 4, 5, 6, 7, 8, 9), (2, 4, 5, 6, 7, 8, 9), …, (2, 3, 4, 5, 6, 7, 8) respectively.

[0066] For each second data set under each first data set, the data of the N-k sensors contained in the first data set is reconstructed according to the second data set, to obtain the first reconstructed data corresponding to the second data set. Finally, N-k groups of first reconstructed data can be obtained for each first data set. The method for reconstructing the sensor data in the embodiments of the present application is not specifically limited, for example, methods based on deep learning, mathematical interpolation, statistical learning, etc. can be used.

[0067] In some embodiments, the reconstruction of the data of the corresponding N-k sensors according to each second data set to obtain the first reconstructed data comprises:

[0068] The N-k second data sets are input into a pre-trained detection model, so that the detection model reconstructs the data of the corresponding N-k sensors according to each second data set to generate the first reconstructed data corresponding to each second data set.

[0069] In some embodiments, only normal training samples are used for training in the training stage of the detection model. The normal training sample refers to a sample set in which all N sensors are in a normal working state and the collected data is accurate and reliable.

[0070] It should be noted that the conventional fault positioning method is usually based on sample data when the sensor fails to position the fault, but in a complex system (such as a large number of sensors N), the combination of sensor failure conditions grows exponentially, and it is difficult to traverse all fault conditions in actual engineering, resulting in a low accuracy of sensor fault positioning. The present application is based on a detection model, which reconstructs the data of the corresponding N-k sensors using partial sensor original data, and further positions the sensor fault based on the reconstructed data. Since the detection model only needs to use normal training samples for training in the training stage, it does not need to rely on sensor failure data, which can not only reduce the data collection cost, but also improve the accuracy of sensor fault positioning.

[0071] In some embodiments, the detection model is optimized by minimizing a loss function including reconstruction error and regression error in the training stage.

[0072] That is, the loss function of the detection model includes two parts: reconstruction error and regression error. The weighted joint loss function can be used to simultaneously optimize the two tasks of reconstruction error and regression value. It should be noted that the research goal of sensor layout can be expressed as formula (1):

[0073] f:x→y (Formula 1)

[0074] In formula (1), f represents the regression model, which can learn the mapping relationship between sensor data and target values ​​from training data through supervised learning (such as neural network and support vector regression); x represents the input of the regression model f. In a system containing sensors 1, 2, ..., 9, x is a vector composed of the data of 9 sensors, x = [x1, x2, x3, x4, x5, x6, x7, x8, x9]; y represents the output of the regression model f, that is, the regression value, that is, the target value predicted by the regression model f (such as physical quantities such as temperature and pressure or system status).

[0075] The above detection model is specifically used for:

[0076] S1021: For each second data group, preprocess the second data group;

[0077] Exemplarily, the preprocessing steps include, but are not limited to, data cleaning, normalization / standardization, etc., which are used to normalize the input data to meet the requirements of the detection model and improve the stability and final performance (robustness) of the detection model.

[0078] S1022: extracting features from the preprocessed data using a feature encoder;

[0079] The feature encoder captures a compact and meaningful feature vector (or feature tensor) of the information required for data reconstruction and regression prediction.

[0080] S1023: Predicting the regression value corresponding to the second data group through a regressor based on the extracted features, and reconstructing the data of the corresponding Nk sensors through a feature decoder to generate first reconstructed data corresponding to the second data group.

[0081] The output of the detection model includes the reconstructed data of the sensor and the predicted regression value. The features extracted by the feature encoder are applied to both reconstruction and regression tasks.

[0082] S103: Obtain a first reconstruction error corresponding to the first data group based on the obtained Nk groups of first reconstructed data and the corresponding original data of the Nk sensors.

[0083] In each round of fault locating, for each first data group, a first reconstruction error corresponding to the first data group is obtained according to the obtained N-k first reconstruction data and the original data of the N-k sensors contained in the first data group.

[0084] In some embodiments, a reconstruction error between each first reconstruction data and the original data of the corresponding N-k sensors can be calculated, and then the maximum value of all the calculated reconstruction errors is selected as the first reconstruction error.

[0085] S104: If there is a first data group whose corresponding first reconstruction error does not exceed the error threshold, the excluded k sensors in the first data group are determined as fault sensors; otherwise, k is increased by 1 and the next round of fault locating is performed until the fault sensors are located or k reaches a set threshold.

[0086] In each round of fault locating, the first reconstruction error corresponding to each first data group is compared with a pre-set error threshold. If the first reconstruction error corresponding to a certain first data group exceeds the error threshold, it means that there is a fault sensor in the N-k sensors contained in the first data group; if the first reconstruction error corresponding to a certain first data group does not exceed the error threshold, it means that all the N-k sensors contained in the first data group are normal sensors.

[0087] Therefore, in the embodiments of the present application, in each round of fault locating, if there is a first data group whose corresponding first reconstruction error does not exceed the error threshold among the multiple first data groups, it means that all the N-k sensors contained in the first data group are normal sensors, that is, it can be determined that the k sensors excluded by the first data group in the N sensors are fault sensors; if there is no first data group whose corresponding first reconstruction error does not exceed the error threshold among the multiple first data groups, it means that the number of fault sensors is not k, then k is increased by 1 within a set threshold range, and the next round of fault locating is performed in the manner described in S101 to S104, and so on, until the specific fault sensors are located after multiple rounds of fault locating or k reaches the set threshold.

[0088] In some embodiments, the error threshold described above can be determined by the following steps, specifically including:

[0089] S1041: Obtain multiple groups of sample data, wherein each group of sample data contains normal data of N sensors;

[0090] Multiple groups of sample data for calculating the error threshold are obtained. These sample data come from the case where all N sensors work normally and the data is accurate and reliable. It should be noted that the error threshold calculated based on the normal data can reflect the error range of the "reconstruction ability" of the detection model in the ideal case.

[0091] S1042: generating, by using the detection model, third reconstruction data corresponding to the ith set of sample data after excluding the jth sensor;

[0092] For each set of sample data (a total of M sets), the jth (1≤j≤N) sensor of the N sensors is excluded to simulate the case where the jth (1≤j≤N) sensor of the N sensors fails, thereby forming a new "input sample data". Each "input sample data" actually only contains data of N-1 sensors.

[0093] Input all "input sample data" into the detection model to enable the detection model to reconstruct data of the N sensors according to each "input sample data" respectively, and generate corresponding third reconstruction data R ij , where i=1, 2, …, M, and j=1, 2, …, N.

[0094] S1043: calculating the mean and the standard deviation of all third reconstruction data, and determining the error threshold according to the mean and the standard deviation.

[0095] In some embodiments, the calculation formula of the error threshold can be as shown in formula (2):

[0096] TH=mean(R ij )+3*std(R ij ) (formula 2)

[0097] In formula (2), TH represents the error threshold; mean(R ij ) represents the mean of the third reconstruction data R ij , that is, the average level of the third reconstruction data R ij ; std(R ij ) represents the standard deviation of the third reconstruction data R ij , that is, the dispersion degree of the third reconstruction data R ij around the mean. In formula (2), the coefficient before std(R ij ) is 3. It should be noted that the coefficient before std(R ij ) can be set according to actual needs, including but not limited to 2, 3, etc. It is calculated that most of the third reconstruction errors of the sample data will fall within the range of the error threshold calculated according to formula (2), and therefore, any reconstruction error exceeding this error threshold is considered to be abnormal and is caused by the presence of a faulty sensor.

[0098] In some embodiments, the sensor layout research target can be shown in equation (1), where y is the regression value to be predicted. The detection model mentioned above is a dual-task model, whose output includes reconstruction error and regression value y. In this framework, the determination logic of the set threshold can include:

[0099] Based on the detection model, determine the maximum k value that can still predict the correct regression value y using N-(k+1) sensor data when the system is excluded k sensors; determine the maximum k value as the set threshold.

[0100] In embodiments of the present application, by excluding k different sensors from N sensors each time in each round of fault localization process, a plurality of first data groups containing N-k sensor raw data are obtained; for each first data group, based on the redundant information between the sensors, the data of the corresponding N-k sensors is reconstructed through the raw data of part of the sensors; according to the raw data and the reconstructed data of the corresponding N-k sensors, the first reconstruction error corresponding to the first data group is calculated; if there is a first data group in the current round of fault localization process whose first reconstruction error does not exceed the error threshold, the k sensors excluded in the first data group are determined as the faulty sensors; otherwise, k is increased by 1 and the next round of fault localization is performed until the faulty sensors are located or k reaches the set threshold, so as to improve the accuracy of sensor fault localization. In addition, since the present application does not need to rely on high-frequency time series analysis, it can be applied to the fault localization scene of low sampling rate nuclear level sensors.

[0101] In some embodiments, for "in the case where it is determined that there is a faulty sensor in the N sensors" described in S101, the determination that there is a faulty sensor in the N sensors can be made by the following steps, specifically including:

[0102] S1021: excluding one different sensor from the N sensors each time to obtain N third data groups containing N-1 sensor raw data;

[0103] For example, in a system containing sensors 1, 2, …, 9, one different sensor is excluded from the 9 sensors each time, so as to obtain C(9, 8) = 9 third data groups containing 8 sensor raw data, which are (2, 3, 4, 5, 6, 7, 8, 9), (1, 3, 4, 5, 6, 7, 8, 9), …, (1, 2, 3, 4, 5, 6, 7, 8) respectively.

[0104] S1022: reconstructing the data of the N sensors according to each third data group respectively to obtain second reconstruction data;

[0105] For each third data group, the data of N sensors are reconstructed according to the third data group to obtain the second reconstructed data corresponding to the third data group, and finally N groups of second reconstructed data can be obtained.

[0106] In some embodiments, the detection model described in the embodiment of S102 may be used to obtain the second reconstructed data, and the specific steps are not repeated here.

[0107] S1013: Obtain a second reconstruction error based on the obtained N groups of second reconstructed data and the original data of the N sensors;

[0108] In some embodiments, a reconstruction error between each second reconstructed data and the original data of N sensors may be calculated, and then a maximum value may be selected from all calculated reconstruction errors as the second reconstruction error.

[0109] S1014: If the second reconstruction error exceeds the error threshold, it is determined that there is a faulty sensor among the N sensors.

[0110] The error threshold may be determined in the manner described in S1041 to S1043. If the obtained second reconstruction error exceeds the error threshold, it is determined that a faulty sensor exists among the N sensors.

[0111] A sensor fault location method provided in an embodiment of the present application is described below in detail. The sensor fault location method is applied to a system including N sensors, where the N sensors have redundant information due to regular arrangement.

[0112] The research objective of sensor layout can be shown as formula (1) above.

[0113] Here, x = [x1, x2, x3, x4, x5, x6, x7, x8, x9], and y is the regression value. Typically, during training, f is obtained. During testing, the predicted value y′ for the test sample x′ is y′ = f(x′). However, if x′ includes the value of a faulty sensor, the predicted y′ is unreliable, requiring anomaly detection.

[0114] like Figure 2 As shown, Figure 2 This is a flow chart of sensor anomaly detection provided by an embodiment of the present application.

[0115] Assuming N=9, raw data of 9 sensors (ie, all sensor samples) are acquired, and a different sensor is excluded from the 9 sensors each time to obtain 9 third data groups containing raw data of 8 sensors (ie, partial sensors).

[0116] The detection model is named Less-sensor To More-sensor for Abnormal Detection, L2M-AD. The loss function of the detection model includes two parts of reconstruction error and regression error, and the two tasks of reconstruction error and regression value can be optimized synchronously through a weighted joint loss function. The input of the detection model is partial sensor values, and the output includes all sensor values. The purpose is to reconstruct all sensor values by using partial normal sensor values under the premise that there is redundancy in sensors. The output also includes regression values, that is, to ensure that the output features of the encoder contain all information, reconstruct all sensor values, and require the regression target to be minimized at the same time as data reconstruction.

[0117] The detection model is named Less-sensor To More-sensor for Abnormal Detection, L2M-AD. The loss function of the detection model includes two parts of reconstruction error and regression error, and the two tasks of reconstruction error and regression value can be optimized synchronously through a weighted joint loss function. The input of the detection model is partial sensor values, and the output includes all sensor values. The purpose is to reconstruct all sensor values by using partial normal sensor values under the premise that there is redundancy in sensors. The output also includes regression values, that is, to ensure that the output features of the encoder contain all information, reconstruct all sensor values, and require the regression target to be minimized at the same time as data reconstruction.

[0118] The reconstruction error between each second reconstruction data and the original data of the nine sensors is calculated, and the maximum value is selected as the second reconstruction error from all the calculated reconstruction errors.

[0119] In the test process, the results in Table 1 are obtained. Therefore, if the reconstruction error exceeds the error threshold, it indicates that there is a faulty sensor.

[0120] Table 1: Relationship between reconstruction error and sensor state in the test process

[0121] System state Partial sensor data state Full sensor data state Reconstruction error Normal Normal Normal Small Fault Normal Fault Large Fault Fault Fault Large

[0122] In the case where it is determined that there is a faulty sensor in the nine sensors, a multi-round progressive fault location strategy can be used. As shown in Figure 3 , the flowchart of a sensor fault location provided by an embodiment of the present application is shown in Figure 3 .

[0123] The initial value of k is 1, and in the first round of fault location process, the following steps are performed:

[0124] Each time, one different sensor is excluded from the nine sensors to obtain C(9,8) = 9 first data groups of original data of 8 sensors.

[0125] For each first data group, one different sensor is excluded from the 8 sensors corresponding to the first data group each time to obtain 8 second data groups of original data of 7 sensors.

[0126] The eight second data groups under the first data group are input into a detection model. The detection model reconstructs the data of the eight sensors contained in the first data group according to each second data group, obtains first reconstruction data corresponding to the second data group, and finally the first data group can obtain eight groups of first reconstruction data.

[0127] The reconstruction error between each first reconstruction data and the original data of the eight sensors corresponding to the first data group is calculated, and the maximum value is selected as the first reconstruction error corresponding to the first data group from all the calculated reconstruction errors.

[0128] If there is a first data group corresponding to the first reconstruction error that does not exceed the error threshold among the nine first data groups, the excluded sensor in the first data group is determined as the faulty sensor, and the process ends.

[0129] If the first reconstruction errors corresponding to the nine first data groups all exceed the error threshold, it means that the number of faulty sensors is not one, then k is increased by one. Since the updated k does not reach the set threshold (assuming the set threshold is 2), the second round of fault localization is performed.

[0130] In the second round of fault localization, the following steps are performed:

[0131] Each time, two different sensors are excluded from the nine sensors to obtain C(9,7)=36 first data groups containing original data of seven sensors.

[0132] For each first data group, one different sensor is excluded from the seven sensors corresponding to the first data group to obtain seven second data groups containing original data of six sensors.

[0133] The seven second data groups under the first data group are input into a detection model. The detection model reconstructs the data of the seven sensors contained in the first data group according to each second data group, obtains first reconstruction data corresponding to the second data group, and finally the first data group can obtain seven groups of first reconstruction data.

[0134] The reconstruction error between each first reconstruction data and the original data of the seven sensors corresponding to the first data group is calculated, and the maximum value is selected as the first reconstruction error corresponding to the first data group from all the calculated reconstruction errors.

[0135] If there is a first data group corresponding to the first reconstruction error that does not exceed the error threshold among the 36 first data groups, the two excluded sensors in the first data group are determined as the faulty sensors, and the process ends.

[0136] If the first reconstruction errors corresponding to the nine first data groups all exceed the error threshold, it indicates that the number of faulty sensors is not two, and k is added by one. Since the updated k reaches the set threshold, the next round of fault positioning is not performed, and the process ends.

[0137] In this embodiment, in each round of fault positioning process, k different sensors are excluded from the N sensors each time to obtain a plurality of first data groups of raw data of N-k sensors; for each first data group, the data of the corresponding N-k sensors is reconstructed based on the redundancy information between the sensors through the raw data of part of the sensors; the first reconstruction error corresponding to the first data group is calculated according to the raw data and the reconstructed data of the corresponding N-k sensors; if there is a first data group whose first reconstruction error does not exceed the error threshold in the current round of fault positioning process, the k sensors excluded in the first data group are determined as faulty sensors; otherwise, k is added by one and the next round of fault positioning is performed until the faulty sensors are located or k reaches the set threshold. By converting the fault positioning into an anomaly detection problem, the accuracy of sensor fault positioning can be improved. In addition, since the present application does not depend on high-frequency time series analysis, it can be applied to the fault positioning scene of low sampling rate nuclear level sensors.

[0138] The above describes the method provided by the present application. The device provided by the present application is described below:

[0139] Please refer to Figure 4 , a structural schematic diagram of a sensor fault positioning device provided by an embodiment of the present application. The sensor fault positioning device is applied to a system comprising N sensors, and the N sensors have redundancy information through regular arrangement.

[0140] As Figure 4 indicated, the device can include:

[0141] The exclusion unit 410 is configured to, in a case where it is determined that there is a faulty sensor in the N sensors, exclude k different sensors from the N sensors each time in each round of fault positioning process to obtain a plurality of first data groups of raw data of N-k sensors, where the initial value of k is 1.

[0142] The reconstruction unit 420 is configured to, for each first data group, exclude one different sensor from the corresponding N-k sensors each time to obtain a plurality of second data groups of raw data of N-(k+1) sensors, and reconstruct the data of the corresponding N-k sensors according to each second data group to obtain first reconstruction data.

[0143] The error calculation unit 430 is configured to obtain first reconstruction errors corresponding to the first data groups according to the obtained N-k groups of first reconstruction data and the original data of the corresponding N-k sensors.

[0144] The fault positioning unit 440 is configured to determine the excluded k sensors in the first data group as the faulty sensors if there is a first data group corresponding to the first reconstruction error that does not exceed the error threshold; otherwise, add 1 to k and perform the next round of fault positioning until the faulty sensors are located or k reaches a set threshold.

[0145] Optionally, the apparatus further comprises an anomaly detection unit, which is specifically configured to:

[0146] obtain N third data groups each containing original data of N-1 sensors by excluding one different sensor from the N sensors each time;

[0147] reconstruct the data of the N sensors according to each third data group respectively to obtain second reconstruction data;

[0148] obtain second reconstruction errors according to the obtained N groups of second reconstruction data and the original data of the N sensors;

[0149] If the second reconstruction errors exceed the error threshold, it is determined that there is a faulty sensor in the N sensors.

[0150] The functions and effects of the units in the apparatus are specifically described in the implementation process of the corresponding steps in the above method, and will not be described here.

[0151] The embodiments of the present application further provide a hardware structure. Referring to Figure 5 , Figure 5 The electronic device structure provided by the embodiments of the present application. As Figure 5 shown, the hardware structure can include a processor and a machine readable storage medium, the machine readable storage medium stores machine executable instructions that can be executed by the processor; the processor is used to execute the machine executable instructions to realize the method disclosed in the above examples of the present application.

[0152] Based on the same application concept as the above method, the embodiments of the present application further provide a machine readable storage medium, the machine readable storage medium stores a plurality of computer instructions, and the computer instructions are executed by a processor to realize the method disclosed in the above examples of the present application.

[0153] Exemplarily, the machine-readable storage medium described above can be any electronic, magnetic, optical, or other physical storage apparatus, and can contain or store information such as executable instructions, data, and the like. For example, the machine-readable storage medium can be a Random Access Memory (RAM), a volatile memory, a non-volatile memory, a flash memory, a storage drive (e.g., a hard drive), a solid-state drive, any type of storage disc (e.g., an optical disc, a DVD, and the like), or similar storage media, or a combination thereof.

[0154] It should be noted that, in the present document, relational terms such as "first" and "second", and the like, are used solely to distinguish one entity or action from another entity or action, without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. In other words, any of the elements listed in the description of the above-mentioned process, method, article, or apparatus can be present or omitted, unless otherwise specified.

[0155] The above description is merely illustrative of the application, and not restrictive. Any modification, equivalent replacement, improvement, and the like made within the spirit and principle of the application shall be included in the scope of the application.

Claims

1. A sensor fault location method, characterized in that: Applied to a system comprising N sensors, wherein the N sensors have redundant information due to regular arrangement, the method comprises: When it is determined that a faulty sensor exists among the N sensors, in each round of fault location, k different sensors are excluded from the N sensors each time to obtain a plurality of first data groups including raw data of Nk sensors, where an initial value of k is 1; For each first data group, exclude one different sensor from the corresponding Nk sensors at a time to obtain Nk second data groups containing raw data of N-(k+1) sensors, and reconstruct the data of the corresponding Nk sensors based on each second data group to obtain first reconstructed data; Obtaining a first reconstruction error corresponding to the first data group based on the obtained Nk groups of first reconstructed data and the corresponding original data of the Nk sensors; If there is a first data group whose corresponding first reconstruction error does not exceed the error threshold, the k sensors excluded from the first data group are determined to be faulty sensors; otherwise, k is incremented by 1 and the next round of fault location is performed until the faulty sensor is located or k reaches the set threshold.

2. The method according to claim 1, characterized in that The method further comprises: excluding a different sensor from the N sensors each time, to obtain N third data groups containing raw data of N-1 sensors; reconstructing the data of the N sensors according to each third data group to obtain second reconstructed data; Obtaining a second reconstruction error based on the obtained N groups of second reconstructed data and the original data of the N sensors; If the second reconstruction error exceeds the error threshold, it is determined that there is a faulty sensor among the N sensors.

3. The method according to claim 1, characterized in that Reconstructing the data of the corresponding Nk sensors according to each second data group to obtain first reconstructed data includes: The Nk second data groups are input into a pre-trained detection model, so that the detection model reconstructs the data of the corresponding Nk sensors according to each second data group, and generates first reconstructed data corresponding to each second data group.

4. The method according to claim 3, characterized in that The detection model is optimized during the training phase by minimizing a loss function including reconstruction error and regression error. The detection model is specifically used to: For each second data group, preprocessing the second data group; Perform feature extraction on the preprocessed data through the feature encoder; According to the extracted features, the regression value corresponding to the second data group is predicted by a regressor, and the data of the corresponding Nk sensors are reconstructed by a feature decoder to generate first reconstructed data corresponding to the second data group.

5. The method according to claim 3, characterized in that The method further comprises: Acquire multiple sets of sample data, where each set of sample data contains normal data of N sensors; Using the detection model, generating third reconstructed data corresponding to the i-th group of sample data after excluding the j-th sensor; The mean and standard deviation of all third reconstructed data are calculated, and the error threshold is determined according to the mean and the standard deviation.

6. The method according to claim 1, characterized in that Obtaining a first reconstruction error corresponding to the first data group based on the obtained Nk groups of first reconstructed data and the corresponding original data of the Nk sensors includes: Calculating a reconstruction error between each first reconstructed data and the original data of the corresponding Nk sensors; From all calculated reconstruction errors, the maximum value is selected as the first reconstruction error.

7. A sensor fault location device, characterized in that: Applied to a system comprising N sensors, wherein the N sensors have redundant information due to regular arrangement, the apparatus comprises: an excluding unit configured to, when determining that a faulty sensor exists among the N sensors, exclude k different sensors from the N sensors in each round of fault location, thereby obtaining a plurality of first data groups including raw data of Nk sensors, where an initial value of k is 1; a reconstruction unit configured to exclude, for each first data group, a different sensor from the corresponding Nk sensors at a time, to obtain Nk second data groups containing raw data of N-(k+1) sensors, and reconstruct the data of the corresponding Nk sensors based on each second data group to obtain first reconstructed data; an error calculation unit, configured to obtain a first reconstruction error corresponding to the first data group based on the obtained Nk groups of first reconstructed data and the corresponding original data of the Nk sensors; a fault location unit configured to determine, if there is a first data group whose corresponding first reconstruction error does not exceed the error threshold, k sensors excluded from the first data group as faulty sensors; otherwise, increment k by 1 and execute the next round of fault location until the faulty sensor is located or k reaches a set threshold.

8. The device according to claim 7, characterized in that The device further includes an anomaly detection unit, which is specifically configured to: excluding a different sensor from the N sensors each time, to obtain N third data groups containing raw data of N-1 sensors; reconstructing the data of the N sensors according to each third data group to obtain second reconstructed data; Obtaining a second reconstruction error based on the obtained N groups of second reconstructed data and the original data of the N sensors; If the second reconstruction error exceeds the error threshold, it is determined that there is a faulty sensor among the N sensors.

9. An electronic device, characterized in that: The system comprises a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor is configured to execute the machine-executable instructions to implement the method according to any one of claims 1 to 6.

10. A machine-readable storage medium, characterized in that The machine-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images