Sensor screening method and device, sensor data reconstruction method and system

Abstract

The present invention discloses a method for selecting sensors based on sensor data reconstruction technology, comprising: selecting a target sensor set from candidate sensors according to sensor data feasibility, and / or sensor failure probability, and / or sensor importance. Furthermore, a sensor selection device, a sensor data reconstruction method and a system based on sensor data reconstruction technology are also provided. The method can select a reasonable sensor application range that requires data reconstruction to meet the limited computing resources of the data computing and analysis system and the safety of the nuclear power plant.

Description

Technical Field

[0001] The present invention belongs to the technical field of nuclear power equipment detection, and in particular relates to a sensor screening method and device based on sensor data reconstruction technology, and a sensor data reconstruction method and system. Background Art

[0002] Sensor data reconstruction technology refers to the online intelligent monitoring of nuclear power plant sensors and the replacement of faulty sensor signals, so that nuclear power units have the ability to automatically recover after sensor failure within a certain period of time.

[0003] Sensor data reconstruction technology plays an important role in reducing the probability of unplanned shutdowns and the cost of unit construction and operation and maintenance. However, there are many sensors in nuclear power plants, and the support and influence of each sensor on the operation of the unit vary. In addition, the computing resources of the data calculation and analysis system are limited. Therefore, how to reasonably determine the range of sensors that need data reconstruction technology in nuclear power plants is an urgent problem to be solved. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a sensor screening method and device, a sensor data reconstruction method and system based on sensor data reconstruction technology in view of the above-mentioned deficiencies in the prior art, which can screen out a reasonable sensor application range that requires data reconstruction to meet the limited computing resources of the data calculation and analysis system and the safety of the nuclear power plant.

[0005] In a first aspect, an embodiment of the present invention provides a method for screening sensors based on sensor data reconstruction technology, including: screening a target sensor set from candidate sensors according to sensor data feasibility, and / or sensor failure probability, and / or sensor importance.

[0006] Preferably, screening out a target sensor set from candidate sensors based on sensor data feasibility, and / or sensor failure probability, and / or sensor importance includes: screening out a first sensor set from candidate sensors within a default value range based on sensor data feasibility; selecting sensors having a failure probability greater than a failure threshold from the first sensor set to obtain a second sensor set; and screening out the target sensor set from the second sensor set based on sensor importance.

[0007] Preferably, the method of filtering out the first sensor set from the alternative sensors within the default value range according to the feasibility of sensor data includes: filtering out sensors connected to the distributed control system from the alternative sensors within the default value range to obtain a third sensor set; judging whether the response time of the sensors in the third sensor set is less than a first preset threshold, and filtering out the sensors whose judgment result is that the response time is less than the first preset threshold, to obtain a fourth sensor set; judging whether the packet loss rate of data collection of the sensors in the fourth sensor set is less than a second preset threshold, and filtering out the sensors whose judgment result is that the packet loss rate is less than the second preset threshold, to obtain the first sensor set.

[0008] Preferably, the screening out of sensors whose judgment results are that the packet loss rate is less than the second preset threshold value to obtain the first sensor set includes: screening out sensors whose judgment results are that the packet loss rate is less than the second preset threshold value from the fourth sensor set to obtain the fifth sensor set; screening out sensors whose measurement objects are any one of temperature, pressure, flow, liquid level, concentration, and neutron flux from the fifth sensor set to obtain the sixth sensor set; judging whether the sensors in the sixth sensor set can collect historical operating data above the preset operating conditions and the preset acquisition frequency, and the amount of historical operating data is greater than the third preset threshold value, and screening out sensors whose judgment results are that they can collect historical operating data above the preset operating conditions and the preset acquisition frequency, and the amount of historical operating data is greater than the third preset threshold value, thereby obtaining the first sensor set.

[0009] Preferably, selecting sensors with failure probability greater than a failure threshold from the first sensor set to obtain a second sensor set specifically includes: classifying the sensors in the first sensor set according to a measurement principle to obtain W i Type sensor, 1≤i≤14; Get W i Type sensor has N failure modes and corresponding failure probabilities of N failure modes, where N is a positive integer; summing the corresponding failure probabilities of N failure modes, we get W i The total failure probability of the type sensor; select W whose total failure probability is greater than the failure threshold i Type sensor and get W from the first sensor set i The sensors corresponding to the type sensor are used to obtain a second sensor set.

[0010] Preferably, the obtaining W i N failure modes of sensors of the same type and the corresponding failure probabilities of the N failure modes, specifically including: collecting W in the first sensor set i The historical operation data of the sensor corresponding to the type sensor; the collected W iThe historical operation data of the type sensor is frequency thinned; the historical operation data after frequency thinning is cut according to the preset time length to obtain W i Sample set C of sensor type i ; From the sample set C i Get W i Failure sample set F of sensor type i ; According to N failure modes, W i Failure sample set F of sensor type i All subsamples F ik Classification, get W i The N failure modes of the type sensor include missing value, stuck value, sudden value change and drift value. Calculate W i The failure probability of N failure modes of type sensor is W i The failure probabilities corresponding to the N failure modes of the type sensor.

[0011] Preferably, the sample set C i Get W i Failure sample set F of sensor type i , specifically including: S1, from the sample set C i Select M subsamples C of zero-failure data from ij As training samples, the first state monitoring model is obtained, j = 1, 2, ..., n, M < n, and the state monitoring model is used to determine whether the sample is a failure sample; S2, all sub-samples C ij Input into the first state monitoring model to obtain all subsamples C ij S3, filter out the sub-sample C corresponding to the judgment result of failure or not ij , get the sample set Z i ; S4, the sample set Z i As training samples, train the first state monitoring model to obtain the second state monitoring model, and clear the sample set Z i ; S5, iterate S2-S4 h times to obtain the h+2th state monitoring model, where h is a positive integer; S6, all subsamples C ij Input to the h+2th state monitoring model to obtain all subsamples C ij S7, filter out the subsample C corresponding to the judgment result of failure ij , we get W i Failure sample set F of sensor type i .

[0012] Preferably, the method of calculating W according to N failure modes i Failure sample set F of sensor typei All samples F ik Classification, get W i N failure modes of type sensor, including: calculation of W i Failure sample set F of type sensor i All subsamples F ik The mean, variance, and average of the slopes of two adjacent data points; if the mean is less than the fourth preset threshold, then the subsample F ik The failure mode is missing values; if the variance is less than the fifth preset threshold, then the subsample F ik The failure mode is numerical stagnation; if the average value of the slopes of two adjacent data points is greater than the sixth preset threshold, then the subsample F ik The failure mode is value mutation; otherwise, the subsample F ik The failure mode is numerical drift; the statistical failure sample set F i The subsamples F corresponding to the N failure modes in ik The number of W i Type sensor has N failure modes. The calculation W i The failure probability of N failure modes of type sensor is W i The corresponding failure probabilities of the N failure modes of the type sensor include: i The subsamples F corresponding to the N failure modes in ik The number and sample set C i Neutron Sample C ij The failure probability of each of the N failure modes is calculated based on the ratio of the total number of failure modes.

[0013] Preferably, the method of screening out the target sensor set from the second sensor set according to the importance of sensors specifically includes: obtaining important control functions of each sensor in the second sensor set to construct a first mapping table, the first mapping table including sensor names, important control functions, and a mapping relationship between the two; obtaining the consequences of each important control function in the first mapping table in the event of failure based on engineering experience, the consequences including serious consequences and general consequences; screening out sensors in the first mapping table corresponding to the important control functions that produce serious consequences in the event of failure, to obtain the target sensor set.

[0014] In the second aspect, an embodiment of the present invention also provides a sensor data reconstruction method, which uses the screening method described in the first aspect to screen out a target sensor set, and monitors the signals of all sensors in the target sensor set online; when a fault signal occurs in any sensor in the target sensor set, the fault signal is replaced so that the nuclear power unit has the ability to automatically recover after a sensor failure within a preset time.

[0015] In a third aspect, an embodiment of the present invention also provides a sensor screening device based on sensor data reconstruction technology, including a screening module. The screening module is used to screen out a target sensor set from candidate sensors based on sensor data feasibility, and / or sensor failure probability, and / or sensor importance. The screening module includes a first screening unit, a second screening unit, and a third screening unit. The first screening unit is used to screen out a first sensor set from candidate sensors within a default value range based on sensor data feasibility; the second screening unit is connected to the first screening unit and is used to select sensors with a failure probability greater than a failure threshold from the first sensor set to obtain a second sensor set; the third screening unit is connected to the second screening unit and is used to screen out the target sensor set from the second sensor set based on sensor importance.

[0016] Preferably, the second screening unit comprises a classification component, an acquisition component, a calculation component and a screening component. The classification component is connected to the first screening unit and is used to classify the sensors in the first sensor set according to the measurement principle to obtain W i Type sensor, 1≤i≤14. Acquisition component, connected with classification component, used to obtain W i The N failure modes of the type sensor and the corresponding failure probabilities of the N failure modes, N is a positive integer. The calculation component is connected to the acquisition component to sum the corresponding failure probabilities of the N failure modes to obtain W i The total failure probability of the type sensor. The screening component is connected to the calculation component and the first screening unit, and is used to select W whose total failure probability is greater than the failure threshold. i Type sensor and get W from the first sensor set i The sensors corresponding to the type sensor are used to obtain a second sensor set.

[0017] In a fourth aspect, an embodiment of the present invention further provides a sensor data reconstruction system, comprising the sensor screening device based on the sensor data reconstruction technology described in the third aspect, as well as a monitoring module and a replacement module. The sensor screening device based on the sensor data reconstruction technology is used to screen out a target sensor set. The monitoring module is connected to the sensor screening device based on the sensor data reconstruction technology, and is used to perform online monitoring of the target sensor set screened by the sensor screening device based on the sensor data reconstruction technology and output monitoring results. The replacement module is connected to the monitoring module, and is used to replace the fault signal when a fault signal appears in the monitoring result, so that the nuclear power unit has the ability to automatically recover after a sensor failure within a preset time.

[0018] The sensor screening method and device based on the sensor data reconstruction technology, the sensor data reconstruction method and system of the present invention select a target sensor set from the candidate sensors according to the sensor data feasibility, and / or the sensor failure probability, and / or the sensor importance, as the sensor set within the application scope of the data reconstruction technology. The application of the data reconstruction technology to the sensor set selected according to the above conditions can meet the limited computing resources of the data computing and analysis system and the safety of the nuclear power plant. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 : is a flow chart of a sensor screening method based on sensor data reconstruction technology according to Embodiment 1 of the present invention;

[0020] Figure 2 : is a structural diagram of a sensor screening device based on sensor data reconstruction technology according to embodiment 3 of the present invention;

[0021] Figure 3 : is a structural diagram of a second screening unit of a sensor screening device based on sensor data reconstruction technology according to Embodiment 3 of the present invention;

[0022] Figure 4 : A structural diagram of a sensor data reconstruction system according to embodiment 4 of the present invention. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0024] Embodiment 1:

[0025] like Figure 1 As shown, this embodiment provides a method for selecting sensors based on sensor data reconstruction technology, including: selecting a target sensor set from candidate sensors according to sensor data feasibility, and / or sensor failure probability, and / or sensor importance. The order of selecting sensors according to any of the three screening conditions is not unique, and the selected target sensor set can be applied to the sensor data reconstruction technology to meet the limited computing resources of the data computing and analysis system and the safety of the nuclear power plant.

[0026] Optionally, selecting a target sensor set from candidate sensors according to sensor data feasibility, and / or sensor failure probability, and / or sensor importance includes:

[0027] Step 101 : Screening out a first sensor set from candidate sensors within a default value range according to sensor data feasibility.

[0028] Optionally, step 101 specifically includes steps S11-S13:

[0029] Step S11, selecting sensors connected to the distributed control system from candidate sensors within a default value range to obtain a third sensor set.

[0030] Step S12, determining whether the response time of the sensors in the third sensor set is less than the first preset threshold, and selecting sensors whose response time is less than the first preset threshold to obtain a fourth sensor set.

[0031] Step S13, determining whether the packet loss rate of data collection of the sensors in the fourth sensor set is less than the second preset threshold, and screening out sensors whose packet loss rate is less than the second preset threshold to obtain the first sensor set.

[0032] Optionally, step S13 specifically includes steps S131-S133:

[0033] Step S131 , selecting sensors whose judgment results show that the packet loss rate is less than the second preset threshold from the fourth sensor set, to obtain a fifth sensor set.

[0034] Step S132, selecting sensors whose measuring objects are any one of temperature, pressure, flow, liquid level, concentration, and neutron flux from the fifth sensor set, to obtain a sixth sensor set.

[0035] Step S133, determine whether the sensors in the sixth sensor set can collect historical operating data above the preset operating conditions and the preset collection frequency, and the amount of historical operating data is greater than the third preset threshold, and filter out sensors whose judgment results are that they can collect historical operating data above the preset operating conditions and the preset collection frequency, and the amount of historical operating data is greater than the third preset threshold, thereby obtaining the first sensor set.

[0036] In this embodiment, the RCV system (chemical and volumetric system) is taken as an example for explanation. According to the nuclear power engineering document, sensors within the original default value range of the RCV system are obtained as candidate sensors, totaling 34 sensors, as shown in Table 1:

[0037] Table 1

[0038]

[0039]

[0040] The sensor data feasibility of the 34 sensors in Table 1 is analyzed, where the sensor data feasibility includes the sensor data transmission availability and the sensor operation data availability.

[0041] First, we analyze whether all 34 sensors in Table 1 meet the sensor data transmission availability (i.e., signal transmission requirements). The judgment conditions are as follows: (1) Whether the sensor has been connected to the nuclear power plant DCS system; (2) Whether the response time from sensor data acquisition to output is less than the threshold value X 1 ; (3) Is the packet loss rate of sensor data collection less than the threshold value X? 2 .

[0042] If a sensor cannot meet the signal transmission requirements, the sensor is removed from the candidate sensors; if a sensor meets the above three judgment conditions at the same time, it continues to be retained in the candidate sensors. In this embodiment, after analysis, it is found that all 34 sensors in Table 1 are connected to the DCS system, and the response time of each sensor is less than the threshold value X 1 , the packet loss rate of each sensor data collection is less than the threshold X 2 , so all 34 sensors are retained in the candidate sensors.

[0043] Secondly, the availability of sensor operation data (i.e., whether the amount of historical operation data is sufficient) is analyzed. The historical operation data includes normal operation data and abnormal operation data. The judgment conditions are as follows: (1) Whether the object measured by the sensor is the slowly varying data with inertial characteristics of the fluid process system of the nuclear power plant, including: temperature, pressure, flow, liquid level, concentration, and neutron flux; (2) Whether the number of data points of normal historical data and abnormal historical data collected by the sensor for working conditions above 20% Pn and with a collection frequency of once per second or more is greater than a given threshold value X 3 .

[0044] If a sensor cannot meet the sensor operation data availability requirement, the sensor is removed from the candidate sensors; if a sensor meets both of the above judgment conditions, it continues to be retained in the candidate sensors. In this embodiment, after analysis, the 34 sensors in Table 1 are classified into four categories: temperature, pressure, flow, and liquid level, and the 34 sensors can collect historical operation data with a collection frequency of once per second under working conditions above 20% Pn, and the number of historical operation data is greater than the threshold value X 3 , so all 34 sensors in Table 1 are retained in the candidate sensors, so that the first sensor set includes the 34 sensors in Table 1. It should be noted that since the sensor data reconstruction technology can only be applied after the reactor is started to 20% full power, the preset working condition in this implementation is set to 20% Pn (20% full power). In addition, the acquisition frequency of the DCS in this embodiment is once per second. If the data sampling frequency is too low, it will affect the accuracy of the analysis results. Therefore, the preset acquisition frequency in this implementation is set to once per second.

[0045] Step 102: Select sensors whose failure probability is greater than a failure threshold from the first sensor set to obtain a second sensor set.

[0046] Optionally, step 102 specifically includes steps S21-S24:

[0047] Step S21: classify the sensors in the first sensor set according to the measurement principle to obtain W i Type sensor, 1≤i≤14.

[0048] Usually, sensors are classified according to the type of parameter they measure, and are classified as Q d Class, d = 1, 2, ..., 6. As shown in Table 2, there are 6 types of measurement parameters, including: temperature, pressure, flow, liquid level, concentration, neutron flux, a total of 6 categories. For each parameter type, according to the measurement principle, there are 14 categories in total, see Table 2 below for details.

[0049] Table 2

[0050]

[0051] In this embodiment, the 34 sensors in the first sensor set are classified according to the measurement principle types shown in Table 2 to obtain W i The number of sensors is 4, namely thermocouple temperature sensor W 1 , Differential pressure sensor W 5 , Dynamic pressure flow sensor W 6 , differential pressure level sensor W 11 The classification results are shown in Table 3 below:

[0052] Table 3

[0053]

[0054]

[0055] Step S22, obtain W i N failure modes of type sensor and the corresponding failure probabilities of the N failure modes, where N is a positive integer.

[0056] Optionally, get W i The N failure modes of sensors of the type and the corresponding failure probabilities of the N failure modes specifically include: collecting W in the first sensor set i The historical operation data of the sensor corresponding to the type sensor; the collected W i The historical operation data of the type sensor is frequency thinned; the historical operation data after frequency thinning is cut according to the preset time length to obtain W i Sample set C of sensor typei ; From the sample set C i Get W i Failure sample set F of sensor type i ; According to N failure modes, W i Failure sample set F of sensor type i All subsamples F ik Classification, get W i The N failure modes of the type sensor include missing value, stuck value, sudden value change and drift value. Calculate W i The failure probability of N failure modes of type sensor is W i The failure probabilities corresponding to the N failure modes of the type sensor.

[0057] Among them, from the sample set C i Get W i Failure sample set F of sensor type i , specifically including:

[0058] S1, from the sample set C i Select M subsamples C of zero-failure data from ij As training samples, training is performed to obtain a first state monitoring model, where j = 1, 2, ..., n, and M < n. The state monitoring model is used to determine whether the sample is a failure sample;

[0059] S2, all subsamples C ij Input into the first state monitoring model to obtain all subsamples C ij The judgment result of whether the failure is or not;

[0060] S3, filter out the subsample C corresponding to the judgment result of failure being no ij , get the sample set Z i ;

[0061] S4, the sample set Z i As training samples, train the first state monitoring model to obtain the second state monitoring model, and clear the sample set Z i ;

[0062] S5, iteratively execute S2-S4 h times to obtain the h+2th state monitoring model, where h is a positive integer;

[0063] S6, all subsamples C ij Input to the h+2th state monitoring model to obtain all subsamples C ij The judgment result of whether the failure is or not;

[0064] S7, filter out the sub-sample C corresponding to the judgment result of failure being yes ij , we get Wi Failure sample set F of sensor type i .

[0065] In this embodiment, W is collected in sequence i The historical operation data of the sensors corresponding to the type sensors (i.e., the 34 sensors in the first sensor set) are frequency-thinned at a frequency of once per second, and 1 minute of data is taken as a sample. That is, the frequency-thinned data is cut according to the preset time length (1 minute) to obtain a sample set C containing 60 data points per sample. i For example, in this embodiment, 600,000 data points are obtained for each W-type sensor. 6 For example, all dynamic pressure and flow data collected by the RCV system from the DCS are collected once per second, so there is no need to perform frequency thinning. The collected historical operation data is cut to obtain the sample set C 6 is a 10000×60 data matrix, where subsample C 6j j=1,2,…,10000. 400 subsample data with no obvious failures are manually selected from the 10000×60 data matrix (i.e., a 400×60 data matrix), and substituted into the kernel regression classification model for training (the model outputs 0 for normal samples and 1 for failure samples), and the trained kernel regression classification model is obtained; then the 10000×60 data matrix is ​​substituted into the trained kernel regression classification model, and the state of each group of 1×60 is output (the output is a 10000×1 matrix; 0 represents a normal sample, and 1 represents a failure sample), as shown below:

[0066]

[0067] Subsequently, the normal samples output by the model in the 10000×60 data matrix are used as training samples to retrain the kernel regression classification model, and the final kernel regression classification model is obtained by iteration; the 10000×60 data matrix is ​​substituted into the final kernel regression classification model, and each group of 1×60 states is output (the output is a 10000×1 matrix; 0 represents a normal sample and 1 represents a failure sample); all 1×60 sub-samples that are output as failure samples are combined into a 3213×60 failure sample set F 6 (3213 is the number of failed sub-samples of dynamic pressure flow sensor), thus obtaining W 6 Failure sample set F of sensor type 6 In addition, W is obtained according to the above method. 1 , W 5 , W 11 The corresponding failure sample sets of type sensors are F 1 、F5 、F 11 .

[0068] Among them, according to N failure modes, W i Failure sample set F of sensor type i All samples F ik Classification, get W i N failure modes of type sensor, including: calculation of W i Failure sample set F of type sensor i All subsamples F ik The mean, variance, and average of the slopes of two adjacent data points; if the mean is less than the fourth preset threshold, then the subsample F ik The failure mode is missing values; if the variance is less than the fifth preset threshold, then the subsample F ik The failure mode is numerical stagnation; if the average value of the slopes of two adjacent data points is greater than the sixth preset threshold, then the subsample F ik The failure mode is value mutation; otherwise, the subsample F ik The failure mode is numerical drift; the statistical failure sample set F i The subsamples F corresponding to the N failure modes in ik The number of W i N failure modes of type sensor.

[0069] In this embodiment, the four W i The failure sample set of the type sensor is classified into patterns to obtain W i The number of sub-samples corresponding to the four failure modes of the type of sensor. 6 Type as an example: Find the 3213×60 failure sample set F 6 In each subsample F 6k The mean, variance, and average slope of two adjacent data points of (1×60). The dynamic pressure flow sensor W is obtained by calculation. 6 The 3213 failure sub-samples of type sensors included 104 missing values, 178 stuck values, 189 sudden value changes, and 2742 drifted values.

[0070] Among them, calculate W i The failure probability of N failure modes of type sensor is W i The corresponding failure probabilities of the N failure modes of the type sensor include: i The subsamples F corresponding to the N failure modes in ik The number and sample set C i Neutron Sample C ijThe failure probability of each of the N failure modes is calculated based on the ratio of the total number of failure modes.

[0071] In this embodiment, four W are calculated respectively. i The failure probabilities corresponding to the four failure modes of type sensors. i If the number of data of failure mode a is greater than 0, the failure probability p is calculated using statistical methods. ia If F i If the number of data of failure mode a is 0, then its failure probability is temporarily set to 0. 6 Type as an example: 104 missing values, 178 stuck values, 189 sudden changes, and 2742 drift values ​​are substituted into the statistical algorithm to calculate the failure probability of each failure mode. For example, F 6 The failure probability p of the failure mode (missing value) 61 It is 104 / 10000.

[0072] Step S23, summing the failure probabilities corresponding to the N failure modes to obtain W i Total failure probability of sensor type.

[0073] Step S24, select W whose total failure probability is greater than the failure threshold i Type sensor and get W from the first sensor set i The sensors corresponding to the type sensor are used to obtain a second sensor set.

[0074] In this embodiment, four W are calculated respectively. i The total failure probability p of type sensor i =∑ a＝1，2...4 p ia If a W i If the total failure probability of the sensor of type W is greater than the failure threshold, then the sensor W is retained in the first sensor set. i All sensors corresponding to the type sensor; otherwise, delete the W in the first sensor set i All sensors corresponding to the type sensor. 4 W are obtained by calculation i The total failure probability of the sensors of the four types did not appear to be extremely low. i All 34 sensors corresponding to the type sensor are retained to form a second sensor set.

[0075] Step 103: Filter out a target sensor set from the second sensor set according to the sensor importance.

[0076] Optionally, step 103 specifically includes steps S31-S33:

[0077] Step S31, obtaining the important control function of each sensor in the second sensor set to construct a first mapping table, the first mapping table including the sensor name, the important control function, and the mapping relationship between the two.

[0078] Step S32, obtaining the consequences of failure of each important control function in the first mapping table according to engineering experience, the consequences including serious consequences and general consequences.

[0079] Step S33, filtering out sensors corresponding to important control functions that may produce serious consequences in the event of failure in the first mapping table, and obtaining a target sensor set.

[0080] In this embodiment, according to the "Function Analysis" file of the nuclear power plant, a list of important control functions of each sensor in the second sensor set is obtained: B i1 , B i2 ,...B ij , ..., B in , where B ij represents the jth important control function of the i-th sensor. For each important control function B ij , based on engineering experience, analyze this important control function B ij The possible consequences of failure include serious consequences and general consequences. Calculate the number of serious consequences as follows: ij , where s ij It represents the jth serious consequence of the i-th sensor. The serious consequences include the following two aspects:

[0081] (1) Serious consequences of the second type of working conditions, including:

[0082] ·Uncontrolled lifting of the regulating rod assembly when the reactor is subcritical;

[0083] ·Uncontrolled lifting of the regulating rod group during reactor operation;

[0084] Improper position control or rod drop of one or a group of control rod assemblies;

[0085] Uncontrolled dilution of boric acid;

[0086] Partial loss of reactor coolant flow;

[0087] ·Start-up of an outage loop;

[0088] Total loss of load and / or turbine shutdown due to accident;

[0089] Loss of main water supply;

[0090] The main feed water system of the steam generator is not operating normally;

[0091] Excessive load increase (at full power);

[0092] A brief depressurization of the reactor coolant system;

[0093] A safety valve in the secondary circuit is opened by mistake;

[0094] ·The safety injection system is activated accidentally.

[0095] (2) Leading to serious consequences of violating the operating limit conditions (LCO) in the nuclear power plant operating technical specifications.

[0096] The number of serious consequences s of all important control functions of the i-th sensor in the second sensor set i =∑ j＝1，2，…，n s ij If i If the value is greater than or equal to 1, the i-th sensor is regarded as a key sensor signal and is retained in the second sensor set; otherwise, the i-th sensor is deleted from the second sensor set. In this embodiment, the mapping relationship between the 34 sensors in the second sensor set and the important control functions is shown in Table 4:

[0097] Table 4

[0098]

[0099]

[0100] For each important control function in Table 4, the possible consequences of failure of the important control function are analyzed based on engineering experience, and finally 15 sensors with serious consequences of failure are screened as shown in Table 5 to form a target sensor set. Subsequently, data reconstruction technology can be applied to the 15 sensors in the target sensor set.

[0101] Table 5

[0102]

[0103] The sensor screening method based on the sensor data reconstruction technology provided in this embodiment has the following beneficial effects:

[0104] (1) Economic efficiency: It improves the feasibility of sensor data reconstruction technology in nuclear power plants, screens out sensors that are not technically feasible, have low failure probability, or have minor consequences due to failure, and prioritizes the allocation of limited computing resources to important sensor data reconstruction functions, thereby saving unit computing costs and improving economic efficiency; targeted screening of sensors can improve the performance of subsequent sensor data reconstruction models, so that subsequent reconstruction models can enable faulty sensors to have strong automatic recovery functions, allowing the unit to continue to operate at full power. Therefore, it can replace the traditional "periodic test" and conduct unified and targeted maintenance during planned shutdowns, thereby reducing operation and maintenance costs.

[0105] (2) Safety: The collection and classification of sensor failure data is conducive to building models during the reconstruction of sensor data, which can improve the performance of the model and enable the reconstruction model to accurately repair nuclear power plant sensor failures, thereby avoiding the power plant system from triggering automatic actions based on erroneous unit status signals to malfunction or refuse to operate, thereby improving the operational reliability of sensors. Limited computing resources can be provided to important sensors to reduce the response time of sensor reconstruction, and avoid the power plant system sensor failure, which may cause erroneous unit status signals to trigger automatic actions due to lack of response time, resulting in serious consequences, thereby improving the operational reliability of sensors.

[0106] Embodiment 2:

[0107] This embodiment provides a sensor data reconstruction method, which is applied to a nuclear power plant. The method includes:

[0108] Step 201: Use the sensor screening method based on sensor data reconstruction technology described in Example 1 to screen out a target sensor set.

[0109] Step 202: Online monitoring of signals of all sensors in the target sensor set.

[0110] Step 203, when a fault signal appears in any sensor in the target sensor set, the fault signal is replaced so that the nuclear power unit has the ability to automatically recover after the sensor failure within a preset time.

[0111] Embodiment 3:

[0112] like Figure 2 As shown, this embodiment provides a sensor screening device based on sensor data reconstruction technology, including a screening module 3. The screening module 3 is used to screen a target sensor set from candidate sensors according to sensor data feasibility, and / or sensor failure probability, and / or sensor importance.

[0113] Optionally, the screening module 3 includes a first screening unit 31 , a second screening unit 32 and a third screening unit 33 .

[0114] The first screening unit 31 is used to screen out the first sensor set from the candidate sensors within the default value range according to the feasibility of the sensor data. The second screening unit 32 is connected to the first screening unit 31 and is used to select sensors with a failure probability greater than a failure threshold from the first sensor set to obtain the second sensor set. The third screening unit 33 is connected to the second screening unit 32 and is used to screen out the target sensor set from the second sensor set according to the importance of the sensors.

[0115] Optionally, the first screening unit includes a judgment component. The judgment component is used to screen out sensors connected to the distributed control system from the candidate sensors within the default value range to obtain a third sensor set, and to judge whether the response time of the sensors in the third sensor set is less than the first preset threshold, and to screen out sensors whose judgment result is that the response time is less than the first preset threshold to obtain a fourth sensor set, and to judge whether the packet loss rate of data acquisition of the sensors in the fourth sensor set is less than the second preset threshold, and to screen out sensors whose judgment result is that the packet loss rate is less than the second preset threshold from the fourth sensor set to obtain a fifth sensor set, and to screen out sensors whose measurement objects are any one of temperature, pressure, flow, liquid level, concentration, and neutron flux from the fifth sensor set to obtain a sixth sensor set, and to judge whether the sensors in the sixth sensor set can collect historical operation data above the preset working condition and the preset acquisition frequency, and the amount of historical operation data is greater than the third preset threshold, and to screen out sensors whose judgment result is that the historical operation data above the preset working condition and the preset acquisition frequency can be collected, and the amount of historical operation data is greater than the third preset threshold, so as to obtain the first sensor set.

[0116] Alternatively, if Figure 3 As shown, the second screening unit 32 includes a classification component 321 , an acquisition component 322 , a calculation component 323 and a screening component 324 .

[0117] The classification component 321 is connected to the first screening unit 31 and is used to classify the sensors in the first sensor set according to the measurement principle to obtain W i Type sensor, 1≤i≤14.

[0118] The acquisition component 322 is connected to the classification component 321 and is used to obtain W i N failure modes of type sensor and the corresponding failure probabilities of the N failure modes, where N is a positive integer.

[0119] The calculation component 323 is connected to the acquisition component 322 and is used to sum the failure probabilities corresponding to the N failure modes to obtain W i Total failure probability of sensor type.

[0120] The screening component 324 is connected to the calculation component 323 and the first screening unit 31, and is used to select W whose total failure probability is greater than the failure threshold. i Type sensor and get W from the first sensor set i The sensors corresponding to the type sensor are used to obtain a second sensor set.

[0121] Optionally, the acquisition component is used to collect W in the first sensor set. i The historical operation data of the sensor corresponding to the type sensor is collected according to the preset frequency. i The historical operation data of the type sensor is frequency thinned, and the historical operation data after frequency thinning is cut according to the preset time length to obtain W i Sample set C of sensor type i , and is also used to extract the sample set C i Get W i Failure sample set F of sensor type i , according to N failure modes, W i Failure sample set F of sensor type i All subsamples F ik Classification, get W i The N failure modes of the type sensor include missing value, stuck value, sudden value change and drift value, and are used to calculate W i The failure probability of N failure modes of type sensor is W i The failure probabilities corresponding to the N failure modes of the type sensor.

[0122] It should be noted that the functions of the modules, units and components in the sensor screening device based on the sensor data reconstruction technology of this embodiment are as described in the first embodiment.

[0123] Embodiment 4:

[0124] like Figure 4 As described above, this embodiment provides a sensor data reconstruction system, including the sensor screening device based on the sensor data reconstruction technology described in Example 3, as well as a monitoring module 41 and a replacement module 42.

[0125] A sensor screening device based on sensor data reconstruction technology is used to screen out a target sensor set.

[0126] The monitoring module 41 is connected to the sensor screening device based on the sensor data reconstruction technology, and is used to perform online monitoring on the target sensor set screened by the sensor screening device based on the sensor data reconstruction technology and output the monitoring result.

[0127] The replacement module 42 is connected to the monitoring module 41 and is used to replace the fault signal when a fault signal appears in the monitoring result, so that the nuclear power unit has the ability to automatically recover after the sensor failure within a preset time.

[0128] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for screening sensors based on sensor data reconstruction technology, characterized in that: include: According to the feasibility of sensor data, sensor failure probability and sensor importance, the target sensor set is selected from the candidate sensors. Among them, the target sensor set is selected from the candidate sensors according to the sensor data feasibility, sensor failure probability and sensor importance, including: Filtering a first sensor set from candidate sensors within a default value range according to sensor data feasibility; Selecting sensors whose failure probability is greater than a failure threshold from the first sensor set to obtain a second sensor set; filtering the target sensor set from the second sensor set according to the sensor importance, Selecting sensors whose failure probability is greater than a failure threshold from the first sensor set to obtain a second sensor set specifically includes: Classify the sensors in the first sensor set according to the measurement principle to obtain Type Sensor, 1 14; Get N failure modes of the type sensor and the corresponding failure probabilities of the N failure modes, where N is a positive integer; Sum the failure probabilities corresponding to N failure modes and we get Total failure probability of sensor type; Select the total failure probability greater than the failure threshold Type sensor and get it from the first sensor collection The sensor corresponding to the type sensor, obtains the second sensor set, The acquisition N failure modes of the type sensor and the corresponding failure probabilities of the N failure modes, specifically including: Collect the first sensor set Historical operating data of sensors corresponding to the type sensor; According to the preset frequency Frequency thinning based on historical operating data of type sensors; According to the preset time length, the historical operation data after frequency thinning is cut into data to obtain Sample collection of type sensor ; From the sample collection Get Failure sample set of sensor type ; According to N failure modes Failure sample set of sensor type All subsamples in Classification, get N failure modes of the type sensor, including missing value, stuck value, sudden value change and drift of value; calculate The failure probability of N failure modes of type sensor is obtained The failure probabilities corresponding to the N failure modes of the type sensor.

2. The method for selecting sensors based on sensor data reconstruction technology according to claim 1, characterized in that: The step of selecting the first sensor set from candidate sensors within a default value range according to the feasibility of the sensor data includes: Selecting sensors connected to the distributed control system from candidate sensors within a default value range to obtain a third sensor set; Determine whether the response time of the sensors in the third sensor set is less than the first preset threshold, and select the sensors whose response time is less than the first preset threshold to obtain a fourth sensor set; It is determined whether the packet loss rate of data collection of the sensors in the fourth sensor set is less than the second preset threshold, and the sensors having a judgment result that the packet loss rate is less than the second preset threshold are screened out to obtain the first sensor set.

3. The method for selecting sensors based on sensor data reconstruction technology according to claim 2, characterized in that: The step of filtering out sensors whose packet loss rate is less than a second preset threshold value to obtain a first sensor set includes: Filtering out sensors whose judgment results show that the packet loss rate is less than the second preset threshold from the fourth sensor set, to obtain a fifth sensor set; Select sensors whose measurement objects are any one of temperature, pressure, flow, liquid level, concentration, and neutron flux from the fifth sensor set to obtain a sixth sensor set; Determine whether the sensors in the sixth sensor set can collect historical operating data above the preset operating conditions and the preset collection frequency, and the amount of historical operating data is greater than the third preset threshold, and filter out sensors whose judgment results are that they can collect historical operating data above the preset operating conditions and the preset collection frequency, and the amount of historical operating data is greater than the third preset threshold, thereby obtaining the first sensor set.

4. The method for selecting sensors based on sensor data reconstruction technology according to claim 3, characterized in that: The sample set Get Failure sample set of sensor type , specifically including: S1, from the sample set Select M subsamples of zero-failure data from As training samples, the first state monitoring model is obtained. =1,2,…,n,M n, the condition monitoring model is used to determine whether the sample is a failure sample; S2, all subsamples Input to the first condition monitoring model to obtain all subsamples The judgment result of whether the failure is or not; S3, filter out the sub-samples corresponding to the judgment result of failure , get the sample set ; S4, the sample collection Use the training samples to train the first state monitoring model, obtain the second state monitoring model, and clear the sample set. ; S5, iterate S2-S4 h times to obtain the h+2th state monitoring model, where h is a positive integer; S6, all subsamples Input to the h+2th state monitoring model to obtain all subsamples The judgment result of whether the failure is or not; S7, filter out the sub-samples corresponding to the judgment result of failure being yes ,get Failure sample set of sensor type .

5. The method for selecting sensors based on sensor data reconstruction technology according to claim 4, characterized in that: According to the N failure modes Failure sample set of sensor type All samples in Classification, get N failure modes of type sensors, including: calculate Failure sample set of type sensor All subsamples in The mean, variance, and average of the slopes of two adjacent data points; If the mean is less than the fourth preset threshold, then the subsample The failure mode of is missing value; If the variance is less than the fifth preset threshold, the subsample The failure mode is numerical stuck; If the average value of the slopes of two adjacent data points is greater than the sixth preset threshold, then the subsample The failure mode of is numerical mutation; Otherwise, the subsample The failure mode of is numerical drift; Statistical failure sample set The corresponding subsamples of N failure modes in The number of N failure modes of sensor type, The calculation The failure probability of N failure modes of type sensor is obtained The failure probabilities corresponding to the N failure modes of the type sensor include: According to the failure sample set The corresponding subsamples of N failure modes in The number and sample set Neutron Sample The failure probability of each of the N failure modes is calculated based on the ratio of the total number of failure modes.

6. The method for selecting sensors based on sensor data reconstruction technology according to claim 5, characterized in that: The step of selecting the target sensor set from the second sensor set according to the sensor importance specifically includes: Obtaining important control functions of each sensor in the second sensor set to construct a first mapping table, wherein the first mapping table includes sensor names, important control functions, and a mapping relationship between the two; Obtain the consequences of failure of each important control function in the first mapping table according to engineering experience, including serious consequences and general consequences; Sensors corresponding to important control functions in the first mapping table that would produce serious consequences in the event of failure are screened out to obtain a target sensor set.

7. A sensor data reconstruction method, characterized in that: The target sensor set is screened out using the screening method described in any one of claims 1 to 6, Online monitoring of the signals of all sensors in the target sensor set; When a fault signal appears in any sensor in the target sensor set, the fault signal is replaced so that the nuclear power unit has the ability to automatically recover after sensor failure within a preset time.

8. A sensor screening device based on sensor data reconstruction technology, characterized in that: Including filtering modules, The screening module is used to screen the target sensor set from the candidate sensors according to the feasibility of sensor data, sensor failure probability and sensor importance. The screening module includes a first screening unit, a second screening unit and a third screening unit. A first screening unit is used to screen out a first sensor set from candidate sensors within a default value range according to the feasibility of sensor data. a second screening unit connected to the first screening unit, and configured to select sensors with a failure probability greater than a failure threshold from the first sensor set to obtain a second sensor set; a third screening unit, connected to the second screening unit, for screening the target sensor set from the second sensor set according to sensor importance, the second screening unit comprising a classification component, an acquisition component, a calculation component and a screening component, A classification component is connected to the first screening unit and is used to classify the sensors in the first sensor set according to the measurement principle to obtain Type Sensor, 1 14. The acquisition component is connected to the classification component to obtain N failure modes of type sensor and the corresponding failure probabilities of N failure modes, N is a positive integer, The calculation component is connected to the acquisition component to sum the failure probabilities corresponding to N failure modes and obtain The total failure probability of the sensor type, A screening component is connected to the calculation component and the first screening unit, and is used to select the total failure probability greater than the failure threshold. Type sensor and get it from the first sensor collection The sensor corresponding to the type sensor, obtains the second sensor set, The acquisition component is used to collect the first sensor set The historical operation data of the sensor corresponding to the type sensor is collected according to the preset frequency The historical operation data of the type sensor is frequency thinned, and the historical operation data after frequency thinning is cut according to the preset time length to obtain Sample collection of type sensor , from the sample set Get Failure sample set of sensor type , according to N failure modes Failure sample set of sensor type All subsamples in Classification, get N failure modes of the type sensor, including missing value, stuck value, sudden value change and drift value, as well as, for calculating The failure probability of N failure modes of type sensor is obtained The failure probabilities corresponding to the N failure modes of the type sensor.

9. A sensor data reconstruction system, comprising the sensor screening device based on the sensor data reconstruction technology according to claim 8, a monitoring module and a replacement module, A sensor screening device based on sensor data reconstruction technology is used to screen out a target sensor set. A monitoring module is connected to the sensor screening device based on the sensor data reconstruction technology, and is used to perform online monitoring on the target sensor set screened by the sensor screening device based on the sensor data reconstruction technology and output the monitoring result. The replacement module is connected to the monitoring module and is used to replace the fault signal when a fault signal appears in the monitoring result, so that the nuclear power unit has the ability to automatically recover after sensor failure within a preset time.

Images