A soft redundancy and soft measurement control system based on artificial intelligence

Abstract

The present invention relates to a soft redundancy and soft measurement control system based on artificial intelligence, belonging to the field of intelligent nuclear power technology, the control system includes a state estimation and anomaly detection module and a voting module, the input signal of the state estimation and anomaly detection module is a plurality of instrument measurement signals with mutual correlation in the nuclear power plant system and equipment, and the state estimation signal, anomaly diagnosis signal and model reliability evaluation signal of the current monitoring instrument are obtained according to the instrument measurement signal by using artificial intelligence technology; the input signal of the voting module includes all or part of the instrument measurement signal, the state estimation mean signal, the anomaly diagnosis signal and the model reliability evaluation signal, and the output signal is the post-voting signal. The present invention effectively feeds back the state estimation and anomaly detection results, can optimize the alarm logic of the nuclear power plant control system, improve the operator's judgment level on the operating status of the nuclear power plant, thereby improving the safety and economy of the operation of the nuclear power plant.

Description

Technical Field

[0001] The present invention relates to the field of intelligent nuclear power technology, and in particular to a soft redundancy and soft measurement control system based on artificial intelligence. Background Art

[0002] As competition in the energy market becomes increasingly fierce, nuclear power plants must optimize their reliability and economy. The rapid development and widespread application of artificial intelligence technology have proven that AI-based decision-making support systems can help operators accurately assess the current status, safety margins, and expected operational deviations of nuclear power plants, thereby reducing human errors. From a long-term perspective, its goal can even gradually replace operators and ultimately achieve "unmanned" or "less-manned" duty at nuclear power plants, thereby improving the economy and safety of nuclear power plants.

[0003] In the field of nuclear power plant control systems, traditional alarm methods are generally based on the following three methods: 1) Instrument measurement values ​​are directly used for static threshold alarms; 2) Alarms are performed using voting devices, that is, multiple redundant instrument measurement values ​​are compared with each other to determine which instrument signal is abnormal, and the average value of the remaining valid signals is calculated and output; 3) Other types of alarms, such as alarms designed based on reactor physics and thermal hydraulic design principles. For the above three design situations, when the instrument itself degrades or the system operating status fluctuates, false alarms will occur, which will interfere with the operator's assessment of the current status of the nuclear power plant. The reasons are mainly as follows: 1) The alarm signal of a single instrument can only reflect the abnormal manifestation (symptoms), but cannot reflect the actual abnormal cause (cause), and when the instrument is abnormal, a false alarm will occur; 2) The static threshold alarm method can only achieve effective alarm under steady-state operation, and the alarm threshold is based on design analysis, not based on the actual state of the nuclear power plant (different power plants have different states). Under transient conditions, non-rated operating conditions, and emergency shutdown conditions, a large number of false alarms may occur, and even alarm avalanche problems may occur; 3) The voters used in traditional redundant instruments have weak abnormality monitoring capabilities. Summary of the invention

[0004] The purpose of the present invention is to provide a soft redundancy and soft measurement control system based on artificial intelligence in the current field of nuclear power plant instrumentation control, which can optimize the alarm logic of the nuclear power plant control system and improve the operator's judgment level on the operating status of the nuclear power plant, thereby improving the safety and economy of the operation of the nuclear power plant.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0006] A soft redundancy and soft measurement control system based on artificial intelligence, including a state estimation and anomaly detection module and a voter module;

[0007] The state estimation and anomaly detection module, whose input signal is a plurality of instrument measurement signals with mutual correlation in the nuclear power plant system and equipment, obtains the state estimation signal of the current monitoring instrument according to the instrument measurement signal by using artificial intelligence technology, and outputs an abnormal diagnosis signal and a model reliability evaluation signal according to the instrument measurement signal and the state estimation signal;

[0008] The input signal of the voting module includes all or part of the instrument measurement signal, the state estimation mean signal, the abnormal diagnosis signal and the model reliability evaluation signal, and its output signal is a post-voting signal.

[0009] Furthermore, in the artificial intelligence-based soft redundancy and soft measurement control system as described above, the state estimation signal is a state estimation value of the instrument measurement signal.

[0010] Furthermore, in the artificial intelligence-based soft redundancy and soft measurement control system as described above, the state estimation and abnormality detection module obtains the abnormal diagnosis signal based on the residual between all instrument measurement signals and their corresponding state estimation signals; when the abnormal diagnosis signal has no alarm, the state estimation signal participates in the voting process of the voting module; when the abnormal diagnosis signal indicates that an alarm has occurred in one or more of the instrument measurement signals, the corresponding instrument measurement signal does not participate in the voting process of the voting module.

[0011] Furthermore, in the artificial intelligence-based soft redundancy and soft measurement control system as described above, the state estimation and anomaly detection module obtains the model reliability assessment signal based on the preprocessed real-time data and the state estimation signal.

[0012] Furthermore, in the artificial intelligence-based soft redundancy and soft measurement control system as described above, the model reliability evaluation signal is a continuous value between [0,1], which is divided into three levels: high, medium and low according to the set threshold. When the model reliability is high or medium, the state estimation and anomaly detection module is in normal operation; when the model reliability is low, all output signals of the state estimation and anomaly detection module are unavailable.

[0013] Further, in the artificial intelligence-based soft redundancy and soft measurement control system as described above, the state estimation mean signal is the mean of all the state estimation signals.

[0014] Furthermore, in the artificial intelligence-based soft redundancy and soft measurement control system as described above, the voting module includes an N+1 strategy voter, a (N-1)+1 strategy voter and a soft measurement strategy voter, wherein N represents the number of redundant instruments input into the original voting device.

[0015] Furthermore, in the artificial intelligence-based soft redundancy and soft measurement control system as described above, the N+1 strategy voter adds one state estimation mean signal on the basis of the original strategy voter; the (N-1)+1 strategy voter reduces one instrument measurement signal on the basis of the original strategy voter and adds one state estimation mean signal; the soft measurement strategy voter only uses the state estimation signal.

[0016] Furthermore, in the soft redundancy and soft measurement control system based on artificial intelligence as described above, the three voters in the voting module all follow the following voting strategy: when the model reliability assessment signal level is high or medium and the abnormal diagnosis signal has no alarm, all input signals and state estimation signals in the same redundant channel are averaged to obtain the post-voting signal; when the model reliability assessment signal level is high or medium and the abnormal diagnosis signal has an alarm, the input signal corresponding to the alarm signal is eliminated, and the remaining non-abnormal input signals and the state estimation signal are averaged to obtain the post-voting signal; when the model reliability assessment signal level is low, the state estimation signal and the abnormal diagnosis signal are unavailable, and the original voter is used for voting to obtain the post-voting signal.

[0017] The artificial intelligence-based soft redundancy and soft measurement control system of the present invention has the following significant technical effects:

[0018] The present invention is based on artificial intelligence state estimation and anomaly detection technology, uses the actual normal operation data of the nuclear power plant, integrates multi-instrument data and system multi-operation state parameters to achieve soft redundancy or soft measurement, and can change the existing nuclear power plant control logic without reducing the system reliability, optimize the alarm mode and alarm capability, and greatly reduce the system construction cost and operation and maintenance cost, thereby achieving the purpose of improving the safety and economy of nuclear power plant operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A structural block diagram of a soft redundancy and soft measurement control system based on artificial intelligence provided in an embodiment of the present invention;

[0020] Figure 2 This is the liquid level control logic diagram of the steam generator in a nuclear power plant. DETAILED DESCRIPTION

[0021] The present invention is further described below in conjunction with specific embodiments and the accompanying drawings.

[0022] The present invention provides a soft redundancy and soft measurement control system structure based on artificial intelligence. Figure 1As shown, the control system includes a state estimation and anomaly detection module A01 and a voter module B01, wherein the input signal of the state estimation and anomaly detection module A01 is a number of instrument measurement signals A02 with high cross-correlation in the nuclear power plant system and equipment, and its output signal includes: a state estimation signal A05, an anomaly diagnosis signal A03 and a model reliability evaluation signal A04 for the instrument measurement signal A02. The input signal of the voter module B01 includes all or part of the instrument measurement signal A02, the state estimation mean signal B02, the anomaly diagnosis signal A03 and the model reliability evaluation signal A04, and its output signal is the voted signal B03.

[0023] The state estimation and anomaly detection module A01 uses artificial intelligence technology to obtain the state estimation signal A05 of the current monitoring instrument based on the instrument measurement signal A02 with high cross-correlation in the nuclear power plant, and can determine whether the current instrument has an abnormality through model uncertainty, thereby obtaining an abnormal diagnosis signal A03.

[0024] The state estimation signal A05 is the state estimation value of all instrument signals in the input signal. When the model reliability evaluation signal A04 is at a high or medium level, the state estimation accuracy is still high even if the input data is abnormal.

[0025] The abnormal diagnosis signal A03 is determined based on the residual between all input signals and their corresponding state estimation signal A05. When the abnormal diagnosis signal A03 has no alarm, the state estimation signal A05 will participate in the voting process of the voting module B01; when the abnormal diagnosis signal A03 indicates that one or more instrument signals have an alarm, the corresponding input signal will not participate in the voting process of the voting module B01.

[0026] The state estimation and anomaly detection module A01 uses the preprocessed real-time data and the state estimation signal A05 to evaluate the reliability of the model and outputs the model reliability evaluation signal A04. The model reliability evaluation signal A04 is a continuous value between [0,1]. The larger the value of the model reliability evaluation signal, the higher the credibility of the state estimation signal A05 and the abnormal diagnosis signal A03. Conversely, the lower the credibility of the state estimation signal A05 and the abnormal diagnosis signal A03. The model reliability evaluation signal A04 can be set to three levels: high, medium, and low according to different thresholds. When the model reliability is high, the state estimation and abnormality detection module A01 is in normal operation, and the state estimation signal A05 and the abnormality diagnosis signal A03 have high credibility; when the model reliability is medium, the state estimation and abnormality detection module A01 is in normal operation, and the state estimation signal A05 and the abnormality diagnosis signal A03 have medium credibility; when the model reliability is low, the state estimation signal A05 and the abnormality diagnosis signal A03 have low credibility, and the voter module B01 will no longer receive the state estimation signal A05 and the abnormality diagnosis signal A03 and trigger a low reliability alarm signal.

[0027] The voting module B01 includes three types of voting devices with decision strategies, namely: N+1 strategy voting device, (N-1)+1 strategy voting device and soft measurement strategy voting device, where N represents the number of redundant instruments input to the original voting device. The N+1 strategy voting device adds a state estimation mean signal B02 on the basis of the original strategy voting device; the (N-1)+1 strategy voting device reduces one instrument signal on the basis of the original strategy voting device and adds one state estimation mean signal B02; the soft measurement strategy voting device only uses the state estimation signal A05.

[0028] The three voters in the voting module B01 all follow the following voting strategy: when the model reliability assessment signal A04 is high or medium and there is no alarm in the abnormal diagnosis signal A03, all input signals and state estimation signal A05 in the same redundant channel are averaged to obtain the post-voting signal B03; when the model reliability assessment signal A04 is high or medium and there is an alarm in the abnormal diagnosis signal A03, the input signal corresponding to the alarm signal is eliminated, and the remaining non-abnormal input signals and the state estimation signal A05 are averaged to obtain the post-voting signal B03; when the model reliability assessment signal A04 is low, the state estimation signal A05 and the abnormal diagnosis signal A03 are unavailable, and the original voter is used for voting to obtain the post-voting signal B03.

[0029] The specific implementation scheme of the artificial intelligence-based soft redundancy and soft measurement control system in the liquid level control system of the steam generator in a nuclear power plant is described below.

[0030] The liquid level control logic of the steam generator in a nuclear power plant is as follows: Figure 2 As shown, the main feedwater flow measurement channel is composed of three redundant instruments: 043MD, 046MD and 052MD, and the steam generator level measurement channel is composed of three redundant instruments: 010MN, 052MN and 058MN. This embodiment will specifically explain the implementation plan of soft redundancy and soft measurement control system based on artificial intelligence for this redundant instrument configuration. The specific implementation method is as follows:

[0031] Taking the three redundant instruments of the main feed water flow measurement channels 043MD, 046MD and 052MD as an example, the real-time monitoring data of these three instruments and the data of other monitoring instruments with high cross-correlation in the nuclear power plant will be input into the state estimation and anomaly detection module A01, thereby obtaining the state estimation signal A05, anomaly diagnosis signal A03 and a model reliability assessment signal A04 for each instrument, and these signals will be input into the voter module B01 for subsequent calculations.

[0032] When the voter module B01 uses the N+1 voting strategy, the voter module B01 first determines the level of the model reliability assessment signal A04. If it is at a high or medium level, the state estimation signal A05 output by the state estimation and abnormality monitoring module A01 is used. Then, according to the abnormal diagnosis signal A03 corresponding to each instrument, it is determined whether the instrument is abnormal. If 043MD, 046MD and 052MD are normal, the state estimation value A05 of all instruments and the real-time monitoring values ​​of 043MD, 046MD and 052MD are averaged to obtain the post-voting signal B03 for output; if the abnormal diagnosis signal of a certain instrument (for example, 043MD) alarms, and the other signals are normal, the voter module B01 will remove the real-time monitoring signal of 043MD, and then calculate the average of the state estimation value A05 of all instruments and the real-time monitoring values ​​of the instruments without abnormalities (046MD, 052MD). The post-voting signal B03 is output; if the abnormal diagnosis signals of two instruments (such as 043MD and 046MD) alarm, and the other signals have no alarm, the voting module B01 will remove the real-time monitoring signals of 043MD and 046MD, and average the state estimation values ​​A05 of all instruments and the real-time monitoring values ​​of the instrument without abnormality (052MD), so as to obtain the post-voting signal B03 for output; if the abnormal diagnosis signals of the three instruments are all alarmed, the voting module B01 will remove the real-time monitoring signals of all instruments, and only average the state estimation values ​​A05 of all instruments, so as to obtain the post-voting signal B03 for output. If the level of the model reliability evaluation signal A04 is low, the state estimation signal A05 and the abnormal diagnosis signal A03 are not available. At this time, the voting module B01 will use the original voting machine to vote, and only read the real-time monitoring values ​​of 043MD, 046MD and 052MD to obtain the post-voting signal B03.

[0033] When the voter module B01 uses the (N-1)+1 voting strategy, one measurement channel can be reduced, leaving only two measuring instruments (e.g., 043MD and 046MD). The voter module B01 first determines the level of the model reliability assessment signal A04. If it is at a high or medium level, the state estimation signal A05 output by the state estimation and abnormality monitoring module A01 is used. Then, according to the abnormal diagnosis signal A03 corresponding to each instrument, it is determined whether the instrument is abnormal. If 043MD and 046MD are normal, the state estimation value A05 of all instruments and the real-time monitoring values ​​of 043MD and 046MD are averaged to obtain the post-voting signal B03 for output; if the abnormal diagnosis signal of a certain instrument (e.g., 043MD) alarms, and the other signals are normal, the voter module B01 will remove the real-time monitoring signal of 043MD, and then calculate the average of the state estimation value A05 of all instruments and the real-time monitoring value of the instrument 046MD without abnormality, so as to obtain the post-voting signal. B03 is output; if the abnormal diagnosis signals of the two instruments (043MD and 046MD) both alarm, the voting module B01 will calculate the average of the state estimation values ​​A05 of all instruments after eliminating the real-time monitoring signals of 043MD and 046MD, so as to obtain the post-voting signal B03 for output; if the level of the model reliability evaluation signal A04 is low, the state estimation signal A05 and the abnormal diagnosis signal A03 are not available. At this time, the voting module B01 will use the original voting machine for voting, and only read the real-time monitoring values ​​of 043MD and 046MD to obtain the post-voting signal B03.

[0034] When the voter module B01 uses the soft measurement voting strategy, the measurement channel can be reduced to only one instrument (for example, 043MD). The voter module B01 first determines the level of the model reliability assessment signal A04. If it is at a high or medium level, the state estimation signal A05 output by the state estimation and abnormality monitoring module A01 is used. Then, according to the abnormal diagnosis signal A03, it is determined whether the instrument is abnormal. If 043MD is not abnormal, the state estimation value A05 of the instrument and the real-time monitoring value of 043MD are averaged to obtain the post-voting signal B03 for output; if the abnormal diagnosis signal of a certain instrument alarms, the voter module B01 will directly output the state estimation value A05 of the instrument after removing the real-time monitoring signal of 043MD to obtain the post-voting signal B03; if the level of the model reliability assessment signal A04 is low, the state estimation signal A05 and the abnormal diagnosis signal A03 are not available. At this time, the real-time monitoring value of 043MD is directly read to obtain the post-voting signal B03.

[0035] The artificial intelligence-based soft redundancy and soft measurement control system provided by the present invention effectively feeds back the state estimation and abnormality detection results, optimizes the nuclear power plant control logic based on the original voter of the nuclear power plant control system, and realizes "N+1" redundant control logic, or "(N-1)+1" redundant control logic, or soft measurement control logic, thereby improving the alarm capability and the reliability of the closed-loop control system, and at the same time can also improve the safety and economy of the nuclear power plant, and has broad application prospects.

[0036] The above embodiments are merely illustrative of the present invention, and the present invention may also be implemented in other specific ways or other specific forms without departing from the gist or essential features of the present invention. Therefore, the described embodiments should be considered illustrative rather than restrictive in any respect. The scope of the present invention should be described by the appended claims, and any changes equivalent to the intent and scope of the claims should also be included within the scope of the present invention.

Claims

1. A soft redundancy and soft measurement control system based on artificial intelligence, characterized in that: Includes state estimation and anomaly detection modules and voting module; The state estimation and anomaly detection module, whose input signal is a plurality of instrument measurement signals with mutual correlation in the nuclear power plant system and equipment, obtains the state estimation signal of the current monitoring instrument according to the instrument measurement signal by using artificial intelligence technology, and outputs an abnormal diagnosis signal and a model reliability evaluation signal according to the instrument measurement signal and the state estimation signal; The input signal of the voting module includes all or part of the instrument measurement signal, the state estimation mean signal, the abnormal diagnosis signal and the model reliability evaluation signal, and its output signal is a post-voting signal. The voting module includes an N+1 strategy voter, a (N-1)+1 strategy voter and a soft measurement strategy voter, wherein N represents the number of redundant instruments input to the original voting module; The three voters in the voting module all follow the following voting strategy: when the model reliability assessment signal level is high or medium and there is no alarm for the abnormal diagnosis signal, all input signals in the same redundant channel and the state estimation signal are averaged to obtain the post-voting signal; when the model reliability assessment signal level is high or medium and there is an alarm for the abnormal diagnosis signal, the input signal corresponding to the alarm signal is eliminated, and the remaining non-abnormal input signals and the state estimation signal are averaged to obtain the post-voting signal; when the model reliability assessment signal level is low, the state estimation signal and the abnormal diagnosis signal are unavailable, and the original voter is used for voting to obtain the post-voting signal.

2. The artificial intelligence-based soft redundancy and soft measurement control system according to claim 1 is characterized in that: The state estimation signal is a state estimation value of the instrument measurement signal.

3. The artificial intelligence-based soft redundancy and soft measurement control system according to claim 2 is characterized in that: The state estimation and anomaly detection module obtains the anomaly diagnosis signal according to the residual between all instrument measurement signals and their corresponding state estimation signals; when the anomaly diagnosis signal has no alarm, the state estimation signal participates in the voting process of the voting module; When the abnormal diagnosis signal indicates that an alarm occurs in one or more of the instrument measurement signals, the corresponding instrument measurement signal does not participate in the voting process of the voting module.

4. The artificial intelligence-based soft redundancy and soft measurement control system according to any one of claims 1 to 3, characterized in that: The state estimation and anomaly detection module obtains the model reliability assessment signal according to the preprocessed real-time data and the state estimation signal.

5. The artificial intelligence-based soft redundancy and soft measurement control system according to claim 4 is characterized in that: The model reliability evaluation signal is a continuous value between [0, 1], and is divided into three levels: high, medium, and low according to the set threshold. When the model reliability is high or medium, the state estimation and anomaly detection module is in normal operation; when the model reliability is low, all output signals of the state estimation and anomaly detection module are unavailable.

6. The artificial intelligence-based soft redundancy and soft measurement control system according to claim 1, characterized in that: The state estimation mean signal is the mean of all the state estimation signals.

7. The artificial intelligence-based soft redundancy and soft measurement control system according to claim 6 is characterized in that: The N+1 strategy voter adds one state estimation mean signal on the basis of the original strategy voter; the (N-1)+1 strategy voter reduces one instrument measurement signal on the basis of the original strategy voter and adds one state estimation mean signal; the soft measurement strategy voter only uses the state estimation signal.

Images