Intelligent monitoring method for DCS operation of nuclear power plant

Abstract

The invention particularly relates to an intelligent monitoring method for DCS operation of a nuclear power plant, which is suitable for intelligent supervision of use of tools for preventing human errors in the field of the nuclear power plant, and performs frame-by-frame image recognition on a current working equipment code through a video stream output by a DCS; performing frame-by-frame image recognition on the state of the current equipment and the state of an alarm picture through a video stream output by the DCS; providing a character prompt of the current operation step through a display device; an audio stream is extracted through a microphone or a telephone interface, and a record of a worker is judged through voice recognition; frame-by-frame image recognition and mouse recognition are carried out through a video stream output by the DCS to judge the current pointed position; judging whether parameters need to be input or not by combining frame-by-frame images with regulations through a video stream output by the DCS; according to the invention, intelligent behavior supervision and operation reminding are carried out on the whole behavior process of the operator, so that the purposes of reducing cognitive load and preventing human errors are achieved.

Description

Technical Field

[0001] The present invention specifically relates to an intelligent monitoring method for DCS operation of a nuclear power plant, which is suitable for intelligent supervision of the use of tools for preventing human errors on site in a nuclear power plant, adopts an intelligent system to guide staff to operate equipment and check status, and adopts artificial intelligence technology to make intelligent judgments on the use of tools for preventing human errors by staff. Background Art

[0002] As a complex human-machine system, preventing human errors is a key issue that needs to be paid attention to in nuclear power operations. With the rapid development of intelligent technology, human error prevention has also gradually developed from traditional human defense to technical defense and intelligent defense, with the goal of further improving personnel efficiency and human reliability. According to statistics, the main control room of a typical nuclear power plant handles more than 30,000 operating procedures each year, including more than 1,000 high-frequency and high-risk operations. In order to avoid human errors, each operation contains at least 3 or more human defense tools. However, human-related incidents are still unavoidable. Once a mistake occurs, the economic losses caused are at least tens of millions. After analyzing the root causes, it was found that most of them focused on human-related incidents, indicating that traditional "human defense" has encountered bottlenecks, which is also a problem and pain point that needs to be solved urgently in the nuclear power industry.

[0003] With the digital and intelligent transformation of nuclear power, the demand for intelligent assistance to prevent human errors has shown a strong growth trend. Through communication with different professionals of nuclear power owners and analysis of human events, it was found that the main problems existing in the DCS procedure operation process in the main control room of nuclear power plants are as follows: failure to confirm the operation object in accordance with the requirements of the procedures; failure to confirm the status response in accordance with the requirements of the procedures; lack of supervision or invalid supervision of key steps in the procedures; loss of key information in the three-way communication process leading to misunderstandings; skipping or missing steps; some high-frequency operations are too fast, bypassing the use of human error tools.

[0004] Therefore, according to Rasmussen's SRK model, tasks can be divided into skill-based (SB) tasks, rule-based (RB) tasks and knowledge-based (KB) tasks. Based on the fact that most of the existing DCS operation task scenarios in the main control room of nuclear power plants are rule-based tasks, it is necessary to conduct research from the perspective of artificial intelligence supervision and augmented reality prompts for preventing human errors in the entire process of procedure execution based on rule-based task scenarios and tools for preventing human errors, so as to achieve the goal of reducing the mental workload of procedure operators, reducing the labor cost of supervisors / guardians, and effectively reducing human errors.

[0005] Currently, speech recognition, gesture recognition, target detection, and image classification technologies in artificial intelligence are developing rapidly, and augmented reality technology is gradually entering the industrial world, laying a good technical foundation for supervising the use of tools by operators to prevent human errors. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent monitoring method for DCS operation of a nuclear power plant, so as to realize intelligent behavior supervision and operation reminder for the entire behavior process of the operator, thereby achieving the purpose of reducing cognitive load and preventing human errors.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is:

[0008] An intelligent monitoring method for DCS operation of a nuclear power plant includes: performing frame-by-frame image recognition of the current working equipment code through the video stream output by the DCS; performing frame-by-frame image recognition of the current equipment status and alarm screen status through the video stream output by the DCS; providing text prompts of the current operation steps through a display device; extracting audio streams through a microphone or telephone interface, and determining the staff's vote counting through voice recognition; performing frame-by-frame image recognition through the video stream output by the DCS to perform mouse recognition to determine the current pointed position; performing frame-by-frame image recognition and combining with procedures through the video stream output by the DCS to determine whether parameters need to be input; performing frame-by-frame image recognition to determine values ​​and mouse click areas through the video stream output by the DCS; and performing frame-by-frame image recognition of the equipment status after the operation through the video stream output by the DCS.

[0009] The display device connected through the edge inference terminal provides text prompts of the current operation steps.

[0010] The audio stream is extracted by connecting the microphone or telephone interface of the edge inference terminal, and voice recognition determines the staff's vote counting.

[0011] The video stream output by DCS is combined with the procedure frame by frame to determine whether parameters need to be input, and the parameter input box is monitored.

[0012] If this work step requires supervision by others, repeat the above content.

[0013] If this work step requires three-way communication, call three-way communication voice recognition for judgment.

[0014] Each step is implemented as follows: image recognition of the consistency between the current DCS page and the page corresponding to the corresponding step in the procedure; image recognition of the working equipment code in the DCS page, and comparison of the recognition result with the preset number in the procedure; identification of the current equipment status through image recognition algorithm: image segmentation of the equipment image, use of target detection method to detect all parameters on the equipment, and comparison of the status with the preset status in the procedure; text prompts for the current operation steps are provided through the display terminal; and the staff's vote counting is determined through voice recognition: voice recognition outputs text, segmentation is performed according to the word segmentation library, and compared with the key words in the procedure. After the key word comparison, a sequence comparison is performed, and if there is no correct comparison within 60 seconds, the system will automatically confirm the correct answer. If the comparison or comparison result is wrong, the vote counting error is prompted; the mouse recognition is performed through image recognition DCS video stream to determine the current pointed position, and the mouse operation intention is double-determined. The first method: through the pre-marked operation button area, once entered, the intention to operate a certain button is determined; the second method: three consecutive frames of images are used for OCR recognition of the encoding of the button pointed by the mouse, which is recorded as a double judgment. If the two judgment times occur within 500ms, it is determined as the current mouse operation intention to prevent OCR misjudgment; the display terminal connected to the edge reasoning terminal provides prompt results of the work steps; the device status after the operation is determined by image recognition: the corresponding parameters of the image recognition device are compared with the expected values.

[0015] Judgment of three-way communication: 1) Determine through voiceprint that different voices correspond to different staff members; 2) Voice recognition outputs text; 2) Segment words according to the word segmentation library; 3) Compare the word segments of different staff members with the keywords in the regulations. If there are no keywords, it indicates an abnormality; 4) If step 3) is passed, extract the dialogue text sequence where the keyword appears for the first time. The sequence is three sentences: one sentence of text from person A, one sentence of text from person B, and one sentence of text from person A; 5) Determine whether the same keyword appears in the three sentences. If so, select the next keyword and repeat step 4) until they are all correct. Otherwise, it will prompt that the three-way communication is abnormal.

[0016] The identification methods of all steps are made into independent modules. If the sequence of procedure steps changes, it is only necessary to adjust the positions of multiple modules of the corresponding step operation objects in the program execution sequence or combine them.

[0017] The beneficial effects achieved by the present invention are:

[0018] Compared with the traditional human-defense-based anti-human error concept, the present invention places more emphasis on using artificial intelligence technology to assist in improving the operator's safety risk perception and reminding the operator of the work process. It can more effectively replace traditional human defense measures with intelligent means for human failure points, and provide operators with more assistance and support, thereby achieving the goal of truly reducing human errors and improving personnel efficiency. The method and system can provide effective operation guidance and supervision for operators. During the entire operation process, the artificial intelligence system assumes the role of supervision and can effectively provide operators with operation guidance and real-time feedback, thereby avoiding human error problems during the operation of procedures such as operating object errors, skipped steps, and insufficient status monitoring. At the same time, it can reduce the mental workload of on-site staff, strengthen attention, and improve execution efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 To prevent human error, monitor the flow chart;

[0020] Figure 2 To monitor the system process intelligently;

[0021] Figure 3 Assist in checking the consistency between the operation steps and the actual execution;

[0022] Figure 4 Monitoring process for important operations;

[0023] Figure 5 Guardianship proceedings for temporary orders;

[0024] Figure 6 It is a three-way communication monitoring process;

[0025] Figure 7 This is a connection diagram for DCS monitoring equipment. DETAILED DESCRIPTION

[0026] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] Human nature (internal) and situational characteristics (external) determine that most human errors in nuclear power plants are unintended, that is, caused by careless and negligent actions. The artificial intelligence-based nuclear power plant anti-human error tool uses a supervision method based on rule-based tasks to mainly enhance individual attention and personnel efficiency in the three links of "perception", "thinking" and "action", and solves the problem of human errors in the unintentional human-computer interaction process, namely: (1) object selection errors. For example: operation / display object selection / reading errors; (2) sequence errors. For example: procedure skipping or missing steps, etc.; (3) time errors. For example: switch operation is too early or too late; (4) completion quality errors. For example: valve adjustment opening is too little or too much.

[0028] By analyzing and summarizing the human error problems existing in the procedural operation process of the above-mentioned nuclear power plant human error events, taking the rule-based tasks (DCS operations) of nuclear power plants as the research object, combining the rules for preventing human errors and the concept of intelligent prevention of human errors, using new technologies combined with artificial intelligence, based on operating procedures, using voice recognition and image recognition technology, combined with effective human-computer interaction, a supervision method and corresponding system for the use of DCS operation prevention tools were designed.

[0029] The present invention adopts a variety of innovative methods and designs: (1) The failure modes in the use of human error prevention tools are analyzed, and targeted intelligent supervision and prompting methods are designed for each failure mode. (2) Specific program inspection processes and methods are designed for the use of different human error prevention tools such as counting votes, pointing to the correct object, and three-way communication. (3) An intelligent human error prevention model for nuclear power plants is established.

[0030] The present invention mainly involves the following specific technical contents: (1) The present invention analyzes the behavior patterns of operators in different types of work steps and develops specific intelligent supervision processes and methods. (2) In response to the above-mentioned different human failure problems, the present invention has developed intelligent supervision methods and implementation algorithms based on the fusion of multiple methods such as image recognition, voice recognition, image classification, and augmented reality prompts. (3) The present invention has designed a method for loosely coupling the work steps in the procedure with the intelligent supervision algorithm to prevent human errors. Through free combination, the intelligent supervision steps to prevent human errors can be mapped to the work steps without code editing, so that one-time development can be achieved to adapt to different operating procedures of the same equipment or even new procedures modified in the future.

[0031] The detailed description of the technical solution of the present invention is as follows:

[0032] (1) The present invention is divided into the following steps to prevent human error:

[0033] ·Identify the current working device code frame by frame through the video stream output by DCS.

[0034] ·Identify the current device status and alarm screen status through the frame-by-frame image recognition algorithm of the video stream output by DCS.

[0035] The display device connected through the edge inference terminal provides text prompts of the current operation steps.

[0036] The audio stream is extracted through the microphone or telephone interface connected to the edge inference terminal, and voice recognition determines the staff's vote counting.

[0037] ·Perform frame-by-frame image recognition on the video stream output by DCS to identify the mouse and determine the current pointing position.

[0038] ·The video stream output by DCS is combined with the procedure to determine whether parameters need to be input frame by frame, and the parameter input box is monitored.

[0039] ·Perform frame-by-frame image recognition and determine the value and mouse click area through the video stream output by DCS.

[0040] The device status after performing frame-by-frame image recognition operations on the video stream output by DCS.

[0041] If this work step requires supervision by others, follow Figure 1 Execute, the system repeats the above content.

[0042] If this work step requires three-way communication, call the three-way communication voice recognition for judgment.

[0043] (2) Each step in (1) above is implemented as follows:

[0044] Image recognition of the consistency between the current DCS page and the corresponding page of the corresponding step in the procedure

[0045] · Image recognition DCS page working equipment code will be compared with the procedure preset number.

[0046] Identify the current state of the device through image recognition algorithms: 1) Segment the device image; 2) Use target detection methods to detect all parameters on the device; 3) Compare the state with the preset state of the procedure.

[0047] The display terminal connected through the edge inference terminal provides text prompts of the current operation steps.

[0048] Determine the staff's vote counting through voice recognition: 1) Voice recognition outputs text; 2) Word segmentation is performed based on the word segmentation library; 3) Compare with the key words in the regulations; 4) Perform sequence comparison after keyword comparison; 5) If there is no correct comparison within 60 seconds or the comparison result is wrong, it will prompt that the vote counting is wrong.

[0049] ·Use image recognition to identify the current position of the mouse through DCS video stream, and use double judgment to judge the mouse operation intention. The first method: through the pre-marked operation button area, once entering, it is determined that the intention to operate a certain button. The second method: three consecutive frames of images are OCR recognized in the code of the button pointed by the mouse, which is recorded as double judgment. If the two judgment times occur within 500ms, it is judged as the current mouse operation intention to prevent OCR misjudgment.

[0050] The display terminal connected through the edge reasoning terminal provides prompt results of the work steps.

[0051] Determine the device status after operation through image recognition: Compare the corresponding parameters of the device with the expected values ​​through image recognition.

[0052] · Judgment of three-way communication: 1) Determine through voiceprint that different voices correspond to different staff members; 2) Voice recognition outputs text; 2) Segment words according to the word segmentation library; 3) Compare the word segments of different staff members with the keywords (possibly multiple) in the regulations. If there are no keywords, it indicates an abnormality; 4) If step 3) is passed, extract the dialogue text sequence where the keyword appears for the first time. The sequence is three sentences: one sentence of text from staff A, one sentence of text from staff B, and one sentence of text from staff A; 5) Determine whether the same keyword appears in the three sentences. If so, select the next keyword and repeat step 4) until they are all correct. Otherwise, it indicates that the three-way communication is abnormal.

[0053] (3) The present invention makes the identification methods of all the steps in (1) into independent modules. If the sequence of program steps in the system changes, it is only necessary to adjust the positions of the multiple modules of the corresponding step operation objects in the program execution sequence or combine them, without the need to develop a new program.

Claims

1. An intelligent monitoring method for DCS operation of a nuclear power plant, characterized in that: Identify the current working equipment code frame by frame through the video stream output by DCS; Identify the current equipment status and alarm screen status frame by frame through the video stream output by DCS; Provide text prompts of the current operation steps through the display device; extract audio streams through microphone or telephone interface, and use voice recognition to determine the staff's vote counting; perform frame-by-frame image recognition and mouse recognition through the video stream output by DCS to determine the current pointing position; perform frame-by-frame image recognition and combine the procedures through the video stream output by DCS to determine whether parameters need to be input; The video stream output by DCS is used to identify the value and mouse click area frame by frame; The device status after frame-by-frame image recognition operation is performed on the video stream output by DCS.

2. The intelligent monitoring method for DCS operation of a nuclear power plant according to claim 1, characterized in that: The display device connected through the edge inference terminal provides text prompts of the current operation steps.

3. The intelligent monitoring method for DCS operation of a nuclear power plant according to claim 1, characterized in that: The audio stream is extracted by connecting the microphone or telephone interface of the edge inference terminal, and voice recognition determines the staff's vote counting.

4. The intelligent monitoring method for DCS operation of a nuclear power plant according to claim 1, characterized in that: The video stream output by DCS is combined with the procedure frame by frame to determine whether parameters need to be input, and the parameter input box is monitored.

5. The intelligent monitoring method for DCS operation of a nuclear power plant according to claim 1, characterized in that: If this work step requires supervision by others, repeat the above content.

6. The intelligent monitoring method for DCS operation of a nuclear power plant according to claim 1, characterized in that: If this work step requires three-way communication, call three-way communication voice recognition for judgment.

7. The intelligent monitoring method for DCS operation of a nuclear power plant according to claim 1, characterized in that: Each step is implemented as follows: image recognition of the consistency between the current DCS page and the page corresponding to the corresponding step in the procedure; image recognition of the working equipment code in the DCS page, and comparison of the recognition result with the preset number in the procedure; identification of the current equipment status through the image recognition algorithm: image segmentation of the equipment image, use of target detection method to detect all parameters on the equipment, and comparison of the status with the preset status in the procedure; providing text prompts for the current operation steps through the display terminal; determining the staff's vote count through voice recognition: voice recognition outputs text, performs word segmentation according to the word segmentation library, and compares it with the key words in the procedure. After the key word comparison, a sequential comparison is performed. If there is no correct comparison within 60 seconds or the comparison result is wrong, a vote count error is prompted; The DCS video stream is used to identify the mouse and determine the current position. The mouse operation intention is determined by double determination. The first method is to identify the intention to operate a certain button once entering the pre-marked operation button area. The second method is to perform OCR recognition on the code of the button pointed by the mouse in three consecutive frames of images, which is recorded as double determination. If the two determination times occur within 500ms, it is determined as the current mouse operation intention to prevent OCR misjudgment. The display terminal connected through the edge reasoning terminal provides prompt results of the work steps; the device status after the operation is determined through image recognition: the corresponding parameters of the device are compared with the expected values ​​through image recognition.

8. The intelligent monitoring method for DCS operation of a nuclear power plant according to claim 6, characterized in that: Judgment of three-way communication: 1) Determine through voiceprint that different voices correspond to different staff members; 2) Voice recognition outputs text; 2) Segmentation is performed based on the word segmentation library; 3) The word segmentation of different staff members is compared with the keywords in the regulations. If there are no keywords, it indicates an abnormality; 4) If step 3) is passed, the conversation text sequence where the keyword appears for the first time is extracted, and the sequence is three sentences: one sentence from person A, one sentence from person B, and one sentence from person A; 5) Determine whether the same keyword appears in the three sentences. If so, select the next keyword and repeat step 4) until all are correct. Otherwise, it will prompt that the three-way communication is abnormal.

9. The intelligent monitoring method for DCS operation of a nuclear power plant according to claim 1, characterized in that: The identification methods of all steps are made into independent modules. If the sequence of procedure steps changes, it is only necessary to adjust the positions of multiple modules of the corresponding step operation objects in the program execution sequence or combine them.

Images