Operation guidance method, guidance system and equipment for nuclear power plant operators

Abstract

The present invention discloses a method, a guidance system and a device for guiding operators of nuclear power plants. The method comprises: obtaining a first structured feature vector according to the current operation of the operator, and obtaining a second structured feature vector according to the operation log of the nuclear power plant in a preset time period before the current moment; after combining each event in a preset event set with the first structured feature vector and the second structured feature vector respectively, inputting a preset prediction model to obtain the probability of occurrence of each event in the event set caused by the current operation; and guiding the operator to continue or abandon the current operation by using the probability of occurrence of each event. The present invention predicts the events that may be caused by each operation of the operator in real time online by combining each current operation of the operator with the operating status of the nuclear power plant before the current moment, so as to assist the operator in issuing correct control commands and improve the safety of nuclear power plant operation.

Description

Technical Field

[0001] The invention belongs to the technical field of nuclear power, and in particular relates to an operation guidance method, a guidance system and equipment for operators of a nuclear power plant. Background Art

[0002] During the startup, shutdown and accident handling of a nuclear power plant, a large number of functions, processes and equipment conditions require manual control and intervention by the operator in the main control room. If the operator is inexperienced or makes operational errors, it may cause abnormalities in the unit or even incidents that affect safe operation.

[0003] In the prior art, there is a lack of effective prediction methods for events that may be caused by the operator's operation. Whether the operation is correct is judged only based on experience or specifications, and the operational safety needs to be improved. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a nuclear power plant operator operation guidance method, guidance system and equipment in view of the above-mentioned deficiencies in the prior art, so as to realize real-time online prediction of events that may be caused by each operation of the operator, assist the operator to issue correct control commands, and improve the operation safety of the nuclear power plant.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present invention provides a method for guiding operators of a nuclear power plant, comprising:

[0007] Acquire a first structured feature vector according to the current operation of the operator, and acquire a second structured feature vector according to the operation log of the nuclear power plant within a preset time period before the current moment;

[0008] After combining each event in the preset event set with the first structured feature vector and the second structured feature vector respectively, the combination is input into a preset prediction model to obtain the probability of occurrence of each event in the event set caused by the current operation;

[0009] The occurrence probability of each event is used to guide the operator to continue or abandon the current operation.

[0010] Preferably, the prediction model is obtained by the following steps:

[0011] Each acquired nuclear power plant historical operation log is encoded after marking the event type to form a third structured feature vector;

[0012] Embed an event causal relationship mining network in the basic recurrent neural network to obtain a prediction network;

[0013] The third structured feature vector is input into the prediction network for training to obtain the prediction model.

[0014] Preferably, the step of encoding each acquired nuclear power plant historical operation log after marking the event type to form a third structured feature vector specifically includes:

[0015] Read the historical operation logs in the nuclear power plant historical operation log database, and mark the event type of each historical operation log;

[0016] Filter the annotated historical operation logs and delete the historical operation logs corresponding to the same event type that are less than a preset time interval;

[0017] Each filtered historical operation log is encoded into the third structured feature vector of a preset structure.

[0018] Preferably, the marking of the event type of each historical operation log specifically includes:

[0019] Using regular expressions to match and replace the historical operation logs to obtain historical operation logs in similar data formats;

[0020] The Levenstein distance is used to measure the similarity of any two historical operation logs in similar data formats. When the Levenstein distance is less than the preset threshold, the two historical operation logs are classified as the same event type and marked with the corresponding event type ID.

[0021] Preferably, the filtering of the annotated historical operation logs to delete the historical operation logs corresponding to the same event type that are less than a preset time interval specifically includes:

[0022] For any historical operation log, search for all other historical operation logs marked with the same event type;

[0023] The time interval between the historical operation log and the other historical operation logs found is calculated, and if any time interval is less than the preset time interval, the historical operation log is deleted.

[0024] Preferably, encoding each filtered historical operation log into the third structured feature vector of a preset structure specifically includes:

[0025] Mapping the event type annotated in each filtered historical running log into the third structured feature vector using one-hot encoding;

[0026] Mapping the text-like elements in each filtered historical operation log into the third structured feature vector using dictionary coding;

[0027] The numerical elements in each filtered historical operation log are directly mapped into the third structured feature vector.

[0028] Preferably, the basic recurrent neural network specifically adopts a long short-term memory network LSTM, and the event causal relationship mining network specifically adopts a relational memory core RMC.

[0029] Preferably, the step of inputting the third structured feature vector into the prediction network for training to obtain the prediction model specifically includes:

[0030] Combining the third structured feature vectors adjacent to each other in time order into data pairs;

[0031] Inputting the data pairs into the event causal relationship mining network in sequence to obtain a causal relationship index for each group of data pairs;

[0032] Each set of data pairs and their causal relationship index are sequentially input into the basic recurrent neural network for training to obtain the prediction model.

[0033] Preferably, after combining each event in the preset event set with the first structured feature vector and the second structured feature vector, respectively, and inputting into a preset prediction model to obtain the probability of occurrence of each event in the event set caused by the current operation, specifically includes:

[0034] Preliminarily combining all event types mapped in the third structured feature vector to form an event set;

[0035] Using the second structured feature vector and the first structured feature vector as data of the first k-1 time nodes;

[0036] Any event in the event set is used as data of the kth time node, and after being arranged and combined with the data of the first k-1 time nodes in chronological order, the data are sequentially input into the prediction model to obtain the probability of occurrence of the event caused by the current operation;

[0037] Each event in the event set is traversed to obtain the occurrence probability of each event in the event set caused by the current operation.

[0038] Preferably, the using the occurrence probability of each event to guide the operator to continue or abandon the current operation specifically includes:

[0039] Arranging and displaying each event and its probability of occurrence in the order of the probability of occurrence from high to low on the graphical user interface of the operator display;

[0040] A human-machine interaction option is provided for the operator to choose to continue or abandon the current operation, so as to receive and execute the operator's instruction to select the human-machine interaction option.

[0041] Preferably, the events specifically include: device response events and risk events;

[0042] The arranging and displaying each event and its probability of occurrence in order from high to low probability of occurrence specifically includes:

[0043] The device response events and the risk events are arranged in descending order according to their probability of occurrence, and are displayed in a preset list respectively.

[0044] In a second aspect, the present invention provides an operation guidance system for operators of a nuclear power plant, comprising:

[0045] An acquisition module, used to acquire a first structured feature vector according to the current operation of the operator, and to acquire a second structured feature vector according to the operation log of the nuclear power plant within a preset time period before the current moment;

[0046] a prediction module connected to the acquisition module, and configured to combine each event in a preset event set with the first structured feature vector and the second structured feature vector, respectively, and input the combined results into a preset prediction model to obtain the probability of occurrence of each event in the event set caused by the current operation;

[0047] The guidance module is connected to the prediction module and is used to guide the operator to continue or abandon the current operation by using the occurrence probability of each event.

[0048] Preferably, the system further comprises: a prediction model acquisition module, wherein the prediction model acquisition module specifically comprises:

[0049] An encoding unit, used for encoding each acquired historical operation log of the nuclear power plant after marking the event type to form a third structured feature vector;

[0050] A network unit for embedding an event causal relationship mining network in a basic recurrent neural network to obtain a prediction network;

[0051] A training unit is connected to the encoding unit and the network unit, and is used to input the third structured feature vector into the prediction network for training to obtain the prediction model.

[0052] Preferably, the training unit specifically includes:

[0053] a pairing subunit, used for combining the third structured feature vectors adjacent to each other in time order into data pairs;

[0054] A causal relationship subunit, connected to the pairing subunit, for sequentially inputting the data pairs into the event causal relationship mining network to obtain a causal relationship index for each group of data pairs;

[0055] The training subunit is connected to the causal relationship subunit and is used to input each group of data pairs and their causal relationship index into the basic recurrent neural network in sequence for training to obtain the prediction model.

[0056] Preferably, the prediction module specifically includes:

[0057] An event set unit, used to pre-combine all event types mapped in the third structured feature vector to form an event set;

[0058] A previous time node unit, used for taking the second structured feature vector and the first structured feature vector as data of the previous k-1 time nodes;

[0059] A prediction unit connected to the event set unit and the previous time node unit, used to take any event in the event set as data of the kth time node, and after arranging and combining with the data of the previous k-1 time nodes in chronological order, sequentially input them into the prediction model to obtain the probability of occurrence of the event caused by the current operation;

[0060] The traversal unit is connected to the prediction unit and is used to traverse each event in the event set to obtain the occurrence probability of each event in the event set caused by the current operation.

[0061] Preferably, the guidance module specifically includes:

[0062] A display unit, for arranging and displaying each event and its probability of occurrence in the order of the probability of occurrence from high to low on a graphical user interface of an operator display;

[0063] The interactive unit is connected to the display unit and is used to provide a human-computer interaction option for the operator to choose to continue or abandon the current operation, so as to receive and execute the operator's instruction to select the human-computer interaction option.

[0064] In a third aspect, the present invention provides a nuclear power plant operator operating device, comprising:

[0065] The operator operating device of a nuclear power plant is used for the operator to issue operations to realize the operation control of the nuclear power plant;

[0066] The nuclear power plant operator operation guidance system as described above is electrically connected to the nuclear power plant operator operation device and is used to guide the operator's operation.

[0067] The nuclear power plant operator operation guidance method, guidance system and equipment provided by the present invention predict the events that may be caused by each operation of the operator in real time online by combining each current operation of the operator with the operating status of the nuclear power plant before the current moment, and feed back the prediction results to the operator in time as the basis for the operator to finally issue the operation control command, so as to assist the operator to issue the correct control command and improve the safety of nuclear power plant operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a flowchart of the operation guidance method for nuclear power plant operators in an embodiment of the present invention;

[0069] Figure 2 It is a structural schematic diagram of a nuclear power plant operator operating device in an embodiment of the present invention;

[0070] Figure 3 It is a flowchart of a method for guiding operators of a nuclear power plant in another embodiment of the present invention;

[0071] Figure 4 A schematic diagram of an operation confirmation window for a nuclear power plant operator in another embodiment of the present invention;

[0072] Figure 5 The figure is a schematic diagram of the structure of the operation guidance system for operators of a nuclear power plant in an embodiment of the present invention. DETAILED DESCRIPTION

[0073] The following will be combined with the accompanying drawings in the present invention to clearly and completely describe the technical solutions in the invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of the present invention.

[0074] In the description of the present invention, it should be noted that the directions or positional relationships indicated by “upper” and the like are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience and simplification of the description. They do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they cannot be understood as limitations on the present invention.

[0075] In the description of the present invention, the terms “first” and “second” are used for descriptive purposes only and should not be understood as indicating or implying relative importance.

[0076] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "connect", "set", "install", "fix" and the like should be understood in a broad sense, for example, it can be a fixed connection or a detachable connection, or an integral connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0077] In the description of the present invention, each unit and module involved may correspond to only one physical structure, or may be composed of multiple physical structures, or multiple units and modules may be integrated into one physical structure; the units and modules involved may be implemented by software or by hardware, for example, the units and modules may be located in a processor.

[0078] In the description of the present invention, the functions and steps marked in the flowcharts and block diagrams of the present invention may occur in an order different from that marked in the drawings if there is no conflict.

[0079] In order to facilitate the understanding of the present invention, the data recording of existing nuclear power plants is first explained. As the number of nuclear power units in operation in my country has increased significantly, a large amount of experience and log data has been accumulated during nuclear power operation and simulator training, providing a data basis for risk prediction of nuclear power plants.

[0080] Based on the log data of the historical operation of the nuclear power plant and the simulator training, the present invention proposes to mine the event association in the data from the above data, establish a prediction model, and for each manual operation of the operator, combine the current operating status of the nuclear power plant and the historical operating status to predict the operation result in real time online, and feed back the prediction result to the operator in time as the basis for issuing the final control command, so that the response status and safety risks of the nuclear power plant can be predicted before the operator issues the command, and the auxiliary information can be fed back to the operator in time so that the operator can issue the correct control command according to the prediction result, which is conducive to realizing the ex ante control of risks and is of great significance to improving the safety and economy of nuclear power operation.

[0081] Embodiment 1:

[0082] like Figure 1 As shown, Embodiment 1 of the present invention provides a method for guiding operators of a nuclear power plant, which is applied to Figure 2 The nuclear power plant operator operation guidance system 1 shown, the method comprises:

[0083] S1. Acquire a first structured feature vector according to the current operation of the operator, and acquire a second structured feature vector according to the operation log of the nuclear power plant within a preset time period before the current moment.

[0084] Specifically, in this embodiment, if Figure 2 As shown, the nuclear power plant operator operation guidance system 1 and the nuclear power plant operator operation device 2 are connected to form the nuclear power plant operator operation equipment 01, which is arranged in the main control room of the nuclear power plant. Each time the operator operates the nuclear power plant operator operation device 2, the data of the current operation control command is sent to the nuclear power plant operator operation guidance system 1. The nuclear power plant operator operation guidance system 1 checks whether the current operation will cause a risk event, and analyzes and warns the possible risks in advance. Before each control command is issued, the prediction result is displayed as auxiliary information on the display of the nuclear power plant operator operation device 2 in time, so that the operator can make a second confirmation of the current operation control command to reduce the possibility of operator misoperation and guide the operator to issue safe and reliable control commands to the nuclear power unit.

[0085] More specifically, if Figure 3 As shown, since each operation of the operator may trigger an event that is related to both the current operation and the current operating status of the nuclear power plant, the operating status can be obtained based on the operating logs within a period of time before the current moment. Therefore, in this embodiment, when the nuclear power plant operator operation guidance system 1 receives the operator's current operation issued by the nuclear power plant operator operating device 2, a first structured feature vector is generated for the current operation, and at the same time, the operating logs within the previous n hours stored in the nuclear power plant operation log database are retrieved to generate a second structured feature vector, so as to obtain the continuous operating status of the nuclear power plant within n hours before the current moment, as a calculation basis for the subsequent prediction of events that will occur in the nuclear power plant at the next moment.

[0086] S12: After combining each event in the preset event set with the first structured feature vector and the second structured feature vector respectively, the combination is input into a preset prediction model to obtain the occurrence probability of each event in the event set caused by the current operation.

[0087] Specifically, in this embodiment, if Figure 3 As shown, in order to predict the events that will occur in the nuclear power plant at the next moment, the probability of various common important events in the nuclear power plant is calculated to determine. First, a closed event set needs to be defined in advance, in which various important events that may occur in the nuclear power plant are recorded, especially risk events, that is, the event set corresponds to at least a list of risk events, which can be obtained based on past experience and data. Then, by combining each event in the event set with the first structured feature vector and the second structured feature vector, and then inputting the preset prediction model, the operator's current operation may trigger the probability of each event at the next moment through the continuous operation status of the nuclear power plant within n hours before the current moment. Events with a high probability of occurrence are more likely to be triggered at the next moment.

[0088] In an optional embodiment, the prediction model is obtained by the following steps:

[0089] Each acquired nuclear power plant historical operation log is encoded after marking the event type to form a third structured feature vector;

[0090] Embed an event causal relationship mining network in the basic recurrent neural network to obtain a prediction network;

[0091] The third structured feature vector is input into the prediction network for training to obtain the prediction model.

[0092] Specifically, in this embodiment, if Figure 3 As shown in the figure, in order to obtain the probability of occurrence of each event that may be caused by the operator's current operation, it is necessary to first obtain a prediction model. This prediction model constructs the operation log in the nuclear power plant instrumentation and control system or simulator into a third structured feature vector for machine learning, and tries to retain necessary information in the characterization process to achieve the characterization expression of historical log data; establishes a causal relationship analysis and risk event prediction network based on a recurrent neural network (RNN), utilizes the information mining ability of the recurrent neural network for time series data, takes the third structured feature vector as the data basis, encodes the causal law of the nuclear power plant state into time series information, and stores it in the hidden state of the recurrent neural network, and adaptively mines the causal relationship between the operator's operation and the equipment response and risk events of the nuclear power plant; fully applies the causal relationship analysis results, combined with the continuous operation status of the nuclear power plant within n hours before the current moment, to obtain a reasonable prediction of the equipment response and risk events of the nuclear power plant at the next moment.

[0093] In an optional embodiment, the encoding of each acquired nuclear power plant historical operation log after marking the event type to form a third structured feature vector specifically includes:

[0094] Read the historical operation logs in the nuclear power plant historical operation log database, and mark the event type of each historical operation log;

[0095] Filter the annotated historical operation logs and delete the historical operation logs corresponding to the same event type that are less than a preset time interval;

[0096] Each filtered historical operation log is encoded into the third structured feature vector of a preset structure.

[0097] Specifically, in this embodiment, if Figure 3As shown in the figure, the historical operation log can be read from the nuclear power plant historical operation log database. In order to construct the operation log in the nuclear power plant instrumentation and control system or simulator into the third structured feature vector, the historical operation log needs to be preprocessed, including: first, identifying similar log information, classifying similar types of events and marking them with the same ID, each ID indicating a different event type; second, the log data needs to be filtered. There are a large number of redundant records in the operation of the nuclear power plant, and some parameters and fault information are recorded repeatedly. Therefore, the redundant information needs to be filtered to reduce the space requirement for model training. Finally, the text content can be converted into structured data to form a third structured feature vector that can be used for training.

[0098] In an optional embodiment, marking the event type of each historical operation log specifically includes:

[0099] Using regular expressions to match and replace the historical operation logs to obtain historical operation logs in similar data formats;

[0100] The Levenstein distance is used to measure the similarity of any two historical operation logs in similar data formats. When the Levenstein distance is less than the preset threshold, the two historical operation logs are classified as the same event type and marked with the corresponding event type ID.

[0101] Specifically, in this embodiment, the operation log is usually text data, consisting of a series of words and numbers, which records the operating status of each device in the nuclear power system and the variable parameters of the instrument, as well as the semantic information describing the event. Due to the differences in its source and recording method, each historical operation log has a large difference in the number of parameters, location, event description method, etc. Therefore, in order to classify and label the event types in the nuclear power plant, first use regular expressions to replace the variables in the text data, match the variables and parameters in the text data and replace them with the same symbol to improve the similarity of the text format. Then use the Levenstein distance to measure the similarity of any two replaced texts. The smaller the Levenstein distance, the higher the similarity of the two log texts. A similarity threshold is defined. If the Levenstein distance is less than the threshold, the two text messages are classified as the same type of event and marked as the same event type ID.

[0102] In an optional embodiment, filtering the annotated historical operation logs to delete the historical operation logs corresponding to the same event type that are less than a preset time interval specifically includes:

[0103] For any historical operation log, search for all other historical operation logs marked with the same event type;

[0104] The time interval between the historical operation log and the other historical operation logs found is calculated, and if any time interval is less than the preset time interval, the historical operation log is deleted.

[0105] Specifically, in this embodiment, after the events are categorized and labeled, the redundant log data is filtered out to remove the same events that are recorded repeatedly in a short period of time. The specific method is that for any historical operation log, the remaining logs labeled with the same event type are searched, the time interval between the two is calculated and a threshold is set, and if the time interval is less than the threshold, the data is deleted, otherwise the data is retained.

[0106] In an optional embodiment, encoding each filtered historical operation log into the third structured feature vector of a preset structure specifically includes:

[0107] Mapping the event type annotated in each filtered historical running log into the third structured feature vector using one-hot encoding;

[0108] Mapping the text-like elements in each filtered historical operation log into the third structured feature vector using dictionary coding;

[0109] The numerical elements in each filtered historical operation log are directly mapped into the third structured feature vector.

[0110] Specifically, in this embodiment, the annotated and filtered historical operation logs are encoded to obtain a third structured feature vector for network training, and each log corresponds to a feature vector. The event type is mapped to the feature vector using one-hot encoding, text elements are encoded as discrete values ​​using dictionary mapping, and numerical elements are directly mapped to values ​​in the vector. It should be noted that Figure 3 As shown, the method of obtaining the first structured feature vector and the second structured feature vector is similar to this, but because the operator's current operation and the operation logs within the previous n hours have the same source and the same format, no preprocessing is required.

[0111] In an optional embodiment, the basic recurrent neural network specifically adopts a long short-term memory network LSTM, and the event causal relationship mining network specifically adopts a relational memory core RMC.

[0112] Specifically, in this embodiment, if Figure 3As shown in the figure, the event prediction model of the nuclear power plant embeds the event causal relationship mining network in the basic recurrent neural network. The basic recurrent neural network is based on LSTM (long-short term memory), which takes the output of the hidden layer of the previous time node as the input of the current time node, updates it, and outputs it to the next time node. The causal relationship mining network is based on RMC (Relational Memory Core), which is used to calculate the causal relationship index between two log data, measure the degree of causal association between the two data, and adaptively mine the causal laws between log data.

[0113] In an optional embodiment, inputting the third structured feature vector into the prediction network for training to obtain the prediction model specifically includes:

[0114] Combining the third structured feature vectors adjacent to each other in time order into data pairs;

[0115] Inputting the data pairs into the event causal relationship mining network in sequence to obtain a causal relationship index for each group of data pairs;

[0116] Each set of data pairs and their causal relationship index are sequentially input into the basic recurrent neural network for training to obtain the prediction model.

[0117] Specifically, in this embodiment, in the prediction network of the event causal relationship mining network embedded in the basic recurrent neural network, at each time node, two data with adjacent occurrence times in the log sequence are taken as a set of data pairs, which are input as time-sorted log data pairs, and the feature vectors of the two data are input into the RMC network to obtain the association matrix, which is processed to obtain the quantitative causal relationship index. The data pairs and causal relationship index are input into the LSTM network, and the input needs to meet the strict time sequence. If the log data within a period of time is X, Y, Z in order of time, R XY To represent the causal relationship between X and Y data, we need to use {X,Y,R XY},{Y,Z,R YZ}, the data pairs are input into the LSTM-based basic recurrent neural network to train the prediction network to obtain a trained prediction model. Furthermore, in order to reduce the space complexity of model training, the data can be divided into several fixed-size subsets and input into the model for iterative optimization of parameters, and the data set can be traversed multiple times for training.

[0118] In an optional embodiment, after combining each event in the preset event set with the first structured feature vector and the second structured feature vector, respectively, and inputting the combined data into a preset prediction model to obtain the probability of occurrence of each event in the event set caused by the current operation, specifically includes:

[0119] Preliminarily combining all event types mapped in the third structured feature vector to form an event set;

[0120] Using the second structured feature vector and the first structured feature vector as data of the first k-1 time nodes;

[0121] Any event in the event set is used as data of the kth time node, and after being arranged and combined with the data of the first k-1 time nodes in chronological order, the data are sequentially input into the prediction model to obtain the probability of occurrence of the event caused by the current operation;

[0122] Each event in the event set is traversed to obtain the occurrence probability of each event in the event set caused by the current operation.

[0123] Specifically, in this embodiment, the trained network can be used for event prediction. After marking the event type from the historical operation log, an event set is formed from the marked historical events. The occurrence probabilities of all events in the event set are calculated, and all possible events are sent to the prediction network one by one together with the data at the current time and before the current time, and the causal relationship index is output as the occurrence probability of these events. More specifically, the operator's current moment operation is encoded as a first structured feature vector, and the operation log within the previous n hours is found and encoded as a second structured feature vector. The second structured feature vector and the first structured feature vector are jointly input into the first k-1 time nodes of the prediction model in chronological order, and the entire event set is traversed, and each time an event in the event set is used as the input of the kth time node, and the output of each event in the event set at the kth time node is calculated in sequence, which represents the possible occurrence probability of the specific event. Furthermore, for example, the second structured feature vector is sorted by time as a and b, the first structured feature vector is c, and the event set is {E1, E2, E3}, then the data needs to be input into the prediction model in the order of {a, b, c, E1}. The causal relationship index between the two data can be obtained at the output layer at the current time point, and the hidden layer saves the state information of the previous time, integrates the causal relationship index, updates the event state and passes it to the next time node, that is, outputs the causal relationship index R in sequence. ab , R bc , R cE1 , where the causality index R of the last set of data pairs is cE1That is the probability of event E1. Then, the probability of event E2 and E3 is predicted in the same way, and the probability of each event in the event set is obtained.

[0124] S13. Use the occurrence probability of each event to guide the operator to continue or abandon the current operation.

[0125] Specifically, in this embodiment, if Figure 3 As shown, after obtaining the prediction result of the occurrence probability of each event, it is necessary to feed back the prediction result to the operator in a timely manner. The present invention provides feedback in the form of a confirmation window. After the operator gives the current operation and before issuing each control command, a secondary confirmation prompt pops up, and at the same time displays a visual interface of the risk prediction result, presenting clear auxiliary judgment results and interactive logic, helping the operator to quickly understand the operation risks and issue control commands quickly.

[0126] In an optional embodiment, the using the probability of occurrence of each event to guide the operator to continue or abandon the current operation specifically includes:

[0127] Arranging and displaying each event and its probability of occurrence in the order of the probability of occurrence from high to low on the graphical user interface of the operator display;

[0128] A human-machine interaction option is provided for the operator to choose to continue or abandon the current operation, so as to receive and execute the operator's instruction to select the human-machine interaction option.

[0129] Specifically, in this embodiment, if Figure 4 As shown in the figure, in the prediction result feedback link, the graphical user interface and human-computer interaction logic of the nuclear power plant safety risk prediction system are designed, all events in the event set are sorted according to the probability of occurrence, and several events that are most likely to occur are arranged and displayed. At the same time, the operator is asked whether to continue the current operation. If he chooses to continue, the control command will continue to be executed. If he chooses to return, the execution will be abandoned, the secondary confirmation window will be closed, and the previous menu will be returned, so as to guide the operator to operate correctly and reduce the possibility of operator error operation.

[0130] In an optional embodiment, the events specifically include: device response events and risk events;

[0131] The arranging and displaying each event and its probability of occurrence in order from high to low probability of occurrence specifically includes:

[0132] The device response events and the risk events are arranged in descending order according to their probability of occurrence, and are displayed in a preset list respectively.

[0133] Specifically, in this embodiment, if Figure 4As shown, the predicted equipment response events and safety risk events in each link of the nuclear power plant are displayed in the form of a list. When the operator performs each operation, the prediction model predicts the probability of all possible equipment responses and risk events. In the pop-up secondary confirmation window, all equipment responses and risk events predicted based on the operator's current operation and their probability of occurrence are displayed to help the operator make judgments.

[0134] Embodiment 2:

[0135] like Figure 5 As shown, Embodiment 2 of the present invention provides a nuclear power plant operator operation guidance system 1, comprising:

[0136] An acquisition module 11 is used to acquire a first structured feature vector according to the current operation of the operator, and to acquire a second structured feature vector according to the operation log of the nuclear power plant within a preset time period before the current moment;

[0137] A prediction module 12, connected to the acquisition module 11, for combining each event in a preset event set with the first structured feature vector and the second structured feature vector, respectively, and inputting the combined data into a preset prediction model to obtain the probability of occurrence of each event in the event set caused by the current operation;

[0138] The guidance module 13 is connected to the prediction module 12 and is used to guide the operator to continue or abandon the current operation by using the occurrence probability of each event.

[0139] In an optional embodiment, the nuclear power plant operator operation guidance system 1 further includes: a prediction model acquisition module, and the prediction model acquisition module specifically includes:

[0140] An encoding unit, used for encoding each acquired historical operation log of the nuclear power plant after marking the event type to form a third structured feature vector;

[0141] A network unit for embedding an event causal relationship mining network in a basic recurrent neural network to obtain a prediction network;

[0142] A training unit is connected to the encoding unit and the network unit, and is used to input the third structured feature vector into the prediction network for training to obtain the prediction model.

[0143] In an optional embodiment, the encoding unit specifically includes:

[0144] The marking subunit is used to read the historical operation logs in the nuclear power plant historical operation log database and mark the event type of each historical operation log;

[0145] A filtering subunit, connected to the marking subunit, for filtering the marked historical operation logs and deleting the historical operation logs corresponding to the same event type that are less than a preset time interval;

[0146] The encoding subunit is connected to the filtering subunit and is used to encode each filtered historical operation log into the third structured feature vector of a preset structure.

[0147] In an optional embodiment, the marking subunit specifically includes:

[0148] A matching and replacing subunit, used for matching and replacing the historical operation logs using regular expressions to obtain historical operation logs in similar data formats;

[0149] The similarity measurement subunit is connected to the matching and replacement subunit, and is used to use the Levenshtein distance to measure the similarity of any two historical operation logs in similar data formats. When the Levenshtein distance is less than a preset threshold, the two historical operation logs are classified as the same event type and marked with the corresponding event type ID.

[0150] In an optional embodiment, the filtering subunit specifically includes:

[0151] The search subunit is used to search for all other historical operation logs marked with the same event type for any historical operation log;

[0152] The deletion subunit is connected to the search subunit and is used to calculate the time interval between the historical operation log and the other historical operation logs searched, and if any time interval is less than the preset time interval, the historical operation log is deleted.

[0153] In an optional embodiment, the encoding subunit specifically includes:

[0154] A first mapping subunit, configured to map the event type annotated by each filtered historical operation log into the third structured feature vector by using one-hot encoding;

[0155] A second mapping subunit, used to map the text elements in each filtered historical operation log into the third structured feature vector using dictionary coding;

[0156] The third mapping subunit is used to directly map the numerical elements in each filtered historical operation log into the third structured feature vector.

[0157] In an optional embodiment, in the network unit, the basic recurrent neural network specifically adopts a long short-term memory network LSTM, and the event causal relationship mining network specifically adopts a relational memory core RMC.

[0158] In an optional embodiment, the training unit specifically includes:

[0159] a pairing subunit, used for combining the third structured feature vectors adjacent to each other in time order into data pairs;

[0160] A causal relationship subunit, connected to the pairing subunit, for sequentially inputting the data pairs into the event causal relationship mining network to obtain a causal relationship index for each group of data pairs;

[0161] The training subunit is connected to the causal relationship subunit and is used to input each group of data pairs and their causal relationship index into the basic recurrent neural network in sequence for training to obtain the prediction model.

[0162] In an optional embodiment, the prediction module 12 specifically includes:

[0163] An event set unit, used to pre-combine all event types mapped in the third structured feature vector to form an event set;

[0164] A previous time node unit, used for taking the second structured feature vector and the first structured feature vector as data of the previous k-1 time nodes;

[0165] A prediction unit connected to the event set unit and the previous time node unit, used to take any event in the event set as data of the kth time node, and after arranging and combining with the data of the previous k-1 time nodes in chronological order, sequentially input them into the prediction model to obtain the probability of occurrence of the event caused by the current operation;

[0166] The traversal unit is connected to the prediction unit and is used to traverse each event in the event set to obtain the occurrence probability of each event in the event set caused by the current operation.

[0167] In an optional embodiment, the guidance module 13 specifically includes:

[0168] A display unit, for arranging and displaying each event and its probability of occurrence in the order of the probability of occurrence from high to low on a graphical user interface of an operator display;

[0169] The interactive unit is connected to the display unit and is used to provide a human-computer interaction option for the operator to choose to continue or abandon the current operation, so as to receive and execute the operator's instruction to select the human-computer interaction option.

[0170] In an optional embodiment, the events specifically include: device response events and risk events;

[0171] The display subunit specifically includes:

[0172] A device response display subunit, used to arrange the device response events in descending order of their occurrence probability and display them in a preset list;

[0173] The risk event display subunit is used to arrange the risk events in order of their occurrence probability from high to low, and then display them in another preset list.

[0174] Embodiment 3:

[0175] like Figure 2 As shown, Embodiment 3 of the present invention provides a nuclear power plant operator operating device 01, comprising:

[0176] The nuclear power plant operator operating device 2 is used for the operator to issue operations to realize the operation control of the nuclear power plant;

[0177] The nuclear power plant operator operation guidance system 1 as described in Example 2 is electrically connected to the nuclear power plant operator operation device 2 and is used to guide the operator's operation.

[0178] The nuclear power plant operator operation guidance method, guidance system and equipment provided in embodiments 1-3 of the present invention predict the events that may be caused by each operation of the operator in real time online by combining each current operation of the operator with the operating status of the nuclear power plant before the current moment, and promptly feed back the prediction results to the operator as the basis for the operator to finally issue the operation control command, so as to assist the operator to issue the correct control command and improve the safety of nuclear power plant operation.

[0179] 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 guiding operators of a nuclear power plant, characterized in that: include: Acquire a first structured feature vector according to the current operation of the operator, and acquire a second structured feature vector according to the operation log of the nuclear power plant within a preset time period before the current moment; After combining each event in the preset event set with the first structured feature vector and the second structured feature vector respectively, inputting the preset prediction model to obtain the probability of occurrence of each event in the event set caused by the current operation at the next moment, specifically includes: The second structured feature vector is sorted as the data of the first k-2 time nodes according to time, the first structured feature vector is used as the data of the k-1th time node, and an event in the event set is used as the data of the kth time node, and is input into the prediction model, The prediction model sequentially outputs the causal relationship index of the data of each two adjacent time nodes, and fuses the causal relationship index output by the output layer at each time point in the hidden layer and passes it to the next time point, until the causal relationship index of the data of the k-1th time node and the data of the kth time node is output, which is the probability of the current operation causing the event at the next moment. Traversing each event in the event set, and obtaining the probability of the current operation triggering each event in the event set at the next moment; Using the occurrence probability of each event to guide the operator to continue or abandon the current operation; The prediction model is specifically obtained through the following steps: Each acquired nuclear power plant historical operation log is encoded after marking the event type to form a third structured feature vector; Embed an event causal relationship mining network in the basic recurrent neural network to obtain a prediction network; The third structured feature vector is input into the prediction network for training to obtain the prediction model.

2. The method according to claim 1, characterized in that The method of encoding each acquired nuclear power plant historical operation log after marking the event type to form a third structured feature vector specifically includes: Read the historical operation logs in the nuclear power plant historical operation log database, and mark the event type of each historical operation log; Filter the annotated historical operation logs and delete the historical operation logs corresponding to the same event type that are less than a preset time interval; Each filtered historical operation log is encoded into the third structured feature vector of a preset structure.

3. The method according to claim 2, characterized in that The event type of each historical operation log is marked, specifically including: Using regular expressions to match and replace the historical operation logs to obtain historical operation logs in similar data formats; The Levenstein distance is used to measure the similarity of any two historical operation logs in similar data formats. When the Levenstein distance is less than the preset threshold, the two historical operation logs are classified as the same event type and marked with the corresponding event type ID.

4. The method according to claim 2, characterized in that: The filtering of the annotated historical operation logs to delete the historical operation logs corresponding to the same event type that are less than a preset time interval specifically includes: For any historical operation log, search for all other historical operation logs marked with the same event type; The time interval between the historical operation log and the other historical operation logs found is calculated, and if any time interval is less than the preset time interval, the historical operation log is deleted.

5. The method according to claim 2, characterized in that: The step of encoding each filtered historical operation log into the third structured feature vector of a preset structure specifically includes: Mapping the event type annotated in each filtered historical running log into the third structured feature vector using one-hot encoding; Mapping the text-like elements in each filtered historical operation log into the third structured feature vector using dictionary coding; The numerical elements in each filtered historical operation log are directly mapped into the third structured feature vector.

6. The method according to claim 1, characterized in that The basic recurrent neural network specifically adopts the long short-term memory network LSTM, and the event causal relationship mining network specifically adopts the relational memory core RMC.

7. The method according to any one of claims 1 to 6, characterized in that: The step of inputting the third structured feature vector into the prediction network for training to obtain the prediction model specifically includes: Combining the third structured feature vectors adjacent to each other in time order into data pairs; Inputting the data pairs into the event causal relationship mining network in sequence to obtain a causal relationship index for each group of data pairs; Each set of data pairs and their causal relationship index are sequentially input into the basic recurrent neural network for training to obtain the prediction model.

8. The method according to claim 7, characterized in that Before combining each event in the preset event set with the first structured feature vector and the second structured feature vector respectively and inputting the combined data into a preset prediction model to obtain the probability of occurrence of each event in the event set caused by the current operation, the method further includes: All event types mapped in the third structured feature vector are combined in advance to form an event set.

9. The method according to claim 1, characterized in that: The using the occurrence probability of each event to guide the operator to continue or abandon the current operation specifically includes: Arranging and displaying each event and its probability of occurrence in the order of the probability of occurrence from high to low on the graphical user interface of the operator display; A human-machine interaction option is provided for the operator to choose to continue or abandon the current operation, so as to receive and execute the operator's instruction to select the human-machine interaction option.

10. The method according to claim 9, characterized in that The events specifically include: device response events and risk events; The arranging and displaying each event and its probability of occurrence in order from high to low probability of occurrence specifically includes: The device response events and the risk events are arranged in descending order according to their probability of occurrence, and are displayed in a preset list respectively.

11. A nuclear power plant operator operation guidance system, characterized in that: include: An acquisition module, used to acquire a first structured feature vector according to the current operation of the operator, and to acquire a second structured feature vector according to the operation log of the nuclear power plant within a preset time period before the current moment; A prediction module, connected to the acquisition module, is used to combine each event in the preset event set with the first structured feature vector and the second structured feature vector, respectively, and input the combination into a preset prediction model to obtain the probability of occurrence of each event in the event set caused by the current operation at the next moment, specifically used to: The second structured feature vector is sorted as the data of the first k-2 time nodes according to time, the first structured feature vector is used as the data of the k-1th time node, and an event in the event set is used as the data of the kth time node, and is input into the prediction model, The prediction model sequentially outputs the causal relationship index of the data of each two adjacent time nodes, and fuses the causal relationship index output by the output layer at each time point in the hidden layer and passes it to the next time point, until the causal relationship index of the data of the k-1th time node and the data of the kth time node is output, which is the probability of the current operation causing the event at the next moment. Traversing each event in the event set, and obtaining the probability of the current operation triggering each event in the event set at the next moment; a guidance module, connected to the prediction module, for guiding the operator to continue or abandon the current operation by using the probability of occurrence of each event; The system further includes: a prediction model acquisition module, which specifically includes: An encoding unit, used for encoding each acquired historical operation log of the nuclear power plant after marking the event type to form a third structured feature vector; A network unit for embedding an event causal relationship mining network in a basic recurrent neural network to obtain a prediction network; A training unit is connected to the encoding unit and the network unit, and is used to input the third structured feature vector into the prediction network for training to obtain the prediction model.

12. The system according to claim 11, characterized in that The training unit specifically includes: a pairing subunit, used for combining the third structured feature vectors adjacent to each other in time order into data pairs; A causal relationship subunit, connected to the pairing subunit, for sequentially inputting the data pairs into the event causal relationship mining network to obtain a causal relationship index for each group of data pairs; The training subunit is connected to the causal relationship subunit and is used to input each group of data pairs and their causal relationship index into the basic recurrent neural network in sequence for training to obtain the prediction model.

13. The system according to claim 12, characterized in that The system further comprises: The event set unit is used to pre-combine all event types mapped in the third structured feature vector to form an event set.

14. The system according to claim 11, characterized in that The guidance module specifically includes: A display unit, for arranging and displaying each event and its probability of occurrence in the order of the probability of occurrence from high to low on a graphical user interface of an operator display; The interactive unit is connected to the display unit and is used to provide a human-computer interaction option for the operator to choose to continue or abandon the current operation, so as to receive and execute the operator's instruction to select the human-computer interaction option.

15. A nuclear power plant operator operating device, characterized in that: include: The operator operating device of a nuclear power plant is used for the operator to issue operations to realize the operation control of the nuclear power plant; The nuclear power plant operator operation guidance system as described in any one of claims 11 to 14 is electrically connected to the nuclear power plant operator operation device and is used to guide the operator's operation.

Images