Nuclear power plant equipment predictive evaluation and maintenance method and system based on self-attention

Abstract

The invention relates to the technical field of equipment life prediction, and provides a nuclear power plant equipment predictive evaluation and maintenance method and system based on self-attention, and the method comprises the steps: obtaining multi-source sensing data of equipment through a sliding window mechanism, and obtaining an input time window sequence after preprocessing; reasoning the time window sequence data through a model to obtain a state vector for regression analysis, and generating the residual service life of the equipment; performing visualization processing on the attention matrix to obtain an equipment health state interpretation map of the attention weight, and generating interpretation information; and fusing the obtained residual service life of the equipment and the corresponding interpretation information to generate an equipment maintenance strategy suggestion. According to the method, the sliding window and the Transform model are creatively fused, deep modeling and interpretability analysis of multi-source equipment data are achieved, and the equipment health state atlas is constructed based on an attention mechanism and used for accurate prediction and maintenance strategy generation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment life prediction, in particular to a nuclear power plant equipment predictive evaluation maintenance method and system based on self-attention. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] In high-safety-level industrial scenarios such as nuclear power plants, the health status of key equipment is directly related to the safety, reliability and economy of the system. The remaining useful life (RUL) prediction technology of equipment, as an important means to realize intelligent maintenance and predictive maintenance, can significantly improve the efficiency of equipment operation and maintenance, reduce the risk of unplanned downtime, and promote the strategic transformation from traditional post-maintenance to predictive maintenance. There are special challenges in data collection and processing in nuclear power plants: many monitoring points, long data transmission link and high collection frequency, resulting in large data volume, high real-time requirement and strict processing accuracy requirement, which puts higher requirements on the deployment architecture and response efficiency of the RUL prediction technology.

[0004] The current mainstream remaining useful life (RUL) prediction methods mainly include physical modeling, statistical analysis and deep learning based paths. Among them, the physical modeling method relies on the mathematical description of the degradation mechanism of the equipment, which is often difficult to cover the multi-source coupling and strong nonlinear degradation behavior in complex systems, and has limited applicability and scalability; the statistical analysis method shows insufficient modeling capability when facing high-dimensional variables, multi-condition disturbance and nonlinear degradation trend, and the prediction accuracy is unstable; while the deep learning model (such as LSTM) has made progress in sequence modeling, but still has weak modeling ability for long-term time series dependence, is prone to gradient disappearance in the training process, and has poor result interpretability, which is difficult to meet the transparency and trustworthiness requirements of high-safety-level industrial systems. In addition, most of the existing RUL prediction schemes assume continuous data, clear structure and stable processing environment, which is difficult to adapt to the industrial characteristics of long-distance sensor deployment, unstable data link, high-frequency update of massive time series data and other industrial characteristics in nuclear power plants and other practical application scenarios, and lack the deployment ability and processing efficiency of coordination with the distributed control system (Distributed Control System, DCS). SUMMARY

[0005] In order to solve the above problems, the application provides a nuclear power plant equipment predictive evaluation maintenance method and system based on self-attention, which innovatively combines sliding window and Transformer model, realizes deep modeling and explainability analysis of multi-source equipment data, and constructs an equipment health state atlas based on an attention mechanism, thereby realizing equipment life prediction and maintenance.

[0006] In order to achieve the above purpose, the application adopts the following technical solutions: One or more embodiments provide a nuclear power plant equipment predictive evaluation maintenance method based on self-attention, including the following steps: A sliding window mechanism is used to obtain multi-source sensing data of the nuclear power plant equipment, and the preprocessed data is constructed into an input time window sequence; The time window sequence data is inferred by a Transformer model to obtain a state vector for regression analysis, thereby generating equipment remaining useful life; The attention matrix obtained in the inference process of the attention mechanism of the Transformer model is visualized to obtain an equipment health state explanation atlas of attention weights; and a dynamic knowledge graph is constructed to generate explanation information for abnormal deviation points in the equipment health state explanation atlas; The obtained equipment remaining useful life and corresponding explanation information are fused to generate equipment maintenance strategy suggestions.

[0007] One or more embodiments provide a nuclear power plant equipment predictive evaluation maintenance system based on self-attention, including: A sliding window unit configured to use a sliding window mechanism to obtain multi-source sensing data of the nuclear power plant equipment, and to construct an input time window sequence after preprocessing; An inference unit configured to infer time window sequence data by a Transformer model to obtain a state vector for regression analysis, thereby generating equipment remaining useful life; An explanation unit configured to visualize the attention matrix obtained in the inference process of the attention mechanism of the Transformer model to obtain an equipment health state explanation atlas of attention weights; and to construct a dynamic knowledge graph to generate explanation information for abnormal deviation points in the equipment health state explanation atlas; A fusion unit configured to fuse the obtained equipment remaining useful life and corresponding explanation information to generate equipment maintenance strategy suggestions.

[0008] Compared with the prior art, the application has the following beneficial effects: The application overcomes the problems of insufficient modeling capability, poor nonlinear feature processing and poor interpretability in traditional RUL prediction technology, and improves the prediction accuracy and reliability. By deeply integrating the dynamic sliding window mechanism and the Transformer model, the structural stability and time consistency management of multi-dimensional time series input data are realized, the perception ability of the model to long sequence and multi-variable equipment degradation trend is significantly improved, and the prediction stability and robustness are enhanced. The attention matrix visualization technology is introduced to realize the transparency of the prediction process, the system can automatically identify the key time period and key feature channel in the input data, realize the self-explaining modeling of the equipment abnormal state and its evolution mode, and enhance the transparency and expert understanding of the model output.

[0009] The advantages of the present application and the advantages of additional aspects will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0010] The accompanying drawings, which form a part of this specification, are included to provide a further understanding of the application, and are incorporated in and constitute a part of this specification. The embodiments of these drawings are set forth to explain the present application and are not limiting of the present application.

[0011] Figure 1 is a flow chart block diagram of the equipment predictive maintenance and life assessment method of embodiment 1 of the present application; Figure 2 is a schematic diagram of the edge-center joint modeling mechanism in the equipment predictive maintenance and life assessment method of embodiment 1 of the present application; Figure 3 is a schematic diagram of the inference process based on the Transformer model of embodiment 1 of the present application; Figure 4 is a flow chart of the equipment health state interpretation atlas generation method of embodiment 1 of the present application. DETAILED DESCRIPTION

[0012] The present application will be further described below in conjunction with the drawings and embodiments.

[0013] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the present application belongs.

[0014] It should be noted that the terminology used herein is for describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. It should be noted that, without conflict, the various embodiments and features within those embodiments can be combined with each other. The embodiments will now be described in detail with reference to the accompanying drawings.

[0015] Example 1 In one or more of the technical solutions disclosed in the embodiments, such as Figures 1 to 4 As shown, the predictive assessment and maintenance method for nuclear power plant equipment based on self-attention includes the following steps: Step 1: Use a sliding window mechanism to acquire multi-source sensor data from nuclear power plant equipment, and construct an input time window sequence after preprocessing; Step 2: Use the Transformer model to infer the time window sequence data, obtain the state vector, perform regression analysis, and generate the remaining service life of the equipment; Step 3: Visualize the attention matrix obtained during the inference process of the Transformer model's attention mechanism to obtain a device health status interpretation graph with attention weights; construct a dynamic knowledge graph to generate explanatory information for abnormal deviations in the device health status interpretation graph; Step 4: Combine the obtained remaining service life of the equipment and its corresponding explanatory information to generate equipment maintenance strategy recommendations; In this embodiment, the problems of insufficient modeling capability, poor nonlinear feature processing and poor interpretability in traditional RUL prediction technology are overcome, and the prediction accuracy and reliability are improved. By deeply integrating the dynamic sliding window mechanism with the Transformer model, the structure stability and time consistency management of multi-dimensional time series input data are realized, which significantly improves the perception ability of the model to long sequence and multi-variable equipment degradation trend, and enhances the prediction stability and robustness. The attention matrix visualization technology is introduced to realize the transparency of the prediction process, and the system can automatically identify the key time period and key feature channel in the input data, realize self-explaining modeling of equipment abnormal state and its evolution mode, and enhance the transparency and expert interpretability of the model output. Further, by constructing a device health state explanation atlas integrating trend curve, abnormal frequency and risk level distribution, the current state and historical evolution path of the device are presented comprehensively, and the auditability and auxiliary decision-making ability in engineering scenarios are enhanced. Combined with the RUL prediction result and the explanation information, a maintenance strategy is formulated to realize the refinement and intelligentization of predictive maintenance, which is especially suitable for high safety level industrial scenarios such as nuclear power plants with high frequency data update, numerous devices and complex state, and improves the operation efficiency and safety of the system.

[0016] In step 1, a real-time sliding input implementation method for equipment remaining useful life (RUL) prediction is provided, which is suitable for the inference input construction of Transformer type model, especially suitable for high frequency, remote and unstable data collection environment in nuclear power plant, and has the characteristics of strong real-time, high anti-interference ability and good prediction continuity; Multi-source sensor data are collected by field sensors, including real-time collection of multi-dimensional operating parameters from nuclear power plant equipment, including temperature, vibration, current, flow, pressure and other operating parameter data, which are transmitted to edge computing devices through industrial communication protocols (such as OPC UA, Modbus); Nuclear power plant equipment, including but not limited to main pump, cooling system, generator, etc.; The data processing in step 1 is realized in the edge computing device, which is deployed in the DCS node or industrial control rack of the nuclear power plant, has high performance data processing and low delay response ability, and is responsible for receiving, caching and preprocessing field sensor data to ensure data integrity and real-time performance.

[0017] In the DCS edge side of the nuclear power plant, the traditional fixed-beat FIFO sliding window will mix different working conditions such as start-stop, power-up, steady state and SCRAM in the same window when processing sensor data stream, resulting in working condition aliasing and feature drift, thereby reducing the discrimination and robustness of abnormal detection and remaining useful life (RUL) modeling.

[0018] To address the aforementioned issues, this embodiment proposes an event-aligned sliding window construction method. Its core lies in introducing operating condition signal recognition events as window division trigger boundaries to achieve dynamic window alignment and segmentation sampling.

[0019] Specifically, the sliding window mechanism for acquiring multi-source sensor data from a device can be configured to be implemented in an edge device, such as... Figure 2 As shown, it includes the following steps: Step 11: Synchronously monitor the received multi-source sensor data, DCS operating condition signals, and interlocking signals to identify events; DCS operating condition signals and interlock signals identify events including equipment or system start-up and shutdown, load increase and decrease, circuit or valve position switching, interlock action and protection action, etc. Specifically, after timestamp-based dejitting and merging, the identified events of DCS operating condition signals and interlocking signals are recorded as follows: ; in, For event timestamps; Event types include start / stop, power increase / decrease, circuit switching, interlocking, and protection actions.

[0020] Step 12: According to the identified event occurrence points Based on the event type, the time period before and after the event occurrence point is set as the transition window, and the time period after the event occurs is divided into the steady state window. The transition window is smaller than the steady state window. The size of the transition window and the steady state window are set to different sizes according to the event type. The time window without events is set as the fixed window, resulting in a dynamically changing time window. Specifically, when an event is identified At that time, two types of windows are generated based on event alignment: Transition window: With a higher sampling rate Resample the entire channel process to capture rapid changes before and after an event; Steady-state window: With a lower sampling rate It collects and extracts statistical quantities such as mean, variance, and spectral energy to characterize the stable trend after an event.

[0021] No-event intervals follow the set beat Slide to generate a regular window to ensure continuous reasoning.

[0022] in: The duration of observation before the event. This refers to the post-event transition observation period. To stabilize the delay duration, Indicates the steady-state window length; Before an event, such as the start-stop of the main pump, valve switching, power-up instruction, backtracking to seconds of data, and incorporating this data into the transition window. In this way, early signs or slow drift before the event trigger can be captured; for example, 30 seconds of vibration and temperature trends before the event trigger; continuing to retain seconds of data after the event trigger to form a complete transition window. This part can record the fast response characteristics of the device after the event occurs; for example, the transient changes in pressure and flow within 10 seconds after valve switching; After an event, the steady state is not immediately opened, but is delayed for seconds to ensure that the system response is complete and the key quantities enter the relatively stable interval. This delay avoids mixing strong transient fluctuations into the steady state analysis. For example, after the main pump is accelerated for 60 seconds, the current and pressure fluctuations gradually subside, and at this time it is considered to be in a steady state. After the system enters the stable stage, the length of time for data collection is The window is mainly used to extract statistical features such as mean, variance, and spectral energy to reflect the long-term running trend of the device after the event. For example, 5 minutes of temperature and vibration data after entering the steady state.

[0023] Step 13, based on the dynamic updating time window, the obtained multi-source sensor data is divided to obtain a plurality of time window data, and each time window data is preprocessed as an input time window sequence.

[0024] Further, the method for preprocessing the input time window sequence includes the following steps: Step 101, for the obtained time window data, align the data of different source sensors according to the time window, mask mark and interpolate the missing values in each time window to make the data length consistent; The data of this embodiment is calculated at the same time as the data is received, and the data is divided according to the time window. After division, the time alignment is performed through this step; Step 1011, align the time stamp of each sensor data with the grid of the time window, and unify the data captured at different sampling frequencies to one time window. If the time difference between data points is within the allowable error, it is considered to be continuous, and the discontinuous is interpolated and completed by mask to make the data length consistent; Specifically, the mask data can be estimated and completed by linear interpolation or Kalman update; Step 1012, for each time window data, check whether each type of sensor data is missing field by field; Specifically, it can be judged whether temperature, vibration, current, pressure, etc. are missing in the event window. If missing, it is filled by mask interpolation; Step 102, for each time window, construct the data of each data point into a triple data including metadata, mask label and confidence, to obtain a multivariate sensor state vector of each time window; Specifically, all data points are constructed into triple inputs ; wherein, is a numerical value, may be original data or mask data; is a mask label, missing or interpolated and set to 0; is a confidence; The determination method of the confidence is to set the original data to high confidence, set the interpolated data to medium confidence, and set the estimated numerical value to low confidence, and write into the triple data. Transition window prohibits cross-event backfilling to avoid introducing false context.

[0025] Further, the obtained triple data can be updated and managed by window: each window buffer is updated according to the first-in first-out (FIFO) principle: for each active window maintain a circular buffer. When a new sample arrives, generate and then press it in to ensure the consistency of the feature matrix dimension and time sequence.

[0026] Step 103, for the triple data of each time window, encode the mask label and confidence of each data point in the window, encode the action mode, event type and time position corresponding to the data of each time window, and splice the metadata of the triple data after encoding, to obtain Token of each time window as input time window sequence; In the above embodiment, after completing the dynamic update, time alignment and data completion of the sliding window, the multivariate sensor state vector of each time is further enhanced and encoded to form a unified model input sequence.

[0027] Step 1031, normalize the metadata in the triple data; Specifically, at time , the sensor state vector contains multiple physical measurement point values such as temperature, vibration, current, pressure, etc.; after the sensor state vector is normalized, a standardized vector is formed to ensure that data of different dimensions are comparable; Optionally, the normalization of the sensor state vector can use Min-Max or Z-score, etc. Step 1032, extract the mask label and confidence data in the triple data, calculate the weighted value as the first encoding feature ; Specifically, the missing indication and the confidence weight are concatenated into an enhanced token (Token) through a function for dynamically adjusting the attention of each channel in the model inference process. The function is used to fuse the missing indication and the confidence into part of the enhanced Token, which is represented as: ; wherein is a learnable mapping matrix.

[0028] Step 1033, identify the type of time window, encode the time window type to obtain the second encoding feature ; In this embodiment, the time window type includes a transition window, a steady state window and a no event window; Step 1034, identify the event type corresponding to the time window, and encode to obtain a time type sub-vector ; This step encodes the trigger event category corresponding to the time step, such as load rise and valve switching, to enhance the model's ability to recognize event-driven degradation features.

[0029] Step 1035, encode the timestamp position of the data to obtain a time position encoding sub-vector ; so that the model can maintain the perception of the sequence order; Step 1036, sequentially concatenate the normalized metadata with the first encoding feature, the second encoding feature , the time type sub-vector, and the time position encoding sub-vector to obtain an enhanced token (Token), and then combine the input time window sequence according to each time window; Finally, the enhanced Token formed at time is defined as: ; wherein the symbol represents a vector concatenation operation. The structure of the input time window sequence ensures that the input sequence not only contains numerical information, but also introduces data quality, running mode, event category and time position information, which can fully represent the running state of the nuclear power plant equipment.

[0030] Through the above enhanced Token construction steps, the embodiment realizes unified expression of multi-source and multi-dimensional sensor data without relying on manual setting of rules, thereby providing complete and interpretable input basis for Q / K / V vector generation and attention calculation in the subsequent Transformer encoder.

[0031] Through the above improvements, the window alignment and classification sampling mechanism under event triggering is realized in view of the reality that there are various different working conditions in nuclear power plants, which can alleviate the working condition aliasing problem at the source, and strengthen the timing details in the transition state and the stability of the trend in the steady state, thereby significantly improving the distinguishability and robustness of nuclear power plant edge side anomaly detection and RUL prediction.

[0032] Step 2, input the time window sequence into the Transformer model for inference, perform multi-head self-attention calculation on the input time window sequence data, generate a predicted query vector (Q), a running state feature vector (K), and a historical information carrier vector (V) through linear transformation, perform dot product matching, Softmax normalization and attention weighted fusion to generate a current device health state vector.

[0033] In step 2, the time window sequence data is inferred by the Transformer model, as shown in Figure 3 , including the following steps: Step 21, input the time window sequence data into the attention mechanism module, and generate a query vector Query, a key vector Key and a value vector Value through three groups of linear projection weight matrices , and respectively: ; This step provides a vector basis for the attention mechanism and completes the initial feature expression of the sequence information. This structure allows the model to automatically learn the implicit dependency between device states without manually setting state rules.

[0034] Step 22, calculate the attention weight of each time point using the multi-head attention mechanism, and perform softmax function normalization to obtain the attention weight vector of each time point; Calculate the correlation score by calculating the inner product of the query vector Query and the key vector KeyQuery: ; wherein, represents the query vector at the jth time point; represents the key vector at the jth time point; represents the dimension; After normalization by the softmax function, the attention weight vector at each time point is obtained : ; In the above embodiment, the dot product attention mechanism is used to explicitly model the state information between different time segments. The attention weight is obtained by calculating the similarity between Query and Key, and then the corresponding Value is weighted and fused to form the dynamic understanding representation of the current device state. This mechanism has explicit time weight output capability, which is the basis for ensuring model interpretability and long-range dependency modeling capability. It adaptively learns the complex nonlinear relationship between device states from multi-dimensional time series data, significantly reducing the dependence on manually constructed degradation indicators, threshold settings or rule logic configuration, and improving the modeling automation level and development efficiency.

[0035] Step 23, weight and fuse the attention weight vector and the value vector to obtain the state vector of the fused context information at the current time point : ; State representation features Capture the degradation trend features of the device throughout the cycle, i.e., the degradation information of the key device.

[0036] An alternative technical solution replaces the attention mechanism with a convolutional neural network (CNN) to extract features; the attention mechanism of Transformer realizes the dependency modeling between time slices, and a convolutional neural network can also be used to extract features from time series and input into a regression layer to complete prediction.

[0037] Specifically, a one-dimensional convolution kernel is constructed to slide and extract local features from device historical state data, and then a multi-layer convolution and pooling structure is used to model the degradation trend. This solution is particularly suitable for devices with short prediction cycles and obvious feature changes, and has fast model training speed and convenient deployment.

[0038] This solution implicitly models the time dependency in the convolution kernel weights, although it does not have explicit and interpretable attention output, but it performs well in inference efficiency and small data scenarios, and has application feasibility.

[0039] Step 24, multi-head attention splicing and feature integration, the state vectors obtained by all h attention heads are spliced along the feature dimension to form a global comprehensive feature : ; This splicing operation enables the model to simultaneously aggregate features of multiple sensor signals at different time scales and physical dimensions, enhancing the ability to perceive the complex state of nuclear power equipment.

[0040] Step 25, the global comprehensive features after splicing are mapped linearly Dimension alignment is performed, and a unified format feature vector is output , so as to input subsequent module processing.

[0041] Step 26, residual connection is performed on the feature vector , the original input is added, and normalization processing is performed to obtain a global state vector : ; This operation retains the direct path of the original features of the device, prevents information loss in deep model training, and ensures the integrity of the expression of nuclear power plant operation data.

[0042] Layer normalization is performed on to balance the mean and variance of the feature dimension, improve the robustness and generalization ability of the model to high-dimensional complex nuclear signals (especially in the context of multi-feature coupling), and the formula is: ; The final output is the global state vector that fuses multi-head attention and nuclear power multi-dimensional operation features, which has both local degradation features and global health information, and is used as the input of the subsequent feedforward network to extract complex degradation patterns and RUL prediction. After multi-head self-attention splicing, residual connection and normalization processing, the global state vector output by the Transformer model already contains the comprehensive features of the multi-dimensional sensor signals of the nuclear power equipment, such as the joint change features of the main pump vibration, cooling water temperature, and motor current.

[0043] In order to further extract possible complex degradation patterns of the equipment, such as long-term vibration decay trend, temperature abnormal fluctuation, and electrical abnormality accompanying load change, the is input into the feedforward neural network module.

[0044] Step 27, the obtained global state vector is processed by the feedforward network feature extraction, full connection, activation, residual connection, and layer normalization to obtain the final state vector ; Specifically, the feedforward network includes two full connection layers and ReLU activation function layers, has strong non-linear fitting ability, and can identify degradation patterns and coupled features. The output of the feedforward network is connected again with residual connection and layer normalization to form a stable feature representation.

[0045] Further, RUL regression prediction is performed on the extracted state vector an input regression prediction module, outputting a remaining useful life (RUL) prediction value of the equipment in the current state; Optionally, the regression prediction module adopts a single-layer or multi-layer fully connected network, and the formula is expressed as: ; wherein, is a regression mapping function, which can accurately predict the remaining useful life value according to the historical data and the current features of the equipment after training.

[0046] In some embodiments, the uncertainty range of the prediction result can also be estimated by a Bayesian regression layer, an MC Dropout method or a double-head network structure, and the upper and lower confidence limits of the RUL, such as a 95% confidence interval, are output.

[0047] Specifically, the Bayesian regression layer is introduced, the Bayesian regression structure is introduced in the output layer of the Transformer model, and the Transformer model simultaneously outputs the mean prediction value μ and the variance prediction value σ of the RUL. 2 In the training process, the loss function based on the negative log-likelihood is optimized, so as to realize the adaptive modeling of the prediction confidence range; and the confidence interval can be expressed as: ; This deformation scheme is suitable for systems with extremely high safety levels, and can assist operation and maintenance personnel in evaluating the model and risk boundary, improving the decision-making robustness, and is especially suitable for high-safety-level scenarios such as nuclear power plants, whose operation and maintenance strategy tends to be the conservative principle of early maintenance rather than risk omission.

[0048] In the above process, the complete process from multi-dimensional sensor data stream, real-time feature extraction, self-attention calculation, multi-head information fusion, nonlinear feature recognition and remaining life prediction is realized.

[0049] Further technical solutions, in the training process of the Transformer model, a data set containing complete degradation life records is used, the real RUL is used as a supervision label, the model parameters are continuously optimized by minimizing the loss function, and the loss function includes a prediction loss term, a long-term error penalty term and a confidence modeling term. The loss function formula is as follows: ; wherein, represents the real life of the ith sample, i.e., the label; represents the prediction value of the ith sample; represents a dynamic weight coefficient; represents a confidence modeling term, represents an uncertainty loss coefficient, denotes the coefficient of the influence of the long-term error on the total loss; denotes the prediction value of the step and the label, used to calculate the penalty of the long-term prediction error; denotes the prediction value of the step and the label, used to calculate the penalty of the long-term prediction error; The loss function comprises: a prediction loss term : used to judge the accuracy of the prediction by using the weighted mean square error; a confidence modeling term : used to encourage the model to explicitly estimate the prediction uncertainty and improve the adaptability to unknown states; a long-term error penalty term : used to suppress the drift of the model in judging future trends and improve the prediction robustness; The Transformer model of the embodiment can be periodically iteratively trained to support online incremental learning and multi-device migration.

[0050] Compared with the traditional recurrent neural network (such as LSTM) which models the previous state in time sequence, the embodiment introduces the multi-head attention mechanism of the Transformer, so that the model can automatically learn the dependency relationship and attention weight between time points during the training process, thereby breaking the strict time sequence restriction, establishing a global context dependency within a long sequence, and improving the response capability to key state changes. In the actual industrial environment, the sensor signal is easily affected by noise, delay or data jitter. Through the dynamic attention mechanism, the attention weight of the data at different times can be adaptively adjusted during the inference stage, thereby having strong robustness and error suppression capability, and significantly improving the stability and credibility of the prediction output.

[0051] Since the attention mechanism generates an explicit attention matrix during the inference process, the embodiment can visualize the attention matrix to generate an attention weight atlas to generate explanation information, which is used to assist the operation and maintenance personnel to understand the basis of the model prediction, and enhance the trust and actual availability of the system in the nuclear power plant scene. In the embodiment, the attention explanation mechanism in the nuclear power plant background specifically: the attention matrix generated in the foregoing steps is used to extract key sensor features and time periods related to RUL prediction, such as the time window corresponding to the change of the main pump vibration frequency, the duration of the abnormal cooling temperature, and the explanation information is output together with the RUL prediction result. In this way, not only the device life prediction can be given, but also intuitive degradation mode explanation and decision basis can be provided for the operation and maintenance personnel.

[0052] Further, in step 3, the attention matrix obtained by the attention mechanism of the Transformer model during the inference process is visualized to obtain an equipment health state explanation atlas of the attention weight, and an explanation information generation method is as follows:Figure 4 As shown, comprising the following steps: Step 31, obtaining the attention matrix output by the Transformer model, based on the combination of time steps and feature channels with high weight values in the attention matrix, reversely positioning the input data region concentrated by the attention of the Transformer model as a high attention region, that is, identifying the key working condition signals and their corresponding abnormal patterns on which the judgment result of the Transformer model is significantly dependent; For example, if the "cooling temperature rise" is given a very high attention weight in a certain period of time, it means that the change of this variable plays a decisive role in model prediction, which can be explained as a potential "abnormal signal" or "key working condition"; Further, as shown in Figure 2 The attention matrix can be uploaded and stored and processed through the set central platform; Step 32, based on the feature channels and time windows of the high attention region, extracting the abnormal features of the key equipment, identifying the abnormal behavior patterns with physical meaning, extracting the degradation features and their time sequence information to construct a degradation feature set; Specifically, the construction process of the degradation feature set includes: Step 321, according to the high attention region obtained in step 31, selecting time points with weight values significantly higher than the set threshold, and taking the time points as the center to intercept a fixed length time window in the original multi-dimensional input sequence, and extracting the corresponding original sensor data subsequence; Step 322, mapping the channel index to the physical quantity: mapping the feature channel index in the attention matrix to the actual physical quantity name corresponding to the sensor, such as the lateral vibration of the main pump, the outlet temperature of the cooling system, the current signal, etc., to clarify the physical meaning of each high weight signal and the equipment part it belongs to; Step 323, performing feature extraction operation: performing multi-dimensional feature calculation on the original sensor data subsequence intercepted in step 321 to obtain the extracted feature A; Among them, the feature A includes but is not limited to: Time domain statistics, including mean, maximum, variance, change rate, etc.; Frequency domain indicators, including power spectral density, frequency peak value, etc.; Variability indicators, including the number of mutation points, skewness, kurtosis, etc. Outlier detection results, such as outlier judgment based on IQR and z-score.

[0053] Step 324, reference baseline difference analysis: comparing the extracted current feature A with the historical feature statistical range of the equipment under normal working conditions to evaluate its deviation degree, and taking the signal features with over-limit or significant abnormalities as abnormal signal items; Step 325, generating a structured abnormal feature set: based on the signal items judged as abnormal and their corresponding data and device record information, a degradation feature set is constructed; Optionally, the content of the degradation feature set can include: abnormal feature name and dimension, abnormal starting and duration, belonging device subsystem, abnormal amplitude or confidence score, and labeled feature category; The degradation feature set will serve as the input basic data for subsequent health state atlas generation, risk level assessment, alarm rule matching, etc.

[0054] In the above steps, the position of the model's attention is found in step 31 by obtaining the attention matrix, and in step 32, the results of step 31 are used to further extract device abnormal features; Step 33, fuse the abnormal features in the degradation feature set with the historical operation data to construct a device health state interpretation atlas of the Transformer model attention weight; Specifically, the device health state interpretation atlas includes device degradation trend curves, abnormal frequency distribution, and risk distribution maps, etc. The atlas intuitively shows the current state and historical evolution path of the device, assisting experts in understanding and operation and maintenance judgment; Step 331, align abnormal features with historical working conditions: align the degradation feature set generated in step 32 with the historical operation data in the time dimension, establish a mapping index according to the device, working condition parameters and time stamp, and form a related data set of the current state and historical evolution path; Step 332, degradation trend modeling: for each key device feature in the degradation feature set, combine the corresponding historical data to fit the evolution curve of the feature over time, and superimpose the current abnormal position to generate a device degradation trend graph, which is used to represent the dynamic evolution process of the device health state; Wherein, the fitting of the evolution curve can adopt linear regression, polynomial trend line or moving average method, etc. Step 333, abnormal frequency statistics: based on historical data, the occurrence frequency of various abnormal events in different time periods and different working conditions is counted to construct an abnormal intensity heat map or a failure frequency map to reveal high-risk areas and frequent patterns.

[0055] Step 334, risk level distribution calculation: combine the abnormal feature type, severity and duration information to calculate the health score or risk level of the device in each time period, and construct a risk level distribution map to visualize the risk concentration trend of the device operation state; Step 335, fuse and integrate the device degradation trend graph, failure frequency graph and risk level distribution graph based on a unified time axis with the device identifier to generate a device health state interpretation atlas of the nuclear power plant to provide health state interpretation information.

[0056] Step 336, outputting the global state vector of the Transformer model Compared with the health baseline, the health state information of the equipment at each time is obtained, the health deviation degree is judged, and is associated with each data point of the nuclear power equipment health state interpretation graph to obtain the final nuclear power equipment health state interpretation graph; Specifically, in the embodiment, the Transformer model outputs a global state quantity after inputting each time window, and a series of state representation vectors can be obtained as the time window continuously slides: ; The sequence corresponds to the health state of the equipment at different times.

[0057] The deviation degree of the current state relative to the health baseline , and the calculation formula is: ; Wherein, represents the health baseline, which can be determined by historical health sample mean, artificial calibration, etc. Through the above steps, the obtained equipment health state interpretation graph at least contains the following information: display the historical evolution trajectory of the key feature vector of the equipment; mark the position and influence range of the key abnormal event set ; visualize the deviation degree of the current state relative to the health baseline ; Further, if the deviation degree is large, the reason needs to be explored, and a dynamic knowledge graph can be used to further generate interpretation information.

[0058] In the above embodiment, the visual nuclear power equipment health state interpretation graph is given, which directly represents the running track before and after the change of the equipment health state through the change curve, and directly displays the change trend; in order to generate dynamic explanation that can give textual representation, the embodiment also proposes a method of dynamic knowledge graph to realize dynamic interpretation of data and generate health state interpretation information fused with text description; Further, in order to realize the explanation alignment of the knowledge level, when the deviation degree of the global state vector displayed in the equipment health state interpretation graph is greater than the set value, it is identified as an abnormal deviation point, and a dynamic knowledge graph is constructed for the abnormal deviation point to generate health state interpretation information described in text; For the abnormal deviation point in the equipment health state interpretation graph, a dynamic knowledge graph is constructed based on the abnormal deviation point to generate interpretation information, including the following steps: Step 361, construct a directed graph with device nodes, component nodes, failure mode nodes and working condition nodes as nodes, and the time sequence correlation weight between nodes as edges As a dynamic knowledge graph: ; Wherein: the node set , respectively represents the device node, component node, failure mode node and working condition node; the edge set , represents the time sequence correlation between devices, components, failure modes and working conditions; The time sequence correlation weight of each edge, the calculation formula is: ; Wherein, is the attention score, which is calculated by the multi-head attention mechanism of the Transformer model, and is used to measure the degree of dependence of input feature u on input feature v. The sum of all is equal to 1, and the calculation formula is as follows: ; Wherein, is the Query vector, which is mapped from the input feature ; for example, the input feature may be a certain sensor signal or time feature; is the Key vector, which is mapped from the input feature ; represents the dot product of two input features, which measures the correlation between input feature u and input feature v; represents the normalization factor, which prevents the inner product value from being too large when the dimension is too large; represents the operation of converting all correlation scores into a probability distribution; is the event intensity, which is used to represent the actual impact of an abnormal event on the running state of the device at a certain time; In a specific embodiment, the event intensity can be determined based on one or more of the following factors: The overrun amplitude, the degree to which the monitoring signal exceeds the set threshold, the event intensity increases with the increase of the overrun degree; The duration, the time length of the abnormal state maintaining the abnormal state, the event intensity increases with the increase of the duration.

[0059] The frequency of occurrence, the number of times of abnormal event occurrence in a single time window; when the number of times of abnormal event occurrence in a certain time window is large, the event intensity increases with the increase of the frequency of occurrence.

[0060] In practical applications, the event intensity can also be determined by a weighted combination of the above factors to comprehensively reflect the influence of abnormal events in terms of amplitude, duration and frequency.

[0061] Step 362, based on the constructed directed graph , an explanation path with time evolution characteristics, i.e., textual explanation information, is generated to represent the correspondence of the device from the current state to the potential failure mode, as follows: ; Among them, represents an explanation path, which is used to represent the reasoning chain and its intensity of the health state of the device at time t; represents a device node; represents a component node, which represents a key component in the device; represents a failure mode node, which represents a possible failure mode of the component, such as wear, corrosion, crack, jam, leakage, etc.; represents a working condition node, which represents the working condition environment of the device during operation, such as start-stop, power-up, steady-state operation, emergency shutdown (SCRAM), high temperature and high pressure operation, etc.

[0062] is the edge weight, which represents the intensity of the association relationship between the device, the component, the failure mode and the working condition at time .

[0063] Step 363, obtain the operation personnel feedback data and operation annotation results as feedback, and merge the feedback information with the information of the knowledge graph in a union set manner to dynamically update the knowledge graph ; According to the health state explanation information, combined with the device type, such as pump, cooling, pipeline, valve, electrical, etc., a maintenance alarm rule matching the degradation trend is established, and the maintenance strategy suggestion and alarm signal are pushed to the DCS monitoring interface of the nuclear power plant in real time, and the corresponding operation strategy suggestion is generated, such as immediate shutdown inspection, advance maintenance arrangement, regular monitoring, etc., which is checked, confirmed and responded by the on-site operation personnel or the main control room operator.

[0064] Specifically, the operation personnel feedback data and operation annotation results will be fed back to the Transformer model learning pool to realize the dynamic update of the knowledge graph : ; Among them: represents a set of feedback and operation annotation from the operation personnel, which can include confirming the alarm, modifying the alarm level, manually labeling the failure cause, etc.; Dynamic updating can make the graph structure and weights Optimization is continued over time to reflect the real degradation process of the device and the human-computer interaction experience.

[0065] Further, the dynamic update of the knowledge graph includes updating or supplementing the attributes of the nodes, updating the node set, and updating the edge set, which are described as follows. 1) At time , there is a knowledge graph in the system. Among them, the node set contains device nodes, component nodes, fault mode nodes, and working condition nodes; the edge set represents the association relationship between the above nodes, and the edge has attribute information, including attention score, event intensity, risk level, etc.

[0066] 2) At time , a new feature set is obtained. Among them: is the newly identified node information, including the potential fault mode obtained by model inference, the component state labeled by the operation and maintenance personnel, and the newly added working condition category; is the newly generated or corrected edge information, including the correspondence between abnormal events and nodes, the dependence relationship corresponding to the attention score, the causal relationship confirmed by operation and maintenance feedback, etc.

[0067] 3) The update rule is: ; Specifically, the following three cases are included: When the nodes in already exist in , update or supplement the attributes of the nodes, for example, add artificial confirmation marks or correct health status labels; When contains new devices, components, or fault modes, add them to the node set ; When contains new association relationships, add them to the edge set ; when overlaps with existing edges, update the attributes of the edge such as weight value, risk label, and timestamp.

[0068] Through the above union method, the atlas can not only maintain historical evolution information, but also dynamically absorb new model inference results and operation personnel feedback, realizing rolling update and continuous expansion of the atlas over time. Thus, the atlas can fully reflect the historical trajectory and current state of the device health status at any moment, and support subsequent explanation path generation and risk assessment.

[0069] Further, according to the health state explanation information, combined with the type of nuclear power equipment such as pump, cooling, pipeline, valve, electrical equipment, etc., maintenance alarm rules matched with the equipment degradation trend are established, and alarm signals are generated according to the alarm triggering conditions and corresponding risk levels.

[0070] Specifically, combined with historical equipment failure records and maintenance experience library, the risk level of the current state is determined according to the equipment operation characteristics; In step 4, based on the risk level determined by the health state explanation information, device maintenance strategy suggestions and corresponding alarm thresholds are generated; Specifically, the device maintenance strategy suggestions include immediate shutdown inspection, advance maintenance arrangement, regular monitoring, etc. The alarm threshold is used to trigger the alarm system of the nuclear power plant distributed control system (DCS).

[0071] Real-time pushing of the device maintenance strategy suggestions and alarm signals to the DCS monitoring interface of the nuclear power plant can be realized, and the on-site operation personnel or main control room operator can check, confirm and respond. The operation personnel or main control room operator can manually confirm, adjust or mark the maintenance suggestions and alarm signals generated by the system according to the actual situation, forming feedback data and operation records. The operation personnel feedback data and operation marking results are returned to the Transformer model learning pool for continuous optimization of model parameters and decision logic, so as to improve the accuracy and adaptability of subsequent attention explanation and maintenance alarm strategy.

[0072] Through the above process, the nuclear power plant equipment realizes closed-loop management from real-time data acquisition, abnormal feature identification, health state explanation, to risk assessment and maintenance alarm generation, significantly improving the accuracy and efficiency of predictive maintenance of nuclear power plants.

[0073] Further technical solutions also include a model correction mechanism based on asynchronous feedback, which includes the following steps: Step 51, model inference and output: the Transformer model deployed on the edge node or the center server outputs the RUL prediction result at the current time after receiving the latest sliding window input, while retaining the attention weight information inside the model; Step 52, result pushing and manual marking: the prediction result and the health state explanation information of the attention weight information output by the Transformer model are manually confirmed, adjusted and corrected. Specifically, the model prediction results and the corresponding attention visualization information are pushed to the terminal interface of the operation and maintenance personnel or the central DCS monitoring platform in the form of an interpretable graphical interface. The operation and maintenance personnel can manually confirm, evaluate or correct the model output results in combination with the actual working conditions, equipment maintenance history and experience knowledge. The corrective annotation can include false positives, time period corrections, weight anomaly explanations, etc. Step 53, feedback information is returned asynchronously: the prediction results after manual confirmation adjustment and corrective annotation are returned to the Transformer model in the form of asynchronous data stream, and the Transformer model is fine-tuned and trained as a fine-tuning data set. The modified model in this embodiment is used for subsequent inference tasks, which realizes continuous absorption of field experience information and update of prediction strategy under the premise of uninterrupted service, avoids the problem of long update cycle and slow response of traditional offline training, and significantly improves the practicality and synchronicity of the model in the engineering field.

[0074] Further technical solutions, an edge-center joint modeling mechanism is adopted, specifically: including edge nodes and center nodes. Edge node: a lightweight Transformer model is deployed on the edge node of the nuclear power plant industrial distributed control system (DCS). This edge node model can use real-time data collected by on-site sensors to perform dynamic and small-scale model parameter updates and real-time inference through local computing devices in a sliding window manner, ensuring that the model prediction has extremely low computational delay and can quickly respond to changes in the field state.

[0075] The center node, i.e., the center platform, is used to perform lightweight synchronization functions, such as obtaining attention weight patterns and discriminant logic summary information from Transformer model outputs of various edge nodes to realize cross-site data feature and weight collaboration and sharing.

[0076] The scheme of the embodiment realizes the structural stability and time consistency management of multi-dimensional time sequence input data by deeply integrating the dynamic sliding window mechanism with the Transformer model, significantly improves the perception ability of the model to the degradation trend of long sequence and multi-variable equipment, and enhances the prediction stability and robustness. Meanwhile, with the help of the attention mechanism of the Transformer, the system can automatically identify the key time period and key feature channel in the input data, realize self-explaining modeling of the equipment abnormal state and its evolution mode, and enhance the transparency and expert understandability of the model output. On this basis, the system further constructs a health state atlas integrating trend curve, abnormal frequency and risk level distribution, comprehensively presents the current state and historical evolution path of the equipment, and enhances the auditability and auxiliary decision-making ability in engineering scenarios. In order to realize the continuous collaborative optimization of the model and operation experience, the asynchronous feedback correction mechanism is introduced, so that the artificial labeling information can be returned to the edge node asynchronously and drive the model to update slightly, overcoming the lag problem of traditional model offline retraining. In addition, the scheme supports the uncertainty quantization output of RUL prediction results, provides confidence interval expression based on Bayesian regression, MC Dropout or double output structure, and improves the confidence evaluation ability and robust decision support ability of the model in high-risk scenarios.

[0077] Embodiment 2 Based on embodiment 1, the present embodiment provides 8. A self-attention-based nuclear power plant equipment predictive evaluation maintenance system, comprising: a sliding window unit configured to obtain multi-source sensor data of the nuclear power plant equipment using a sliding window mechanism, and construct an input time window sequence after preprocessing; an inference unit configured to infer the time window sequence data through a Transformer model to obtain a state vector for regression analysis, and generate a remaining useful life of the equipment; an explanation unit configured to visualize the attention matrix obtained by the attention mechanism of the Transformer model in the inference process to obtain an equipment health state explanation atlas of attention weights, and construct a dynamic knowledge graph to generate explanation information for abnormal deviation points in the equipment health state explanation atlas; a fusion unit configured to fuse the obtained remaining useful life of the equipment and the corresponding explanation information to generate a device maintenance strategy suggestion.

[0078] Further, the multi-source sensor data of the nuclear power plant equipment is obtained using a sliding window mechanism, and an input time window sequence is constructed after preprocessing, including the following steps: synchronously listen to the received multi-source sensor data, DCS working condition signals and interlocking signals, and identify events; according to the identified event occurrence points According to the event type, the time period set before and after the event occurrence point is set as a transition window, and the time set after the event occurrence is divided into a steady state window, the transition window is smaller than the steady state window, the sizes of the transition window and the steady state window are set to be different according to the event type, and the time period without events is set as a fixed window as a time window, so as to obtain a dynamically changing time window. Based on the dynamically updated time window, the obtained multi-source sensing data is divided to obtain a plurality of time window data, and each time window data is preprocessed as an input time window sequence.

[0079] Further, the attention matrix obtained by the attention mechanism of the Transformer model in the reasoning process is visualized to obtain a device health state interpretation atlas of attention weight, and an explanation information generation method comprises the following steps: An attention matrix output by a Transformer model is obtained, and based on the combination of time steps and feature channels with high weight values in the attention matrix, an input data region concentrated in the attention degree of the Transformer model is reversely located as a high attention region; Based on the feature channel and the time window of the high attention region, the abnormal features of the key equipment are extracted, the abnormal behavior patterns with physical meaning are identified, the degradation feature set is constructed by extracting the equipment degradation features and the time sequence information of the equipment degradation features; The abnormal features in the degradation feature set are fused with the historical operation data to construct a device health state interpretation atlas of the attention weight of the Transformer model.

[0080] It should be noted that each module in the embodiment corresponds to each step in Embodiment 1 one by one, and the specific implementation process is the same, which will not be repeated here.

[0081] The above only describes the preferred embodiments of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0082] The above describes the specific embodiments of the present application in combination with the drawings, but is not a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications or variations made by those skilled in the art on the basis of the technical solutions of the present application without creative labor are still within the protection scope of the present application.

Claims

1. A self-attention based predictive assessment maintenance method for nuclear power plant equipment, characterized in that, The method comprises the following steps: A sliding window mechanism is used to obtain multi-source sensing data of nuclear power plant equipment, and after preprocessing, an input time window sequence is constructed; The time window sequence data is inferred through a Transformer model to obtain a state vector for regression analysis, and the remaining useful life of the equipment is generated; The attention matrix obtained by the attention mechanism of the Transformer model in the inference process is visualized to obtain an equipment health state interpretation atlas of attention weights; for abnormal deviation points in the equipment health state interpretation atlas, a dynamic knowledge graph is constructed to generate interpretation information; The obtained remaining useful life of the equipment and the corresponding interpretation information are fused to generate equipment maintenance strategy suggestions.

2. The self-attention-based nuclear power plant equipment prognostic evaluation maintenance method of claim 1, wherein: A sliding window mechanism is used to obtain multi-source sensing data of nuclear power plant equipment, and after preprocessing, an input time window sequence is constructed, comprising the following steps: The received multi-source sensing data, DCS working condition signals and interlocking signals are synchronously monitored to identify events; According to the identified event occurrence points, the time period set before and after the event occurrence points is set as a transition window, and the time period set after the event occurrence is divided into a steady state window, the transition window is smaller than the steady state window, and the sizes of the transition window and the steady state window are set to be different according to the event type, and when there is no event, a fixed window is set as a time window to obtain a dynamically changing time window; Based on the dynamically updated time window, the obtained multi-source sensing data is divided to obtain a plurality of time window data, and each time window data is preprocessed as an input time window sequence.

3. The self-attention-based predictive evaluation and maintenance method for nuclear power plant equipment according to claim 1, characterized in that: The time window sequence data is inferred through a Transformer model, comprising the following steps: The time window sequence data is input into an attention mechanism module to generate query vectors, key vectors and value vectors through three sets of linear projection weight matrices respectively; The multi-head attention mechanism is used to calculate the attention weight of each time point, and the softmax function is used for normalization to obtain the attention weight vector of each time point; The attention weight vector is fused with the value vector to obtain a state vector of the fused context information at the current time point : Perform multi-head attention concatenation and feature integration to obtain the state vectors from all attention heads. By concatenating along the feature dimensions, a global comprehensive feature is formed; The global comprehensive features after splicing are linearly mapped dimension alignment is performed, and a uniform format feature vector is output ; The residual connection is performed, the feature vector is added to the original input , and normalization processing is performed to obtain a global state vector ; The global state vector obtained is After the full connection operation, the activation operation, the residual connection, and the layer normalization processing through the feature extraction of the feedforward network, the final state vector is obtained .

4. The self-attention based nuclear power plant equipment prognostic evaluation maintenance method of claim 1, wherein: In the training process of the Transformer model, the loss function includes a prediction loss term, a long-term error penalty term and a confidence modeling term.

5. The self-attention-based predictive evaluation and maintenance method for nuclear power plant equipment according to claim 1, characterized in that: The attention matrix obtained by the attention mechanism of the Transformer model in the inference process is visualized to obtain an equipment health state interpretation atlas of attention weights, and a method for generating interpretation information comprises the following steps: The Transformer model output attention matrix is obtained, and based on the combination of time steps and feature channels with high weight values in the attention matrix, the input data region in the Transformer model attention concentration set is located reversely as a high attention region; Based on the feature channel and time window of the high attention area, the abnormal features of the key equipment are extracted, the abnormal behavior patterns with physical meaning are identified, the degradation features and their time sequence information are extracted to construct a degradation feature set; The abnormal features in the degradation feature set are fused with the historical operation data to construct a device health state interpretation atlas of the Transformer model attention weight.

6. The self-attention-based nuclear power plant equipment prognostic evaluation maintenance method of claim 5, wherein: The abnormal features in the degradation feature set are fused with the historical operation data to construct a device health state interpretation atlas of the Transformer model attention weight, including the following steps: Align the generated degradation feature set with the historical operation data in the time dimension, establish a mapping index according to the equipment, working condition parameters and time stamp, and form a correlation data set of the current state and historical evolution path; For each key equipment feature in the degradation feature set, combine the corresponding historical data to fit the evolution curve of the feature over time, and superimpose the current abnormal position to generate an equipment degradation trend chart; Based on the historical data, the occurrence frequency of each type of abnormal event in different time periods and under different working conditions is counted to construct an abnormal intensity heat map or a fault frequency chart; Combined with the abnormal feature type, severity and duration information, the health score or risk level of the equipment in each time period is calculated, and a risk level distribution chart is constructed; The equipment degradation trend chart, fault frequency chart and risk level distribution chart are integrated based on the same time axis and equipment identifier to construct a device health state interpretation atlas. The global state vector output by the Transformer model The health state information of the device at each time is obtained by comparing with the health baseline, the health deviation degree is judged, and each data point of the nuclear power device health state interpretation atlas is associated to obtain the final nuclear power device health state interpretation atlas.

7. The self-attention-based predictive evaluation maintenance method for nuclear power plant equipment according to claim 1, characterized in that: For the abnormal deviation points in the device health state interpretation atlas, an explanation information is generated based on the dynamic knowledge graph constructed by the abnormal deviation points, including the following steps: A directed graph is constructed as a dynamic knowledge graph with equipment nodes, component nodes, fault mode nodes and working condition nodes as nodes, and the time sequence correlation weight between nodes as edges; Based on the constructed directed graph, an explanation path with time evolution features is generated as the health state interpretation information in the form of text description, which represents the corresponding relationship between the current state of the equipment and the potential fault mode; The operation and maintenance personnel feedback data and operation annotation results are obtained as feedback, and the feedback information is merged with the information of the knowledge graph in a set union manner for dynamic updating of the knowledge graph.

8. A self-attention based nuclear power plant equipment prognostic evaluation maintenance system, characterized in that, It includes: A sliding window unit configured to obtain multi-source sensing data of the nuclear power plant equipment using a sliding window mechanism, and to construct an input time window sequence after preprocessing; An inference unit configured to infer the time window sequence data through a Transformer model to obtain a state vector for regression analysis and generate a remaining useful life of the equipment; An explanation unit configured to visualize the attention matrix obtained by the attention mechanism of the Transformer model during inference to obtain a device health state interpretation atlas of attention weight, and to construct a dynamic knowledge graph to generate explanation information for abnormal deviation points in the device health state interpretation atlas; A fusion unit configured to fuse the obtained remaining useful life of the equipment and its corresponding explanation information to generate a device maintenance strategy suggestion.

9. The self-attention-based nuclear power plant equipment predictive assessment maintenance system of claim 8, wherein: The multi-source sensor data of the nuclear power plant equipment is obtained by using a sliding window mechanism, and after preprocessing, an input time window sequence is constructed, including the following steps: Synchronously listening to the received multi-source sensor data, DCS working condition signals and interlocking signals, and identifying events; According to the identified event occurrence point, setting a transition window before and after the event occurrence point according to the event type, and dividing a stable window within a set time after the event occurrence, the transition window being smaller than the stable window, the sizes of the transition window and the stable window being set to be different according to the event type, and when there is no event, a fixed window is set as a time window, thereby obtaining a dynamically changing time window; Based on the dynamically updated time window, the obtained multi-source sensor data is divided to obtain a plurality of time window data, and for each time window data, preprocessing is performed to obtain an input time window sequence.

10. The self-attention-based nuclear power plant equipment predictive assessment maintenance system of claim 8, wherein: The attention matrix obtained by the attention mechanism of the Transformer model in the reasoning process is visualized to obtain a device health state interpretation atlas of attention weights, and an explanation information generation method includes the following steps: Obtaining the attention matrix output by the Transformer model, and based on the combination of time steps and feature channels with high weight values in the attention matrix, reversely locating the input data region with high attention in the Transformer model as a high attention region; Based on the feature channel and the time window of the high attention region, abnormal features of the key equipment are extracted, abnormal behavior patterns with physical meaning are identified, degradation features and their time sequence information are extracted to construct a degradation feature set; Fusing the abnormal features in the degradation feature set with historical operation data to construct a device health state interpretation atlas of the attention weights of the Transformer model.

Images