Nuclear power plant anomaly detection method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of nuclear power plant intelligent operation and maintenance, and discloses a nuclear power plant anomaly detection method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining first operation data of a nuclear power unit; inputting the first operation data into the stacked auto-encoder to obtain a first data feature of the first operation data and a first reconstruction error of the first operation data; performing clustering analysis on the first data feature based on a Gaussian mixture model to obtain a minimum distance between the first data feature and a clustering center of a plurality of target clusters obtained by clustering of the Gaussian mixture model; wherein the plurality of target clusters are obtained by clustering second data features obtained by the stacked auto-encoder for a plurality of pieces of second operation data by the Gaussian mixture model; and determining an anomaly detection result of the nuclear power unit based on a comparison relationship between the first reconstruction error and a reconstruction error threshold value and a comparison relationship between the minimum distance and a distance threshold value. The method can solve the problem of low anomaly detection accuracy of the nuclear power plant.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for nuclear power plants, specifically to methods, devices, electronic equipment, and storage media for detecting anomalies in nuclear power plants. Background Technology

[0002] Nuclear power plants operate in highly complex and safety-critical environments where even minor anomalies can escalate into catastrophic failures. Therefore, accurate and timely anomaly detection is crucial for maintaining operational integrity, protecting personnel safety, and preventing environmental hazards. However, traditional monitoring methods typically wait for anomalies to reach a large, fixed threshold before triggering an alarm. This makes it difficult to detect subtle anomalies in the complex, high-dimensional data generated by nuclear power plant systems in a timely and accurate manner. By this time, accidents have often already occurred, resulting in low accuracy in anomaly detection at nuclear power plants. Summary of the Invention

[0003] This invention provides a method, apparatus, electronic device, and storage medium for detecting anomalies in nuclear power plants, in order to solve the problem of low accuracy in detecting anomalies in nuclear power plants.

[0004] In a first aspect, the present invention provides a method for detecting anomalies in a nuclear power plant, comprising: Obtain the initial operating data of the nuclear power unit; The first running data is input into a stacked autoencoder to obtain the first data feature of the first running data and the first reconstruction error of the first running data; Cluster analysis is performed on the first data feature based on Gaussian mixture model to obtain the minimum distance between the first data feature and the cluster centers of multiple target clusters obtained by Gaussian mixture model clustering; wherein, the multiple target clusters are obtained by Gaussian mixture model clustering the second data features obtained by the stacked autoencoder for multiple second running data respectively; Based on the comparison between the first reconstruction error and the reconstruction error threshold, and the comparison between the minimum distance and the distance threshold, the abnormal detection result of the nuclear power unit is determined.

[0005] Secondly, the present invention provides a nuclear power plant anomaly detection device, comprising: The data acquisition module is used to acquire the first operating data of the nuclear power unit; The feature extraction module is used to input the first running data into the stacked autoencoder to obtain the first data feature of the first running data and the first reconstruction error of the first running data. The clustering analysis module is used to perform clustering analysis on the first data features based on a Gaussian mixture model to obtain the minimum distance between the first data features and the cluster centers of multiple target clusters obtained by the Gaussian mixture model; wherein, the multiple target clusters are obtained by the Gaussian mixture model clustering the second data features obtained by the stacked autoencoder for multiple second running data respectively; An anomaly detection module is used to determine the anomaly detection result of the nuclear power unit based on the comparison relationship between the first reconstruction error and the reconstruction error threshold, and the comparison relationship between the minimum distance and the distance threshold.

[0006] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the nuclear power plant anomaly detection method of the first aspect or any corresponding embodiment described above.

[0007] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the nuclear power plant anomaly detection method of the first aspect or any corresponding embodiment described above.

[0008] The nuclear power plant anomaly detection method proposed in this invention acquires the first operating data of the nuclear power unit, providing comprehensive and accurate basic data for subsequent anomaly detection. The first operating data is input into a stacked autoencoder for feature extraction, effectively capturing complex nonlinear relationships within the data. This yields the first data features and the first reconstruction error of the first operating data, automatically extracting features without manual intervention. Furthermore, the first reconstruction error quantifies the degree to which the first operating data deviates from the normal pattern, sensitively reflecting abnormal operating conditions. Next, the first data features are input into a Gaussian mixture model for cluster analysis. This yields the minimum distance between the first data features and the cluster centers of multiple target clusters obtained from the Gaussian mixture model, effectively covering different normal operating modes and reflecting the distribution of the first operating data in the feature space. Finally, by combining the comparison between the first reconstruction error and the reconstruction error threshold, and the comparison between the minimum distance and the distance threshold, the anomaly detection result of the nuclear power unit is determined. This dual-threshold comparison effectively improves the accuracy of nuclear power plant anomaly detection. Attached Figure Description

[0009] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of the first process of a nuclear power plant anomaly detection method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a nuclear power plant anomaly detection architecture according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the historical training of a stacked autoencoder according to an embodiment of the present invention; Figure 4 This is a schematic diagram of reconstruction quality assessment according to an embodiment of the present invention; Figure 5 This is a schematic diagram of nuclear power plant anomaly determination according to an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the change of the target cluster number with the class index value according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the GMM clustering results according to an embodiment of the present invention; Figure 8 This is a first schematic diagram of the simulation verification test results according to an embodiment of the present invention; Figure 9(a) is a second schematic diagram of the simulation verification test results according to an embodiment of the present invention; Figure 9(b) is a third schematic diagram of the simulation verification test results according to an embodiment of the present invention; Figure 9(c) is a fourth schematic diagram of the simulation verification test results according to an embodiment of the present invention; Figure 9(d) is a fifth schematic diagram of the simulation verification test results according to an embodiment of the present invention; Figure 10 This is a structural block diagram of a nuclear power plant anomaly detection device according to an embodiment of the present invention; Figure 11 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0013] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0014] Early anomaly detection methods in nuclear power plants relied on threshold settings for key parameters such as temperature and pressure, often resulting in numerous false alarms due to noise or transient fluctuations. These methods also lacked adaptability to complex scenarios. While statistical methods such as Principal Component Analysis (PCA) can reduce dimensionality, they struggle to handle nonlinear relationships in high-dimensional data. The field of anomaly detection in industrial systems has evolved from rule-based systems to machine learning-driven methods. The increasingly sophisticated instrumentation and data acquisition capabilities of modern nuclear power plants have created opportunities for data-driven anomaly detection methods—whose anomaly detection capabilities far surpass those of traditional methods.

[0015] Deep learning methods excel at modeling complex patterns, with autoencoders learning to reconstruct normal data and marking samples with high reconstruction errors as anomalous. However, they can misclassify rare but normal variations. Furthermore, relying solely on reconstruction errors may lead to the omission of subtle deviations in structurally similar data.

[0016] Clustering-based schemes (such as Gaussian Mixture Models (GMMs) and k-means clustering) have been widely used for probabilistic modeling of normal operating conditions in complex industrial systems. Specifically, while k-means clustering is suitable for spherical or highly separated data clusters, its limitations in handling complex nonlinear distributions make it less suitable for the diverse and often noisy data commonly found in nuclear power plant operations. In contrast, GMMs, through probabilistic allocation mechanisms, can more flexibly adapt to naturally random or overlapping data distributions. GMMs approximate a multimodal distribution through a weighted combination of Gaussian components, providing a flexible framework for density estimation. Its parameters are typically estimated using the expectation-maximization algorithm. In contrast, while k-means clustering is computationally efficient, its assumption of isotropic covariance structure limits its ability to model complex operating conditions.

[0017] To overcome the above limitations, this invention proposes a hybrid framework that integrates stacked autoencoders and GMM. The stacked autoencoders are used to reduce the dimensionality of nonlinear features, and the GMM is combined for probabilistic modeling. The deep feature extraction and probabilistic clustering techniques are used to improve the efficiency of anomaly detection in nuclear power plants.

[0018] According to an embodiment of the present invention, a method for detecting anomalies in a nuclear power plant is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0019] This embodiment provides a method for detecting anomalies in nuclear power plants, which can be used in electronic devices such as servers and computers. Figure 1 This is a flowchart of a nuclear power plant anomaly detection method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain the first operating data of the nuclear power unit.

[0020] Specifically, sensors are used to collect raw operating data of nuclear power units under normal operating conditions from multiple dimensions.

[0021] Optionally, the parameters of the raw operating data may include one or more of the following: pressure, temperature, flow rate, liquid level, radioactivity, neutron flux, and subcooling, or other types of operating data, without specific limitations.

[0022] Furthermore, referring to the raw operating data shown in Table 1, since the raw operating data have different magnitudes and units (such as pressure in MPa and temperature in °C), standardization processing is used to normalize all raw operating data to ensure that the contribution of the raw operating data is consistent in anomaly detection.

[0023] Table 1. Example of raw running data

[0024] In practical applications, the StandardScaler function of the Scikit-Learn library is used to standardize the raw operating data. The first operating data of the nuclear power unit is obtained through the following formula:

[0025] Where μ is the mean of each raw operating data point, and σ is the standard deviation of each raw operating data point. After standardization of the raw operating data, the first operating data of the nuclear power unit satisfies the following conditions: mean 0 and standard deviation 1.

[0026] Step S102: Input the first running data into the stacked autoencoder to obtain the first data features of the first running data and the first reconstruction error of the first running data.

[0027] Among them, the Stacked Autoencoder (SAE) is a deep learning model composed of multiple autoencoder (AE) layers stacked together. It extracts hierarchical features and performs data dimensionality reduction from massive amounts of normal data through unsupervised feature learning.

[0028] See Figure 2 The stacked autoencoder comprises an input layer, a decoding layer, an output layer, and at least two encoding layers. The input layer is used to input the operating data of the nuclear power unit. The at least two encoding layers are used to extract features from the input data to obtain first data features. The first encoding layer is connected to the input layer; its input includes the input operating data; the (i+1)th encoding layer's input includes the output of the ith encoding layer; and the last encoding layer's output includes the first data features. It should be noted that each encoding layer has multiple neurons, and the number of neurons in each layer can be the same or different, depending on the feature dimensions of the operating data to be extracted. For example, if the operating data to be extracted has 79 feature dimensions, the stacked autoencoder has two encoding layers: the first encoding layer has 32 neurons, and the second encoding layer has 16 neurons. The decoding layer is connected to the output of the last encoding layer, and its input includes the first data features. The decoding layer is used to reconstruct the first data features output by the at least two encoding layers to obtain the reconstruction result. The output layer is connected to the decoding layer. The input of the output layer includes the reconstruction result output by the decoding layer, and the output layer is used to output the reconstruction result obtained by the decoding layer.

[0029] The first reconstruction error is used to characterize the deviation between the first running data and the reconstruction result.

[0030] Optionally, the first reconstruction error can be characterized by mean square error, root mean square error, or mean absolute error; no specific limitation is made here.

[0031] For example, the SAE is implemented using Keras's functional API. The input layer is configured to match the feature dimensions of the first run data (e.g., 79 features), which are selected based on expertise in nuclear power plant operations. The encoder contains two fully connected layers with 32 and 16 neurons respectively. The ReLU (Rectified Linear Unit) activation function is used to mitigate the vanishing gradient problem and introduce non-linearity, enabling the SAE to capture complex relationships in the data.

[0032] For example, SAE is compiled using the Adam optimizer (learning rate = 0.001) and trained by minimizing the mean squared error (MSE) loss. The reconstruction error of SAE is defined as follows:

[0033] Where, x k The input running data (such as the first running data). The result is the reconstruction of the input running data, where N is the total number of input running data in the current round.

[0034] To prevent overfitting, an early stopping strategy can be used for SAE training, with a patience value set to 5 epochs and 20% of the training data reserved for validation. See also Figure 3 The training history graph of SAE shown below, from Figure 3 It can be seen that SAE converges quickly in the initial iterations and then tends to stabilize, indicating that it has effectively learned the underlying data patterns.

[0035] Specifically, the first running data is input into the stacked autoencoder, and the first data feature of the first running data is obtained after processing by the encoding layer, and the first reconstruction error of the first running data is obtained after processing by the decoding layer.

[0036] In practical applications, SAE reconstruction quality is evaluated using multiple metrics. (See also...) Figure 4 ,from Figure 4As can be seen, the SAE's mean squared error is 0.0061, root mean square error is 0.0778, and mean absolute error is 0.0266, indicating that SAE has high reconstruction accuracy. Furthermore, the high Pearson correlation coefficient of SAE is 0.9960, indicating a strong linear correlation between the first running data and the reconstructed data; the coefficient of determination of SAE is 0.9921, indicating that the reconstructed data almost completely reproduces all the characteristics of the original running data. The histogram shows that most errors are concentrated near zero, indicating that the errors of SAE are extremely small and random. In summary, SAE has good reconstruction quality, effectively identifying deviations as potential anomalies while reconstructing the first running data.

[0037] Step S103: Perform cluster analysis on the first data feature based on the Gaussian mixture model to obtain the minimum distance between the first data feature and the cluster centers of multiple target clusters obtained by the Gaussian mixture model.

[0038] Among them, multiple target clusters are obtained by Gaussian mixture model clustering the second data features obtained by stacked autoencoder for multiple second running data respectively.

[0039] The Gaussian mixture model includes multiple Gaussian distributions.

[0040] Step S104: Based on the comparison between the first reconstruction error and the reconstruction error threshold, and the comparison between the minimum distance and the distance threshold, determine the anomaly detection result of the nuclear power unit.

[0041] The nuclear power plant anomaly detection method provided in this embodiment acquires the first operating data of the nuclear power unit, providing comprehensive and accurate basic data for subsequent anomaly detection. The first operating data is input into a stacked autoencoder for feature extraction, effectively capturing complex nonlinear relationships within the operating data. This yields the first data features and the first reconstruction error of the first operating data, automatically extracting features without manual intervention. Furthermore, the first reconstruction error quantifies the degree to which the first operating data deviates from the normal pattern, sensitively reflecting abnormal operating conditions. Next, the first data features are input into a Gaussian mixture model for cluster analysis, obtaining the minimum distance between the first data features and the cluster centers of multiple target clusters obtained by the Gaussian mixture model. This effectively covers different normal operating modes and reflects the distribution of the first operating data in the feature space. Finally, by combining the comparison between the first reconstruction error and the reconstruction error threshold, and the comparison between the minimum distance and the distance threshold, the anomaly detection result of the nuclear power unit is determined. This dual-threshold comparison significantly improves the accuracy of nuclear power plant anomaly detection.

[0042] In some alternative implementations, the reconstruction error threshold is the maximum value among the second reconstruction errors obtained by the stacked autoencoder for multiple second running data.

[0043] Specifically, multiple second running data are input into the SAE as training data, the second reconstruction errors corresponding to each second running data are sorted, and the maximum value is used as the reconstruction error threshold.

[0044] In some alternative implementations, the distance threshold is the maximum of the target distances between each second data feature and the cluster centers of the plurality of target clusters.

[0045] Specifically, the target distances between each second data feature and the cluster centers of multiple target clusters are sorted, and the maximum value among the target distances is used as the distance threshold.

[0046] In some optional implementations, the step S104 above, which determines the anomaly detection result of the nuclear power unit based on the comparison between the first reconstruction error and the reconstruction error threshold, and the comparison between the minimum distance and the distance threshold, includes: If the first reconstruction error is greater than the reconstruction error threshold and the minimum distance is greater than the distance threshold, then the abnormal detection result of the nuclear power unit is determined to be an abnormality of the nuclear power unit.

[0047] Specifically, when the first reconstruction error is less than or equal to the reconstruction error threshold, regardless of whether the minimum distance is greater than or less than or equal to the distance threshold, the nuclear power unit is not considered to have experienced an anomaly. When the first distance is less than or equal to the distance threshold, regardless of whether the first reconstruction error is greater than or less than or equal to the distance threshold, the nuclear power unit is not considered to have experienced an anomaly.

[0048] Understandably, unless the first reconstruction error is greater than the reconstruction error threshold and the minimum distance is greater than the distance threshold at the same time, the anomaly detection results of the nuclear power unit will be that the nuclear power unit has not experienced any anomalies.

[0049] For example, if the first data feature x i If the first reconstruction error exceeds the reconstruction error threshold and the minimum distance threshold, it is considered an outlier. See [link to relevant documentation]. Figure 5 The "abnormal area" (diagonal striped area) in the upper right quadrant is shown below, and its judgment rules are as follows:

[0050] Among them, the reconstruction error threshold and distance threshold All settings are conservative. This represents the anomaly detection result for the i-th first data feature; This is an indicator function; a value of 1 indicates that an anomaly has been detected, while a value of 0 indicates that no anomaly has been detected. The first reconstruction error is the i-th first data feature; This is the minimum distance between the i-th first data feature and the cluster centers of multiple target clusters obtained by Gaussian mixture model clustering, that is, the minimum distance between the i-th first data feature and the cluster center of the nearest target cluster.

[0051] The nuclear power plant anomaly detection method provided in this embodiment determines the anomaly detection results of the nuclear power unit based on the comparison relationship between the first reconstruction error and the reconstruction error threshold, and the comparison relationship between the minimum distance and the distance threshold. This ensures that only samples with both reconstruction error and spatial distribution deviation are marked, greatly reducing the false alarm rate. It can effectively filter out abnormal fluctuations of a single indicator caused by non-faults and improve the reliability of the detection results.

[0052] In some alternative implementations, the nuclear power plant anomaly detection method further includes: Step a1: Obtain multiple sets of second running data.

[0053] The second set of operational data consists of historical operational data.

[0054] Specifically, the acquisition of the second running data is described in step S101 above, and will not be repeated here. Step a2: Input multiple second running data into the stacked autoencoder to obtain the second data features corresponding to the multiple second running data.

[0055] Specifically, the acquisition of the second data feature is described in step S102 above, and will not be repeated here. Step a3: Based on the Gaussian mixture model, cluster analysis is performed on the second data features corresponding to multiple second running data to obtain multiple target clusters.

[0056] Specifically, the second data features corresponding to the second running data are input into the Gaussian mixture model, and cluster analysis is performed on the second data features according to the number of candidate clusters to obtain multiple target clusters.

[0057] In some optional implementations, step a3 above, based on a Gaussian mixture model, performs cluster analysis on the second data features corresponding to multiple second running data to obtain multiple target clusters, including: Step a31: Obtain the number of multiple candidate clusters.

[0058] Optionally, the number of candidate clusters can be any natural number greater than 2, without any specific limitation.

[0059] Step a32: For each number of candidate clusters, the Gaussian mixture model is evaluated based on the third data features corresponding to multiple third running data, and at least one clustering index value of the Gaussian mixture model is obtained under the number of candidate clusters.

[0060] The third data features corresponding to the multiple third running data can be obtained by processing the third running data using a stacked encoder or by processing it using a variational autoencoder; no specific limitation is made here.

[0061] Optionally, the third running data can be the same as the second running data, or it can be other running data, without specific limitations here.

[0062] Specifically, for each number of candidate clusters, the Gaussian mixture model is evaluated based on the third data feature and the likelihood function to obtain at least one clustering index value of the Gaussian mixture model under the number of candidate clusters.

[0063] Step a33: Determine the target number of clusters from multiple candidate cluster numbers based on at least one clustering index value.

[0064] The number of target clusters represents the number of normal operating modes.

[0065] Step a34: Based on the target cluster number and Gaussian mixture model, perform cluster analysis on the second data features corresponding to multiple second running data to obtain multiple target clusters.

[0066] Specifically, the second data features corresponding to multiple second running data are input into the GMM, and cluster analysis is performed on the second data features according to the target cluster number to obtain multiple target clusters corresponding to multiple target cluster numbers.

[0067] In some optional implementations, at least one clustering criterion value includes a Bayesian information criterion value, an Akaike information criterion value, and a silhouette coefficient. The step a33 above, determining the target number of clusters from multiple candidate cluster numbers based on at least one clustering criterion value, includes: Step a331 involves weighted fusion of the Bayesian information criterion value, Akaike information criterion value, and silhouette coefficient under the same number of candidate clusters to obtain the clustering evaluation results of the Gaussian mixture model under each number of candidate clusters.

[0068] The clustering evaluation results are characterized by clustering scores.

[0069] Specifically, the target number of clusters varies with the clustering metric value. See also Figure 6 To balance the complexity and fitting performance of Gaussian Mixture Models (GMMs), a weighted fusion of the Bayesian information criterion, Akaike information criterion, and silhouette coefficients is performed for the same number of candidate clusters. The clustering evaluation results of the Gaussian Mixture Model for each number of candidate clusters are obtained using the following formula:

[0070] in, The values ​​represent the clustering evaluation results of the Gaussian mixture model under the corresponding number of candidate clusters; BIC represents the Bayesian information criterion value. The weights of the Bayesian information criterion values ​​are represented by AIC; AIC represents the Akaike information criterion values. The weights of the Akaike Information Criterion values; For the profile coefficient, The weights are for the silhouette coefficients.

[0071] It's important to note that balancing statistical standards and clustering quality requires considering both the clustering metrics and actual performance of the GMM. Both BIC and AIC are based on likelihood functions, but their trade-off mechanisms differ: BIC more severely penalizes complexity through the logarithm of N (ln N), favoring a simpler model; while AIC prioritizes fit quality with a lighter penalty. Giving them equal weights can balance simplicity and fit. However, since clustering quality is the ultimate goal, the silhouette coefficient, which directly measures separation and compactness, is preferred to ensure practical performance.

[0072] Furthermore, a lower BIC indicates a better GMM clustering effect. BIC is obtained using the following formula:

[0073] Where m is the number of parameters in the GMM, and m is calculated based on the sample dimension and the number of target clusters; N is the number of samples; and L is the likelihood value.

[0074] Specifically, the likelihood value is obtained using the following formula:

[0075] Where X is the set of the first data features.

[0076] Furthermore, the probability density of the Gaussian mixture model is obtained through the following formula:

[0077] in, For the model parameters in the Gaussian mixture model The probability density of the i-th first data feature; These are the model parameters for the Gaussian mixture model, containing all... , K represents the number of candidate clusters. The mixing weights are the weights of the k-th Gaussian distribution; Let be the mean vector of the k-th Gaussian distribution; Let be the covariance matrix of the k-th Gaussian distribution; Let be the probability density function of the k-th Gaussian distribution.

[0078] Furthermore, AIC is similar to BIC, but it imposes a lighter penalty on the complexity of the GMM, balancing goodness of fit and model simplicity. AIC is obtained through the following formula:

[0079] Furthermore, the silhouette coefficient evaluates the clustering effect of GMM by assessing the compactness within clusters and the separation between clusters. Its value ranges from -1 to 1; a higher silhouette coefficient indicates better clustering. The silhouette coefficient is obtained using the following formula:

[0080] Where a(i) is the intra-cluster dissimilarity of the i-th first feature data, that is, the average distance of the i-th first feature data to all other sample points in its cluster; b(i) is the inter-cluster dissimilarity of the i-th first feature data, that is, the minimum value among the average distances of the i-th first feature data to all first feature data in each other cluster. Let be the contour coefficient of the i-th first feature data.

[0081] Step a332: Based on the clustering evaluation results, determine the target number of clusters from multiple candidate cluster numbers.

[0082] Specifically, the number of candidate clusters corresponding to the highest clustering evaluation result is taken as the target number of clusters.

[0083] In practical applications, the weights in the GMM are assigned as α (30%), β (30%), and γ (40%). Cluster scores are calculated based on these weights, and the highest cluster score corresponds to a number of clusters of 4, meaning the target number of clusters is 4. Figure 7 It can be seen that each target cluster (marked with different colors and center points) exhibits a reasonable degree of separation, indicating that when the number of target clusters is 4, the clustering effect of GMM is better.

[0084] The nuclear power plant anomaly detection method provided in this embodiment obtains the clustering evaluation results of the Gaussian Mixture Model (GMM) for each number of candidate clusters by weighted fusion of the Bayesian information criterion value, the Akaike information criterion value, and the silhouette coefficient under the same number of candidate clusters, and determines the target number of clusters based on the clustering results. By flexibly adjusting the weight allocation, a comprehensive and accurate evaluation of the GMM is achieved from multiple dimensions, avoiding the limitations of a single indicator.

[0085] As a specific application embodiment of this invention, a nuclear power unit simulation design verification platform is used to collect first operating data to evaluate the effectiveness of combining a stacked autoencoder (SAE) with a gas turbine generator (GMM) in anomaly detection. The dataset is divided into a training dataset and a test dataset. The training dataset is generated based on the normal operating conditions of the nuclear power unit simulation design verification platform under 80%-100% power output, reflecting peak load regulation scenarios. The sampling interval is 1 second (1 Hz); the total sample size is 10,000 (approximately 2.78 hours of operating data); the feature dimensions are 79 (e.g., temperature, pressure, flow rate, etc.); the collection range is operating data under normal operating conditions. The test dataset also comes from the simulation design verification platform, using the same sampling interval (1 Hz), but inserts a steam generator heat transfer tube rupture (SGTR) event to evaluate anomaly detection performance. The total number of samples in the test dataset is 9404. The first 3870 samples are normal operation, the SGTR event is introduced in the 3871st sample, and the subsequent 5534 samples record the system response. Based on the detection architecture combining stacked autoencoders and GMM proposed in this invention, anomalies in nuclear power plants are detected. Test results are available in [reference needed]. Figure 8 Normal samples (pink dots) form densely overlapping clusters, while abnormal samples (cyan dots) are significantly separated from the main normal clusters, with edge overlap only at the boundaries, verifying the clustering ability of GMM. Figures 9(a) and 9(c) show the overall distribution of all test samples. Based on the combination of distance threshold and mean square error threshold, normal and abnormal sample instances can be clearly distinguished. Figures 9(b) and 9(d) show magnified views of the first detected abnormal point, highlighting the ability of the method of this invention to accurately identify deviations at their initial appearance.

[0086] It is worth noting that, as shown in Figures 9(a) and 9(c), several data points in the 2000-4000 range exceed the mean squared error threshold. However, under the dual-threshold anomaly detection framework, these points are correctly identified as normal data, demonstrating that the detection strategy combining the reconstruction error threshold and the distance threshold effectively reduces the false alarm rate while maintaining detection accuracy. In the test dataset generated by the simulation platform, the anomaly sample starts at 3871. Figures 9(b) and 9(d) show that the detection system successfully identified anomaly point 3871, proving that the method of this invention has the ability to detect anomalies immediately at the onset of operational anomalies, ensuring a timely response to potential deviations. Tests based on the nuclear power unit simulation design verification platform show that the method integrating the stacked autoencoder and Gaussian mixture model, combined with the dual-threshold strategy, can achieve robust identification with zero false alarms in the initial stage of a steam generator tube rupture (SGTR) event, effectively achieving a balance between anomaly detection sensitivity and reliability.

[0087] This embodiment also provides a nuclear power plant anomaly detection device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0088] This embodiment provides a nuclear power plant anomaly detection device, such as... Figure 10 As shown, it includes: The data acquisition module 1001 is used to acquire the first operating data of the nuclear power unit.

[0089] The feature extraction module 1002 is used to input the first running data into the stacked autoencoder to obtain the first data features of the first running data and the first reconstruction error of the first running data.

[0090] The clustering analysis module 1003 is used to perform clustering analysis on the first data features based on the Gaussian mixture model, and to obtain the minimum distance between the first data features and the cluster centers of multiple target clusters obtained by the Gaussian mixture model; wherein, the multiple target clusters are obtained by the Gaussian mixture model clustering the second data features obtained by the stacked autoencoder for multiple second running data respectively.

[0091] The anomaly detection module 1004 is used to determine the anomaly detection result of the nuclear power unit based on the comparison relationship between the first reconstruction error and the reconstruction error threshold, and the comparison relationship between the minimum distance and the distance threshold.

[0092] In some alternative implementations, the nuclear power plant anomaly detection device is also used to reconstruct the error threshold as the maximum value among the second reconstruction errors obtained by the stacked autoencoder for multiple second operating data.

[0093] In some alternative implementations, the nuclear power plant anomaly detection device is also configured to use a distance threshold that is the maximum value of the target distances between each of the second data features and the cluster centers of a plurality of target clusters.

[0094] In some optional implementations, the anomaly detection module 1004 includes: An anomaly detection unit is used to determine that the anomaly detection result of the nuclear power unit is an anomaly if the first reconstruction error is greater than the reconstruction error threshold and the minimum distance is greater than the distance threshold.

[0095] In some alternative implementations, the nuclear power plant anomaly detection device further includes: The first processing module is used to acquire multiple sets of second-running data.

[0096] The second processing module is used to input multiple second running data into the stacked autoencoder to obtain the second data features corresponding to the multiple second running data.

[0097] The third processing module is used to perform cluster analysis on the second data features corresponding to multiple second running data based on the Gaussian mixture model, so as to obtain multiple target clusters.

[0098] In some alternative implementations, the third processing module includes: The candidate number acquisition unit is used to obtain the number of multiple candidate clusters.

[0099] The evaluation unit is used to evaluate the Gaussian mixture model based on the third data features corresponding to multiple third running data for each candidate cluster number, and obtain at least one clustering index value of the Gaussian mixture model under the candidate cluster number.

[0100] The quantity determination unit is used to determine the target number of clusters from multiple candidate cluster quantities based on at least one clustering index value.

[0101] The clustering analysis unit is used to perform clustering analysis on the second data features corresponding to multiple second running data based on the target cluster number and Gaussian mixture model, so as to obtain multiple target clusters.

[0102] In some optional implementations, at least one clustering index value includes a Bayesian information criterion value, an Akaike information criterion value, and a silhouette coefficient; the quantity determination unit includes: The weighted fusion subunit is used to weight and fuse the Bayesian information criterion value, Akaike information criterion value and silhouette coefficient under the same number of candidate clusters to obtain the clustering evaluation results of the Gaussian mixture model under each number of candidate clusters.

[0103] The quantity determination subunit is used to determine the target number of clusters from multiple candidate cluster numbers based on the cluster evaluation results.

[0104] The nuclear power plant anomaly detection device provided in this embodiment of the invention can execute the nuclear power plant anomaly detection method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as in the corresponding embodiments described above, and will not be repeated here.

[0105] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0106] The following is a detailed reference. Figure 11The diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, a graphics processing unit, etc.) 1101, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1102 or a program loaded from memory 1108 into random access memory (RAM) 1103. The RAM 1103 also stores various programs and data required for the operation of the electronic device. The processor 1101, ROM 1102, and RAM 1103 are interconnected via a bus 1104. An input / output (I / O) interface 1105 is also connected to the bus 1104.

[0107] Typically, the following devices can be connected to I / O interface 1105: input devices 1106 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 1107 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 1108 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1109. Communication device 1109 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 11 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0108] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 1109, or installed from a memory 1108, or installed from a ROM 1102. When the computer program is executed by the processor 1101, it performs the functions defined in the nuclear power plant anomaly detection method of the embodiments of the present invention.

[0109] Figure 11 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0110] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the nuclear power plant anomaly detection method shown in the above embodiments is implemented.

[0111] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0112] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for detecting anomalies in a nuclear power plant, characterized in that, The method includes: Obtain the initial operating data of the nuclear power unit; The first running data is input into a stacked autoencoder to obtain the first data feature of the first running data and the first reconstruction error of the first running data; Cluster analysis is performed on the first data feature based on Gaussian mixture model to obtain the minimum distance between the first data feature and the cluster centers of multiple target clusters obtained by Gaussian mixture model clustering; wherein, the multiple target clusters are obtained by Gaussian mixture model clustering the second data features obtained by the stacked autoencoder for multiple second running data respectively; Based on the comparison between the first reconstruction error and the reconstruction error threshold, and the comparison between the minimum distance and the distance threshold, the abnormal detection result of the nuclear power unit is determined.

2. The method according to claim 1, characterized in that, The reconstruction error threshold is the maximum value among the second reconstruction errors obtained by the stacked autoencoder for the plurality of second running data.

3. The method according to claim 1, characterized in that, The distance threshold is the maximum value among the target distances between each of the second data features and the cluster centers of the plurality of target clusters.

4. The method according to claim 1, characterized in that, The determination of the anomaly detection result of the nuclear power unit based on the comparison relationship between the first reconstruction error and the reconstruction error threshold, and the comparison relationship between the minimum distance and the distance threshold, includes: If the first reconstruction error is greater than the reconstruction error threshold, and the minimum distance is greater than the distance threshold, then the abnormal detection result of the nuclear power unit is determined to be that the nuclear power unit has an abnormality.

5. The method according to claim 1, characterized in that, The method further includes: Obtain the plurality of second running data; The plurality of second running data are respectively input into the stacked autoencoder to obtain the second data features corresponding to the plurality of second running data; Based on the Gaussian mixture model, cluster analysis is performed on the second data features corresponding to the multiple second running data to obtain the multiple target clusters.

6. The method according to claim 5, characterized in that, Based on the Gaussian mixture model, cluster analysis is performed on the second data features corresponding to the multiple second running data to obtain the multiple target clusters, including: Obtain the number of multiple candidate clusters; For each number of candidate clusters, the Gaussian mixture model is evaluated based on the third data features corresponding to multiple third running data, and at least one clustering index value of the Gaussian mixture model is obtained under the number of candidate clusters. Based on the at least one clustering index value, the target number of clusters is determined from the plurality of candidate cluster numbers; Based on the target cluster size and the Gaussian mixture model, cluster analysis is performed on the second data features corresponding to the multiple second running data to obtain the multiple target clusters.

7. The method according to claim 6, characterized in that, The at least one clustering index value includes a Bayesian information criterion value, an Akaike information criterion value, and a silhouette coefficient; determining the target number of clusters from the plurality of candidate cluster numbers based on the at least one clustering index value includes: The Bayesian information criterion value, the Akaike information criterion value, and the silhouette coefficient are weighted and fused under the same number of candidate clusters to obtain the clustering evaluation results of the Gaussian mixture model under each number of candidate clusters; Based on the clustering evaluation results, the target number of clusters is determined from the plurality of candidate cluster numbers.

8. A nuclear power plant anomaly detection device, characterized in that, The device includes: The data acquisition module is used to acquire the first operating data of the nuclear power unit; The feature extraction module is used to input the first running data into the stacked autoencoder to obtain the first data feature of the first running data and the first reconstruction error of the first running data. The clustering analysis module is used to perform clustering analysis on the first data features based on a Gaussian mixture model to obtain the minimum distance between the first data features and the cluster centers of multiple target clusters obtained by the Gaussian mixture model; wherein, the multiple target clusters are obtained by the Gaussian mixture model clustering the second data features obtained by the stacked autoencoder for multiple second running data respectively; An anomaly detection module is used to determine the anomaly detection result of the nuclear power unit based on the comparison relationship between the first reconstruction error and the reconstruction error threshold, and the comparison relationship between the minimum distance and the distance threshold.

9. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the nuclear power plant anomaly detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the nuclear power plant anomaly detection method according to any one of claims 1 to 7.

Images