A Method for Abnormal Detection of Plant Equipment Based on Time Series Prediction Large Model

Abstract

The present invention relates to the technical field of industrial production, and in particular to a method for detecting abnormal conditions of plant equipment based on a time series prediction large model, including: S100, obtaining the operation status data of the plant equipment to be detected, and performing data preprocessing on the operation status data of the plant equipment to be detected to obtain the preprocessed data; S200, inputting the preprocessed data into a pre-trained time series prediction large model to obtain the high-dimensional features corresponding to the preprocessed data, the results of abnormal condition detection, and the data prediction values; the time series prediction large model includes: a high-dimensional feature acquisition unit and an abnormal condition detection and prediction unit. The present invention realizes the accurate monitoring, abnormal condition detection, and early prediction of the equipment operation status, and improves the accuracy and timeliness of equipment abnormal condition detection.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial production, and in particular, to a method for detecting abnormal conditions of plant equipment based on a large time series prediction model. Background Art

[0002] In modern industrial production, the normal operation of equipment is crucial for ensuring production efficiency and safety. However, during long-term operation, various abnormal conditions may occur in the equipment, such as faults, wear, aging, etc. If these abnormalities are not detected and processed in a timely manner, they may lead to serious production accidents and economic losses. Therefore, equipment abnormal detection has become an important research topic in industrial production.

[0003] Traditional equipment abnormal detection methods mainly rely on manual experience and regular inspections. These methods are not only time-consuming and labor-intensive, but also prone to missed detections and false detections. With the development of sensor technology and data acquisition technology, detecting abnormalities by monitoring the event sequence data of equipment operating status has become a new trend. However, when dealing with complex industrial equipment data, existing event sequence prediction methods have insufficient modeling capabilities for complex time series data, unable to effectively capture subtle changes in equipment operating status, resulting in insufficient accuracy and timeliness of abnormal detection, and poor generality of the method.

[0004] Therefore, there is an urgent need for a method for detecting abnormal conditions of plant equipment based on a large time series prediction model. Summary of the Invention

[0005] (I) Technical Problems to be Solved

[0006] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method for detecting abnormal conditions of plant equipment based on a large time series prediction model, which solves the technical problems in the prior art of insufficient modeling capabilities for complex time series data, unable to effectively capture subtle changes in equipment operating status, resulting in insufficient accuracy and timeliness of abnormal detection, and poor generality of the method.

[0007] (II) Technical Solutions

[0008] To achieve the above object, the main technical solutions adopted by the present invention include:

[0009] An embodiment of the present invention provides a method for detecting abnormal conditions of plant equipment based on a large time series prediction model, including:

[0010] S100. Obtain the operating status data of the plant equipment to be detected, and perform data preprocessing on the operating status data of the plant equipment to be detected to obtain preprocessed data;

[0011] S200. Input the preprocessed data into a pre-trained large time series prediction model to obtain the high-dimensional features corresponding to the preprocessed data, the results of anomaly detection, and the data prediction values;

[0012] The large time series prediction model includes: a high-dimensional feature acquisition unit and an anomaly detection and prediction unit.

[0013] Optionally, the method further includes:

[0014] S300. When the result of the anomaly detection is no anomaly, determine whether the data prediction value is within the upper and lower threshold ranges. If not, report an anomaly; if so, proceed to step S400;

[0015] S400. Obtain a prediction curve based on the data prediction value, and determine whether the prediction curve deviates from the historical curve. If it deviates, proceed to step S500;

[0016] The historical curve is the curve corresponding to the preprocessed data;

[0017] S500. Perform a linear mapping on the high-dimensional features corresponding to the preprocessed data to obtain a vector of a specific dimension. Match the vector of the specific dimension with the corresponding feature vectors in the pre-constructed feature library one by one to obtain the cosine similarity of the two vectors. If the cosine similarity is less than the pre-set threshold, report an anomaly;

[0018] The vector of the specific dimension is a vector with dimensions (1, 1024);

[0019] The pre-set threshold is set according to specific requirements.

[0020] Optionally, the upper and lower threshold values are obtained according to parameter limit boundaries, expert experience boundaries, or by the dynamic sliding window statistical method.

[0021] Optionally, the pre-constructed feature library includes features with dimensions (1, 1024) corresponding to the key data of different plant area equipment under different working conditions and different time periods;

[0022] The key data of different plant area equipment under different working conditions and different time periods are determined by the clustering method and the specific requirements of the plant area;

[0023] Among them, the key data of different plant area equipment under different working conditions and different time periods are input into the high-dimensional feature acquisition unit to obtain the corresponding high-dimensional features; the corresponding high-dimensional features are linearly mapped to obtain features with dimensions (1, 1024) corresponding to the key data of different working conditions and different time periods.

[0024] Optionally, before S100, it further includes:

[0025] S000. Train the large time series prediction model using the training data set to obtain a trained large time series prediction model;

[0026] The training data set includes: a normal operation state data training set and an abnormal operation state data set;

[0027] The normal operation state data training set includes the data of normal operation of each device in each factory; the abnormal operation state data set includes the data of abnormal operation of each device in each factory.

[0028] Optionally, the S000 specifically includes:

[0029] S010. Input the data of each time series in the normal operation state data training set into the corresponding high-dimensional feature acquisition unit, use the MLM technology, randomly select a certain period of time for masking processing in each iteration, and perform forward prediction of the model; use the Adam optimizer, and repeat training for multiple rounds until the MSE loss is minimized;

[0030] S020. Input the data of each time series in the normal operation state data training set and the abnormal operation state data training set into the corresponding abnormal data training set, use the mean square error as the prediction loss, the cross entropy as the abnormal detection loss, use Adam as the optimizer, and repeat training for multiple rounds until the mean square error and the cross entropy are minimized.

[0031] Optionally, the S200 specifically includes:

[0032] S210. Input the preprocessed data into the high-dimensional feature acquisition unit to obtain the high-dimensional features corresponding to the preprocessed data;

[0033] S220. Input the high-dimensional features corresponding to the preprocessed data into the abnormal detection and prediction unit to obtain the results of abnormal detection and the predicted data values.

[0034] Optionally, the high-dimensional feature acquisition unit uses the time series Transformer architecture as the basic architecture, and the high-dimensional feature acquisition unit includes: a time series data processing layer, a time series data embedding layer, a learnable prompt embedding layer, a time correlation attention mechanism layer, and a space correlation attention mechanism layer.

[0035] Optionally, the abnormal detection and prediction unit includes: a high-dimensional time series feature extraction layer and a multi-task head layer.

[0036] Optionally, in the S100, the data preprocessing of the operation state data of the plant equipment to be detected includes:

[0037] Delete the invalid data in the operating status data of the plant equipment to be detected under non-operating status, and convert the operating status data of the plant equipment after deleting the invalid data under non-operating status into a normal distribution interval to obtain the preprocessed data.

[0038] (III) Beneficial effects

[0039] The beneficial effects of the present invention are as follows: An abnormal detection method for plant equipment based on a time series prediction large model of the present invention extracts the internal relationship features of tag numbers through the self-attention mechanism of the Transformer architecture, extracts the relationship features between tag numbers through the cross-attention mechanism, and uses prompt embedding to guide the learning of data features under different conditions of different equipment, and can better learn the relationships between data with different frequencies of different plants and different equipment; at the same time, dynamic normalization variance and dynamic attention scaling factors are proposed to dynamically process the data, so as to better balance data with different distributions, and greatly improve the accuracy of abnormal detection under different equipment and conditions; the present invention proposes a time correlation attention mechanism, which can give higher weights to closer time points when calculating the attention mechanism; in addition, the multi-layer detection mechanism of the present invention has a wider coverage, can greatly improve the accuracy of abnormal detection, and at the same time enhances the real-time performance to help maintenance personnel respond quickly; furthermore, the method of the present invention is applicable to various types of industrial equipment and is not limited by equipment types and operating environments. Description of the drawings

[0040] Figure 1 It is a schematic flow chart of an abnormal detection method for plant equipment based on a time series prediction large model according to Embodiment 1 of the present invention;

[0041] Figure 2 It is a flow block diagram of an abnormal detection method for plant equipment based on a time series prediction large model according to Embodiment 2 of the present invention;

[0042] Figure 3 It is a schematic diagram of the time series prediction large model according to Embodiment 2 of the present invention. Detailed implementation manners

[0043] In order to better explain the present invention for easy understanding, the present invention will be described in detail below with reference to the drawings through specific implementation manners.

[0044] Time series data: Time series data, which specifically refers to data with time attributes continuously collected by equipment sensors / manuals in the present invention.

[0045] Tag number: The name of the data collected by different sensors, which is also called a feature in some places. In the present invention, in order to distinguish it from the features in the model hidden space, the tag number is specifically used to represent the name of the data collected by different data sources.

[0046] An abnormal detection method for plant equipment based on a large time series prediction model proposed by an embodiment of the present invention aims to solve the problems of insufficient modeling ability for complex time series data, poor model migration ability, and scarce abnormal data in the prior art. The present invention adopts a large time series prediction model based on the Transformer architecture. Compared with other algorithms, through hybrid pre-training and fine-tuning training, it can effectively capture complex patterns in equipment operation data, break through the data gap between different data distributions in different factories, and greatly improve the accuracy of abnormal detection. The present invention adopts a multi-layer detection mechanism, considering finer coverage and wider range, and greatly improves the accuracy of abnormal detection. In addition, by real-time monitoring the equipment operation status and performing multi-level detection, the detected abnormalities can be accurately classified, which is more conducive to business applications.

[0047] To better understand the above technical solutions, the exemplary embodiments of the present invention will be described in more detail with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more clear and thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0048] Embodiment 1

[0049] See Figure 1 , an abnormal detection method for plant equipment based on a large time series prediction model in this embodiment includes:

[0050] Step S100: Obtain the operation status data of the plant equipment to be detected, and perform data preprocessing on the operation status data of the plant equipment to be detected to obtain the preprocessed data;

[0051] In this embodiment, the operation status data is temperature data, pressure data, vibration data, etc.; in addition to the tag numbers of the equipment itself, such as temperature data, vibration data, pressure data, liquid level data, current data, valve opening data, etc., it can also be the tag numbers of the equipment input and output, such as inlet flow, outlet flow, etc.

[0052] Step S200: Input the preprocessed data into a pre-trained large time series prediction model to obtain the high-dimensional features corresponding to the preprocessed data, the results of abnormal detection, and the data prediction values;

[0053] The large time series prediction model includes: a high-dimensional feature acquisition unit and an abnormal detection and prediction unit.

[0054] The present invention can process various data, has universality, and at the same time has a certain ability to automatically find correlations, and does not depend on the artificially constructed tag number relationships.

[0055] In this embodiment, the operating status data of the equipment is collected in real time by sensors or workers installed on the equipment. The collected operating status data needs to be uploaded to the data management platform finally, and the platform automatically converts the data format into the format required for input to the specified model.

[0056] For example, the data input into the large time series prediction model needs to be in a normal distribution interval. At this time, after data cleaning, the data management platform will automatically perform data normalization processing to convert the data after data cleaning into a normal distribution interval.

[0057] In the specific implementation process, data cleaning mainly refers to cleaning invalid data in non-operating states such as the collected data when the equipment is in a shutdown state and the unstable data during the startup process of the sensor.

[0058] When the result of anomaly detection indicates that there is an anomaly, an alarm is issued in a timely manner, and the detected anomaly is roughly classified. At the same time, corresponding handling suggestions are provided to help maintenance personnel respond and solve problems quickly.

[0059] A method for detecting anomalies in plant equipment based on a large time series prediction model according to this embodiment solves the problems of insufficient modeling ability for complex time series data, poor model migration ability, and scarce anomaly data in the prior art by introducing the large time series prediction model technology based on the Transformer architecture, realizes precise monitoring and anomaly detection of the equipment operating status, and improves the accuracy and timeliness of equipment anomaly detection.

[0060] Embodiment 2

[0061] See Figure 2 , a method for detecting anomalies in plant equipment based on a large time series prediction model according to this embodiment includes:

[0062] Step S100: Obtain the operating status data of the plant equipment to be detected, and perform data preprocessing on the operating status data of the plant equipment to be detected to obtain the preprocessed data.

[0063] In this embodiment, the operating status data is temperature data, pressure data, vibration data, etc.; in addition to the tag numbers of the equipment itself, such as temperature data, vibration data, pressure data, liquid level data, current data, valve opening data, etc., it can also be the tag numbers of the input and output of the equipment, such as inlet flow rate, outlet flow rate, etc.

[0064] Step S200: Input the preprocessed data into the pre-trained large time series prediction model to obtain the high-dimensional features corresponding to the preprocessed data, the result of anomaly detection, and the data prediction value.

[0065] The large time series prediction model includes a high-dimensional feature acquisition unit and an anomaly detection and prediction unit.

[0066] Step S300: When the result of anomaly detection is no anomaly, determine whether the data prediction value is within the upper and lower threshold ranges. If not, report an anomaly; if so, proceed to step S400.

[0067] The upper and lower threshold values are obtained according to parameter limit boundaries, expert experience boundaries, or by the dynamic sliding window statistical method.

[0068] Step S400: Based on the data prediction value, obtain a prediction curve and determine whether the prediction curve deviates from the historical curve. If it deviates, proceed to step S500.

[0069] The historical curve is the curve corresponding to the preprocessed data.

[0070] Step S500: Linearly map the high-dimensional features corresponding to the preprocessed data to obtain a vector of a specific dimension. Match the vector of the specific dimension with the corresponding feature vectors in the pre-built feature library one by one to obtain the cosine similarity of the two vectors. If the cosine similarity is less than the pre-set threshold, report an anomaly.

[0071] The vector of the specific dimension is a vector with a dimension of (1, 1024); the pre-set threshold is set according to specific requirements.

[0072] In the specific implementation process, the logic for judging whether the data is abnormal is as follows: The anomaly detection model outputs the anomaly detection result and the data prediction value. If the output anomaly detection result is an anomaly at this time, directly report the anomaly; if the output anomaly detection result is no anomaly, then it needs to be judged in combination with the upper and lower threshold values. If the device has clear parameter limit boundaries or expert experience boundaries, select this boundary as the upper and lower threshold values. If not, use the dynamic sliding window statistical method to determine the upper and lower threshold values. If the data prediction value exceeds the threshold, report a warning; if it does not exceed the threshold, judge whether the prediction curve corresponding to the data prediction value deviates from the historical curve in the past period of time. Here, the data points of the historical curve and the prediction curve are clustered, and not belonging to the same category means possible deviation. When the result does not deviate, return normal and do not report an anomaly; if it deviates, feature comparison is required, and it is matched with the normal data of this tag number. If the cosine similarity is less than the pre-set threshold, report an anomaly, otherwise return normal and do not report an anomaly.

[0073] The pre-set threshold is specifically set according to specific requirements.

[0074] After reporting an exception, according to business requirements, alarms can be classified into real-time alarms and predictive alarms. They can be classified into different process alarms such as temperature, flow rate, pressure, etc., into equipment alarms such as motors, electrical, mechanical, etc., and also into various index alarms to meet various business requirements.

[0075] In this embodiment, the pre-constructed feature library includes features in the (1, 1024) dimension corresponding to the key data of equipment in different factories under different working conditions and different time periods.

[0076] The key data of equipment in different factories under different working conditions and different time periods are determined by clustering methods and the specific requirements of the factories where they are located.

[0077] Among them, the key data of equipment in different factories under different working conditions and different time periods are input into the high-dimensional feature acquisition unit to obtain the corresponding high-dimensional features; the corresponding high-dimensional features are linearly mapped to obtain features in the (1, 1024) dimension corresponding to the key data of different working conditions and different time periods.

[0078] In the specific implementation process, the key data of equipment in different factories under different working conditions and different time periods are the operation data during normal operation. And the feature library will be updated periodically.

[0079] In the specific implementation process, before step S100, it also includes:

[0080] Step S000: Use the training data set to train the time series prediction large model to obtain a trained time series prediction large model.

[0081] The training data set includes: a normal operation state data training set and an abnormal operation state data set.

[0082] The normal operation state data training set includes the normal operation data of each device in each factory; the abnormal operation state data set includes the abnormal operation data of each device in each factory.

[0083] In the specific implementation process, the training data set is the historical operation data of each device in each factory, including normal and abnormal data. The data comes from different factories and different devices, and the data of each device is used as a set; the data is divided into normal data and abnormal data.

[0084] Step S000 specifically includes:

[0085] Step S010: Input the data of each time series in the normal operation state data training set into the corresponding high-dimensional feature acquisition unit. Adopt the MLM (masked language modeling) technology, randomly select a certain period of time for masking in each iteration, and perform forward prediction of the model; adopt the Adam optimizer, and repeat the training for multiple rounds until the MSE loss is minimized.

[0086] Among them, MLM is a commonly used pre-training technology in NLP. Given an input sequence, randomly select a part for masking, and then the model needs to predict the masked part according to the context of the sequence; in the pre-training of time series tasks, by randomly masking a part of the time series, this masked part is predicted through a multivariate context sequence, so as to improve the feature extraction ability of the high-dimensional feature acquisition unit for time series.

[0087] In the specific training process, in each training iteration, first randomly select a certain period of time for masking the multivariate time series to determine the part to be predicted; then perform forward prediction of the model and calculate the MSE loss between the output of the model and the original unmasked true sequence; then calculate the gradient through backpropagation, and select the Adam optimizer to update the model parameters. Repeat the training for multiple rounds until the MSE loss is minimized and stop the training.

[0088] Step S020: Input the data of each time series in the normal operation state data training set and the abnormal operation state data training set into the corresponding abnormal data training set. Adopt the mean square error as the prediction loss, the cross entropy as the anomaly detection loss, and use Adam as the optimizer. Repeat the training for multiple rounds until the mean square error and the cross entropy are minimized.

[0089] In the training process of the anomaly detection and prediction unit, for the anomaly detection task, data with labels is required. Label the abnormal data, and at the same time select normal data with a similar amount for labeling. For the prediction task, no labeling is required.

[0090] In the specific training process, it is the same as the training process of the high-dimensional feature acquisition unit. In each training iteration, data is sent in, the model infers, the loss is calculated, the gradient is calculated through backpropagation, the model is updated, and the iteration is repeated. Stop training after reaching the target; use the mean square error as the prediction loss, use the cross entropy as the detection loss, and use Adam as the optimizer.

[0091] In this embodiment, step S200 specifically includes:

[0092] Step S210: Input the preprocessed data into the high-dimensional feature acquisition unit to obtain the high-dimensional features corresponding to the preprocessed data.

[0093] Step S220: Input the high-dimensional features corresponding to the preprocessed data into the anomaly detection and prediction unit to obtain the results of anomaly detection and the predicted data values.

[0094] Among them, the high-dimensional feature acquisition unit uses the time series Transformer architecture as the basic architecture. The high-dimensional feature acquisition unit includes: a time series data processing layer, a time series data embedding layer, a learnable prompt embedding layer, a temporal correlation attention mechanism layer, and a spatial correlation attention mechanism layer.

[0095] The anomaly detection and prediction unit includes: a high-dimensional time series feature extraction layer and a multi-task head layer.

[0096] In this embodiment, the time embedding layer of the high-dimensional feature acquisition unit is used to add time embeddings to the data of each time step to capture the time signal of the time series data. And the time embedding layer is a learnable linear embedding layer, which maps the timestamp to a vector space of a fixed dimension, and its dimension is the same as the feature dimension after encoding the time series data, which is convenient for feature splicing.

[0097] The time series data processing layer divides the time series data into blocks according to the set patch size, then performs feature mapping, and uses sine position encoding for position feature extraction. Finally, the time feature, data feature, and position feature are spliced together as the feature of this patch of data.

[0098] The learnable prompt embedding layer is used to capture the correlation between different devices or signals in the unified data set by introducing learnable prompts specific to the data set. The data features of all different patches from the same data set are spliced with the same prompt embedding to enhance the generalization ability of the model.

[0099] The temporal correlation attention mechanism layer uses the self-attention mechanism of the Transformer architecture to extract the internal relationship features of the signals, and uses the cross-attention mechanism to extract the relationship features between the signals. In order to better learn the relationship between data with different frequencies of different devices in different factories, the present invention also proposes a dynamic normalized variance and a dynamic attention scaling factor, so as to better balance data with different distributions. The method is to perform clustering analysis on the data, and determine the variance and scaling factor according to the clustering category of the data. Different clustering categories can have different variances and scaling factors.

[0100] In order to better solve the situation of doping of high-frequency and low-frequency signals in industrial equipment, the high-dimensional feature acquisition unit also introduces a spatial correlation attention mechanism layer. When calculating the attention weight, a higher weight is given to the closer time points, which solves the time dependence problem in the processing of high-frequency and low-frequency signals, and enhances the ability of the high-dimensional feature acquisition model to capture the dynamics of the time series.

[0101] In this embodiment, referring to Figure 3 the network architecture diagram of the large time series prediction model, considering the direct or indirect correlation relationships between devices, a time dimension encoder and a tag dimension encoder are designed. Considering the use of mixed training with data from different factories, an independent prompt guidance is designed. Considering multi-task shared features and multi-task independent operation, a multi-head structure is designed.

[0102] In this embodiment, in step S100, the data preprocessing of the operation status data of the plant equipment to be detected includes:

[0103] Deleting the invalid data in the non-operating state from the operation status data of the plant equipment to be detected, and converting the operation status data of the plant equipment after deleting the invalid data in the non-operating state into a normal distribution interval to obtain the preprocessed data.

[0104] In this embodiment, a high-dimensional feature acquisition unit and an anomaly detection and prediction unit based on the Transformer architecture are adopted. Compared with other algorithms, through hybrid pre-training and fine-tuning training, it can effectively capture the complex patterns in the equipment operation data, break through the data gap between different data distributions in different factories, and greatly improve the accuracy of anomaly detection; the present invention adopts a multi-layer detection mechanism, considering a finer and wider coverage, and greatly improves the accuracy of anomaly detection; in addition, by real-time monitoring the equipment operation status and performing multi-level detection, the detected anomalies are accurately classified, which is more conducive to business applications.

[0105] Embodiment 3

[0106] A method for detecting anomalies in plant equipment based on a large time series prediction model in this embodiment. It is assumed that in a large manufacturing plant, the production line includes multiple key equipment, such as compressors, conveyor belts, motors, cooling systems, etc.

[0107] During the implementation process, first, the operation status data of the plant equipment is obtained, such as the data collected by temperature sensors, pressure sensors, and vibration sensors installed on compressors, conveyor belts, motors, and cooling systems, etc., and the above data is uploaded to the data processing platform; at the same time, the operation status of the equipment regularly recorded by workers every day, such as operation duration, maintenance situation, etc., is also uploaded to the data processing platform.

[0108] Then, filter out the data in the shutdown state and during the sensor startup process in the data processing platform, retain the effective operation data, and perform data normalization processing on the effective operation data to convert the data into values between 0 and 1 to ensure the consistency of model input.

[0109] Input the processed data into the trained large time series prediction model for prediction and monitoring, including algorithm prediction, over-limit prediction results, and unmatched feature pairing, etc.

[0110] When an anomaly occurs, real-time alarms and prediction warnings are issued, and the anomaly situations are classified into process alarms such as temperature, flow rate, and pressure, as well as equipment alarms such as motors, electrics, and machinery. In addition, based on the classified anomaly situations, the system provides handling suggestions, such as checking sensors, adjusting equipment parameters, or performing equipment maintenance, etc.

[0111] Through the above steps, potential problems can be discovered in advance through prediction warnings, avoiding downtime losses caused by equipment failures; and through the provided handling suggestions, maintenance personnel can quickly locate the problems and take measures, shortening the fault handling time.

[0112] Furthermore, advance prediction warnings can avoid a large number of alarms that may affect on-site traceability when a failure occurs. In addition, by recording alarm and handling historical data, it helps the factory optimize the equipment maintenance strategy.

[0113] In a specific implementation manner, for the method for detecting plant equipment using a large time series prediction model of the present invention, when performing model training, a hybrid training technique is used. In addition to the generality of the Transformer architecture itself, dynamic normalization and dynamic scaling are used in data processing, and time-related attention weighting is used in attention calculation, which can enhance the general ability and make the method of the present invention applicable to various types of industrial equipment, regardless of the equipment type and operating environment.

[0114] In this embodiment, by using the method for detecting anomalies in plant equipment based on a large time series prediction model, the downtime caused by equipment failures is reduced, the operating efficiency and production capacity of the production line are improved; the safety and stability of equipment operation are improved, and the equipment maintenance and replacement costs are reduced.

[0115] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality" means two or more unless otherwise specifically defined.

[0116] In the present invention, unless otherwise clearly specified or limited, the terms "mounted", "connected", "coupled", "fixed", etc. shall be construed broadly. For example, it may be a fixed connection, a detachable connection, or integrated; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium; it may be the internal communication of two components or the interaction relationship between two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0117] In the present invention, unless otherwise clearly specified or limited, when a first feature is "on" or "under" a second feature, it may be that the first and second features are in direct contact, or the first and second features are indirectly in contact through an intermediate medium. Moreover, when a first feature is "above", "over" and "on" a second feature, it may be that the first feature is directly above or obliquely above the second feature, or it merely means that the horizontal height of the first feature is higher than that of the second feature. When a first feature is "under", "beneath" and "under" a second feature, it may be that the first feature is directly below or obliquely below the second feature, or it merely means that the horizontal height of the first feature is lower than that of the second feature.

[0118] In the description of this specification, the descriptions of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0119] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An abnormal detection method for plant area equipment based on a large time series prediction model, characterized in that Including: S100. Obtain the operation status data of the plant equipment to be detected, perform data preprocessing on the operation status data of the plant equipment to be detected, and obtain the preprocessed data; S200. Input the preprocessed data into a pre-trained time series prediction large model, and obtain the high-dimensional features, the results of anomaly detection, and the data prediction values corresponding to the preprocessed data; The time series prediction large model includes: a high-dimensional feature acquisition unit and an anomaly detection and prediction unit; among them, the high-dimensional feature acquisition unit uses the temporal Transformer architecture as the basic architecture, and the high-dimensional feature acquisition unit includes: a temporal data processing layer, a temporal data embedding layer, a learnable prompt embedding layer, a temporal correlation attention mechanism layer, and a spatial correlation attention mechanism layer; the anomaly detection and prediction unit includes: a high-dimensional temporal feature extraction layer and a multi-task head layer; The temporal data processing layer divides the preprocessed data into blocks according to the set patch size, then performs feature mapping, and at the same time uses sine position encoding for position feature extraction, and splices the time feature, data feature, and position feature as the feature of this patch of data; The temporal data embedding layer is used to add a temporal embedding to each time step data to capture the time signal of the preprocessed data, and the temporal data embedding layer is also a learnable embedding learning layer, which is used to map the timestamp to a vector space of a fixed dimension; The learnable prompt embedding layer is used to capture the correlation between different devices or signals in the unified data set by introducing learnable prompts specific to the data set; The temporal correlation attention mechanism layer extracts the internal relationship features of the tag number by using the self-attention mechanism of the Transformer architecture, and uses the cross-attention mechanism to extract the relationship features between tag numbers; The spatial correlation attention mechanism layer gives higher weights to closer time points; S300. When the result of the anomaly detection is no anomaly, judge whether the data prediction value is within the upper and lower threshold ranges. If not, report an anomaly; if so, enter step S400; S400. According to the data prediction value, obtain a prediction curve, and judge whether the prediction curve deviates from the historical curve. If it deviates, enter step S500; The historical curve is the curve corresponding to the preprocessed data; S500. Perform linear mapping on the high-dimensional features corresponding to the preprocessed data to obtain a vector of a specific dimension, match the vector of the specific dimension with the corresponding feature vectors in the pre-built feature library one by one, obtain the cosine similarity of the two vectors, and if the cosine similarity is less than the pre-set threshold, report an anomaly; The vector of the specific dimension is a vector with a dimension of (1, 1024); Before S100, it also includes: S000. Train the time series prediction large model with a training data set to obtain a trained time series prediction large model; The training data set includes: a normal operation status data training set and an abnormal operation status data set; The normal operating state data training set includes the normal operating data of each device in each factory; the abnormal operating state data set includes the abnormal operating data of each device in each factory; The S000 specifically includes: S010. Input the data of each time series in the normal operating state data training set into the corresponding high-dimensional feature acquisition unit. Using the MLM technology, randomly select a certain period of time for masking processing in each iteration and perform forward prediction of the model. Use the Adam optimizer and repeat training for multiple rounds until the MSE loss is minimized; S020. Input the data of each time series in the normal operating state data training set and the abnormal operating state data training set into the corresponding abnormal data training set. Use the mean square error as the prediction loss, the cross entropy as the abnormal detection loss, and use Adam as the optimizer. Repeat training for multiple rounds until the mean square error and the cross entropy are minimized.

2. The method for abnormal detection of plant equipment based on a time series prediction large model according to claim 1, wherein The upper and lower threshold values are obtained according to parameter limit boundaries, expert experience boundaries, or by the dynamic sliding window statistical method.

3. The method for abnormal detection of plant equipment based on a time series prediction large model according to claim 1, wherein The pre-constructed feature library includes features of (1, 1024) dimensions corresponding to the key data of different plant equipment under different working conditions and different time periods; The key data of different plant equipment under different working conditions and different time periods are determined by the clustering method and the specific requirements of the plant where they are located; Among them, input the key data of different plant equipment under different working conditions and different time periods into the high-dimensional feature acquisition unit to obtain the corresponding high-dimensional features; perform linear mapping on the corresponding high-dimensional features to obtain features of (1, 1024) dimensions corresponding to the key data of different working conditions and different time periods.

4. The method for abnormal detection of plant area equipment based on a large time series prediction model according to claim 1, wherein, The S200 specifically includes: S210. Input the preprocessed data into the high-dimensional feature acquisition unit to obtain the high-dimensional features corresponding to the preprocessed data; S220. Input the high-dimensional features corresponding to the preprocessed data into the abnormal detection and prediction unit to obtain the results of abnormal detection and the predicted data values.

5. The abnormal detection method for plant area equipment based on the time series prediction large model according to claim 1, characterized in that, In the S100, the data preprocessing of the operating state data of the plant equipment to be detected includes: Delete the invalid data in the non-operating state from the operating state data of the plant equipment to be detected, and convert the operating state data of the plant equipment after deleting the invalid data in the non-operating state into a normal distribution interval to obtain the preprocessed data.

Images