Method, system, device and medium for processing high-temperature high-pressure pipeline monitoring data

Abstract

This invention provides a method, system, equipment, and medium for processing monitoring data of high-temperature and high-pressure pipelines. The processing method includes real-time acquisition of monitoring data collected by multiple sensors on a high-temperature and high-pressure pipeline in a power plant; forming input data from monitoring data of different types of sensors within a preset time period; inputting the input data of the high-temperature and high-pressure pipeline into a pre-trained safety status prediction model; extracting features from the input data through the convolutional layer of the safety status prediction model to obtain a feature matrix; aggregating features along the time dimension through the pooling layer of the safety status prediction model to obtain global features; and performing feature compression and nonlinear mapping on the global features through the fully connected layer of the safety status prediction model to predict the safety status of the high-temperature and high-pressure pipeline. This invention can significantly reduce the amount of data transmitted and stored in pipeline monitoring data.

Description

Technical Field

[0001] This invention relates to the field of high-temperature and high-pressure pipeline data acquisition and processing technology, and in particular to a method, system, equipment and medium for processing high-temperature and high-pressure pipeline monitoring data. Background Technology

[0002] With the ongoing construction of smart power plants, the safe operation monitoring of high-temperature and high-pressure pipelines is being widely applied. Various types of sensors have been deployed on-site to collect real-time pipeline operating status data and transmit it to the power plant's central control network system for analysis and processing. However, the amount of raw data collected by the sensors is enormous. Directly processing this raw data would consume significant computing and communication resources, resulting in high processing latency and low efficiency, making it difficult to meet the needs of real-time pipeline safety early warning. Therefore, improvements are needed. Summary of the Invention

[0003] This invention provides a method, system, equipment, and medium for processing monitoring data of high-temperature and high-pressure pipelines, in order to solve the technical problem that the amount of raw data collected by existing sensors is extremely large, consuming a large amount of computing and communication resources.

[0004] This invention proposes a method for processing monitoring data of high-temperature and high-pressure pipelines, including:

[0005] The system acquires real-time monitoring data from different types of sensors on high-temperature and high-pressure pipelines in power plants, and converts the monitoring data from each sensor within a preset time period into input data. Input data from high-temperature and high-pressure pipelines is fed into a pre-trained safety status prediction model: By using the convolutional layer of the safety status prediction model, features are extracted from the input data to obtain a feature matrix that characterizes the local features of the monitoring data from each sensor. By using the pooling layer of the safety state prediction model, feature aggregation is performed on the feature matrix along the time dimension to obtain global features characterizing the high-temperature and high-pressure pipeline. The safety status of the high-temperature and high-pressure pipeline is predicted by performing feature compression and nonlinear mapping on the global features through the fully connected layer of the safety status prediction model.

[0006] In one embodiment of the present invention, the real-time acquisition of monitoring data collected by different types of sensors on the high-temperature and high-pressure pipeline of the power plant, and the conversion of the monitoring data of each sensor within a preset time period into input data, includes: Real-time acquisition of monitoring data collected by different types of sensors on high-temperature and high-pressure pipelines in power plants; determination of the working status of each sensor; retention of monitoring data corresponding to sensors whose working status is normal. The retained monitoring data is filtered to identify normal monitoring data, and the filtered monitoring data within the preset time period is used as input data.

[0007] In one embodiment of the present invention, the number of sensors of the same type inside the high-temperature and high-pressure pipeline is at least two, and they are located at the same monitoring position; The process involves acquiring real-time monitoring data from different types of sensors on the high-temperature and high-pressure pipeline of the power plant, determining the operating status of each sensor, and retaining the monitoring data corresponding to sensors whose operating status is normal. This includes: Real-time acquisition of monitoring data collected by different types of sensors; weighted fusion calculation is performed on all the same type of monitoring data to obtain fused data. Calculate the deviation between the fused data and the monitoring data of each sensor of this type, and determine the working status of each sensor based on the deviation value, retaining the monitoring data corresponding to the sensor whose working status is normal. The weighted fusion calculation satisfies: ; in, To integrate data, The number of sensors of the same type Represented as the first Each sensor at time step Monitoring data, For the first Each sensor at time step Credibility weight, For the first Each sensor operates within a preset time period, and at a specific time step. Previous monitoring data distribution For monitoring data Relative to the distribution of monitoring data The likelihood probability is given below.

[0008] In one embodiment of the present invention, the step of filtering the retained monitoring data to select normal monitoring data and forming input data from the selected monitoring data within a preset time period includes: Select one sensor in normal condition from each type of sensor and obtain the monitoring data corresponding to the selected sensor; For each selected sensor, a comprehensive score of the monitoring data at each time step within a preset time period is calculated using a preset isolated forest model. The comprehensive scores are filtered according to the preset scoring threshold, retaining the monitoring data corresponding to the comprehensive scores less than the scoring threshold, and removing the monitoring data corresponding to the comprehensive scores greater than or equal to the scoring threshold. For the location corresponding to the removed monitoring data, linear interpolation is performed based on the adjacent retained monitoring data to calculate fill data to fill the missing location, so that the length of the monitoring data and fill data of each sensor is consistent. The retained monitoring data of each sensor and the linearly interpolated fill data are combined to form the input data.

[0009] In one embodiment of the present invention, the security state prediction model undergoes architecture optimization before model training, and the architecture optimization method includes: Obtain a training sample set, which includes multiple training data and a security status label value corresponding to each training data; The multiple training data are input into the convolutional layer of the security status prediction model to be built, and the output feature values ​​of each convolutional channel in the convolutional layer are obtained. For each convolutional channel, the frequency of each output feature value among all output feature values ​​is calculated to obtain the marginal distribution of the output feature values ​​of each convolutional channel; the frequency of each security state label value among all security state label values ​​is calculated to obtain the marginal distribution of the security state label values; and the frequency of various combinations of the output feature value and security state label value of each convolutional channel appearing simultaneously among all output feature values ​​and all security state label values ​​is calculated to obtain the joint distribution of the two. For each convolutional channel, the mutual information between its output feature value and the security status label value is calculated based on its edge distribution, the edge distribution of the security status label value, and the joint distribution of its output feature value and the security status label value. The mutual information of each convolutional channel is compared with a preset threshold. Convolutional channels with mutual information greater than or equal to the preset threshold are retained, and convolutional channels with mutual information less than the preset threshold are removed to obtain the removed convolutional layer. After pruning the convolutional layers, the pooling layer and the fully connected layer are connected to obtain the completed safety state prediction model.

[0010] In one embodiment of the present invention, the mutual information between the output feature values ​​of the convolutional channels in the convolutional layer and the security status label value ,satisfy: ; in, For the first The first convolutional layer Mutual information between the output feature values ​​of each convolutional channel and the security status label value The range of values ​​for the security status label. To output the range of values ​​for the feature values, The output feature values ​​of the convolutional channels, For the security status label value, To output feature values With security status label value The joint distribution To output feature values marginal distribution, For security status label value The marginal distribution.

[0011] In one embodiment of the present invention, the security status prediction model is trained by the following method: Obtain a training sample set, which includes multiple monitoring data samples and security status label values ​​corresponding to the monitoring data samples; The monitoring data samples are input into the safety status prediction model to be trained to obtain the corresponding predicted safety status values; Based on the predicted safety status value and the safety status label value, calculate the loss value of the loss function of the safety status prediction model; Based on the loss value, the parameters of the security state prediction model to be trained are adjusted to obtain the pre-trained security state prediction model.

[0012] This invention also proposes a data processing system for monitoring high-temperature and high-pressure pipelines, comprising: The data acquisition unit is used to acquire monitoring data collected by multiple sensors on the high-temperature and high-pressure pipeline of the power plant in real time. The monitoring data of each sensor within a preset time period are aligned according to the time step and then stitched together to form the input data. The data processing unit is used to input the input data of the high-temperature and high-pressure pipeline into the pre-trained safety status prediction model: By using the convolutional layer of the safety status prediction model, features are extracted from the input data to obtain a feature matrix that characterizes the local features of the monitoring data from each sensor. By using the pooling layer of the safety state prediction model, feature aggregation is performed on the feature matrix along the time dimension to obtain global features characterizing the high-temperature and high-pressure pipeline. The safety status of the high-temperature and high-pressure pipeline is predicted by performing feature compression and nonlinear mapping on the global features through the fully connected layer of the safety status prediction model.

[0013] The present invention also proposes an electronic device, the electronic device comprising: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, enable the electronic device to perform the high-temperature and high-pressure pipeline monitoring data processing method as described above.

[0014] The present invention also proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a computer processor, causes the computer to perform the processing method for high-temperature and high-pressure pipeline monitoring data described above.

[0015] The beneficial effects of this invention are as follows: This invention proposes a method, system, equipment, and medium for processing monitoring data of high-temperature and high-pressure pipelines. It acquires monitoring data collected by multiple sensors on the high-temperature and high-pressure pipeline in real time. The monitoring data from each sensor within a preset time period are aligned according to time steps and then concatenated to form corresponding input data, i.e., an input matrix. This input data is then fed into a pre-trained safety state prediction model. Through the convolutional, pooling, and fully connected layers of the safety state prediction model, convolution, pooling, and nonlinear mapping processing of the input data is achieved, thereby obtaining the predicted safety state of the high-temperature and high-pressure pipeline. Therefore, by using the safety state prediction model to extract features and compress and encode the original sensor data, the high-dimensional original monitoring data is transformed into a low-dimensional safety state, significantly reducing the amount of data that needs to be transmitted and stored. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0017] In the attached diagram: Figure 1 This is a schematic diagram illustrating the steps of a method for processing high-temperature and high-pressure pipeline monitoring data according to an embodiment of the present invention.

[0018] Figure 2 This is a structural block diagram of a high-temperature and high-pressure pipeline monitoring data processing system provided in an embodiment of the present invention.

[0019] Figure 3 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0020] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0021] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. The drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0022] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0023] Please see Figures 1 to 3 This invention proposes a method, system, equipment, and medium for processing monitoring data of high-temperature and high-pressure pipelines. This invention relates to the field of energy and power, specifically to a method, system, equipment, and medium for processing monitoring data of high-temperature and high-pressure pipelines, which is applicable to the mechanical safety monitoring and real-time safety status assessment of high-temperature and high-pressure pipelines in power plants.

[0024] Please see Figure 1 This invention proposes a method for processing monitoring data of high-temperature and high-pressure pipelines, which may include the following steps.

[0025] Step S10: Real-time acquisition of monitoring data collected by multiple different types of sensors on the high-temperature and high-pressure pipeline of the power plant, and forming input data from the monitoring data of each sensor within a preset time period.

[0026] Specifically, five types of sensors are installed on the high-temperature and high-pressure pipelines of the power plant: stress sensors, vibration sensors, temperature sensors, displacement sensors, and load sensors. These different types of sensors collect data on pipeline support loads, pipeline vibration frequencies, pipeline stress, pipe wall temperature, and pipeline thermal expansion displacement. Specifically, stress sensors collect the stress values ​​borne by the pipeline wall, vibration sensors collect the vibration amplitude of the pipeline wall, temperature sensors collect the pipeline wall temperature, displacement sensors collect the displacement of the pipeline and its supports, and load sensors collect the load values ​​borne by the supports.

[0027] For example, each sensor collects monitoring data in real time according to a preset sampling frequency: the sampling frequency of the stress sensor and vibration sensor is 100 Hz, the sampling frequency of the temperature sensor and displacement sensor is 10 Hz, and the sampling frequency of the load sensor is 1 Hz. Monitoring data from each sensor in the high-temperature, high-pressure pipeline is received in real time and aligned according to a unified time axis.

[0028] Specifically, using the highest sampling frequency as a benchmark, the time step corresponding to 100Hz is taken as the basic time unit. For sensors with lower sampling frequencies, the data sampling rate is unified to 100Hz through linear interpolation. Data within a preset time period is selected; this preset time period can be set to 1 second, corresponding to 100 consecutive time steps. After aligning the time steps, the monitoring data from all sensors at each time step are concatenated into a data vector in a fixed order. This data vector contains stress, vibration, temperature, displacement, and load values ​​in sequence. The data vectors from the 100 consecutive time steps are stacked in chronological order to form an input data matrix with a dimension equal to the number of time steps multiplied by the sensor type. The input data is a two-dimensional input matrix, where rows correspond to time steps and columns correspond to sensor types.

[0029] In one embodiment of the present invention, in step S10, monitoring data collected by different types of sensors on the high-temperature and high-pressure pipeline of the power plant are acquired in real time, and the monitoring data of each sensor within a preset time period are used to form input data. The step may also include the following steps.

[0030] Step S110: Real-time acquisition of monitoring data collected by different types of sensors on the high-temperature and high-pressure pipeline of the power plant, determination of the working status of each sensor, and retention of the monitoring data corresponding to the sensors whose working status is normal.

[0031] Specifically, at least two sensors of the same type are installed within the high-temperature, high-pressure pipeline, located at the same monitoring position. Real-time monitoring data from multiple sensors on the pipeline is received, and the operating status of each sensor is assessed. Since multiple sensors of the same type are arranged, the average value of the current monitoring data from all sensors of the same type is calculated, and then the percentage deviation of each sensor's monitoring data from this average value is calculated. When the percentage deviation of a sensor exceeds a preset deviation threshold, that sensor is deemed to be in an abnormal operating state and is removed from the system; the monitoring data corresponding to the sensors in a normal operating state are retained.

[0032] In one embodiment of the present invention, in step S110, at least two sensors of the same type are arranged inside the high-temperature and high-pressure pipeline of the power plant.

[0033] The process involves acquiring real-time monitoring data from different types of sensors on the high-temperature and high-pressure pipeline of the power plant, determining the operating status of each sensor, and retaining the monitoring data corresponding to sensors whose operating status is normal. This includes: Step S111: Real-time acquisition of monitoring data collected by different types of sensors; weighted fusion calculation is performed on all the collected monitoring data of the same type to obtain fused data.

[0034] Step S112: Calculate the deviation between the fused data and the monitoring data of each sensor of this type, and determine the working status of each sensor based on the deviation value, and retain the monitoring data corresponding to the sensor whose working status is normal.

[0035] The weighted fusion calculation satisfies: .

[0036] in, To integrate data, The number of sensors of the same type Represented as the first Each sensor at time step Monitoring data, For the first Each sensor at time step Credibility weight, For the first Each sensor operates within a preset time period, and at a specific time step. Previous monitoring data distribution For monitoring data Relative to the distribution of monitoring data The likelihood probability is given below.

[0037] Step S120: Filter the retained monitoring data to select normal monitoring data, and form the selected monitoring data within the preset time period into input data.

[0038] Specifically, the monitoring data from each retained sensor is filtered to remove abnormal data points along the time dimension. For monitoring data collected by each sensor at multiple time steps within a preset time period, an isolated forest model is used to calculate a comprehensive score for each time step; a higher comprehensive score indicates a more abnormal data point. The comprehensive score for each time step is compared with a preset score threshold, retaining monitoring data whose comprehensive score does not exceed the threshold and removing those whose comprehensive score exceeds the threshold. After filtering, all retained normal monitoring data are aligned and concatenated according to time steps to form an input matrix, which serves as the input data for the subsequent safety status prediction model.

[0039] In one embodiment of the present invention, step S120 may include the following steps.

[0040] Step S121: Select a sensor in normal condition from each type of sensor and obtain the monitoring data corresponding to the selected sensor.

[0041] Step S122: For each selected sensor, calculate the comprehensive score of the monitoring data at each time step based on the monitoring data of the sensor at each time step within the preset time period using the preset isolated forest model.

[0042] Step S123: Filter the comprehensive scores according to the preset scoring threshold, retain the monitoring data corresponding to the comprehensive scores less than the scoring threshold, and remove the monitoring data corresponding to the comprehensive scores greater than or equal to the scoring threshold.

[0043] Specifically, when screening monitoring data, for each selected sensor in normal condition, the comprehensive score corresponding to the monitoring data at each time step within a preset time period is compared one by one with a preset scoring threshold to clearly distinguish between normal and abnormal monitoring data: monitoring data with a comprehensive score not exceeding the scoring threshold is retained, as this type of data indicates that the pipeline's operating parameters at the corresponding time step are within the normal range and can be used for subsequent safety status prediction; monitoring data with a comprehensive score exceeding the scoring threshold is removed, as this type of data is abnormal and may be caused by abnormal situations such as instantaneous sensor interference or sudden changes in pipeline operating parameters. Directly retaining it would affect the accuracy of the prediction model, so it must be removed. Step S124: For the location corresponding to the removed monitoring data, perform linear interpolation based on the adjacent retained monitoring data to calculate fill data to fill the missing location, so that the length of the monitoring data and fill data of each sensor is consistent, and combine the retained monitoring data of each sensor with the linearly interpolated fill data to form the input data.

[0044] Specifically, the removal of outlier data can lead to gaps in the time steps of the monitoring data retained by each sensor, resulting in inconsistent data sizes across different sensors. This makes it impossible to directly align and stitch the data to form an input matrix. To address this issue, after removing outlier data, linear interpolation must be performed on the monitoring data retained by all sensors to ensure that the monitoring data from each sensor is continuous and uniform in size within a preset time period.

[0045] For example, for each sensor, the remaining valid monitoring data after removing abnormal data is first sorted out, and the time step node corresponding to each valid data is recorded. For missing nodes in the time step sequence, that is, blank time steps left after removing abnormal data, the two valid monitoring data adjacent to the missing node are used as interpolation benchmarks. The estimated value of the monitoring data corresponding to the missing node is calculated by using a linear interpolation formula, and the estimated value is filled into the missing node to complete the monitoring data.

[0046] The pre-defined isolated forest model satisfies: = ; in, For time step Monitoring data, For time step Monitoring data The corresponding overall score, The total number of isolated trees in the isolated forest model is a preset fixed positive integer, which can be adjusted according to the sample size of the monitoring data. For example, when the sample size of the monitoring data is 1000-5000 groups, m can be 50-100 trees. By using multiple isolated trees in parallel to perform independent anomaly detection on the monitoring data, the randomness of detection by a single tree is avoided, thereby improving the accuracy and stability of the monitoring data screening.

[0047] The isolation tree is a binary discriminant tree constructed based on historical normal monitoring data from sensors. Its construction process is as follows: using historical normal monitoring data as samples, a specific monitoring data feature is randomly selected, and a splitting threshold is randomly chosen within the value range of that feature. This divides the monitoring data within the current node into left and right child nodes. This process of randomly selecting features and setting random splitting thresholds is repeated, continuously performing binary splits on each child node until only a single monitoring data point remains within the node, or the node can no longer be further divided, ultimately forming a complete isolation tree.

[0048] Represented as time step Monitoring data In the The length of the isolated path of an isolated tree represents the degree of deviation between the monitoring data and the normal monitoring data within a preset time period. Specifically, the normal monitoring data... Data that is similar in distribution to historical data and relatively dense requires multiple partitioning steps to isolate, resulting in a longer path; abnormal monitoring data. For example, sudden changes in pipe temperature and pressure can cause the pipe to deviate from its normal distribution, making it easier to quickly identify and thus resulting in a shorter path.

[0049] Specifically, an isolated tree is a binary discriminant tree, with the root node at the top. All monitoring data enters the partitioning process starting from the root node. (Monitoring data) Continuously select values ​​randomly for left and right binary classification until the data can no longer be divided, reaching the lowest-level leaf node. Monitor data. The total number of levels traversed from the root node to a leaf node is the length of the isolated path. (Anomaly monitoring data) Outlier, short path, normal monitoring data Dense and long.

[0050] is the standardization factor, and is a non-zero constant; To monitor the scoring of data in a single isolated tree; This represents the weighting factor corresponding to the current operating conditions of the power plant unit.

[0051] Step S20: Input the input data of the high-temperature and high-pressure pipeline into the pre-trained safety status prediction model.

[0052] Specifically, the input matrix formed in step S10 is input into the safety status prediction model. The safety status prediction model is a pre-trained lightweight convolutional neural network model deployed on [platform name] to process the input matrix and output the safety status of the high-temperature and high-pressure pipeline.

[0053] The structure of the security status prediction model includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer. The input layer is used to receive the input matrix, the convolutional layer is used to extract local features, the pooling layer is used to aggregate temporal information, and the fully connected layer is used to output the prediction result.

[0054] Step S30: Through the convolutional layer of the safety status prediction model, feature extraction is performed on the input data to obtain a feature matrix representing the local features of the monitoring data of each sensor.

[0055] Specifically, the convolutional layer of the safety status prediction model receives the input matrix from step S20, the dimension of which is the number of time steps multiplied by the number of sensor types.

[0056] The convolutional layer employs a two-layer asymmetric decomposition convolution structure. Each convolutional kernel consists of a 1x3 temporal convolution kernel and a 3x1 spatial convolution kernel. The first convolutional layer performs convolution operations on the input matrix. The 1x3 temporal convolution kernel slides along the time dimension to extract the variation pattern of each sensor between adjacent time steps, while the 3x1 spatial convolution kernel slides along the sensor dimension to extract the correlation between different sensors at the same time step.

[0057] The first convolutional layer has 64 output channels. After the first convolutional layer, a feature matrix with dimensions equal to the number of time steps multiplied by 64 is obtained. Each row in this feature matrix corresponds to a time step, and each column corresponds to the output feature of a convolutional channel. Each output feature represents a certain local pattern of the sensor monitoring data at that time step.

[0058] The second convolutional layer continues the convolution operation on the feature matrix output by the first layer, also using 1x3 and 3x1 asymmetric decomposition convolution, with the number of output channels set to 32. After the second convolutional processing, a feature matrix with a dimension of the number of time steps multiplied by 32 is obtained. Each feature value in this feature matrix represents the comprehensive response pattern of multi-sensor data within a certain local time window of the input data.

[0059] Step S40: Through the pooling layer of the safety state prediction model, feature aggregation is performed on the feature matrix along the time dimension to obtain global features characterizing the high temperature and high pressure pipeline.

[0060] Specifically, the pooling layer of the security state prediction model receives the feature matrix output by step S30, and the dimension of the feature matrix is ​​the number of time steps multiplied by 32.

[0061] The pooling layer employs adaptive average pooling to aggregate the feature matrix along the time dimension. Specifically, the pooling layer calculates the arithmetic mean of the eigenvalues ​​of each convolutional channel in the feature matrix over all time steps. For each of the 32 convolutional channels, the average eigenvalue of that channel over 100 time steps is calculated.

[0062] After pooling, the time step dimension is compressed to 1, and the dimension of the feature matrix becomes 1 multiplied by 32, resulting in a feature vector containing 32 eigenvalues. This feature vector is called the global feature because it compresses and fuses information from all time steps within the entire preset time period. Each eigenvalue represents the average response intensity of the corresponding convolutional channel within the entire time window, thus reflecting the overall pipeline status of the high-temperature and high-pressure pipeline during the current time period.

[0063] Step S50: Through the fully connected layer of the safety state prediction model, the global features are compressed and nonlinearly mapped to predict the safety state of the high-temperature and high-pressure pipeline.

[0064] Specifically, the fully connected layer of the safety state prediction model receives the global feature vector output from step S40, which has a dimension of 1 x 32. The fully connected layer performs a nonlinear transformation operation on this feature vector, mapping the 32-dimensional global feature vector to an output vector of a preset dimension. Specifically, the fully connected layer stores a weight matrix, the number of rows of which equals the output dimension and the number of columns equals the input dimension, i.e., 32. The fully connected layer achieves linear compression from the high-dimensional feature space to the low-dimensional output space by multiplying the input feature vector with the weight matrix and adding a bias term.

[0065] The preset dimension of the output vector is determined based on the granularity of the safety status assessment. In this embodiment, the output dimension is set to 1, that is, a continuous value is output as the safety status assessment value of the high-temperature and high-pressure pipeline. The value of this continuous value ranges from 0 to 1. The closer the value is to 0, the better the safety status of the pipeline; the closer the value is to 1, the worse the safety status of the pipeline.

[0066] After completing the linear transformation, the fully connected layer introduces nonlinear mapping capabilities through an activation function, enabling the model to learn complex input-output relationships. The activation function used is the Sigmoid function, which compresses the result of the linear transformation to the interval between 0 and 1, forming the final safety status assessment value. This safety status assessment value can be directly used to determine the safety level of the pipeline. For example, an assessment value less than 0.3 indicates a normal state, an assessment value between 0.3 and 0.7 indicates a warning state, and an assessment value greater than 0.7 indicates a dangerous state. For high-temperature and high-pressure pipelines, the safety status prediction model processes each step independently according to steps S20 to S50, ultimately outputting the corresponding safety status assessment value for the high-temperature and high-pressure pipeline.

[0067] In a specific embodiment, after processing by the fully connected layer, a 5-dimensional safety state feature value is output, where each element represents the probability of a certain type of failure occurring in the high-temperature and high-pressure pipeline. For example, the safety state feature value output by the safety state prediction model can be represented as S={0.12, 0.08, 0.15, 0.05, 0.10}, where the first element 0.12 represents the probability of stress-related failure, the second element 0.08 represents the probability of vibration-related failure, the third element 0.15 represents the probability of temperature-related failure, the fourth element 0.05 represents the probability of displacement-related failure, and the fifth element 0.10 represents the probability of load-related failure.

[0068] When a pipeline exhibits an anomaly, the corresponding failure probability increases significantly. For example, when a pipeline experiences stress exceeding limits, the safety state characteristic value may become S = {0.82, 0.15, 0.63, 0.91, 0.34}, where the probability of stress-related failure rises to 0.82 and the probability of displacement-related failure rises to 0.91, indicating that the pipeline has an 82% probability of experiencing a stress anomaly failure and a 91% probability of experiencing a displacement anomaly failure. This safety state characteristic value is the safety state characterization result of the high-temperature and high-pressure pipeline. Subsequently, these failure probability values ​​can be used to directly determine whether the pipeline is in a normal, warning, or dangerous state. For example, when a failure probability exceeds 0.7, it is determined to be a dangerous state; when it exceeds 0.3 but does not exceed 0.7, it is determined to be a warning state; and when it does not exceed 0.3, it is determined to be a normal state. For high-temperature and high-pressure pipelines, the safety state prediction model processes each step independently from S20 to S50, ultimately outputting the corresponding safety state characteristic value.

[0069] In one embodiment of the present invention, for the security state prediction model in step S20, architecture optimization is performed before model training, and the architecture optimization method includes the following steps.

[0070] Step S210: Obtain a training sample set, which includes multiple training data and a security status label value corresponding to each training data.

[0071] Specifically, a training sample set is obtained, which includes multiple training data points and the corresponding safety status label value for each training data point. The training data is an input matrix formed by preprocessing sensor monitoring data extracted from a historical monitoring database. The safety status label values ​​corresponding to multiple training data points are arranged sequentially to form a safety status label value sequence.

[0072] Step S220: Input the multiple training data into the convolutional layer of the security status prediction model to be constructed, and obtain the output feature value of each convolutional channel in the convolutional layer.

[0073] Specifically, multiple training data points are sequentially input into the convolutional layer of the security status prediction model to be built. The convolutional layer performs a convolution operation on each training data point, and each convolutional channel outputs a feature value for each training data point. For each convolutional channel, the feature values ​​corresponding to different training data points are arranged according to the input order of the training data to form the output feature sequence for that channel. A corresponding output feature sequence is constructed for each convolutional channel in the convolutional layer.

[0074] Step S230: For each convolutional channel, count the frequency of each output feature value among all output feature values ​​to obtain the marginal distribution of the output feature values ​​of each convolutional channel; count the frequency of each security state label value among all security state label values ​​to obtain the marginal distribution of the security state label values; count the frequency of various combinations of the output feature value and security state label value of each convolutional channel appearing simultaneously among all output feature values ​​and all security state label values ​​to obtain the joint distribution of the two.

[0075] Specifically, for each convolutional channel, the frequency of each output feature value in the total output feature values ​​is counted and its proportion is calculated to obtain the marginal distribution of the output feature values ​​for that channel. Simultaneously, the frequency of each security state label value in the total security state label values ​​is counted and its proportion is calculated to obtain the marginal distribution of the security state label values. Furthermore, the frequency of various combinations of the output feature values ​​and security state label values ​​occurring simultaneously in all output feature values ​​and all security state label values ​​is counted and its proportion is calculated to obtain the joint distribution of the two.

[0076] Step S240: For each convolutional channel, calculate the mutual information between its output feature value and the security status label value based on its edge distribution, the edge distribution of the security status label value, and the joint distribution of its output feature value and the security status label value.

[0077] Specifically, for each convolutional channel, the mutual information between the channel's output feature value and the safety status label value is calculated based on its edge distribution, the edge distribution of the safety status label value, and their joint distribution. Mutual information is calculated by summing all possible output feature values ​​and safety status label values, multiplying the joint probability of each combination by the logarithm of the product of the edge probabilities, and then summing the results of all combinations. A larger mutual information value indicates a stronger statistical dependence between the channel's output feature and the safety status label value, making the channel more valuable for predicting pipeline safety status.

[0078] Specifically, the mutual information between the output feature values ​​of the convolutional channels in the convolutional layer and the security status label value. ,satisfy: ; in, For the first The first convolutional layer Mutual information between the output feature values ​​of each convolutional channel and the security status label value The range of values ​​for the security status label. To output the range of values ​​for the feature values, The output feature values ​​of the convolutional channels, For the security status label value, To output feature values With security status label value The joint distribution To output feature values marginal distribution, For security status label value The marginal distribution.

[0079] Step S250: Compare the mutual information of each convolutional channel with a preset threshold, retain convolutional channels with mutual information greater than or equal to the preset threshold, and remove convolutional channels with mutual information less than the preset threshold to obtain the removed convolutional layer.

[0080] Specifically, the mutual information of each convolutional channel is compared with a preset threshold. The preset threshold can be set according to actual application requirements; in this embodiment, it is set to 0.3. For convolutional channels with a mutual information value greater than or equal to 0.3, the features extracted by this channel are considered to be highly correlated with the pipeline safety status and are retained. For convolutional channels with a mutual information value less than 0.3, the features extracted by this channel are considered to be unrelated to the pipeline safety status, and this convolutional channel is pruned from the convolutional layer, resulting in a pruned convolutional layer.

[0081] Step S260: After the pruned convolutional layer, connect the pooling layer and the fully connected layer to obtain the completed safety state prediction model.

[0082] Specifically, a pooling layer and a fully connected layer are connected after the pruned convolutional layer. The pooling layer is used to aggregate features from the feature matrix output by the convolutional layer along the time dimension, compressing features from multiple time steps into a single feature vector. The fully connected layer is used to perform linear transformations and nonlinear mappings on the feature vector output by the pooling layer, outputting the safe state prediction result. This completes the construction of the safe state prediction model.

[0083] In one embodiment of the present invention, the security status prediction model is trained using the following method: Obtain a training sample set, which includes multiple monitoring data samples and the corresponding security status label values ​​of the monitoring data samples.

[0084] The monitoring data samples are input into the safety status prediction model to be trained to obtain the corresponding predicted safety status values.

[0085] Based on the predicted safety status value and the safety status label value, calculate the loss value of the loss function of the safety status prediction model.

[0086] Based on the loss value, the parameters of the security state prediction model to be trained are adjusted to obtain the pre-trained security state prediction model.

[0087] Specifically, a training sample set is obtained, which includes multiple monitoring data samples and the corresponding safety status label value for each sample. The monitoring data samples are input into the safety status prediction model to be trained to obtain the predicted safety status value. The loss value is calculated based on the predicted safety status value and the safety status label value. The model parameters are adjusted based on the loss value. After multiple iterations of training, a pre-trained safety status prediction model is obtained.

[0088] In one embodiment of the invention, the security status feature values ​​output by the security status prediction model are encapsulated into data frames according to a custom Pipe-Msg protocol. The Pipe-Msg protocol, based on the Google Protocol Buffers serialization framework, encodes fields such as security status feature values, device identification, timestamps, and data priority tags into binary data frames. Compared to text formats such as JSON or XML, Protobuf serialization results in smaller data size, faster parsing speed, and supports cross-platform and cross-language data exchange. After encapsulation, the edge computing gateway generates standardized pipeline data frames, which can be directly used for subsequent encrypted transmission and network communication.

[0089] Secondly, standardized pipeline data frames are processed at the dynamic encryption and authentication layer. During initial registration, a hardware fingerprint is generated using the SM3 hash algorithm based on the CPU serial number, memory ID, and network card address. This fingerprint is bound to the device ID and written to the registration list of the internal authentication server. During two-way authentication, the gateway sends the device ID, gateway public key, random number R1, and hardware fingerprint to the authentication server. The server verifies if the device ID is in the registration list and compares it with the hardware fingerprint. If verification is successful, it generates a random number R2, signs it with its private key, and returns it. The gateway verifies the server's signature and then signs its response with its private key. After the server verifies the gateway's signature, an encrypted communication link is established. This authentication mechanism upgrades identity authentication from key-based trust to device-based trust. Even if an attacker steals the gateway's private key, authentication will still fail because the hardware fingerprint cannot be copied.

[0090] After authentication is established, the gateway and the internal network decryption server use a two-factor dynamic key evolution mechanism to generate session keys. The time factor T uses the current Unix timestamp accurate to the minute, and the data status factor C uses a counter of transmitted data frames that increments from zero and resets to zero after reaching one million frames to prevent overflow.

[0091] The key evolution method follows the formula: .

[0092] in, The pre-set 256-bit root key is derived from SM2 key negotiation and stored locally in the TPM security chip with encryption. For bitwise XOR operation, To ensure non-linear coupling between the time factor and the data factor, a 10-bit left shift is performed. This is a key derivation function based on the SM4 algorithm, which can be constructed using SM4-CMAC or SM4-based HMAC.

[0093] Key updates are triggered by either time or data conditions. A time update occurs every five minutes, and a data update occurs every thousand frames transmitted. The key is updated as soon as either condition is met. The gateway and server maintain key synchronization through initial synchronization, heartbeat synchronization every thirty seconds, and a mechanism for recovering from out-of-sync errors. Encryption uses SM4-CTR mode, grouping data frames into 128-bit blocks. An initialization vector is constructed based on a time factor and a data counter. The ciphertext is obtained by XORing the calculated keystream with the plaintext. Integrity verification uses HMAC-SM3, generating a message authentication code which is appended to the ciphertext. The gateway combines and encapsulates the frame header, initialization vector, SM4 ciphertext, and authentication code to generate the final encrypted pipeline data frame, which is then transmitted to the unidirectional transport layer.

[0094] In one embodiment of the present invention, the encrypted data frame is transmitted to a predictive adaptive unidirectional isolation device. This device integrates an edge computing module and a dual-medium transmission unit, supporting both visible light and fiber optic transmission media. The device has a built-in priority identification module that parses the priority tag in the data frame header: for warning-level data, a high-speed visible light channel is enabled, using high-power LEDs and photodiodes for reception, achieving a bandwidth of up to 100 megabits per second and a latency of less than 10 milliseconds; dedicated bandwidth ensures maximum real-time performance. For regular-level data, a reliable fiber optic channel is enabled, using single-mode fiber and optical transceivers, achieving a bandwidth of 10 megabits per second and a latency of 20 to 50 milliseconds; time-division multiplexing is used for multiple data streams to be transmitted in a time-slice rotation.

[0095] Real-time calculation of the Link Health Index (LHI), the calculation formula is as follows: .

[0096] in, The link round-trip time is measured by sending probe frames; one-way delay × 2. Bit Error Rate (CRC32 checksum failure rate). This is the latency jitter (the standard deviation of the latency of 10 consecutive probe frames). , , These are the link round-trip delay threshold, bit error rate threshold, and delay jitter threshold, which are non-zero constants. The relevant thresholds can be set according to the policy. , , The weighting coefficients can be set according to the strategy.

[0097] Specifically, a switching strategy is formulated based on the LHI value: when the LHI is between 0.8 and 1.0, the link is healthy and the current link is maintained; when the LHI is between 0.6 and 0.8, a link degradation warning is issued, and a backup link is started to warm up in preparation for switching; when the LHI is between 0.4 and 0.6, the link is severely degraded, and the system actively switches to the backup link; when the LHI is below 0.4, the link fails, a forced switch is initiated, and an alarm is triggered.

[0098] The device also predicts the LHI trend for the next 5 seconds based on the current LHI using a first-order inertial filter. The prediction formula is as follows: .

[0099] in, This is a predicted link health value for 5 seconds from now. This represents the link health index at the current moment. This represents the rate of change of the link health index over time. For the predicted time interval.

[0100] If the predicted value is expected to drop below 0.6, the handover preparation will be initiated in advance to achieve a seamless handover. For early warning level data, the device adopts a dual-transmit and selective-receive mechanism, with data being transmitted simultaneously from both visible light and fiber optic links. The internal network selects the data that arrives first for processing.

[0101] In one embodiment of the present invention, encrypted data frames are transmitted to an intranet semantic protocol adaptation server for decryption and protocol conversion. The decryption server uses the same key derivation method and the same time factor and data counter input as the edge computing gateway to synchronously generate a session key, and performs SM4-CTR decryption and SM3-MAC integrity verification on the received encrypted data frames. If the verification fails, the server discards the data frame and sends a retransmission request to the gateway, while simultaneously triggering an intranet alarm push to the intelligent operation and maintenance module.

[0102] After decryption, the server performs semantic protocol conversion. Unlike traditional solutions that only implement syntax-level mapping, this invention establishes a dynamic pipeline security semantic mapping table, directly converting physical quantities to security assessment semantics at the protocol conversion layer.

[0103] In one embodiment of the invention, the compliant data, after being converted by the semantic protocol adaptation layer, continues to be transmitted to the intranet pipeline safety assessment platform and intelligent operation and maintenance system. The system performs tasks such as stress over-limit early warning, pipeline health status assessment, and fault source analysis based on accurate and real-time pipeline monitoring data. Stress over-limit early warning predicts the remaining pipeline life based on safety semantic features combined with a finite element analysis model. Fault source analysis correlates data from multiple monitoring points to locate the root cause of the fault, such as locating stress redistribution caused by support failure. Operation and maintenance decision support generates specific maintenance suggestions, such as generating a maintenance instruction that suggests tightening support number three and requiring a shutdown for the next operation.

[0104] The application service layer simultaneously feeds back information such as device operation and fault verification to the edge computing gateway, enabling bidirectional data interaction and dynamic model optimization throughout the entire process.

[0105] Feedback information includes manually reviewed anomaly labels used to verify and correct anomaly judgments in the isolated forest model, and to optimize the dynamic threshold of the improved isolated forest model; fault tracing reports used to confirm the correlation between anomalies and actual faults, and to optimize the operating condition weight allocation strategy; sensor maintenance records used to update sensor fault history and optimize the parameters of the reliability dynamic decay model; and safety status prediction model updates used to optimize feature extraction targets. As operational data continues to accumulate, the anomaly detection accuracy continues to improve, providing decision support for the safe operation of power plant pipelines.

[0106] Please see Figure 2 In one embodiment of the present invention, a data processing system 100 for monitoring high-temperature and high-pressure pipelines is also proposed, including a data acquisition unit 110 and a data processing unit 120.

[0107] The data acquisition unit 110 is used to acquire monitoring data collected by different sensors on the high-temperature and high-pressure pipeline of the power plant in real time, and to form input data from the monitoring data of each sensor within a preset time period.

[0108] The data processing unit 120 is used to input the input data of the high-temperature and high-pressure pipeline into the pre-trained safety status prediction model: By using the convolutional layer of the safety status prediction model, features are extracted from the input data to obtain a feature matrix that characterizes the local features of the monitoring data from each sensor. By using the pooling layer of the safety state prediction model, feature aggregation is performed on the feature matrix along the time dimension to obtain global features characterizing the high-temperature and high-pressure pipeline. The safety status of the high-temperature and high-pressure pipeline is predicted by performing feature compression and nonlinear mapping on the global features through the fully connected layer of the safety status prediction model.

[0109] Please see Figure 3 In one embodiment of the present invention, an electronic device 200 is also proposed. The electronic device 200 may include a memory 210, a processor 220 and a bus, and may also include a computer program stored in the memory 210 and executable on the processor 220, such as a processing program for high temperature and high pressure pipeline monitoring data.

[0110] The memory 210 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 210 can be an internal storage unit of the electronic device 200, such as the portable hard drive of the electronic device 200. In other embodiments, the memory 210 can be an external storage device of the electronic device 200, such as a plug-in portable hard drive, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device 200. Furthermore, the memory 210 can include both internal and external storage units of the electronic device 200. The memory 210 can be used not only to store application software and various types of data installed on the electronic device 200, such as code for processing high-temperature and high-pressure pipeline monitoring data, but also to temporarily store data that has been output or will be output.

[0111] In some embodiments, the processor 220 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 220 is the control unit of the electronic device 200, connecting to various components of the entire electronic device 200 via various interfaces and lines. It executes programs or modules stored in the memory 210 (e.g., processing programs for high-temperature and high-pressure pipeline monitoring data) and calls data stored in the memory 210 to perform various functions and process data of the electronic device 200.

[0112] The processor 220 executes the operating system of the electronic device 200 and various installed applications. The processor 220 executes the applications to implement the steps in the above-described method for processing high-temperature and high-pressure pipeline monitoring data.

[0113] For example, the computer program may be divided into one or more modules, which are stored in the memory 210 and executed by the processor 220 to complete this application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the electronic device 200. For example, the computer program may be divided into a data acquisition unit 110 and a data processing unit 120.

[0114] The integrated unit implemented as a software functional module can be stored in a computer-readable storage medium, which can be non-volatile or volatile. The software functional module, stored in the storage medium, includes several instructions to cause a computer device (which may be a personal computer, computer equipment, or network device, etc.) or processor to execute some functions of the high-temperature and high-pressure pipeline monitoring data processing method described in the various embodiments of this application.

[0115] In summary, this invention proposes a method, system, equipment, and medium for processing monitoring data of high-temperature and high-pressure pipelines. It acquires monitoring data from multiple sensors on the high-temperature and high-pressure pipeline in real time, aligns the monitoring data from each sensor within a preset time period according to time steps, and then concatenates them to form corresponding input data, i.e., an input matrix. This input data is then fed into a pre-trained safety state prediction model. Through the convolutional, pooling, and fully connected layers of the safety state prediction model, convolution, pooling, and nonlinear mapping processing of the input data is achieved, thereby obtaining the predicted safety state of the high-temperature and high-pressure pipeline. Therefore, by using the safety state prediction model to extract features and compress and encode the original sensor data, the high-dimensional original monitoring data is transformed into a low-dimensional safety state, significantly reducing the amount of data that needs to be transmitted and stored.

[0116] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for processing monitoring data of high-temperature and high-pressure pipelines, characterized in that, include: The system acquires real-time monitoring data from different types of sensors on high-temperature and high-pressure pipelines in power plants, and converts the monitoring data from each sensor within a preset time period into input data. Input data from high-temperature and high-pressure pipelines is fed into a pre-trained safety status prediction model: By using the convolutional layer of the safety status prediction model, features are extracted from the input data to obtain a feature matrix that characterizes the local features of the monitoring data from each sensor. By using the pooling layer of the safety state prediction model, feature aggregation is performed on the feature matrix along the time dimension to obtain global features characterizing the high-temperature and high-pressure pipeline. The safety status of the high-temperature and high-pressure pipeline is predicted by performing feature compression and nonlinear mapping on the global features through the fully connected layer of the safety status prediction model.

2. The method for processing high-temperature and high-pressure pipeline monitoring data according to claim 1, characterized in that, The method involves acquiring real-time monitoring data from different types of sensors on the high-temperature and high-pressure pipeline of the power plant, and forming input data from the monitoring data of each sensor within a preset time period, including: Real-time acquisition of monitoring data collected by different types of sensors on high-temperature and high-pressure pipelines in power plants; determination of the working status of each sensor; retention of monitoring data corresponding to sensors whose working status is normal. The retained monitoring data is filtered to identify normal monitoring data, and the filtered monitoring data within the preset time period is used as input data.

3. The method for processing high-temperature and high-pressure pipeline monitoring data according to claim 2, characterized in that, The number of sensors of the same type inside the high-temperature and high-pressure pipeline is at least two, and they are located at the same monitoring position; The process involves acquiring real-time monitoring data from different types of sensors on the high-temperature and high-pressure pipeline of the power plant, determining the operating status of each sensor, and retaining the monitoring data corresponding to sensors whose operating status is normal. This includes: Real-time acquisition of monitoring data collected by different types of sensors; weighted fusion calculation is performed on all the same type of monitoring data to obtain fused data. Calculate the deviation between the fused data and the monitoring data of each sensor of this type, and determine the working status of each sensor based on the deviation value, retaining the monitoring data corresponding to the sensor whose working status is normal. The weighted fusion calculation satisfies: ； in, To integrate data, The number of sensors of the same type Represented as the first Each sensor at time step Monitoring data, For the first Each sensor at time step Credibility weight, For the first Each sensor operates within a preset time period, and at a specific time step. Previous monitoring data distribution For monitoring data Relative to the distribution of monitoring data The likelihood probability is given below.

4. The method for processing high-temperature and high-pressure pipeline monitoring data according to claim 3, characterized in that, The process of filtering the retained monitoring data to identify normal monitoring data, and then using the filtered monitoring data within a preset time period as input data, includes: Select one sensor in normal condition from each type of sensor and obtain the monitoring data corresponding to the selected sensor; For each selected sensor, a comprehensive score of the monitoring data at each time step within a preset time period is calculated using a preset isolated forest model. The comprehensive scores are filtered according to the preset scoring threshold, retaining the monitoring data corresponding to the comprehensive scores less than the scoring threshold, and removing the monitoring data corresponding to the comprehensive scores greater than or equal to the scoring threshold. For the location corresponding to the removed monitoring data, linear interpolation is performed based on the adjacent retained monitoring data to calculate fill data to fill the missing location, so that the length of the monitoring data and fill data of each sensor is consistent. The retained monitoring data of each sensor and the linearly interpolated fill data are combined to form the input data.

5. The method for processing high-temperature and high-pressure pipeline monitoring data according to claim 1, characterized in that, The security status prediction model undergoes architecture optimization before model training, and the architecture optimization method includes: Obtain a training sample set, which includes multiple training data and a security status label value corresponding to each training data; The multiple training data are input into the convolutional layer of the security status prediction model to be built, and the output feature values ​​of each convolutional channel in the convolutional layer are obtained. For each convolutional channel, the frequency of each output feature value among all output feature values ​​is calculated to obtain the marginal distribution of the output feature values ​​of each convolutional channel; the frequency of each security state label value among all security state label values ​​is calculated to obtain the marginal distribution of the security state label values; and the frequency of various combinations of the output feature value and security state label value of each convolutional channel appearing simultaneously among all output feature values ​​and all security state label values ​​is calculated to obtain the joint distribution of the two. For each convolutional channel, the mutual information between its output feature value and the security status label value is calculated based on its edge distribution, the edge distribution of the security status label value, and the joint distribution of its output feature value and the security status label value. The mutual information of each convolutional channel is compared with a preset threshold. Convolutional channels with mutual information greater than or equal to the preset threshold are retained, and convolutional channels with mutual information less than the preset threshold are removed to obtain the removed convolutional layer. After pruning the convolutional layers, the pooling layer and the fully connected layer are connected to obtain the completed safety state prediction model.

6. The method for processing high-temperature and high-pressure pipeline monitoring data according to claim 5, characterized in that, Mutual information between the feature values ​​output by the convolutional channels in the convolutional layer and the security status label value ,satisfy: ； in, For the first The first convolutional layer Mutual information between the output feature values ​​of each convolutional channel and the security status label value The range of values ​​for the security status label. To output the range of values ​​for the feature values, The output feature values ​​of the convolutional channels, For the security status label value, To output feature values With security status label value The joint distribution To output feature values marginal distribution, For security status label value The marginal distribution.

7. The method for processing high-temperature and high-pressure pipeline monitoring data according to claim 1, characterized in that, The security status prediction model is trained using the following method: Obtain a training sample set, which includes multiple monitoring data samples and security status label values ​​corresponding to the monitoring data samples; The monitoring data samples are input into the safety status prediction model to be trained to obtain the corresponding predicted safety status values; Based on the predicted safety status value and the safety status label value, calculate the loss value of the loss function of the safety status prediction model; Based on the loss value, the parameters of the security state prediction model to be trained are adjusted to obtain the pre-trained security state prediction model.

8. A data processing system for monitoring high-temperature and high-pressure pipelines, characterized in that, include: The data acquisition unit is used to acquire monitoring data collected by multiple sensors on the high-temperature and high-pressure pipeline of the power plant in real time. The monitoring data of each sensor within a preset time period are aligned according to the time step and then stitched together to form the input data. The data processing unit is used to input the input data of the high-temperature and high-pressure pipeline into the pre-trained safety status prediction model: By using the convolutional layer of the safety status prediction model, features are extracted from the input data to obtain a feature matrix that characterizes the local features of the monitoring data from each sensor. By using the pooling layer of the safety state prediction model, feature aggregation is performed on the feature matrix along the time dimension to obtain global features characterizing the high-temperature and high-pressure pipeline. The safety status of the high-temperature and high-pressure pipeline is predicted by performing feature compression and nonlinear mapping on the global features through the fully connected layer of the safety status prediction model.

9. An electronic device, characterized in that, The electronic device includes: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the electronic device to implement the method for processing high-temperature and high-pressure pipeline monitoring data as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by the computer's processor, causes the computer to perform the processing method for high-temperature and high-pressure pipeline monitoring data as described in any one of claims 1 to 7.

Images