Remote monitoring system for boiler pressure data

Abstract

The invention relates to the technical field of remote monitoring, in particular to a boiler pressure data remote monitoring system which comprises a data feature extraction module, and the data feature extraction module analyzes time sequence data layer by layer based on boiler pressure data flow, extracts pressure fluctuation information and obtains short-term and long-term change parameters through step-by-step convolution operation. According to the invention, through analyzing time sequence data, extracting short-term and long-term change characteristics of boiler pressure and capturing details of pressure fluctuation, understanding of changes of different time dimensions is enhanced, and in combination with low-dimension processing and multi-term parameter optimization of hidden variables, expression precision of pressure characteristics is improved. According to the dynamic clustering method, through parameter distance judgment and sample information updating, self-adaptive tracking of real-time data changes is achieved, abnormal modes are effectively recognized, feature division is carried out, the time length and frequency analysis of the abnormal states are combined, alarm signals are more targeted, faults are helped to be prevented in time, and the safety and operation stability of the boiler are integrally enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of remote monitoring, and particularly to a remote monitoring system for boiler pressure data. Background Art

[0002] The technical field of remote monitoring refers to a technical system that uses modern communication technology, sensing technology, and data processing technology to perform real-time monitoring and management on physical devices, industrial processes, environmental states, etc. This technical field is widely applied in industrial automation, smart cities, environmental monitoring, energy management, etc. By collecting data through sensors and transmitting the data to a remote control center via wireless or wired networks, it realizes real-time analysis, early warning, and optimization of the operating status of the monitored object, improving management efficiency and operating safety.

[0003] Among them, the remote monitoring system for boiler pressure data refers to an application system based on remote monitoring technology, mainly used for real-time monitoring of pressure parameters during the operation of the boiler. By collecting pressure data and uploading it to a remote monitoring platform, it realizes effective management of the safe operation of the boiler. Its purpose is to promptly detect abnormal operations, prevent equipment failures, improve energy utilization efficiency, and ensure the safety and stability of boiler equipment.

[0004] The existing technology has limitations in extracting the characteristics of boiler pressure data, making it difficult to capture both short-term and long-term fluctuation information simultaneously, resulting in insufficient recognition ability for changes in the operating state. The data processing process is based on fixed threshold judgment, lacking dynamism and being difficult to flexibly handle new types of abnormalities under different operating states. The anomaly detection method is limited to simple over-limit alarms, lacking detailed classification and analysis of anomaly types, making it difficult for managers to respond to potential problems promptly and accurately. These deficiencies increase the lag in identifying potential risks during operation and the passivity of management response, reducing the overall safety guarantee ability of the equipment. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings existing in the prior art, and to propose a remote monitoring system for boiler pressure data.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: A remote monitoring system for boiler pressure data includes:

[0007] The data feature extraction module analyzes the time series data layer by layer based on the boiler pressure data stream, extracts pressure fluctuation information, obtains short-term and long-term change parameters through successive convolution operations, and generates a pressure fluctuation eigenvalue through summation and weighted operations;

[0008] The latent variable mapping module calculates the mean and standard deviation based on the pressure fluctuation characteristic value, combines the distribution parameter sampling latent variables, performs re-parameterization operation and low-dimensional processing, optimizes multiple latent variables, and obtains the latent variable representation value;

[0009] The online dynamic clustering module determines the cluster to which the latent variable belongs through parameter distance operation based on the latent variable representation value, calculates the distance between the latent variable and the cluster center, adjusts the cluster center and updates the sample information, and generates the cluster center value and cluster confidence score;

[0010] The abnormal pattern recognition module determines the distance and confidence of latent variables item by item based on the cluster center value and cluster confidence score, marks abnormal attributes, analyzes and features abnormal samples, and obtains abnormal pattern marking results;

[0011] The status alarm module analyzes the abnormal frequency and duration of the abnormal marking sequence based on the abnormal pattern marking results, determines the duration of the abnormal state by accumulating the abnormal duration parameters in multiple time periods, determines the abnormal level by calling the frequency and duration, and generates alarm signal data.

[0012] The pressure fluctuation characteristic value includes short-term change parameters, long-term change parameters, and weighted sum value. The latent variable representation value includes mean, standard deviation, latent variable sampling value, and low-dimensional latent variable. The cluster center value and cluster confidence score are specifically the cluster center, the distance between the latent variable and the cluster center, and the cluster confidence. The abnormal pattern marking result includes the latent variable distance, confidence score, abnormal attribute mark, and abnormal pattern type. The alarm signal data includes abnormal level, occurrence duration, alarm frequency, and alarm status.

[0013] As a further solution of the present invention, the step of obtaining the pressure fluctuation characteristic value is specifically as follows:

[0014] The boiler pressure data stream is processed into time series segments using sliding window technology. A certain number of data points are extracted from each window and timestamps are recorded to obtain pressure data for a continuous time period and generate time series segmented data.

[0015] Based on the time series segmented data, a convolutional neural network layer is applied to each time period to extract the pressure fluctuation characteristics within the time period, and the short-term and long-term pressure changes are analyzed to obtain short-term and long-term pressure fluctuation parameters;

[0016] The short-term and long-term pressure fluctuation parameters are combined and fused by weighted summation, using the formula:

[0017]

[0018] Calculate and generate pressure fluctuation characteristic values;

[0019] Among them, S represents the eigenvalue of pressure fluctuation, and w s,i represents the weight parameter of short-term pressure fluctuation, and weight allocation is performed for the short-term pressure fluctuation in each time period. w l,i represents the weight parameter of long-term pressure fluctuation, and weight allocation is performed for the long-term pressure fluctuation in each time period. P s,i represents the short-term pressure fluctuation parameter in the i-th time period, and the short-term fluctuation information extracted based on convolution calculation. P l,i represents the long-term pressure fluctuation parameter in the i-th time period, and the long-term fluctuation information extracted based on convolution calculation. α i represents the adjustment coefficient in the i-th time period, which is used to balance the ratio of short-term and long-term fluctuation characteristics.

[0020] As a further solution of the present invention, the specific steps for obtaining the value of the latent variable representation are as follows:

[0021] Based on the eigenvalue of pressure fluctuation, data is extracted and the mean and standard deviation are calculated. An array of mean and standard deviation is formed by combining the timestamps of the data, and at the same time, the distribution characteristics of the initial latent variable are determined to obtain an array of statistical parameters;

[0022] Based on the array of statistical parameters, latent variable sampling is performed using the standard normal distribution, and the latent variable is initially mapped using the sample. The mapping result is standardized to form a standardized latent variable array, which is integrated into a standardized latent variable sample set;

[0023] Based on the standardized latent variable sample set, the reparameterization technique is executed, and the formula is used:

[0024] Z i = μ i + σ i × ∈ i

[0025] The reparameterized latent variable array is calculated;

[0026] Among them, Z i represents the i-th reparameterized latent variable, μ i represents the mean of the i-th data, σ i represents the standard deviation of the i-th data, ∈ i represents a random variable sampled from the standard normal distribution;

[0027] The reparameterized latent variable array is processed in low dimension, and the principal components are selected using the PCA dimensionality reduction technique to remove redundant data, and the value of the latent variable representation is obtained.

[0028] As a further solution of the present invention, the specific steps for obtaining the clustering center value and the clustering confidence score are as follows:

[0029] Based on the latent variable representation values, calculate the Euclidean distance between each latent variable and the cluster centers, record the minimum distance to each cluster center and the corresponding cluster center label, and generate a preliminary clustering determination for the latent variables.

[0030] Using the preliminary clustering determination of the latent variables, recalculate the positions of the multiple cluster centers. By averaging the positions of all latent variables belonging to the same cluster, update the coordinates of the cluster centers to obtain an updated array of cluster center coordinates.

[0031] According to the updated array of cluster center coordinates, recalculate the distance from each latent variable to the new cluster centers, using the formula:

[0032]

[0033] Perform the distance calculation to generate a set of distances between the latent variables and the cluster centers.

[0034] where D i,j represents the distance from the i-th latent variable to the j-th cluster center, z i is the latent variable representation value, c j is the coordinate of the cluster center, and γ represents an adjustment coefficient used to adjust the non-linear relationship between the latent variables and the cluster centers.

[0035] Extract the clustering confidence of the cluster centers from the set of distances between the latent variables and the cluster centers. By calculating the average distance and standard deviation of multiple cluster centers, adjust the weights of each cluster center to generate cluster center values and clustering confidence scores.

[0036] As a further solution of the present invention, the steps for obtaining the abnormal pattern marking result are specifically as follows:

[0037] Based on the cluster center values and clustering confidence scores, calculate the distance from each latent variable to the corresponding cluster center one by one, determine whether the distance exceeds the corresponding confidence threshold, and mark the latent variables that exceed the threshold to obtain a preliminary list of abnormal latent variables.

[0038] Using the preliminary list of abnormal latent variables, analyze the feature distribution of the latent variables, compare the feature deviation between the latent variables and the cluster centers, identify the features that significantly deviate from the normal clustering behavior, and generate a set of abnormal latent variables.

[0039] Analyze the set of abnormal latent variables, calculate the standardized deviation value of multiple abnormal latent variables relative to the cluster centers, using the formula:

[0040]

[0041] Identify the abnormal pattern and generate the abnormal feature analysis result.

[0042] where SDi Represents the standardized deviation value of the i-th abnormal latent variable, v k Is the critical weight of the k-th feature, x i,k Is the value of the latent variable on the k-th feature, μ k Is the average value of the feature at the clustering center, m represents the total number of features, and k represents the current feature number;

[0043] According to the abnormal feature analysis result, classify the abnormal patterns, assign abnormal pattern labels to each abnormal latent variable to identify the behavior pattern of each abnormal latent variable, and obtain the abnormal pattern label result.

[0044] As a further solution of the present invention, the obtaining step of the alarm signal data is specifically as follows:

[0045] Based on the abnormal pattern label result, statistically analyze the time series of the multi-abnormal labeled latent variables, analyze the frequency of the abnormal occurrence and the duration of the multi-abnormal state, record the frequency and duration of the multi-abnormal one by one, and obtain the abnormal frequency and duration state record;

[0046] Using the abnormal frequency and duration state record, cumulatively calculate the abnormal state parameters in multiple time periods one by one, determine the total occurrence duration of the abnormal in multiple time periods, sort out the accumulated time data, and obtain the occurrence duration of the abnormal state;

[0047] Call the occurrence duration of the abnormal state, and combine with the abnormal frequency to determine the abnormal level, using the formula:

[0048]

[0049] Calculate the level of each abnormal state to obtain the abnormal level result;

[0050] Among them, L i Represents the level of the i-th abnormal, f i Represents the frequency of the i-th abnormal, t i Represents the duration of the i-th abnormal, α is an adjustment coefficient associated with the abnormal frequency, used to adjust the influence of the frequency on the abnormal level, β is an adjustment coefficient associated with the abnormal duration, used to adjust the influence of the duration on the abnormal level, γ is a constant offset coefficient in the abnormal level calculation, δ i Is the adjustment parameter of the i-th abnormal, used to compensate for the abnormal characteristic offset;

[0051] Using the abnormal level result, generate corresponding alarm signals according to the differentiated abnormal levels, assign alarm levels to each abnormal state, and generate alarm signal data.

[0052] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0053] In the present invention, by analyzing time series data, short-term and long-term change characteristics of boiler pressure are extracted, details of pressure fluctuations are captured, and the understanding of changes in different time dimensions is enhanced. Combining the low-dimensional processing of latent variables with multi-parameter optimization improves the expression accuracy of pressure characteristics. The dynamic clustering method realizes adaptive tracking of real-time data changes through parameter distance judgment and sample information update, effectively identifies abnormal patterns and conducts feature division. Combining the duration and frequency analysis of abnormal states, the alarm signal is more targeted, helps prevent faults in a timely manner, and overall enhances the safety and operation stability of the boiler. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is the system flow chart of the present invention;

[0055] Figure 2 is the flow chart of the steps for obtaining the pressure fluctuation characteristic values of the present invention;

[0056] Figure 3 is the flow chart of the steps for obtaining the representation values of the latent variables of the present invention;

[0057] Figure 4 is the flow chart of the steps for obtaining the clustering center values and clustering confidence scores of the present invention;

[0058] Figure 5 is the flow chart of the steps for obtaining the abnormal pattern marking results of the present invention;

[0059] Figure 6 is the flow chart of the steps for obtaining the alarm signal data of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0061] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as limiting the present invention. In addition, in the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0062] Embodiment 1

[0063] See also Figure 1 , a boiler pressure data remote monitoring system comprises:

[0064] The data feature extraction module is based on the boiler pressure data stream, analyzes the time series data layer by layer, extracts the pressure fluctuation information, obtains the short-term and long-term variation parameters through convolution operation, and generates the pressure fluctuation characteristic value through summation and weighted operation;

[0065] The latent variable mapping module calculates the mean and standard deviation based on the pressure fluctuation characteristic value, combines the distribution parameter sampling latent variables, performs re-parameterization and low-dimensional processing, optimizes multiple latent variables, and obtains the latent variable representation value;

[0066] The online dynamic clustering module determines the cluster to which a sample belongs based on the latent variable representation value through parameter distance operation, calculates the distance between the latent variable and the cluster center, adjusts the cluster center and updates the sample information, and generates the cluster center value and cluster confidence score.

[0067] The abnormal pattern recognition module determines the distance and confidence of latent variables item by item based on the cluster center value and cluster confidence score, marks abnormal attributes, analyzes and features abnormal samples, and obtains abnormal pattern marking results;

[0068] The status alarm module analyzes the abnormal frequency and duration of the abnormal marking sequence based on the abnormal pattern marking results, determines the duration of the abnormal state by accumulating the abnormal duration parameters in multiple time periods, determines the abnormal level by calling the frequency and duration, and generates alarm signal data.

[0069] The pressure fluctuation characteristic values include short-term change parameters, long-term change parameters, and weighted sum values. The latent variable representation values include mean, standard deviation, latent variable sampling value, and low-dimensional latent variable. The cluster center value and cluster confidence score are specifically the cluster center, the distance between the latent variable and the cluster center, and the cluster confidence. The abnormal pattern marking results include latent variable distance, confidence score, abnormal attribute marking, and abnormal pattern type. The alarm signal data includes abnormal level, occurrence duration, alarm frequency, and alarm status.

[0070] See also Figure 2 , the specific steps for obtaining the pressure fluctuation characteristic value are:

[0071] The boiler pressure data stream is processed into time series segments using sliding window technology. A certain number of data points are extracted from each window and timestamps are recorded to obtain pressure data for a continuous time period and generate time series segmented data.

[0072] Segmenting time series using the sliding window technique is an efficient data processing method. By determining the number of data points and their corresponding timestamps in each window, the short-term variation characteristics of boiler pressure data can be accurately captured. In this process, the selection of each data window is set according to the operation cycle of the boiler and the typical frequency of pressure fluctuations, ensuring that the captured data can represent the actual pressure changes. This method can not only obtain pressure data within continuous time periods but also provide basic data for subsequent analysis. By analyzing the data in these continuous time periods, the operating state of the boiler and its possible abnormal fluctuations can be further understood. The result of this step is a pressure data set containing multiple time periods, each with its corresponding pressure value and time label, laying the foundation for the next in-depth analysis.

[0073] Based on the segmented time series data, apply a convolutional neural network layer to each time period, extract the pressure fluctuation characteristics within the time period, and analyze the short-term and long-term pressure changes to obtain short-term and long-term pressure fluctuation parameters;

[0074] Based on the further processing of the segmented time series data, using a convolutional neural network layer to extract features from the data in each time period is a crucial step. By configuring specific kernel sizes and strides, the convolutional layer can effectively identify short-term and long-term fluctuation patterns in the pressure data. This method utilizes the advantages of convolutional neural networks in feature extraction, capable of automatically learning and identifying important features in time series data. This step not only involves adjusting kernel parameters to adapt to different fluctuation characteristics but also includes selecting appropriate activation functions to enhance the model's non-linear processing ability. Through these operations, the network can output short-term and long-term pressure fluctuation parameters for each time period, providing key input information for subsequent comprehensive analysis.

[0075] Combine the short-term and long-term pressure fluctuation parameters and fuse them through weighted summation, using the formula:

[0076]

[0077] Calculate and generate the pressure fluctuation characteristic value;

[0078] where S represents the pressure fluctuation characteristic value, w s,i represents the weight parameter of short-term pressure fluctuation, and weight allocation is performed for the short-term pressure fluctuation of each time period. w l,i represents the weight parameter of long-term pressure fluctuation, and weight allocation is performed for the long-term pressure fluctuation of each time period. P s,i represents the short-term pressure fluctuation parameter of the i-th time period, based on the short-term fluctuation information extracted by convolutional calculation. P l,iThe long-term pressure fluctuation parameter representing the i-th time period, the long-term fluctuation information extracted based on convolution calculation, α i The adjustment coefficient representing the i-th time period, which is used to balance the ratio of short-term and long-term fluctuation characteristics.

[0079] Formula:

[0080]

[0081] The advantage of the formula is that by adjusting the short-term and long-term weight factors and the adjustment coefficient, it can flexibly reflect the actual situation of the boiler pressure fluctuation and meet the analysis requirements under different operating conditions.

[0082] Detailed explanation of the formula and the derivation process of the formula calculation:

[0083] Assume that in a specific monitoring period, the following data is collected: P s,i =[120, 130, 140], P l,i =

[0084] 110, 115, 120], w s,i =[0.5, 0.3, 0.2], w l,i =[0.5, 0.7, 0.8], α i =[0.05, 0.07, 0.08]

[0085] Calculate each term in the formula:

[0086]

[0087] Sum these values:

[0088] S = 67.13 + 47.62 + 36.11 = 150.86

[0089] This result indicates that the overall pressure fluctuation characteristic value is 150.86, which can comprehensively reflect the pressure fluctuation situation of the boiler during this period and further assist in monitoring and adjusting the boiler operating state.

[0090] Please refer to Figure 3 , and the specific steps for obtaining the values represented by the latent variables are as follows:

[0091] Based on the pressure fluctuation characteristic value, extract data and calculate the mean and standard deviation, form an array of mean and standard deviation in combination with the time stamps of the data, and at the same time determine the distribution characteristics of the initial latent variables to obtain an array of statistical parameters;

[0092] Data extraction based on pressure fluctuation eigenvalues first obtains real-time pressure readings from the sensor system. These readings are recorded in chronological order, and the data points within each time window are analyzed to calculate their statistical properties, including the mean and standard deviation. In this way, the pressure data for each time window is converted into an array of easily processed statistical parameters. This conversion ensures the timeliness and accuracy of the data. Further, these statistical parameters provide the necessary basis for determining the distribution characteristics of latent variables for more advanced analysis and calculation.

[0093] Based on the array of statistical parameters, the standard normal distribution is applied for latent variable sampling, and the samples are used to perform a preliminary mapping of the latent variables. The mapping results are standardized to form a standardized latent variable array, which is integrated into a standardized latent variable sample set.

[0094] Referring to the array of statistical parameters, latent variable sampling begins with the application of the standard normal distribution, where statistical parameters such as the mean and standard deviation are used to define the sampling range and density. The sampling process includes generating random sample points and adjusting the distribution of these sample points according to these parameters. This step simulates the various random fluctuations that may occur in the actual physical process. The standardization process is performed by comparing the deviation of each sample point from the overall data set to ensure that the generated latent variable sample set statistically represents the characteristics of the entire data set.

[0095] Based on the standardized latent variable sample set, the reparameterization technique is performed using the formula:

[0096] Z i =μ i +σ i ×∈ i

[0097] The reparameterized latent variable array is calculated.

[0098] Where, Z i represents the i-th reparameterized latent variable, μ i represents the mean of the i-th data, σ i represents the standard deviation of the i-th data, ∈ i represents a random variable sampled from the standard normal distribution;

[0099] Formula:

[0100] Z i =μ i +σ i ×∈ i

[0101] The advantage of the formula is that, by combining the mean, standard deviation, and a linear combination of standard normal distribution variables, it allows latent variable samples to introduce randomness while maintaining the characteristics of the original data. This method is particularly applicable to scenarios where real-world data needs to be simulated from a theoretical distribution, such as in risk management or prediction models.

[0102] Detailed Explanation of the Formula and the Derivation Process of the Formula Calculation:

[0103] Assume that the mean of the actually monitored pressure data is μ i = 100 Pascals, and the standard deviation is σ i = 15 Pascals, and the random variable ∈ i The value sampled from the standard normal distribution is 0.5, and the calculation process is as follows:

[0104] Z i = 100 + 15 × 0.5 = 107.5

[0105] This result indicates that, based on the given pressure fluctuation eigenvalue, the latent variable value obtained through the reparameterization method is 107.5 Pascals. This result is helpful for further analysis and prediction, such as predicting possible pressure anomalies in a mechanical pressure response model.

[0106] Perform low-dimensional processing on the reparameterized latent variable array, use PCA dimensionality reduction technology to select the principal components, remove redundant data, and obtain the latent variable representation value.

[0107] After completing the reparameterization, the subsequent low-dimensional processing is implemented through principal component analysis technology. This technology optimizes the data representation by identifying the main components in the data and removing noise or unimportant components. This process not only reduces the dimensionality of the data but also enhances the efficiency and accuracy of data processing. Through this technology, the most influential features can be extracted from complex multi-dimensional data, ensuring the accuracy and practicality of the latent variable representation. The obtained latent variable representation value provides precise input for subsequent applications.

[0108] Please refer to Figure 4 , and the specific steps for obtaining the clustering center values and clustering confidence scores are as follows:

[0109] Based on the latent variable representation value, calculate the Euclidean distance between each latent variable and the clustering center, record the minimum distance to each clustering center and the corresponding clustering center label, and generate a preliminary clustering determination for the latent variables;

[0110] Calculate based on the distance between each latent variable and the current cluster center. Using the existing cluster center coordinates, traverse all latent variables and calculate the distance from each latent variable to each cluster center using the Euclidean distance formula. This process involves mathematical operations of squaring and taking the square root to ensure that each latent variable is compared and evaluated to find the nearest cluster center. This operation requires comparing each latent variable one by one to ensure data accuracy and clustering efficiency. By comparing, obtain the minimum distance between each latent variable and the cluster center, thereby providing a basis for subsequent cluster center update and sample classification. Each latent variable is assigned a preliminary cluster label, which is determined based on its distance to the nearest cluster center.

[0111] Using the preliminary cluster determination of the latent variables, recalculate the positions of multiple cluster centers. By averaging the positions of all latent variables belonging to the same cluster, update the coordinates of the cluster centers to obtain an updated array of cluster center coordinates.

[0112] According to the preliminary cluster labels, adjust the positions of each cluster center. In this step, the new position of each cluster center is obtained by calculating the mean of all latent variables belonging to the same cluster. This calculation process includes summing the position vectors of all latent variables belonging to the same cluster and then dividing by the number of latent variables in that cluster to obtain the new cluster center coordinates. This step ensures that the cluster center can truly reflect the center of the cluster data it represents. By continuously iterating this process, the accuracy and efficiency of the clustering algorithm can be optimized. The new cluster center coordinates are used for the next round of clustering operations to ensure the stability and accuracy of clustering.

[0113] According to the updated array of cluster center coordinates, recalculate the distance from each latent variable to the new cluster center using the formula:

[0114]

[0115] Perform distance calculation to generate a set of distances between latent variables and cluster centers.

[0116] where, D i,j represents the distance from the i-th latent variable to the j-th cluster center, z i is the latent variable representation value, c j is the coordinate of the cluster center, and γ represents an adjustment coefficient used to adjust the non-linear relationship between the latent variable and the cluster center.

[0117] Formula:

[0118]

[0119] The advantage of the formula is that by introducing the adjustment coefficient γ, the sensitivity of the distance can be flexibly adjusted, so as to adapt to datasets with different density or distribution characteristics, making the clustering more accurate.

[0120] Detailed explanation of the formula and the derivation process of the formula calculation:

[0121] Assume that a latent variable represents a value of z i = 5.2, and the coordinates of a cluster center are c j = 3.8, and the adjustment coefficient is γ = 0.5, then:

[0122]

[0123] D i,j = 1.63

[0124] This result indicates that the distance between the latent variable and the cluster center is 1.63, and this distance will be used to determine which cluster center the latent variable should be assigned to.

[0125] Extract the clustering confidence of the cluster center from the set of distances between the latent variable and the cluster center. By calculating the average distance and standard deviation of multiple cluster centers, adjust the weight of each cluster center to generate the cluster center value and the clustering confidence score.

[0126] Calculate the clustering confidence of each cluster center, which involves the calculation of the average distance and its standard deviation of the cluster center. The weight and confidence of each cluster center are calculated based on the mean of the distances from all the latent variables inside it to the cluster center. This mean is obtained by summing up the distances from all the latent variables to the cluster center and then dividing by the number of latent variables. In addition, it is necessary to calculate the standard deviation of the distances to evaluate the consistency of the clustering. This process ensures the reliability and stability of the clustering results. By adjusting the weight of each cluster center, the responsiveness and accuracy of the entire clustering process can be optimized. The generated cluster center value and the clustering confidence score will directly affect the formulation of subsequent data processing and analysis strategies.

[0127] Please refer to Figure 5 , and the specific steps for obtaining the abnormal pattern marking result are as follows:

[0128] Based on the cluster center value and the clustering confidence score, calculate the distance from each latent variable to the corresponding cluster center one by one, determine whether the distance exceeds the corresponding confidence threshold, and mark the latent variables that exceed the threshold to obtain a preliminary list of abnormal latent variables;

[0129] Preliminary marking of abnormal latent variables is based on the numerical value of the clustering center and the clustering confidence score. The marking of abnormal latent variables is achieved by comparing the distance of each latent variable to its clustering center with the clustering confidence threshold. If the distance exceeds the threshold, the latent variable is marked as abnormal. This process includes calculating the distance between each latent variable and the clustering center, comparing the distance with the confidence threshold, and marking the latent variables whose distance exceeds the threshold. All these steps require precise data processing and calculation.

[0130] Using the preliminary list of abnormal latent variables, analyze the characteristic distribution of the latent variables, compare the characteristic deviation between the latent variables and the clustering center, identify the characteristics that significantly deviate from the normal clustering behavior, and generate a set of abnormal latent variables.

[0131] After obtaining the preliminary list of abnormal latent variables, the next step is to analyze the characteristic distribution of these latent variables. This involves extracting the characteristics of each abnormal latent variable from the dataset and comparing them with the characteristics of the corresponding clustering center. This comparison includes a direct comparison of the characteristics of the abnormal latent variables and the clustering center to find the latent variables that deviate significantly. This process requires in-depth analysis and processing of the data to ensure that the truly abnormal latent variables can be identified.

[0132] Analyze the set of abnormal latent variables, calculate the standardized deviation value of multiple abnormal latent variables relative to the clustering center, using the formula:

[0133]

[0134] Identify abnormal patterns and generate the results of abnormal feature analysis.

[0135] Among them, SD i represents the standardized deviation value of the i-th abnormal latent variable, v k is the key weight of the k-th feature, x i,k is the value of the latent variable on the k-th feature, μ k is the average value of the feature at the clustering center, m represents the total number of features, and k represents the current feature number.

[0136] Formula:

[0137]

[0138] The advantage of the formula is that it allows the deviation of each abnormal latent variable to be quantified by considering the weights of the features, which helps to distinguish which latent variables behave significantly differently from their clustering group, thus more effectively identifying potential abnormal behavior patterns.

[0139] Detailed explanation of the formula and the derivation process of the formula calculation:

[0140] Assume an abnormal latent variable x iHas m features, and the weight of each feature k is v k , the value of the latent variable on feature k is x i,k , and the average value of the corresponding feature at the cluster center is μ k . Calculate SD i . The process is as follows:

[0141] 1. For each feature k, calculate the absolute value of the difference between x i,k and μ k , and then multiply by the weight v k .

[0142] 2. Sum all the above products.

[0143] 3. Sum the squares of the weights v k , and take the square root.

[0144] 4. Divide the result of step 2 by the result of step 3 to get SD i .

[0145] If an abnormal latent variable has three features with weights 2, 3, 5 respectively, and feature values 0.5, 0.3, 0.1 respectively, while the average values of the cluster centers are 0.4, 0.2, 0.15 respectively, the calculation is as follows:

[0146] Calculate the deviation of each feature multiplied by the weight:

[0147] (2·|0.5 - 0.4|)+(3·|0.3 - 0.2|)+(5·|0.1 - 0.15|) = 0.2 + 0.3 + 0.25 = 0.75

[0148] The square root of the sum of the squares of the weights:

[0149]

[0150] Calculate SD i :

[0151] 0.75 / 6.16 ≈ 0.122

[0152] This result indicates that the deviation value of this abnormal latent variable from the cluster center is 0.122, which helps to judge its degree of abnormality and further determine how to handle this abnormal situation.

[0153] According to the analysis results of abnormal features, classify the abnormal patterns, assign abnormal pattern labels to each abnormal latent variable, identify the behavior patterns of each abnormal latent variable, and obtain the abnormal pattern label results.

[0154] The abnormal mode marking is achieved by finally classifying the set of abnormal latent variables obtained from the analysis in the foregoing steps. This includes grouping the abnormal latent variables according to their deviation types and assigning a clear mode mark to each type of abnormality. This process ensures that the specific mode of each abnormal behavior can be accurately identified and recorded, facilitating targeted measures or in-depth analysis in the future.

[0155] Please refer to Figure 6 , and the steps for obtaining the alarm signal data are specifically as follows:

[0156] Based on the abnormal mode marking results, perform statistics on the time series of multi-abnormal marked latent variables, analyze the frequency of abnormality occurrence and the duration of multi-abnormal states, record the frequency and duration of multi-abnormalities one by one, and obtain the abnormal frequency and duration status record;

[0157] Based on the abnormal mode marking results, when statistically analyzing the frequency and duration of abnormality occurrence, first monitor the timestamps of various marks through an automated system, query and aggregate abnormal events through a database, calculate the number of occurrences of each abnormality per day and the total duration, this process is implemented through SQL queries, and after aggregating the data, perform further time series analysis in the analysis software, conduct trend analysis and periodic tests on the abnormal data, identify the patterns and frequency fluctuations of abnormality occurrence, standardize the data to ensure the consistency of analysis, and through the visualization of abnormal frequency and duration status, enable monitoring personnel to quickly identify key abnormalities and persistent problems. The system will automatically record these abnormal frequency and duration data for subsequent steps to call and analyze.

[0158] Utilize the abnormal frequency and duration status record to accumulate the abnormal status parameters for each of multiple time periods one by one, determine the total occurrence duration of the abnormality within multiple time periods, organize the accumulated time data, and obtain the occurrence duration of the abnormal status;

[0159] When accumulating the abnormal status parameters for multiple time periods, the system will apply time series analysis methods based on the abnormal frequency and duration data generated in the previous step, comprehensively consider the influence of short and long cycles, and calculate the abnormal cumulative indicators under different time windows through moving average and exponential smoothing techniques. This method helps to identify the persistence and temporary fluctuations of abnormalities, perform weighted processing on the data of each time window to reflect that the recent abnormal status has a greater impact on the overall assessment, ensure timeliness and response speed, and through these comprehensive data, accurately locate the time periods and duration states where the abnormal status frequently appears, providing a scientific basis for abnormal level determination.

[0160] Call the occurrence duration of the abnormal status and combine it with the abnormal frequency to determine the abnormal level, using the formula:

[0161]

[0162] Calculate the level of each abnormal state to obtain the abnormal level result;

[0163] where L i represents the level of the i-th abnormality, f i represents the frequency of the i-th abnormality, t i represents the duration of the i-th abnormality, α is an adjustment coefficient associated with the abnormal frequency, used to adjust the influence of the frequency on the abnormal level, β is an adjustment coefficient associated with the abnormal duration, used to adjust the influence of the duration on the abnormal level, γ is a constant offset coefficient in the abnormal level calculation, and δ i is the adjustment parameter of the i-th abnormality, used to compensate for the abnormal characteristic offset;

[0164] Formula:

[0165]

[0166] The advantage of the formula is that by introducing the square terms of the frequency and duration, it emphasizes the importance of the frequency and duration of the abnormality in the level determination. At the same time, the adjustment parameter δ in the denominator i considers the deviation influence of the abnormality, enhancing the adaptability and sensitivity of the model.

[0167] Detailed explanation of the formula and the derivation process of the formula calculation:

[0168] Set f i = 5 times / day (representing the frequency of a certain abnormality occurring per day), t i = 120 minutes / day (representing the cumulative duration of this abnormality per day), α = 0.5, β = 0.3, γ = 10, δ i = 0.1, the calculation is as follows:

[0169]

[0170] The result shows that at the given abnormal frequency and duration, the level of the abnormality is 74.91, which means it is a relatively high-level abnormality, and the monitoring system should handle it with high priority.

[0171] Utilize the abnormal level result to generate corresponding alarm signals according to the differentiated abnormal levels, assign alarm levels to each abnormal state, and generate alarm signal data.

[0172] Classify and label each abnormal status according to the abnormal level result. High-level abnormalities will trigger an immediate alarm signal. The alarm system integrates multiple notification methods, including email, SMS, and in-app push. Each notification method generates detailed alarm information through a preset template, including the abnormal type, start time, duration, and recommended response measures. These templates are carefully designed to ensure the integrity and accuracy of the information. The generation process of the alarm signal is fully automated, which can ensure that the relevant operation and maintenance team is notified immediately after the abnormality occurs, respond to the abnormality at the fastest speed, and reduce possible losses or impacts.

[0173] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as the technical solution content of the present invention is not departed from, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A remote monitoring system for boiler pressure data, characterized in that, The system comprises: The data feature extraction module is based on the boiler pressure data stream, analyzes the time series data layer by layer, extracts the pressure fluctuation information, obtains the short-term and long-term variation parameters by convolution operation, and generates the pressure fluctuation characteristic value by summation and weighted operation; The latent variable mapping module calculates the mean and standard deviation based on the pressure fluctuation characteristic value, combines the distribution parameter sampling latent variables, performs re-parameterization operation and low-dimensional processing, optimizes multiple latent variables, and obtains the latent variable representation value; The online dynamic clustering module determines the cluster to which the latent variable belongs through parameter distance operation based on the latent variable representation value, calculates the distance between the latent variable and the cluster center, adjusts the cluster center and updates the sample information, and generates the cluster center value and cluster confidence score; The abnormal pattern recognition module determines the distance and confidence of latent variables item by item based on the cluster center value and cluster confidence score, marks abnormal attributes, analyzes and features abnormal samples, and obtains abnormal pattern marking results; The status alarm module analyzes the abnormal frequency and duration of the abnormal marking sequence based on the abnormal pattern marking results, determines the duration of the abnormal state by accumulating the abnormal duration parameters in multiple time periods, determines the abnormal level by calling the frequency and duration, and generates alarm signal data.

2. The remote monitoring system for boiler pressure data according to claim 1, characterized in that The pressure fluctuation characteristic value includes short-term change parameters, long-term change parameters, and weighted sum value. The latent variable representation value includes mean, standard deviation, latent variable sampling value, and low-dimensional latent variable. The cluster center value and cluster confidence score are specifically the cluster center, the distance between the latent variable and the cluster center, and the cluster confidence. The abnormal pattern marking result includes the latent variable distance, confidence score, abnormal attribute mark, and abnormal pattern type. The alarm signal data includes abnormal level, occurrence duration, alarm frequency, and alarm status.

3. The remote monitoring system for boiler pressure data according to claim 2, wherein The steps for obtaining the pressure fluctuation characteristic value are specifically as follows: The boiler pressure data stream is processed into time series segments using sliding window technology. A certain number of data points are extracted from each window and timestamps are recorded to obtain pressure data for a continuous time period and generate time series segmented data. Based on the time series segmented data, a convolutional neural network layer is applied to each time period to extract the pressure fluctuation characteristics within the time period, and the short-term and long-term pressure changes are analyzed to obtain short-term and long-term pressure fluctuation parameters; The short-term and long-term pressure fluctuation parameters are combined and fused by weighted summation, using the formula: Calculate and generate pressure fluctuation characteristic values; Among them, S represents the eigenvalue of pressure fluctuation, w s,i represents the weight parameter of short-term pressure fluctuation, and weight distribution is carried out for the short-term pressure fluctuation in each time period. w l,i represents the weight parameter of long-term pressure fluctuation, and weight distribution is carried out for the long-term pressure fluctuation in each time period. P s,i represents the short-term pressure fluctuation parameter in the i-th time period, and the short-term fluctuation information is extracted based on convolution calculation. P l,i represents the long-term pressure fluctuation parameter in the i-th time period, and the long-term fluctuation information is extracted based on convolution calculation. α i represents the adjustment coefficient in the i-th time period, which is used to balance the ratio of short-term and long-term fluctuation characteristics.

4. The remote monitoring system for boiler pressure data according to claim 3, characterized in that, The steps for obtaining the hidden variable representation value are specifically as follows: Based on the pressure fluctuation characteristic value, extract data and calculate the mean and standard deviation, combine the timestamp of the data to form a mean and standard deviation array, and determine the distribution characteristics of the initial latent variables to obtain a statistical parameter array; Based on the statistical parameter array, a standard normal distribution is applied to perform latent variable sampling, and the samples are used to perform preliminary mapping of the latent variables, and the mapping results are standardized to form a standardized latent variable array, which is integrated into a standardized latent variable sample set; Based on the standardized latent variable sample set, a reparameterization technique is performed using the formula: Z i = μ i + σ i × ∈ i Calculate and obtain the reparameterized latent variable array; Among them, Z i represents the i-th reparameterized latent variable, μ i represents the mean of the i-th data, σ i represents the standard deviation of the i-th data, ∈ i represents a random variable sampled from the standard normal distribution; Perform dimensionality reduction on the reparameterized latent variable array, use PCA dimensionality reduction technology to select the principal components, remove redundant data, and obtain the latent variable representation value.

5. The remote monitoring system for boiler pressure data according to claim 4, wherein The specific steps for obtaining the cluster center value and the cluster confidence score are as follows: Based on the latent variable representation value, calculate the Euclidean distance between each latent variable and the cluster center, record the minimum distance to each cluster center and the corresponding cluster center label, and generate a preliminary clustering determination of the latent variables. Using the preliminary clustering determination of the latent variables, recalculate the positions of multiple cluster centers, update the coordinates of the cluster centers by averaging the positions of all latent variables belonging to the same cluster, and obtain an updated array of cluster center coordinates. According to the updated array of cluster center coordinates, recalculate the distance from each latent variable to the new cluster center, using the formula: Perform distance calculation to generate a set of distances between the latent variables and the cluster centers. Among them, D i,j represents the distance from the i-th latent variable to the j-th cluster center, z i is the representation value of the latent variable, c j is the coordinate of the cluster center, and γ represents the adjustment coefficient used to adjust the non-linear relationship between the latent variable and the cluster center; Extract the cluster confidence of the cluster centers from the set of distances between the latent variables and the cluster centers, adjust the weights of each cluster center by calculating the average distance and standard deviation of multiple cluster centers, and generate the cluster center value and the cluster confidence score.

6. The remote monitoring system for boiler pressure data according to claim 5, wherein The specific steps for obtaining the abnormal pattern marking result are as follows: Based on the cluster center value and the cluster confidence score, calculate the distance from each latent variable to the corresponding cluster center one by one, determine whether the distance exceeds the corresponding confidence threshold, and mark the latent variables that exceed the threshold as abnormal to obtain a preliminary list of abnormal latent variables. Using the preliminary list of abnormal latent variables, analyze the feature distribution of the latent variables, compare the feature deviation between the latent variables and the cluster centers, identify the features that significantly deviate from the normal clustering behavior, and generate a set of abnormal latent variables. Analyze the set of abnormal latent variables, calculate the standardized deviation value of multiple abnormal latent variables relative to the cluster center, using the formula: Identify the abnormal pattern and generate the abnormal feature analysis result. Among them, SD i represents the standardized deviation value of the i-th abnormal latent variable, v k is the key weight of the k-th feature, x i,k is the value of the latent variable on the k-th feature, μ k is the average value of the feature at the cluster center, m represents the total number of features, and k represents the current feature number; According to the abnormal feature analysis result, classify the abnormal patterns, assign an abnormal pattern label to each abnormal latent variable to identify the behavior pattern of each abnormal latent variable, and obtain the abnormal pattern marking result.

7. The remote monitoring system for boiler pressure data according to claim 6, wherein The specific steps for obtaining the alarm signal data are as follows: Based on the abnormal pattern marking result, perform statistics on the time series of multiple abnormally marked latent variables, analyze the frequency of abnormal occurrences and the duration of multiple abnormal states, record the frequency and duration of multiple abnormalities one by one, and obtain the abnormal frequency and duration record. Using the abnormal frequency and duration record, accumulate the abnormal state parameters for each of multiple time periods one by one, determine the total occurrence duration of the abnormality in multiple time periods, organize the accumulated time data, and obtain the occurrence duration of the abnormal state. Call the occurrence duration of the abnormal state and combine it with the abnormal frequency to determine the abnormal level, using the formula: Calculate the level of each abnormal state to obtain the abnormal level result. Among them, L i represents the level of the i-th anomaly, f i represents the frequency of the i-th anomaly, t i represents the duration of the i-th anomaly, α is an adjustment coefficient associated with the anomaly frequency, used to adjust the influence of the frequency on the anomaly level, β is an adjustment coefficient associated with the anomaly duration, used to adjust the influence of the duration on the anomaly level, γ is a constant offset coefficient in the anomaly level calculation, δ i is the adjustment parameter of the i-th anomaly, used to compensate for the offset of the anomaly characteristics; Using the abnormal level result, generate corresponding alarm signals according to the differentiated abnormal levels, assign an alarm level to each abnormal state, and generate the alarm signal data.

Images