A natural gas station equipment early warning system and method

Abstract

This invention discloses a natural gas station equipment early warning system and method, belonging to the technical field of oil and gas field station equipment analysis. The natural gas station equipment early warning system of this invention includes a data acquisition module, which transmits the acquired data to a data preprocessing module. The data preprocessing module transmits the preprocessed data to a data mining module. The data mining module is connected to a feature engineering module, which transmits the processed feature set to a feature selection module. The feature selection module selects features and transmits the selected features to an algorithm model module. The algorithm model module is used to construct a multi-level hierarchical adaptive model and is also connected to an early warning module for generating early warning signals. This natural gas station equipment early warning system and method of this invention achieves accurate monitoring and early warning of equipment operating status by combining fuzzy optimization algorithms with a multi-level hierarchical adaptive model.

Description

Technical Field

[0001] This invention relates to the field of oil and gas field station equipment analysis technology, specifically to a natural gas station equipment early warning system and method. Background Technology

[0002] With the continuous growth of global energy demand, oil and natural gas, as major energy sources, occupy an important position in global economic development. The exploration and development of oil and gas fields is a crucial link in obtaining oil and natural gas resources, and the operation and maintenance of the equipment directly affects the production efficiency and safety of oil and gas fields. Oil and gas field equipment is diverse, encompassing drilling equipment, oil production equipment, water injection equipment, storage and transportation equipment, and related auxiliary equipment. These devices operate in high-temperature, high-pressure, and corrosive environments, facing severe operational challenges. Therefore, ensuring the efficient and stable operation of equipment is crucial for guaranteeing the sustainable production of oil and gas fields. During oil and gas field operations, emphasis is placed on eliminating major hidden dangers, establishing a safety prevention and control system, and solidly promoting the modernization of the safety production governance system and governance capabilities. As the core of oil and gas field operations, the reliability of oil and gas field equipment must be strictly guaranteed to ensure the normal operation of the oil and gas field. Therefore, a systematic technical method for analyzing oil and gas field equipment is needed to improve equipment reliability and ensure the normal operation of oil and gas field stations. Summary of the Invention

[0003] The purpose of this invention is to provide a natural gas station equipment early warning system and method to solve the technical problems of poor safety and reliability in the operation of existing oil and gas field equipment.

[0004] To achieve the above objectives, one embodiment of the present invention provides a natural gas station equipment early warning system, including a data acquisition module, which transmits the acquired data to a data preprocessing module, which transmits the preprocessed data to a data mining module, which is connected to a feature engineering module, which transmits the processed feature set to a feature selection module, which selects features and transmits the selected features to an algorithm model module, which is used to construct a multi-level hierarchical adaptive model, and is also connected to an early warning module for generating early warning signals.

[0005] In one preferred embodiment of the present invention, the natural gas station equipment early warning system further includes an iterative optimization module for optimizing a multi-level hierarchical adaptive model, wherein the iterative optimization module is connected to the algorithm model module and the early warning module respectively.

[0006] In one preferred embodiment of the present invention, the data acquisition module includes a historical data acquisition unit and a real-time data acquisition unit.

[0007] In one preferred embodiment of the present invention, the data preprocessing module includes a fuzzy calculation unit and a multivariate Gaussian distribution analysis unit. The fuzzy calculation unit is used to quantify knowledge in fuzzy data, and the multivariate Gaussian distribution analysis unit is used to remove abnormal factors from the original data.

[0008] In one preferred embodiment of the present invention, the multivariate Gaussian distribution analysis unit includes a normalization unit, which is used to control data of different dimensions to the same scale.

[0009] In one preferred embodiment of the present invention, the data mining module includes a label expansion unit and a data mining unit for mining deep correlations about phenomena within the data and correlations between data.

[0010] In one preferred embodiment of the present invention, the feature engineering module parses and splits the original data information using wavelet analysis feature extraction method, providing a high-quality and representative feature set for the feature selection module.

[0011] In one preferred embodiment of the present invention, the iterative optimization module includes an online model unit and an offline model unit. The online model unit optimizes the multi-level hierarchical adaptive model through data accumulation and the accumulation of working conditions, while the offline model unit optimizes the multi-level hierarchical adaptive model by finding the optimal values ​​of all parameters in the algorithm through a global search method.

[0012] In one preferred embodiment of the present invention, the early warning module includes a model parameter configuration unit, a dynamic monitoring and early warning unit, an early warning information tracking and processing unit, a data trend analysis unit, and an early warning information statistical analysis unit.

[0013] This invention also discloses a method for early warning of natural gas station equipment, implemented based on the aforementioned early warning system for natural gas station equipment, comprising the following steps:

[0014] Collect data from natural gas station equipment and preprocess the collected data;

[0015] Feature extraction is performed on the preprocessed data and the original data, and the extracted features are selected.

[0016] A multi-level hierarchical adaptive model is constructed based on the selected features, preprocessed data, and expert labels.

[0017] Early warning for natural gas station equipment is achieved based on a constructed multi-level hierarchical adaptive model.

[0018] One preferred embodiment of the present invention, the early warning method for natural gas station equipment, further includes: optimizing the constructed multi-layer hierarchical adaptive model.

[0019] In one preferred embodiment of the present invention, the preprocessing includes removing outliers from the original data using a multivariate Gaussian distribution analysis algorithm and using a normalization algorithm to control the data of different dimensions to the same scale.

[0020] One preferred embodiment of the present invention involves extracting features from the preprocessed data and the original data, and selecting from the extracted features, including:

[0021] Extract the correlations between phenomena and data from the preprocessed data, and filter features;

[0022] The original data is parsed and split, amplified, and a feature set is provided.

[0023] Feature selection is performed based on the feature set.

[0024] In summary, the beneficial effects of the present invention are as follows:

[0025] 1. The natural gas station equipment early warning system of the present invention collects data through a data acquisition module and transmits the collected data to a data preprocessing module for preprocessing. The preprocessed data is then transmitted to a data mining module for data extraction. A feature engineering module parses and splits the original data to provide a feature set for a feature selection module. The feature selection module selects features and transmits the selected features to an algorithm model module. The algorithm model module learns from the processed data, the features transmitted by the feature selection module, and expert labels to construct a multi-level hierarchical adaptive model. The constructed multi-level hierarchical adaptive model then generates accurate and timely early warning signals to achieve early warning for natural gas station equipment.

[0026] 2. The natural gas station equipment early warning system and method of the present invention, by combining fuzzy optimization algorithm and multi-level hierarchical adaptive model, realizes accurate monitoring and early warning of equipment operating status, thereby identifying potential faults or abnormalities in advance, reducing production interruptions and economic losses caused by equipment failures, and ensuring the safe operation of natural gas stations.

[0027] 3. The natural gas station equipment early warning system of the present invention monitors the production and operation process of the equipment, analyzes and characterizes the types of abnormal operating parameters of key equipment, and promptly alerts on-site personnel to possible abnormal situations in the early stage of equipment abnormality.

[0028] 4. The natural gas station equipment early warning system of the present invention can reduce production stoppages caused by equipment failures, improve production efficiency, reduce operation and maintenance costs, and achieve the effect of cost reduction and efficiency improvement.

[0029] 5. Compared to traditional periodic maintenance, the natural gas station equipment early warning system of this invention enables predictive maintenance, reducing unnecessary maintenance operations and thus saving labor and material costs for maintenance and repair. Simultaneously, it extends equipment lifespan and reduces capital expenditure for equipment replacement.

[0030] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention will be apparent from the effects described in the description and the accompanying drawings. Attached Figure Description

[0031] Figure 1 This is a fuzzy distribution curve diagram in the early warning system for natural gas station equipment of the present invention;

[0032] Figure 2 This is a multivariate Gaussian distribution analysis diagram in the early warning system for natural gas station equipment of the present invention;

[0033] Figure 3 This is a flowchart illustrating the early warning method for natural gas station equipment according to the present invention;

[0034] Figure 4 This is a flowchart of a device anomaly early warning system based on multi-parameter joint analysis in an embodiment of the present invention;

[0035] Figure 5 This is a flowchart illustrating the establishment of the early warning model in an embodiment of the present invention;

[0036] Figure 6 This is a flowchart illustrating the process of establishing a fuzzy relation database in an embodiment of the present invention. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] This invention provides a natural gas station equipment early warning system, including a data acquisition module, which transmits the acquired data to a data preprocessing module, which transmits the preprocessed data to a data mining module, which is connected to a feature engineering module, which transmits the processed feature set to a feature selection module, which selects features and transmits the selected features to an algorithm model module, which is used to construct a multi-level hierarchical adaptive model, and is also connected to an early warning module for generating early warning signals.

[0039] The natural gas station equipment early warning system also includes an iterative optimization module for optimizing the multi-level hierarchical adaptive model. The iterative optimization module is connected to the algorithm model module and the early warning module, respectively.

[0040] Specifically:

[0041] The data acquisition module collects historical and real-time data from natural gas station equipment, providing a data foundation for subsequent modules and ensuring the comprehensiveness and accuracy of the data acquisition. The data acquisition module includes a historical data acquisition unit and a real-time data acquisition unit.

[0042] The historical data acquisition unit is used to collect historical data from natural gas station equipment. Specifically, historical data acquisition involves collecting second-level data over a period of time through multiple on-site surveys, and continuously collecting data on a monthly basis. Simultaneously, collaboration with on-site experts was used to confirm labels and interpret data dimensions. A total of 22 data dimensions were collected (flow rate of sulfur-containing natural gas to the absorber, product gas flow rate, online H2S analysis at the feed gas inlet, feed gas gravity separator level, differential pressure of feed gas filter separator A, differential pressure of feed gas filter separator B, level of feed gas filter separator A, level of feed gas filter separator B, differential pressure of the desulfurization absorber, level of the desulfurization absorber, flow rate of lean liquid into the absorber, flash vapor flow rate, regeneration tower level, desulfurization absorber bottom level regulating valve position, flash liquid regulating valve opening, flow rate (circulation rate) of lean liquid into the absorber, feed gas inlet pressure, flash vapor pressure, regeneration tower differential pressure, reboiler steam flow rate, regeneration tower top temperature, and acid gas flow rate). Seven instances of antifoaming agent addition were also collected.

[0043] The real-time data acquisition unit is used to collect real-time data from natural gas station equipment, collect real-time data from the database to determine whether there is abnormal foaming, and accumulate data by matching classified data with the actual situation for future model optimization.

[0044] The data acquisition module transmits the acquired data to the data preprocessing module. The data preprocessing module uses a multivariate Gaussian distribution analysis algorithm to remove outliers from the original data, reducing the impact of noise on the model. It also employs a normalization algorithm to control the data across different dimensions to maintain consistent dimensions, preventing large weight biases from affecting model performance and ensuring data quality. This guarantees that the data used by the subsequent data mining and feature engineering modules is high-quality and noise-free. The data preprocessing module includes a fuzzy computation unit and a multivariate Gaussian distribution analysis unit. The fuzzy computation unit performs knowledge quantification on fuzzy data, while the multivariate Gaussian distribution analysis unit removes outliers from the original data.

[0045] Fuzzy computing unit: The state parameters characterizing abnormalities in chemical equipment all have their own physical meaning and normal range. To enable comprehensive comparative analysis, normalization is required. Fuzzy distribution method is used to quantify the fuzzy and imprecise knowledge of oil and gas processing experts. Fuzzy distribution refers to the membership function of a fuzzy set A in the real number domain R. Commonly used distribution types include: ascending semi-rectangular distribution, descending semi-rectangular distribution, ascending semi-trapezoidal distribution, and descending semi-trapezoidal distribution, etc., and their membership function distribution curves are shown in the figure. Figure 1 As shown.

[0046] A semi-rectangular distribution:

[0047]

[0048] In the formula, a is the threshold parameter;

[0049] Decreasing semi-rectangular distribution:

[0050]

[0051] In the formula, a is the threshold parameter;

[0052] A semi-trapezoidal distribution:

[0053]

[0054] In the formula, a and b are both threshold parameters;

[0055] Decreasing semi-trapezoidal distribution:

[0056]

[0057] In the formula, a and b are both threshold parameters.

[0058] Multivariate Gaussian distribution analysis unit: The multivariate Gaussian distribution analysis algorithm removes outliers from the original data, reducing the impact of noise on the model. The normalization algorithm is used to control the data of different dimensions to the same scale, avoiding large weight deviations that may affect the model performance.

[0059] The multivariate normal distribution in n dimensions is also called the multivariate Gaussian distribution. It is parameterized by the mean vector μ∈Rn and the covariance matrix Σ∈Rn×n, where Σ≥0 is symmetric and positive semidefinite.

[0060] The covariance of a vector-valued random variable Z is defined as Cov(Z) = E[(ZE[Z])(Z-[Z])T]. This summarizes the concept of the difference of a real-valued random variable. Covariance can also be defined as Cov(Z) = E[ZZT] - (E[Z])(E[Z])T. If X ~ N(μ, Σ), then Cov(X) = Σ. The multivariate Gaussian distribution analysis diagram is shown below. Figure 2 As shown.

[0061] Figure 2 The image shows a Gaussian distribution with a mean of zero (i.e., a 2x1 zero vector) and a covariance matrix Σ = I (a 2x2 identity matrix). A Gaussian distribution with zero mean and identity covariance is also known as the standard normal distribution. The middle plot shows the Gaussian density with zero mean and Σ = 0.6I. By determining the region where the data points are distributed, we can identify whether a data point is an outlier. The definition of the outlier region needs to be determined based on the characteristics of the data itself and the constraints of business requirements.

[0062] The multivariate Gaussian distribution analysis unit includes a normalization unit, which is used to control data of different dimensions to the same scale.

[0063] Normalization unit: Normalization is a way to simplify calculations, that is, to transform a dimensional expression into a dimensionless expression, which becomes a scalar. It makes the absolute value of a physical system's numerical value a relative value relationship, and is an effective way to simplify calculations and reduce the value of quantities.

[0064] There are two main methods: one is Standardization, where the quantized features will follow a standard normal distribution, as shown in the formula:

[0065]

[0066] Where μ is the mean of the corresponding feature and δ is the standard deviation of the corresponding feature. The quantized features will be distributed in the interval [-1, 1].

[0067] The data preprocessing module transmits the preprocessed data to the data mining module. The data mining module acquires the preprocessed data and extracts useful correlations between phenomena and data points, providing valuable information for the subsequent feature engineering module and helping it build more targeted features. The data mining module includes a label expansion unit and a data mining unit for uncovering deep correlations between phenomena within the data and between data points.

[0068] Data mining unit: Uncovering deep correlations within data regarding phenomena and relationships between data points. A common method is correlation analysis, which analyzes two or more correlated variables to measure the degree of their relationship. For correlated elements to be valid, there needs to be a certain connection or probability between them. The most commonly used methods are the covariance method and the correlation coefficient method.

[0069] ① Covariance method

[0070] Covariance measures the population error between two variables. If the two variables have the same trend, the covariance is positive, indicating a positive correlation. If the two variables have opposite trends, the covariance is negative, indicating a negative correlation. If the two variables are independent, the covariance is 0, indicating no correlation. The following is the formula for calculating covariance.

[0071]

[0072] In the formula, X and Y are two random variables;

[0073] ② Correlation coefficient method

[0074] The correlation coefficient is a statistical indicator that reflects the strength of the relationship between variables. The correlation coefficient ranges from 1 to -1. A value of 1 indicates a perfect linear correlation between the two variables, -1 indicates a perfect negative correlation, and 0 indicates no correlation. The closer the value is to 0, the weaker the correlation. The following is the formula for calculating the correlation coefficient:

[0075]

[0076] Formula r xy S represents the sample correlation coefficient. xy S represents the sample covariance. x S represents the sample standard deviation of x. y This represents the sample standard deviation of y. Below are S... xy Covariance and S x and S y The formula for calculating standard deviation. Since it involves the sample covariance and sample standard deviation, the denominator is n-1.

[0077] S xy Formula for calculating sample covariance:

[0078]

[0079] S x Formula for calculating sample standard deviation:

[0080]

[0081] S y Formula for calculating sample standard deviation:

[0082]

[0083] Label expansion unit: Based on existing high-risk labels provided by experts, the label expansion unit extends the labels to the front of high-risk phenomena, allowing subsequent algorithms to learn medium- and low-risk features and achieve anomaly warning. A suitable clustering algorithm is selected to ensure that the expanded data phenomena closely resemble the expert labels; the most commonly used algorithm is k-means.

[0084] The k-means algorithm takes an input k and then divides n data objects into k clusters such that the resulting clusters satisfy the following conditions: objects within the same cluster have high similarity, while objects in different clusters have low similarity. Cluster similarity is calculated using a "centroid" (center of gravity) obtained from the mean of the objects in each cluster.

[0085] The working process of the k-means algorithm is explained below:

[0086] First, k objects are randomly selected from n data objects as initial cluster centers; then, for the remaining objects, they are assigned to the clusters most similar to them (represented by the cluster centers) based on their similarity (distance) to these cluster centers.

[0087] Then, the cluster center of each newly obtained cluster (the mean of all objects in that cluster) is calculated; this process is repeated until the standard measure function begins to converge. There are generally two stopping conditions, which differ depending on the standard measure function used. When using Euclidean distance, the objective function is generally to minimize the sum of squared distances from each object to its cluster centroid, as follows:

[0088]

[0089] When using cosine similarity, the objective function is generally to maximize the sum of cosine similarities from the object to its cluster centroid, as follows:

[0090]

[0091] The standard deviation is generally used as the standard measure function. k clusters have the following characteristics: each cluster is as compact as possible, while the clusters are as far apart as possible from each other.

[0092] The feature engineering module is used to parse and decompose the original data, amplifying information at each level to identify details that best match abnormal operating conditions, thus providing a high-quality, representative feature set for the feature selection module. Specifically, the main function of the feature engineering module is to create latent features based on the existing 22-dimensional original features, achieving data dimensionality enhancement. The enhanced data is the result of parsing and decomposing the original data, amplifying information at each level to identify details that best match abnormal operating conditions. The main method used in feature engineering is wavelet analysis feature extraction.

[0093] As the name suggests, a "wavelet" is a small waveform. The "small" aspect refers to its attenuation; the "wave" aspect refers to its oscillating nature, with its amplitude alternating between positive and negative. Compared to the Fourier transform, the wavelet transform is a localized analysis of time (space) and frequency. It refines the signal (function) step-by-step through scaling and translation operations, ultimately achieving time subdivision at high frequencies and frequency subdivision at low frequencies. It can automatically adapt to the requirements of time-frequency signal analysis, thus focusing on any detail of the signal.

[0094] A central idea of ​​wavelet analysis is multi-resolution, meaning that the decomposition of a signal is performed layer by layer according to different resolutions of detail. Its signal space L2(R) is a square-integrable space; if a function g(t) is an element of this space, then g(t) ∈ L2. Its scaling function, for a family of two-dimensional functions (forming the basis of the space), is expressed as:

[0095]

[0096] For all k∈Z, a space can be spanned:

[0097]

[0098] If f(t)∈Vj, then f(t) can be expressed as:

[0099]

[0100] In other words, f(t) can be represented by a set of basis functions in the Vj space, and this basis function can be set; the larger j is, the higher the resolution. This means that data signals can be analyzed using wavelet analysis for multi-resolution analysis. Signals at low resolution can be combined not only by combining signals at low resolution but also by combining signals at high resolution. The scaling function φj,k(t) spans the V space, and the difference space W between different V spaces is spanned by the wavelet function ψj,k(t), expressed as:

[0101]

[0102] In this way, we construct a set of functions φk(t) and ψj,k(t) that spans the entire L2(R). For any function g(t) ∈ L2(R), it can be written as a series expansion of the scaling function and the wavelet function, that is:

[0103]

[0104] In this expansion, the first summation gives a low-resolution or coarse approximation of g(t);

[0105] In the second summation, as the index j increases, higher or finer resolution functions are continuously added, thus incorporating more detailed information and completing the feature extraction of the data.

[0106] The feature engineering module transmits the processed feature set to the feature selection module, which selects features and then transmits the selected features to the algorithm model module. Specifically, the feature selection module selects from all features obtained by the feature engineering module, identifying features derived from the original features that are more relevant to the anomaly, while excluding features with weak or no correlation. This provides the algorithm model module with an effective feature set, reducing model complexity and improving computational efficiency and prediction accuracy. A commonly used feature selection method is the chi-square test.

[0107] Chi-square test (χ²) 2 test), is a commonly used feature selection method, χ² 2 Used to describe the independence of two events, or the degree of deviation between the actual observed value and the expected value. χ² 2 The larger the value, the greater the deviation between the actual observed value and the expected value, and the weaker the independence between the two events.

[0108] The formula for calculating the chi-square is:

[0109]

[0110] In the formula, t represents the presence or absence of a feature, c represents the class (1 or 0, only binary classification is supported here), N is the observed value, and E is the expected value. Therefore, E11 indicates that feature t is present and class c = 1. After obtaining χ... 2 After setting the value, we also need to set χ. 2 The p-value is converted to a p-value. A p-value is the probability of a sample outcome given that the null hypothesis is true; we can convert it using a simple table lookup. When the p-value and χ²... 2 If all values ​​meet the standard, then the feature can be selected.

[0111] The algorithm model module is used to construct a multi-layered hierarchical adaptive model, and it is also connected to an early warning module for generating early warning signals. Specifically, the algorithm model module learns from the data processed by the aforementioned modules, the features obtained from the feature selection module, and expert labels to construct a multi-layered hierarchical adaptive model, predict equipment failure risks, and make decisions.

[0112] The main function of the algorithm model module is to use big data algorithms to learn from the data processed by the aforementioned modules and expert labels, forming a model that can automatically predict and issue alerts in the future. Different algorithm models have varying adaptability to data, and their predictive performance also differs. Extensive data experience and practical testing are needed to determine the appropriate algorithm selection and integration scheme. Based on the correlation analysis results, a premise confidence rule table is established. The premise confidence and correlation rules table form the foundation and main basis for fuzzy inference calculations, containing all the rules for storage device anomaly warning inference. The algorithm used in this invention is the SVM algorithm model.

[0113] SVM (Support Vector Machine) is a common discriminant method. In the field of machine learning, it is a supervised learning model, typically used for pattern recognition, classification, and regression analysis. Its main ideas can be divided into two points:

[0114] (1) It is designed for the linearly separable case. For the linearly inseparable case, the samples that are linearly inseparable in the low-dimensional input space are transformed into the high-dimensional feature space by using a nonlinear mapping algorithm, making them linearly separable. This makes it possible to use a linear algorithm to perform linear analysis on the nonlinear features of the samples in the high-dimensional feature space.

[0115] (2) It constructs an optimal hyperplane in the feature space based on the structural risk minimization theory, so that the learner achieves global optimization and the expectation in the entire sample space satisfies a certain upper bound with a certain probability.

[0116] In general, the SVM method maps the sample space to a high-dimensional or even infinite-dimensional feature space (Hilbert space) through a nonlinear mapping p, transforming a nonlinearly separable problem in the original sample space into a linearly separable problem in the feature space. Simply put, it's about dimensionality increase and linearization. Dimensionality increase involves mapping samples to a higher-dimensional space, which generally increases computational complexity and can even lead to the "curse of dimensionality," thus it's rarely explored. However, for problems like classification and regression, sample sets that cannot be linearly processed in a low-dimensional sample space can be linearly partitioned (or regressed) in a high-dimensional feature space using a linear hyperplane. While general dimensionality increase leads to computational complexity, the SVM method cleverly solves this problem: by applying the kernel function expansion theorem, the explicit expression of the nonlinear mapping is not needed; because a linear learning machine is built in a high-dimensional feature space, compared to linear models, it not only increases computational complexity by almost nothing but also avoids the "curse of dimensionality" to some extent. All of this is thanks to the theory of kernel function expansion and computation.

[0117] Different kernel functions can generate different SVMs. The four most commonly used kernel functions are as follows:

[0118] (1) Linear kernel function: K(x, y) = x·y;

[0119] In the formula, x and y are input vectors;

[0120] (2) Polynomial kernel function: K(x, y) = [(x·y) + 1] d ;

[0121] In the formula, x and y are input vectors, and d is the degree of the polynomial;

[0122] (3) Radial basis functions:

[0123] In the formula, x and y are input vectors, and d is a parameter;

[0124] (4) Kernel function of two-layer neural network: K(x,y)=tanh(a(x·y)+b);

[0125] In the formula, x and y are input vectors, and a and b are adjustable parameters;

[0126] By using different kernel functions, nonlinear problems can be predicted well. Based on the description of the foaming working condition by business experts, it can be analyzed that foaming anomaly is a typical nonlinear prediction problem. Therefore, the SVM method is proposed to be used for modeling.

[0127] The natural gas station equipment early warning system also includes an iterative optimization module for optimizing a multi-level hierarchical adaptive model. This module is connected to both the algorithm model module and the early warning module. The iterative optimization module continuously adjusts the algorithm model to improve prediction accuracy and model adaptability. During model training and deployment, it repeatedly optimizes model parameters to arrive at the optimal model that adapts to both new and old data. The early warning module uses the predictions from the algorithm model module and the optimized model from the iterative optimization module to generate accurate and timely early warning signals, helping decision-makers take necessary measures to prevent equipment failures.

[0128] The iterative optimization module includes online model units and offline model units. After the online model unit has been online for a certain period of time, it accumulates data and new operating conditions during this period, puts new data into the model to retrain the model, and performs a global search to obtain the optimal model that adapts to both the old and new data. The offline model unit uses a global search method to find the optimal values ​​of all parameters in the algorithm, forming the best model that has learned from historical data.

[0129] The early warning module is used to predict abnormal equipment conditions. Staff at natural gas stations can take timely risk avoidance measures based on the early warning information to eliminate risks in their infancy. In the actual application of the early warning module in natural gas stations, on-site process technicians and managers have the authority to modify the settings of this module. The parameters of the early warning model need to be continuously improved. In the early stage of module operation, experts and on-site technicians should be actively organized to discuss, modify and improve the contents of the fault early warning knowledge base to improve its early warning accuracy.

[0130] Preferably, for abnormal operating conditions based on multi-parameter fusion analysis, the early warning module includes a model parameter configuration unit, a dynamic monitoring and early warning unit, an early warning information tracking and processing unit, a data trend analysis unit, and an early warning information statistical analysis unit.

[0131] The working process of a natural gas station equipment early warning system is as follows: A data acquisition module collects historical and real-time data from the natural gas station equipment and transmits the collected data to a data preprocessing module. The data preprocessing module receives the data transmitted by the data acquisition module and preprocesses the data. The preprocessed data is then transmitted to a data mining module for useful phenomenon correlation and data correlation extraction, providing valuable information for a feature engineering module. Simultaneously, the feature engineering module parses and splits the raw data to provide a high-quality and representative feature set for a feature selection module. The feature selection module receives the feature set provided by the feature engineering module and selects from all features. The algorithm model module receives the features selected by the feature selection module and, based on the data processed by the aforementioned modules, the features obtained from the feature selection module, and expert labels, learns to construct a multi-level hierarchical adaptive model. The iterative optimization module continuously adjusts the algorithm model to improve the prediction accuracy and adaptability of the multi-level hierarchical adaptive model and transmits the results to an early warning module. The early warning module generates accurate and timely early warning signals based on the prediction results of the algorithm model module and the model optimized by the iterative optimization module, helping decision-makers take necessary measures to prevent equipment failures.

[0132] This invention also discloses a method for early warning of natural gas station equipment, implemented based on the aforementioned early warning system for natural gas station equipment, such as... Figure 3 As shown, it includes the following steps:

[0133] Step (1): Collect data from natural gas station equipment and preprocess the collected data; specifically, the data acquisition module collects historical and real-time data from natural gas station equipment and transmits the collected data to the data preprocessing module for preprocessing. The preprocessing uses a multivariate Gaussian distribution analysis algorithm to remove abnormal factors from the original data and a normalization algorithm to control data of different dimensions to the same scale.

[0134] Step (2): Extract features from the preprocessed data and the original data, and select the extracted features; specifically, this includes the following steps:

[0135] Step (201): Extract phenomenon correlations and data correlations from the preprocessed data and filter features; specifically, the data mining module extracts useful phenomenon correlations and data correlations from the preprocessed data to provide valuable information for the feature engineering module.

[0136] Step (202) parses and splits the original data, amplifies the original data, and provides a feature set; specifically, the feature engineering module parses and splits the original data to provide a high-quality and representative feature set for the feature selection module.

[0137] Step (203): Feature selection is performed based on the feature set; specifically, the feature selection module obtains the feature set provided by the feature engineering module and selects all features;

[0138] Step (3): Construct a multi-level hierarchical adaptive model based on the selected features, preprocessed data, and expert labels; specifically, the feature selection module obtains the feature set provided by the feature engineering module and selects all features, the algorithm model module obtains the features selected by the feature selection module, and learns based on the data processed by the aforementioned module, the features obtained from the feature selection module, and the expert labels to construct a multi-level hierarchical adaptive model;

[0139] Step (4): Realize early warning of natural gas station equipment based on the constructed multi-level hierarchical adaptive model; Specifically, the early warning module generates accurate and timely early warning signals based on the constructed multi-level hierarchical adaptive model to help decision-makers take necessary measures to prevent equipment failure.

[0140] Before implementing early warning for natural gas station equipment based on the constructed multi-level hierarchical adaptive model, the constructed multi-level hierarchical adaptive model needs to be optimized. Specifically, the iterative optimization module continuously adjusts the algorithm model to improve prediction accuracy and model adaptability. During the model training and deployment process, the model parameters are repeatedly optimized to obtain the optimal multi-level hierarchical adaptive model that adapts to both new and old data.

[0141] Example

[0142] Taking the "clogging" of a Claus condenser as an example, the main steps are as follows: Figure 4 As shown.

[0143] (1) Feature parameter analysis

[0144] Theoretical analysis shows that the pressure difference between the inlet and outlet of the Claus condenser changes most significantly. However, currently, the basic process control system of natural gas processing plants only monitors two variables for the Claus condenser: the process gas inlet temperature and the process gas outlet temperature, with no pressure monitoring data. Combining theoretical analysis, anomaly records, and historical curves, it can be seen that when the Claus condenser is "blocked," its process gas outlet temperature decreases while the pressure at the air distribution port of its upstream Claus reactor increases. Therefore, the characteristic parameters of a "blocked" Claus condenser are two: the process gas outlet temperature and the pressure at the air distribution port of its upstream Claus combustion furnace.

[0145] (2) Weighting analysis

[0146] Process systems are complex, with numerous parameters affecting their operation, and the exact degree of influence of each parameter on the process unit cannot be known. When an anomaly occurs in a process unit, many variables may change, and these variables play different roles in indicating the anomaly. That is, the correlation between the characteristic parameter and the process unit varies. The greater the correlation, the more obvious the variable's role in indicating the process unit's anomaly, and vice versa.

[0147] Anomalies in the process unit are specific, and there is no fixed model for correlation analysis. They need to be determined based on the actual operating curves of the process system during production. In addition, the experience of production staff is crucial.

[0148] (3) Establishment of rule tables

[0149] Taking the Claus condenser as an example, in the specific implementation, a data table RuleLi001 (001 represents the Claus condenser number) is designed to store the rules. The definition of the table is shown in Table 1 below;

[0150] Table 1: Definition of RuleLi001

[0151] field name Data types Length (bytes) Remark R_Condition varchar 50 The premise of the rules R_Result varchar 50 Conclusion of the rule R_CF float 8 Rule confidence

[0152] Assuming the Claus condenser is “blocked,” the confidence levels for the increase in pressure at the upstream Claus combustion furnace blast inlet and the increase in process gas outlet temperature are 0.7 and 0.3, respectively.

[0153] For example, the storage rule is: IF the pressure at the combustion furnace air outlet rises, THEN Claus condenser “blocked” CF = 0.7; IF the process gas outlet temperature rises, THEN Claus condenser “blocked” CF = 0.3. The storage example table of RuleLi0012 is shown in Table 2.

[0154] Table 2: Example of storage for RuleLi0012

[0155]

[0156] (4) Dynamic monitoring data fuzzification processing

[0157] To overcome the limitations of precise methods, mathematical methods such as membership degree and membership function are used to classify the fuzziness of dynamic monitoring data, and a fact table representing the properties of dynamic monitoring data is established based on this.

[0158] ① Construction of membership functions

[0159] To better quantify the fuzzy and imprecise knowledge of oil and gas processing experts, the fuzzy distribution method is adopted. Fuzzy distribution refers to the membership function of a fuzzy set A in the real number domain R. Commonly used distribution types include: ascending semi-rectangular distribution, descending semi-rectangular distribution, ascending semi-trapezoidal distribution, and descending semi-trapezoidal distribution.

[0160] When the Claus condenser is abnormally "blocked", the characteristic parameters of the combustion furnace air outlet pressure and the process gas outlet temperature both show an upward trend. Assuming that a is the normal operating parameter value and b is the alarm limit, in order to identify the abnormal "blockage" symptoms of the equipment in the early stage of equipment abnormality, a semi-trapezoidal distribution is selected.

[0161] ② Solving for membership degree

[0162] The state parameters characterizing abnormalities in chemical equipment all have their own physical meaning and normal range. In order to enable comprehensive comparative analysis, normalization is required. These values ​​are converted into specific values ​​between [0, 1], where 0 represents the best and 1 represents the worst.

[0163] (4) Early warning model establishment

[0164] Equipment anomalies typically manifest as changes in multiple parameters. However, the values ​​and correlations of these parameters are uncertain and fuzzy. Therefore, an equipment anomaly early warning model is established based on fuzzy mathematics theory. A fuzzy relational database is built, and the confidence level of an equipment anomaly is inferred by matching real-time monitoring data with the database. The confidence level is defined in the range [0, 1], where 1 indicates an anomaly has occurred and 0 indicates no anomaly has occurred. Higher confidence levels require greater attention from the monitor, while lower confidence levels can be disregarded. The flowchart for establishing the early warning model is as follows: Figure 5 As shown.

[0165] The key to early warning based on multi-parameter joint analysis is the construction of a fuzzy relation database. The construction process of a fuzzy relation database is as follows: Figure 6 As shown, the construction of a fuzzy relational database needs to be based on the understanding of the abnormal patterns of equipment. It needs to analyze the types of parameters associated with the equipment from the perspective of the entire process flow. Moreover, these parameters have different fluid dynamics and thermodynamics characteristics, and can characterize the abnormal situation of the equipment with their respective connotations.

[0166] In summary, the natural gas station equipment early warning system and method of the present invention achieves accurate monitoring and early warning of equipment operating status by combining fuzzy optimization algorithm with multi-level hierarchical adaptive model.

[0167] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A natural gas station equipment early warning system, characterized in that: The system includes a data acquisition module that transmits the acquired data to a data preprocessing module, which then transmits the preprocessed data to a data mining module. The data mining module is connected to a feature engineering module, which transmits the processed feature set to a feature selection module. The feature selection module selects features and transmits the selected features to an algorithm model module. The algorithm model module is used to construct a multi-level hierarchical adaptive model and is also connected to an early warning module for generating early warning signals.

2. The early warning system for natural gas station equipment as described in claim 1, characterized in that: The natural gas station equipment early warning system also includes an iterative optimization module for optimizing a multi-level hierarchical adaptive model, which is connected to both the algorithm model module and the early warning module.

3. The early warning system for natural gas station equipment as described in claim 1, characterized in that: The data acquisition module includes a historical data acquisition unit and a real-time data acquisition unit.

4. The early warning system for natural gas station equipment as described in claim 1, characterized in that: The data preprocessing module includes a fuzzy computing unit and a multivariate Gaussian distribution analysis unit. The fuzzy computing unit is used to quantify knowledge in fuzzy data, and the multivariate Gaussian distribution analysis unit is used to remove abnormal factors from the original data.

5. The early warning system for natural gas station equipment as described in claim 4, characterized in that: The multivariate Gaussian distribution analysis unit includes a normalization unit, which is used to control data of different dimensions to the same scale.

6. The early warning system for natural gas station equipment as described in claim 1, characterized in that: The data mining module includes a label expansion unit and a data mining unit for mining deep correlations about phenomena within the data and correlations between data.

7. The early warning system for natural gas station equipment as described in claim 1, characterized in that: The feature engineering module uses wavelet analysis to extract features and parse and split the original data, providing a high-quality and representative feature set for the feature selection module.

8. The early warning system for natural gas station equipment as described in claim 2, characterized in that: The iterative optimization module includes an online model unit and an offline model unit. The online model unit optimizes the multi-level hierarchical adaptive model through data accumulation and the accumulation of working conditions. The offline model unit optimizes the multi-level hierarchical adaptive model by finding the optimal values ​​of all parameters in the algorithm through a global search method.

9. A natural gas station equipment early warning system as described in claim 1, characterized in that: The early warning module includes a model parameter configuration unit, a dynamic monitoring and early warning unit, an early warning information tracking and processing unit, a data trend analysis unit, and an early warning information statistical analysis unit.

10. A method for early warning of natural gas station equipment, implemented based on the early warning system for natural gas station equipment according to any one of claims 1-9, characterized in that, Includes the following steps: Collect data from natural gas station equipment and preprocess the collected data; Feature extraction is performed on the preprocessed data and the original data, and the extracted features are selected. A multi-level hierarchical adaptive model is constructed based on the selected features, preprocessed data, and expert labels. Early warning for natural gas station equipment is achieved based on a constructed multi-level hierarchical adaptive model.

11. A method for early warning of natural gas station equipment as described in claim 10, characterized in that, The early warning method for natural gas station equipment also includes: optimizing the constructed multi-level hierarchical adaptive model.

12. The early warning method for natural gas station equipment as described in claim 10, characterized in that, The preprocessing includes removing outliers from the original data using a multivariate Gaussian distribution analysis algorithm and using a normalization algorithm to control the data of different dimensions to the same scale.

13. A method for early warning of natural gas station equipment as described in claim 10, characterized in that, The step of extracting features from the preprocessed data and the original data, and selecting the extracted features, includes: Extract the correlations between phenomena and data from the preprocessed data, and filter features; The original data is parsed and split, amplified, and a feature set is provided. Feature selection is performed based on the feature set.

Images