A method and device for detecting faults in a natural gas station control system

Abstract

The present invention relates to the field of natural gas station control technology, and specifically provides a method and device for detecting faults in a natural gas station control system. The method comprises: real-time collection of multidimensional operating parameters of natural gas in a natural gas station control system; preprocessing the multidimensional operating parameters to generate feature vectors, inputting the feature vectors into a reversible neural network model, and outputting a reconstruction error, wherein the reversible neural network comprises a network architecture composed of reversible transformation modules, the reversible transformation modules are used to learn the multidimensional operating parameters of the natural gas station control system under normal operating conditions, and the reconstruction error is used to characterize the degree of multidimensional deviation between the multidimensional operating parameters and the operating parameters under normal operating conditions learned by the reversible neural network; when the reconstruction error is greater than a reconstruction error threshold, calculating a joint confidence based on a confidence rule base and the reconstruction error, and generating an interactive report containing the fault type when the joint confidence exceeds the confidence threshold, thereby achieving the purpose of fault diagnosis based on multidimensional real-time operating data.

Description

Technical Field

[0001] The present application relates to the field of natural gas station control technology, and in particular to a method and device for detecting faults in a natural gas station control system. Background Art

[0002] As a key node in the urban gas transmission and distribution system, the control system of the natural gas station monitors multi-dimensional parameters such as pressure, temperature, and flow in real time to deal with faults such as stuck pressure regulating valves, scaling of heat exchangers, and low-temperature blockage of pipelines.

[0003] Existing fault detection technologies primarily rely on threshold-based alarm systems. However, since faults such as a stuck pressure regulating valve and heat exchanger fouling are often caused by the synergistic effects of abnormalities in multiple operating parameters, relying solely on threshold detection can result in numerous false positives and false negatives. For example, when a complex fault actually occurs, it may be missed because multiple operating parameters fail to reach their thresholds simultaneously. Furthermore, when a parameter occasionally exceeds its limit during non-faulty conditions, it can trigger a false positive, reducing the accuracy and reliability of fault detection. Summary of the Invention

[0004] In response to the above problems, the present application provides a natural gas station control system fault detection method and device, which achieves the purpose of fault diagnosis based on multi-dimensional real-time operation data through the dual judgment mechanism of reversible neural network and confidence rule base.

[0005] To achieve the purpose of the present invention, this application provides the following technical solutions:

[0006] In a first aspect, the present application provides a method for detecting faults in a natural gas station control system, the method comprising the following steps:

[0007] Real-time collection of multi-dimensional operating parameters of natural gas in the natural gas station control system, including pressure, temperature and flow;

[0008] Preprocessing the multidimensional operating parameters to generate feature vectors, wherein the preprocessing sequentially includes outlier removal, sliding window mean filtering, and normalization;

[0009] Inputting the feature vector into a reversible neural network model and outputting a reconstruction error, wherein the reversible neural network includes a network architecture composed of a reversible transformation module, the affine coupling layer and the convolution layer of the reversible transformation module are used to learn multidimensional operating parameters of the natural gas station control system under normal operating conditions, and the reconstruction error is used to represent the multidimensional deviation degree between the multidimensional operating parameters and the operating parameters under normal operating conditions learned by the reversible neural network;

[0010] When the reconstruction error is greater than the reconstruction error threshold, a joint confidence is calculated based on the confidence rule base and the reconstruction error. When the joint confidence exceeds the confidence threshold, an interactive report containing the fault type is generated, wherein the confidence rule base is constructed based on the natural gas station equipment mechanism and the fault event case library. The rule antecedents of the confidence rule base include pressure fluctuation rate greater than a first threshold, temperature fluctuation rate greater than a second threshold, and flow fluctuation rate greater than a third threshold. The rule consequents are associated with three types of faults: pressure regulating valve sticking, heat exchanger scaling, and pipeline low-temperature blockage. The correspondence between the rule antecedents and the faults is: the combination of pressure fluctuation anomaly and temperature fluctuation anomaly corresponds to pressure regulating valve sticking, the combination of pressure fluctuation anomaly and flow fluctuation anomaly corresponds to pipeline low-temperature blockage, and the combination of temperature fluctuation anomaly and flow fluctuation anomaly corresponds to heat exchanger scaling. The confidence rule base also includes initial rule confidences corresponding to the three types of faults, and the confidence threshold is the initial rule confidence of the corresponding fault.

[0011] In a possible implementation, the interactive report further includes a warning level, countermeasures, and a timing for implementing the countermeasures. After the joint confidence exceeds a confidence threshold to determine the fault type, the method further includes:

[0012] Match response measures based on the fault type. For example, for a stuck pressure regulating valve, prioritize checking valve core wear and implementing lubrication measures. For a scaling heat exchanger, initiate a chemical cleaning procedure and check the heat exchanger fin clearances. For a low-temperature pipe blockage, increase the power of the heating system.

[0013] When the combined confidence is greater than the confidence threshold and less than a first confidence threshold, increasing the frequency of collecting corresponding operating parameters;

[0014] When the combined confidence is greater than or equal to the first confidence threshold and less than the second confidence threshold, a yellow warning is triggered and the corresponding countermeasure is executed during the next load valley period;

[0015] When the combined confidence is greater than or equal to the second confidence threshold, a red alert is triggered and the corresponding countermeasure is executed.

[0016] In a possible implementation, the confidence rule base is constructed based on the natural gas station equipment mechanism and the fault event case library, and further includes:

[0017] The operating parameters whose pressure fluctuation rate is greater than a first threshold, whose temperature fluctuation rate is greater than a second threshold, and whose flow rate fluctuation rate is greater than a third threshold in the fault event operating data are respectively marked as abnormal pressure fluctuation, abnormal temperature fluctuation, and abnormal flow rate fluctuation, wherein the fault event operating data is extracted from the fault event case library of the natural gas station control system;

[0018] Experts evaluate abnormal pressure, temperature, and flow fluctuations based on the mechanism of natural gas station equipment and the proportion of abnormal fluctuations in each operating parameter, and determine the weight coefficient corresponding to each operating parameter;

[0019] Based on the weight coefficients, an evidential reasoning algorithm is used to determine the initial rule confidence of each fault.

[0020] In a possible implementation, the weight coefficient uses an evidential reasoning algorithm to determine the initial rule confidence of each fault, further comprising:

[0021] Determine the confidence level of pressure fluctuation anomaly, temperature fluctuation anomaly, and flow fluctuation anomaly corresponding to each fault, and establish a fault evidence matrix;

[0022] Calculating the confidence of the original rule of each fault using a weighted summation method according to the fault evidence matrix and the weight coefficient;

[0023] The confidence level of the original rule is adjusted based on the equipment operating status and expert experience to obtain a value range of the confidence level of the initial rule for each fault.

[0024] In a possible implementation, the calculating a joint confidence based on a confidence rule base and the reconstruction error further includes:

[0025] When the fluctuation rate matches the rule antecedent, a feature matching degree is calculated based on the initial rule confidence, the fluctuation rate, and a threshold value matching the fluctuation rate, wherein the fluctuation rate includes a pressure fluctuation rate, a temperature fluctuation rate, or a flow fluctuation rate, and the threshold value corresponding to the fluctuation rate includes the first threshold value, the second threshold value, or the third threshold value, respectively. The feature matching degree μ i = initial rule confidence × min (volatility / corresponding threshold, 1);

[0026] The reconstruction error is quantified as a basic probability distribution function m(A)=1-exp(-λE), where E represents the reconstruction error and λ represents a preset fault sensitivity coefficient, which is set to different values according to different faults in the natural gas station control system;

[0027] The basic probability distribution function m(A) is matched with the feature matching degree μ i Perform weighted fusion to obtain the modified probability distribution value m'(A)=α·m(A)+(1-α)·μ i , and calculate the joint confidence according to Dempster's combination rule:

[0028]

[0029] Among them, α represents the empirical coefficient, A and B represent any two of the three rule antecedents, C represents any one of the three types of faults, and K is the conflict factor. .

[0030] In a possible implementation, A and B represent any two of the three rule antecedents, and C represents any one of the three types of faults, further including:

[0031] When A is abnormal pressure fluctuation and B is abnormal temperature fluctuation, C is stuck pressure regulating valve; when A is abnormal pressure fluctuation and B is abnormal flow fluctuation, C is low temperature blockage of pipeline; when A is abnormal temperature fluctuation and B is abnormal flow fluctuation, C is scaling of heat exchanger, and other combinations of A and B are classified into the conflict factor K.

[0032] In one possible implementation, the method further includes:

[0033] If the joint confidence level is in the range of [0.7β, β], the feature vector and an equally spaced sampling sequence of the multidimensional operating parameters within at least the last 12 hours are input into the LSTM prediction model, where β represents the confidence threshold.

[0034] Perform multi-step predictions using the trained LSTM prediction model to output predicted pressure, temperature, and flow rates every 15 minutes for at least the next three hours.

[0035] If the predicted pressure value exceeds the pressure safety threshold, the pressure regulating valve pre-maintenance alarm is triggered; if the predicted temperature value exceeds the temperature safety threshold, the heat exchanger inspection alarm is triggered; if the predicted flow value exceeds the flow safety threshold, the pipeline blockage inspection alarm is triggered. The safety thresholds are the pressure, temperature and flow when there is no fault event at the natural gas station.

[0036] In one possible implementation, the training process of the LSTM prediction model includes:

[0037] Multidimensional operating parameters of a 30-day continuous fault-free event case library extracted from a natural gas station control system are sampled at 15-minute intervals to construct a training sample set; the training sample set is preprocessed, including removing outliers, performing mean filtering through a sliding window, and normalization;

[0038] A network structure was configured, and the Huber loss function was used for model training. The network structure included an input gate layer, a forget gate layer, and an output gate layer. The input gate layer was used to accept 12-hour historical window data for a total of 48 time steps, and each time step included three feature dimensions: pressure, temperature, and flow. The first forget gate layer contained 64 neurons, and the second forget gate layer contained 64 neurons. The output gate layer contained three dense units corresponding to the prediction of pressure, temperature, and flow, respectively. The Huber loss function was used for model training.

[0039] In a second aspect, the present application provides a natural gas station control system fault detection device, the device comprising:

[0040] A data acquisition module is used to collect multi-dimensional operating parameters of natural gas in the natural gas station control system in real time, wherein the multi-dimensional operating parameters include pressure, temperature and flow;

[0041] A preprocessing module, configured to preprocess the multidimensional operating parameters to generate feature vectors, wherein the preprocessing sequentially includes outlier removal, sliding window mean filtering, and normalization;

[0042] an analysis and processing module, configured to input the feature vector into a reversible neural network model and output a reconstruction error, wherein the reversible neural network includes a network architecture composed of a reversible transformation module, the affine coupling layer and the convolution layer of the reversible transformation module are used to learn multidimensional operating parameters of the natural gas station control system under normal operating conditions, and the reconstruction error is used to characterize the multidimensional deviation between the multidimensional operating parameters and the operating parameters under normal operating conditions learned by the reversible neural network;

[0043] A report generation module is used to calculate a joint confidence based on a confidence rule base and the reconstruction error when the reconstruction error is greater than a reconstruction error threshold, and to generate an interactive report containing the fault type when the joint confidence exceeds the confidence threshold, wherein the confidence rule base is constructed based on the natural gas station equipment mechanism and the fault event case library, the rule antecedents of the confidence rule base include a pressure fluctuation rate greater than a first threshold, a temperature fluctuation rate greater than a second threshold, and a flow fluctuation rate greater than a third threshold, and the rule consequents are associated with three types of faults: pressure regulating valve sticking, heat exchanger scaling, and pipeline low-temperature blockage, the correspondence between the rule antecedents and the faults is: the combination of abnormal pressure fluctuation and abnormal temperature fluctuation corresponds to pressure regulating valve sticking, the combination of abnormal pressure fluctuation and abnormal flow fluctuation corresponds to pipeline low-temperature blockage, and the combination of abnormal temperature fluctuation and abnormal flow fluctuation corresponds to heat exchanger scaling, the confidence rule base also includes initial rule confidences corresponding to the three types of faults, and the confidence threshold is the initial rule confidence of the corresponding fault.

[0044] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which executes the steps of the above-mentioned natural gas station control system fault detection method when executed by a processor.

[0045] An embodiment of the present application provides a method for fault detection in a natural gas station control system. The method collects multidimensional operating parameters such as pressure, temperature, flow, and valve opening in real time, pre-processes them, and inputs them into a reversible neural network model to output a reconstruction error that can characterize the degree of multidimensional deviation between the current operating state and the normal operating condition. When the reconstruction error exceeds the reconstruction error threshold, it is determined that the current operating state may be abnormal, and the confidence rule base is activated for in-depth analysis. Furthermore, the confidence rule base adopts a parameter combination analysis strategy to perform a joint analysis of multiple parameters such as pressure-temperature, pressure-flow, and temperature-flow to obtain a joint confidence. When the joint confidence exceeds the confidence threshold, an interactive report containing the fault type is generated. The present application achieves accurate fault diagnosis of the real-time operating state through the dual judgment mechanism of the reversible neural network and the confidence rule base. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings are used to provide a further understanding of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application.

[0047] Figure 1 This is a flow chart of a method structure for fault detection of a natural gas station control system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0049] The present invention proposes the following technical solutions and corresponding embodiments.

[0050] Example 1

[0051] like Figure 1 The flowchart of the structure of a natural gas station control system fault detection method provided by an embodiment of the present application is shown. The embodiment of the present application provides a natural gas station control system fault detection method, including the following steps:

[0052] The S100 collects multi-dimensional operating parameters of natural gas in the natural gas station control system in real time. These parameters include pressure, temperature, and natural gas flow. Pressure transmitters collect pressure, temperature transmitters collect temperature, and flow meters collect natural gas flow.

[0053] S200 performs preprocessing on the multi-dimensional operating parameters to generate feature vectors, wherein the preprocessing includes outlier removal, sliding window mean filtering and normalization.

[0054] Specifically, the 3σ criterion is first used to eliminate outliers in the multi-dimensional operating parameters, and then a sliding window is used for mean filtering. Finally, the pressure parameter is normalized using the station design pressure value, and the temperature and flow parameters are standardized based on the mean and standard deviation of the operating data of the natural gas station in the recent period (for example, the last 30 days), and each operating parameter is converted into a dimensionless feature vector. The standardization specifically includes the pressure parameter using the station design rated working pressure P design As a normalized basis, calculate the normalized pressure P 1,norm =P1 / P design and normalized pressure difference ΔP norm =ΔP / P design , where the pressure difference between the front and rear ends of the pressure regulator is ΔP = P1- P2, P1 is the main pipeline pressure value measured in real time by the pressure transmitter, and P2 is the pressure after pressure regulation. This processing not only retains the engineering physical meaning of the pressure parameter, but also makes the data of stations with different pressure levels comparable. The temperature parameter is calculated based on the historical data of the last 30 days, with the rolling mean μ and standard deviation σ, and the Z-score standardization is implemented to calculate the standardized temperature t 1,std =(t1-μ) / σ, where t1 is the natural gas fluid temperature measured in real time by the temperature transmitter. The rolling mean and standard deviation are updated on time to track changes in ambient temperature. Flow normalization involves first calculating the normalized flow rate L based on the current flow rate and the maximum flow rate. j , norm =L j / L max Perform linear normalization and combine it with the moving standard deviation σ of the flow parameter L Calculate the normalized flow difference ΔL j,std =ΔL j / σ L , where ΔL j =L j -L max , L j is the instantaneous volume flow rate measured by the flowmeter, L max is the maximum designed flow rate of the pipeline, and the final generated eigenvector is as follows [P 1,norm , ΔP norm , t 1,std , Lj , norm ,ΔL j , std ,…], whose dimension is consistent with the number of parameters, which not only meets the neural network input requirements but also embeds the state characteristics of the equipment operation through the parameter coupling calculation in the preprocessing process (such as the retention of the pressure-flow ratio).

[0055] Therefore, the feature vector generated after the above three-level preprocessing not only retains the dynamic correlation characteristics of the original operating parameters, but also eliminates the influence of measurement noise.

[0056] S300 inputs the feature vector into a reversible neural network model and outputs a reconstruction error. The reversible neural network includes a network architecture composed of reversible transformation modules. The affine coupling layer and convolution layer of the reversible transformation module are used to learn the multidimensional operating parameters of the natural gas station control system under normal operating conditions. The reconstruction error is used to characterize the multidimensional deviation degree between the multidimensional operating parameters and the operating parameters under normal operating conditions learned by the reversible neural network.

[0057] The reversible neural network model is a deep learning architecture based on generative models, commonly used for image synthesis, image reconstruction, and data distribution learning. It uses reversible neural network layers and normalized flow technology to transform complex input data into simple probability distributions, thereby achieving efficient generation and inference tasks.

[0058] It should be noted that the construction of the reversible neural network model and its analysis and processing method include the following steps:

[0059] Firstly, a network architecture consisting of reversible transformation modules is designed. Each reversible transformation module contains a first affine coupling layer and a reversible 1×1 convolutional layer. The first affine coupling layer splits the input feature vector into pressure parameter group, temperature parameter group and flow parameter group.

[0060] Then, the nonlinear mapping relationship between any two sets of parameters is learned through a fully connected neural network, and then the feature channels are reversibly shuffled through a 1×1 convolutional layer to enhance the feature interaction between parameters of different dimensions.

[0061] Then, during the network training phase, a multi-dimensional operating parameter dataset under normal operating conditions is used to optimize the network parameters through maximum likelihood estimation, so that the model can accurately reconstruct the input feature vector.

[0062] Finally, during fault detection, the feature vectors corresponding to the multidimensional operating parameters collected in real time are fed into the trained reversible neural network model, which then outputs the reconstruction error. The affine coupling layer of the reversible neural network model accurately captures the nonlinear relationships between multiple parameters, such as pressure, temperature, and flow. The output reconstruction error not only reflects the degree of abnormality in the overall operating state, but also preserves the fault characteristics of each parameter dimension, improving fault detection accuracy.

[0063] S400: When the reconstruction error is greater than the reconstruction error threshold, a joint confidence is calculated based on the confidence rule base and the reconstruction error, and when the joint confidence exceeds the confidence threshold, an interactive report including the fault type is generated.

[0064] The interactive report includes the fault type. The confidence rule base is built based on the natural gas station equipment mechanism and fault event case library. The rule antecedents of the confidence rule base include pressure fluctuation rate greater than the first threshold, temperature fluctuation rate greater than the second threshold, and flow fluctuation rate greater than the third threshold. The rule consequents are associated with three types of faults: pressure regulating valve sticking, heat exchanger scaling, and pipeline low-temperature blockage. The correspondence between the rule antecedents and the faults is: the combination of abnormal pressure fluctuation and abnormal temperature fluctuation corresponds to pressure regulating valve sticking, the combination of abnormal pressure fluctuation and abnormal flow fluctuation corresponds to pipeline low-temperature blockage, and the combination of abnormal temperature fluctuation and abnormal flow fluctuation corresponds to heat exchanger scaling. The confidence rule base also includes the initial rule confidence corresponding to the three faults, and the confidence threshold is the initial rule confidence of the corresponding fault.

[0065] Among them, a pressure fluctuation rate greater than a first threshold indicates abnormal pressure fluctuation, a temperature fluctuation rate greater than a second threshold indicates abnormal temperature fluctuation, and a flow fluctuation rate greater than a third threshold indicates abnormal flow fluctuation.

[0066] For example, when collecting pressure values of multi-dimensional operating parameters, the collection frequency is 1 Hz, based on a 60s sliding window sampling, and the pressure fluctuation rate is calculated based on the pressure values collected within the sliding window as: (maximum pressure value within the collection period - minimum pressure value within the collection period) / average pressure value within the collection period.

[0067] When collecting temperature values of multi-dimensional operating parameters, the collection period is 5 minutes, and the calculated temperature fluctuation rate is: , the unit of temperature fluctuation rate is ℃ / min.

[0068] When collecting flow values of multi-dimensional operating parameters, the collection period is 30 seconds, and the flow fluctuation rate is calculated as = The number of data collection can be appropriately increased when regulating pressure, when the external temperature changes suddenly, or when adjusting the valve opening.

[0069] The above parameter combination analysis strategy of the confidence rule library is used to examine the combinations of three key parameters: pressure and temperature, pressure and flow, and temperature and flow. The multi-dimensional operating parameters are integrated through evidence theory to calculate the joint confidence for each type of fault. Finally, the fault type is determined by the joint confidence.

[0070] In some embodiments, the confidence rule base is constructed based on the natural gas station equipment mechanism and the fault event case library, and further includes:

[0071] Firstly, the operating parameters in the fault event operation data whose pressure fluctuation rate is greater than the first threshold, temperature fluctuation rate is greater than the second threshold, and flow fluctuation rate is greater than the third threshold are marked as pressure fluctuation anomaly, temperature fluctuation anomaly, and flow fluctuation anomaly, respectively. The fault event operation data are extracted from the fault event case library of the natural gas station control system.

[0072] Then, experts assess abnormal pressure, temperature, and flow fluctuations based on the natural gas station equipment mechanisms and the proportion of abnormal fluctuations in each operating parameter, determining the corresponding weight coefficient for each operating parameter. For example, since pressure fluctuations account for the largest proportion, their weight coefficient is relatively high: 0.6 for pressure fluctuations, 0.3 for temperature fluctuations, and 0.1 for flow fluctuations.

[0073] Finally, based on the weight coefficients, an evidential reasoning algorithm is used to determine the confidence level of the initial rules for each fault. For example, the confidence level for a stuck pressure regulating valve ranges from 0.85 ± 0.05, for heat exchanger fouling from 0.78 ± 0.07, and for low-temperature pipe blockage from 0.92 ± 0.03.

[0074] It should be noted that based on each weight coefficient, the initial rule confidence of each fault is determined using an evidential reasoning algorithm, which further includes:

[0075] The confidence level of pressure fluctuation anomaly, temperature fluctuation anomaly and flow fluctuation anomaly corresponding to each fault are determined respectively, and a fault evidence matrix is established.

[0076] Based on the fault evidence matrix and weight coefficients, a weighted summation method is used to calculate the original rule confidence for each fault. The original rule confidence is adjusted based on the equipment operating status and expert experience to obtain the initial rule confidence value range for each fault. For example, considering equipment operating status such as equipment aging and service life, the original rule confidence is adjusted by ±5% to obtain the initial rule confidence value range. The initial rule confidence value is within the initial rule confidence value range.

[0077] The steps of determining the pressure fluctuation anomaly confidence level, the temperature fluctuation anomaly confidence level, and the flow fluctuation anomaly confidence level corresponding to each fault further include:

[0078] First, extract relevant data of at least 500 recent fault events with clear accident causes from the natural gas station control system, and extract multidimensional operating parameters up to 1 hour before the fault occurred from the relevant data. Then calculate the corresponding operating parameter volatility based on the multidimensional operating parameters, and perform percentile statistical analysis on the parameter volatility to calculate the 50th percentile of the operating parameter volatility corresponding to each fault. The 50th percentile represents the typical benchmark state of the fault. For example, the 50th percentile of the pressure fluctuation rate corresponding to the pressure regulating valve stuck fault is 4.3%, which means that the pressure fluctuation rate does not exceed 4.3% when 50% of the pressure regulating valve stuck faults occur. Finally, the Sigmoid function is used to calculate the confidence level of each fluctuation anomaly: , where x represents the actual operating parameter fluctuation rate, x0 takes the P50 value of the corresponding fault, and y is the sensitivity coefficient set according to the characteristics of each operating parameter.

[0079] In some embodiments, calculating the joint confidence based on the confidence rule base and the reconstruction error in S400 further includes:

[0080] S410 When the fluctuation rate matches the corresponding rule antecedent, the feature matching degree is calculated based on the initial rule confidence, the fluctuation rate, and the threshold value matching the fluctuation rate, wherein the fluctuation rate includes the pressure fluctuation rate, the temperature fluctuation rate, or the flow fluctuation rate, and the threshold value corresponding to the fluctuation rate includes the first threshold value, the second threshold value, or the third threshold value, respectively. The feature matching degree μ i = initial rule confidence × min (volatility / corresponding threshold, 1).

[0081] S420 quantifies the reconstruction error as a basic probability distribution function m(A)=1-exp(-λE), where E represents the reconstruction error and λ represents a preset fault sensitivity coefficient. Different values are set according to different faults in the natural gas station control system. For example, for a stuck pressure regulating valve fault, λ=0.8, for a low-temperature pipeline blockage fault, λ=1.5, and for a heat exchanger structural fault, λ=1.2.

[0082] S430 compares the basic probability distribution function m(A) with the feature matching degree μ i Perform weighted fusion to obtain the modified probability distribution value m'(A)=α·m(A)+(1-α)·μ i , and calculate the joint confidence according to Dempster's combination rule:

[0083]

[0084] Among them, α represents the empirical coefficient, A and B represent any two of the three rule antecedents, and C represents any one of the three types of faults. For example, when A is abnormal pressure fluctuation and B is abnormal temperature fluctuation, C is the stuck pressure regulating valve; when A is abnormal pressure fluctuation and B is abnormal flow fluctuation, C is the low temperature blockage of the pipeline; when A is abnormal temperature fluctuation and B is abnormal flow fluctuation, C is the scaling of the heat exchanger. The remaining combinations of A and B are classified as the conflict factor K. K is the conflict factor, .

[0085] In some embodiments, the interactive report further includes a warning level, countermeasures, and a timing for implementing the countermeasures. Therefore, after determining the fault type when the joint confidence exceeds a confidence threshold at S400, the method further includes:

[0086] S500 matches response measures based on the fault type. For example, for a stuck pressure regulating valve fault, the system prioritizes checking the valve core wear and implementing lubrication measures. For a scaling fault in a heat exchanger, the system initiates a chemical cleaning procedure and checks the gaps between the heat exchanger plates. For a low-temperature pipe blockage fault, the system increases the power of the heating system.

[0087] S600: When the combined confidence level is greater than the confidence threshold and less than the first confidence threshold, the frequency of collecting operating parameters corresponding to the combined confidence level is increased. When the combined confidence level is greater than or equal to the first confidence threshold and less than the second confidence threshold, a yellow alert is triggered and countermeasures are implemented during the next load trough period. When the combined confidence level is greater than or equal to the second confidence threshold, a red alert is triggered and countermeasures are implemented. Interactive reports are pushed in real time through the HMI interface of the site control system and simultaneously uploaded to the cloud-based operation and maintenance management platform to generate electronic work orders.

[0088] This embodiment sets up a fault warning classification mechanism, and realizes the precise scheduling of operation and maintenance resources by setting a three-level response strategy of increasing the frequency of operating parameter collection, associating yellow warning with load valley maintenance, and associating red warning with immediate execution of countermeasures.

[0089] In some embodiments, the fault detection method further comprises the following steps:

[0090] If the joint confidence is in [0.7β, β], the feature vector and the equally spaced sampling sequence of the multidimensional operating parameters in the last at least 12 hours are input to the LSTM prediction model, where β represents the confidence threshold.

[0091] The trained LSTM prediction model is used to perform multi-step predictions and output the pressure prediction values, temperature prediction values, and flow prediction values every 15 minutes for at least the next 3 hours.

[0092] If the predicted pressure value exceeds the pressure safety threshold, the pressure regulating valve pre-maintenance alarm is triggered. If the predicted temperature value exceeds the temperature safety threshold, the heat exchanger inspection alarm is triggered. If the predicted flow value exceeds the flow safety threshold, the pipeline blockage inspection alarm is triggered. The safety thresholds are the pressure, temperature and flow when there is no fault event at the natural gas station.

[0093] Among them, the Long Short-Term Memory (LSTM) network is a special recursive neural network that controls the flow of information through cell states and three gating mechanisms (input gate, forget gate, and output gate). It can capture complex patterns in time series data and learn long-term dependent information. The training process of the LSTM prediction model provided in the embodiment of the application includes:

[0094] Multidimensional operating parameters from a 30-day continuous fault-free event database extracted from a natural gas station control system were sampled at 15-minute intervals to construct a training sample set. This sample set was preprocessed, including outlier removal, mean filtering using a sliding window, and normalization. A network structure was configured, and model training was performed using the Huber loss function. The network structure consists of an input gate layer, a forget gate layer, and an output gate layer. The input gate layer accepts 48 time steps of historical data from a 12-hour window, each containing three feature dimensions: pressure, temperature, and flow. The first forget gate layer contains 64 neurons, the second forget gate layer contains 64 neurons, and the output gate layer contains three dense units. Linear activation functions are used for pressure, temperature, and flow prediction, respectively. The Huber loss function was used for model training, and the Adam optimizer was employed. The LSTM prediction model can predict pressure, temperature, and flow at 15-minute intervals for at least three hours into the future, providing a basis for accident prevention.

[0095] Example 2

[0096] This embodiment provides a natural gas station control system fault detection device, comprising:

[0097] The data acquisition module is used to collect multi-dimensional operating parameters of natural gas in the natural gas station control system in real time. The multi-dimensional operating parameters include pressure, temperature and flow;

[0098] A preprocessing module is used to preprocess the multi-dimensional operating parameters to generate feature vectors, wherein the preprocessing includes outlier removal, sliding window mean filtering and normalization;

[0099] an analysis and processing module, configured to input the feature vector into a reversible neural network model and output a reconstruction error, wherein the reversible neural network includes a network architecture composed of a reversible transformation module, wherein the affine coupling layer and convolution layer of the reversible transformation module are used to learn multidimensional operating parameters of the natural gas station control system under normal operating conditions, and the reconstruction error is used to characterize the multidimensional deviation between the multidimensional operating parameters and the operating parameters under normal operating conditions learned by the reversible neural network;

[0100] A report generation module is used to calculate the joint confidence based on the confidence rule base and the reconstruction error when the reconstruction error is greater than the reconstruction error threshold. When the joint confidence exceeds the confidence threshold, an interactive report containing the fault type is generated. The confidence rule base is constructed based on the natural gas station equipment mechanism and the fault event case library. The rule antecedents of the confidence rule base include pressure fluctuation rate greater than the first threshold, temperature fluctuation rate greater than the second threshold, and flow fluctuation rate greater than the third threshold. The rule consequents are associated with three types of faults: pressure regulating valve sticking, heat exchanger scaling, and pipeline low-temperature blockage. The correspondence between the rule antecedents and the faults is: the combination of abnormal pressure fluctuation and abnormal temperature fluctuation corresponds to pressure regulating valve sticking, the combination of abnormal pressure fluctuation and abnormal flow fluctuation corresponds to pipeline low-temperature blockage, and the combination of abnormal temperature fluctuation and abnormal flow fluctuation corresponds to heat exchanger scaling. The confidence rule base also includes the initial rule confidences corresponding to the three faults, and the confidence threshold is the initial rule confidence of the corresponding fault.

[0101] Example 3

[0102] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the natural gas station control system fault detection method described in the above embodiment are executed.

[0103] The embodiments of the present application provide a method, device, and readable storage medium for fault detection in a natural gas station control system. The method collects multidimensional operating parameters such as pressure, temperature, flow, and valve opening in real time, inputs them into a reversible neural network model after preprocessing, and outputs a reconstruction error that can characterize the degree of multidimensional deviation between the current operating state and the normal operating condition. When the reconstruction error exceeds the reconstruction error threshold, it is determined that the current operating state may be abnormal, and the confidence rule base is activated for in-depth analysis. Furthermore, the confidence rule base adopts a parameter combination analysis strategy to perform a joint analysis of multiple parameters such as pressure-temperature, pressure-flow, and temperature-flow to obtain a joint confidence. When the joint confidence exceeds the confidence threshold, an interactive report containing the fault type is generated. The present application achieves accurate fault diagnosis of the real-time operating state through the dual judgment mechanism of the reversible neural network and the confidence rule base.

[0104] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0105] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0106] Through the description of the above embodiments, it will be readily understood by those skilled in the art that the exemplary embodiments described herein can be implemented via software or via a combination of software and necessary hardware. Therefore, the technical solutions according to the embodiments of the present invention can be embodied in the form of a software product, which can be stored on a non-volatile storage medium (such as a CD-ROM, USB flash drive, or mobile hard drive) or on a network and includes instructions for causing a computing device (such as a personal computer, server, touch terminal, or network device) to execute the methods according to the embodiments of the present invention.

[0107] In the several embodiments provided in this application, it should be understood that the disclosed systems, modules and methods can be implemented in other ways. For example, the module embodiments described above are merely illustrative. For example, the division of units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of modules or units, which can be electrical, mechanical or other forms.

[0108] The above embodiments are intended only to illustrate the technical solutions of the present application and are not intended to limit them. The present application is not limited to the precise structures described above and illustrated in the accompanying drawings, and it cannot be assumed that the specific implementation of the present application is limited to these descriptions. For those skilled in the art of the present application, any changes and modifications made without departing from the concept of the present application should be deemed to fall within the scope of protection of the present application.

Claims

1. A method for detecting faults in a natural gas station control system, characterized in that: The method comprises the following steps: Real-time collection of multi-dimensional operating parameters of natural gas in the natural gas station control system, including pressure, temperature and flow; Preprocessing the multidimensional operating parameters to generate feature vectors, wherein the preprocessing sequentially includes outlier removal, sliding window mean filtering, and normalization; Inputting the feature vector into a reversible neural network model and outputting a reconstruction error, wherein the reversible neural network includes a network architecture composed of a reversible transformation module, the affine coupling layer and the convolution layer of the reversible transformation module are used to learn multidimensional operating parameters of the natural gas station control system under normal operating conditions, and the reconstruction error is used to represent the multidimensional deviation degree between the multidimensional operating parameters and the operating parameters under normal operating conditions learned by the reversible neural network; When the reconstruction error is greater than the reconstruction error threshold, a joint confidence is calculated based on the confidence rule base and the reconstruction error. When the joint confidence exceeds the confidence threshold, an interactive report containing the fault type is generated, wherein the confidence rule base is constructed based on the natural gas station equipment mechanism and the fault event case library. The rule antecedents of the confidence rule base include pressure fluctuation rate greater than a first threshold, temperature fluctuation rate greater than a second threshold, and flow fluctuation rate greater than a third threshold. The rule consequents are associated with three types of faults: pressure regulating valve sticking, heat exchanger scaling, and pipeline low-temperature blockage. The correspondence between the rule antecedents and the faults is: the combination of pressure fluctuation anomaly and temperature fluctuation anomaly corresponds to pressure regulating valve sticking, the combination of pressure fluctuation anomaly and flow fluctuation anomaly corresponds to pipeline low-temperature blockage, and the combination of temperature fluctuation anomaly and flow fluctuation anomaly corresponds to heat exchanger scaling. The confidence rule base also includes initial rule confidences corresponding to the three types of faults, and the confidence threshold is the initial rule confidence of the corresponding fault.

2. The natural gas station control system fault detection method according to claim 1, characterized in that: The interactive report also includes a warning level, countermeasures, and a timing for implementing the countermeasures. After the fault type is determined when the combined confidence exceeds a confidence threshold, the method further includes: Match response measures based on the fault type. For example, for a stuck pressure regulating valve, prioritize checking valve core wear and implementing lubrication measures. For a scaling heat exchanger, initiate a chemical cleaning procedure and check the heat exchanger fin clearances. For a low-temperature pipe blockage, increase the power of the heating system. When the combined confidence is greater than the confidence threshold and less than a first confidence threshold, increasing the frequency of collecting corresponding operating parameters; When the combined confidence is greater than or equal to the first confidence threshold and less than the second confidence threshold, a yellow warning is triggered and the corresponding countermeasure is executed during the next load valley period; When the combined confidence is greater than or equal to the second confidence threshold, a red alert is triggered and the corresponding countermeasure is executed.

3. The natural gas station control system fault detection method according to claim 1, characterized in that: The confidence rule base is constructed based on the natural gas station equipment mechanism and fault event case library, and further includes: The operating parameters whose pressure fluctuation rate is greater than a first threshold, whose temperature fluctuation rate is greater than a second threshold, and whose flow rate fluctuation rate is greater than a third threshold in the fault event operating data are respectively marked as abnormal pressure fluctuation, abnormal temperature fluctuation, and abnormal flow rate fluctuation, wherein the fault event operating data is extracted from the fault event case library of the natural gas station control system; Experts evaluate abnormal pressure, temperature, and flow fluctuations based on the mechanism of natural gas station equipment and the proportion of abnormal fluctuations in each operating parameter, and determine the weight coefficient corresponding to each operating parameter; Based on the weight coefficients, an evidential reasoning algorithm is used to determine the initial rule confidence of each fault.

4. The natural gas station control system fault detection method according to claim 3, characterized in that: The method of determining the initial rule confidence of each fault based on each weight coefficient using an evidential reasoning algorithm further includes: Determine the confidence level of pressure fluctuation anomaly, temperature fluctuation anomaly, and flow fluctuation anomaly corresponding to each fault, and establish a fault evidence matrix; Calculating the confidence of the original rule of each fault using a weighted summation method according to the fault evidence matrix and the weight coefficient; The confidence level of the original rule is adjusted based on the equipment operating status and expert experience to obtain a value range of the confidence level of the initial rule for each fault.

5. The natural gas station control system fault detection method according to claim 1, characterized in that: The calculating of the joint confidence based on the confidence rule base and the reconstruction error further includes: When the fluctuation rate matches the rule antecedent, a feature matching degree is calculated based on the initial rule confidence, the fluctuation rate, and a threshold value matching the fluctuation rate, wherein the fluctuation rate includes a pressure fluctuation rate, a temperature fluctuation rate, or a flow fluctuation rate, and the threshold value corresponding to the fluctuation rate includes the first threshold value, the second threshold value, or the third threshold value, respectively. The feature matching degree μ i = initial rule confidence × min (volatility / corresponding threshold, 1); The reconstruction error is quantified as a basic probability distribution function m(A)=1-exp(-λE), where E represents the reconstruction error and λ represents a preset fault sensitivity coefficient, which is set to different values according to different faults in the natural gas station control system; The basic probability distribution function m(A) is matched with the feature matching degree μ i Perform weighted fusion to obtain the modified probability distribution value m'(A)=α·m(A)+(1-α)·μ i , and calculate the joint confidence according to Dempster's combination rule: ， Among them, α represents the empirical coefficient, A and B represent any two of the three rule antecedents, C represents any one of the three types of faults, and K is the conflict factor. .

6. The natural gas station control system fault detection method according to claim 5, characterized in that: A and B represent any two of the three rule antecedents, and C represents any one of the three types of faults, further including: When A is abnormal pressure fluctuation and B is abnormal temperature fluctuation, C is stuck pressure regulating valve; when A is abnormal pressure fluctuation and B is abnormal flow fluctuation, C is low temperature blockage of pipeline; when A is abnormal temperature fluctuation and B is abnormal flow fluctuation, C is scaling of heat exchanger, and other combinations of A and B are classified into the conflict factor K.

7. The natural gas station control system fault detection method according to claim 1, characterized in that: The method further comprises: If the joint confidence level is in the range of [0.7β, β], the feature vector and an equally spaced sampling sequence of the multidimensional operating parameters within at least the last 12 hours are input into the LSTM prediction model, where β represents the confidence threshold. Perform multi-step predictions using the trained LSTM prediction model to output predicted pressure, temperature, and flow rates every 15 minutes for at least the next three hours. If the predicted pressure value exceeds the pressure safety threshold, the pressure regulating valve pre-maintenance alarm is triggered; if the predicted temperature value exceeds the temperature safety threshold, the heat exchanger inspection alarm is triggered; if the predicted flow value exceeds the flow safety threshold, the pipeline blockage inspection alarm is triggered. The safety thresholds are the pressure, temperature and flow when there is no fault event at the natural gas station.

8. The natural gas station control system fault detection method according to claim 7, characterized in that: The training process of the LSTM prediction model includes: The multi-dimensional operating parameters of the 30-day continuous fault-free event case library extracted from the natural gas station control system are sampled at 15-minute intervals to construct a training sample set; Preprocessing the training sample set, including removing outliers, performing mean filtering and standardization through a sliding window; A network structure was configured, and the Huber loss function was used for model training. The network structure included an input gate layer, a forget gate layer, and an output gate layer. The input gate layer was used to accept 12-hour historical window data for a total of 48 time steps, and each time step included three feature dimensions: pressure, temperature, and flow. The first forget gate layer contained 64 neurons, and the second forget gate layer contained 64 neurons. The output gate layer contained three dense units corresponding to the prediction of pressure, temperature, and flow, respectively. The Huber loss function was used for model training.

9. A natural gas station control system fault detection device, characterized in that: The device comprises: A data acquisition module is used to collect multi-dimensional operating parameters of natural gas in the natural gas station control system in real time, wherein the multi-dimensional operating parameters include pressure, temperature and flow; A preprocessing module, configured to preprocess the multidimensional operating parameters to generate feature vectors, wherein the preprocessing sequentially includes outlier removal, sliding window mean filtering, and normalization; an analysis and processing module, configured to input the feature vector into a reversible neural network model and output a reconstruction error, wherein the reversible neural network includes a network architecture composed of a reversible transformation module, the affine coupling layer and the convolution layer of the reversible transformation module are used to learn multidimensional operating parameters of the natural gas station control system under normal operating conditions, and the reconstruction error is used to characterize the multidimensional deviation between the multidimensional operating parameters and the operating parameters under normal operating conditions learned by the reversible neural network; A report generation module is used to calculate a joint confidence based on a confidence rule base and the reconstruction error when the reconstruction error is greater than a reconstruction error threshold, and to generate an interactive report containing the fault type when the joint confidence exceeds the confidence threshold, wherein the confidence rule base is constructed based on the natural gas station equipment mechanism and the fault event case library, the rule antecedents of the confidence rule base include a pressure fluctuation rate greater than a first threshold, a temperature fluctuation rate greater than a second threshold, and a flow fluctuation rate greater than a third threshold, and the rule consequents are associated with three types of faults: pressure regulating valve sticking, heat exchanger scaling, and pipeline low-temperature blockage, the correspondence between the rule antecedents and the faults is: the combination of abnormal pressure fluctuation and abnormal temperature fluctuation corresponds to pressure regulating valve sticking, the combination of abnormal pressure fluctuation and abnormal flow fluctuation corresponds to pipeline low-temperature blockage, and the combination of abnormal temperature fluctuation and abnormal flow fluctuation corresponds to heat exchanger scaling, the confidence rule base also includes initial rule confidences corresponding to the three types of faults, and the confidence threshold is the initial rule confidence of the corresponding fault.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the natural gas station control system fault detection method according to any one of claims 1 to 8 are executed.

Images