Steam conveying control method and system based on multi-stage pressure regulation and dynamic feedback

Abstract

The invention provides a steam conveying control method and system based on multistage pressure regulation and dynamic feedback, and the method comprises the steps: obtaining real-time data, including pressure, flow and vibration signals, of a multi-source sensor in a steam pipe network, and forming a multi-dimensional data set; after judging that the water hammer phenomenon occurs, analyzing the vibration signal to obtain an abnormal vibration frequency in a pipe network, determining a propagation path of impact waves by combining pressure data, and positioning an area where an impact source is located; performing pressure grading control of partition differentiation according to the affected partitions; acquiring and analyzing the steam flow fluctuation condition of each affected subarea, analyzing the steam flow fluctuation condition of each subarea, determining target flow and priority, and adjusting the opening degree of an adjusting valve; real-time pressure and flow reading of a pipe network are continuously monitored, and setting of pressure adjusting points of all levels and the opening degree of the corresponding adjusting valve are dynamically adjusted so that steam conveying can be stably controlled.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a steam delivery control method and system based on multi-stage pressure regulation and dynamic feedback. Background Art

[0002] As the core infrastructure of energy management in large chemical industrial parks, the steam centralized supply system is directly related to production efficiency, safety and economic benefits. Its stable operation is crucial for ensuring the downstream user demand. Under extreme conditions, such as a drastic fluctuation in the steam load of downstream users, the steam pipe network may encounter water hammer phenomenon, triggering instantaneous high-pressure shock waves, which pose a serious threat to the system reliability. Therefore, studying the time-series response characteristics and dynamic control strategies of water hammer phenomenon is not only an urgent need to improve the resilience of the energy system in chemical industrial parks, but also a key direction to promote the progress of industrial safety technology. Currently, the solutions to water hammer problems mostly rely on static design optimization or single-sensor monitoring. However, these methods often seem inadequate when faced with complex and variable load fluctuations. Static design is difficult to adapt to the instantaneous impact under dynamic working conditions, and single monitoring cannot comprehensively capture multi-source disturbance signals, resulting in a lag in water hammer event recognition and insufficient countermeasures. This limitation is particularly obvious in the scenario of a large chemical industrial park with multi-user and multi-device coupling. The core challenges faced by the research focus on the difficulty of accurately predicting the impact of the instantaneous high-pressure shock wave caused by water hammer on the pipeline system, and the insufficient rapid response ability of the dynamic feedback control system under multi-source information. Specifically, when the water hammer phenomenon occurs, the shock wave may cause local deformation of the pipeline, internal leakage of the valve, oscillation of the sensor readings and loosening of the support system. Existing technologies have obvious shortcomings in shock source positioning and pressure zone regulation. These unsolved technical factors may cause water hammer events to evolve into systematic failures, further exacerbating steam distribution imbalance and equipment damage. Therefore, how to quickly identify water hammer events, accurately locate the shock source based on multi-source sensor information, and effectively control the instantaneous high-pressure shock wave through multi-stage pressure zone regulation and optimized steam flow distribution has become the key problem that this research urgently needs to overcome. Summary of the Invention

[0003] The present invention provides a steam delivery control method based on multi-stage pressure regulation and dynamic feedback, mainly including: Obtain the real-time data of multi-source sensors in the steam pipe network, including pressure, flow rate and vibration signals, to form a multi-dimensional data set; Process the real-time pressure data in the data set to obtain pressure time-series data, determine the pressure change trend, and judge whether an instantaneous high-pressure peak appears. If the high-pressure peak exceeds the preset peak threshold, it is determined that the water hammer phenomenon occurs; After the occurrence of water hammer phenomenon is determined, the vibration signal is analyzed to obtain the abnormal vibration frequency in the pipe network. Combining with the pressure data, the propagation path of the shock wave is determined, and the area where the impact source is located is located; The monitoring data of each sensor in the area where the impact source is located is processed by the multi-source data fusion algorithm, and the propagation speed and attenuation characteristics of the shock wave in different pipe sections are calculated to obtain the impact source position and the affected areas; Carry out pressure grading control with partition differences according to the affected areas; Obtain and analyze the steam flow rate fluctuations in each affected area, analyze the steam flow rate fluctuations in each area, determine the target flow rate and priority, and adjust the opening of the regulating valve; Real-time monitor the pipeline deformation data after adjusting the opening of the regulating valve, identify the local deformation area and the degree of deformation. If the deformation amount exceeds the preset safety value, reduce the pressure load in the local deformation area; Continuously monitor the real-time pressure and flow readings of the pipe network, and dynamically adjust the settings of each pressure regulating point and the opening of the corresponding regulating valve to stably control steam transportation.

[0004] Furthermore, obtain the real-time data of multi-source sensors in the steam pipe network, including pressure, flow rate and vibration signals, to form a multi-dimensional data set, including: collecting pressure sensor data, flow rate sensor data and vibration sensor data from multi-source sensors in the steam pipe network according to a preset sampling period, marking the collected data with time stamps by a data collector, and obtaining normalized sensor data through data normalization processing. Calculate the normal value ranges of pressure, flow rate and vibration based on the historical collected data of each sensor as dynamic adjustment thresholds. When the normalized sensor data exceeds the corresponding normal value range, automatically increase the sampling frequency of the sensor to the preset maximum sampling frequency, and resume to the original sampling frequency after the normal values are continuously collected for a preset number of times. Use the linear interpolation algorithm to repair the missing points in the normalized sensor data, align the data according to the time stamp information of the sensor data, and perform noise reduction processing on the data through moving window mean filtering to obtain smooth sensor data. Build a time series database for the smooth sensor data, perform minute-level, hour-level and day-level aggregation storage on the data, set the data storage period, and automatically archive the data that exceeds the storage period to the historical database. Use the Pearson correlation coefficient calculation method to perform multi-dimensional correlation analysis on the pressure data, flow rate data and vibration data, and establish a feature association table for the data with the correlation coefficient exceeding the preset threshold of 0.8. Select high-correlation feature combinations based on the feature association table, divide the pressure data, flow rate data and vibration data into training data sets according to time windows, and use the random forest regression method to build a multi-dimensional sensor data prediction model.

[0005] Furthermore, the real-time pressure data in the data set is processed to obtain pressure time-series data, determine the pressure change trend, and judge whether an instantaneous high-pressure peak appears. If the high-pressure peak exceeds the preset peak threshold, it is determined that a water hammer phenomenon occurs, including: obtaining the original pressure sampling data sequence from the pipe network pressure sensor, validating the effectiveness of the original pressure sampling data through the data verification rule, removing abnormal sampling points, and obtaining the valid pressure data. A sliding window is used to perform median filtering on the valid pressure data, and the window length is set to one-thousandth of the number of sampling points to calculate the filtered pressure data. The wavelet decomposition algorithm is used to perform three-layer scale decomposition on the filtered pressure data, extract the high-frequency coefficients as the data reference, calculate the mean and standard deviation of the pressure data, set the dynamic baseline threshold as the mean plus or minus three times the standard deviation to obtain the normalized pressure data. Calculate the pressure change slope sequence according to the normalized pressure data, extract the set of pressure peak points, label the points in the peak point sequence that exceed the dynamic baseline threshold, and calculate the amplitude, duration, and pressure rise rate of each over-limit peak point. Construct a water hammer feature database through historical water hammer event records, recording the pressure peak amplitude, duration, and pressure rise rate at the moment of water hammer occurrence. Use a long short-term memory neural network to classify the over-limit peak features, and the input features include the pressure peak amplitude, duration, and pressure rise rate. When any feature exceeds the mean of the corresponding feature in the water hammer feature database, it is determined that a water hammer phenomenon occurs.

[0006] Further, after determining the occurrence of water hammer phenomenon, analyze the vibration signal to obtain the abnormal vibration frequency in the pipe network. Combine the pressure data to determine the propagation path of the shock wave and locate the area where the impact source is located, including: obtaining the original vibration data with a sampling frequency of 1000 Hz from the vibration sensors in the pipe network, filtering the noise of the original vibration data through band-pass filtering with a cut-off frequency of 10 Hz to 200 Hz, and performing spectral analysis on the filtered vibration data using the fast Fourier transform to obtain the vibration frequency spectrum. According to the statistical analysis of historical vibration data, obtain the background frequency distribution range during the normal operation of the pipe network, set the upper threshold of the background frequency, mark the points exceeding the threshold in the frequency spectrum, perform frequency clustering processing on the marked points to generate a set of abnormal vibration frequencies, and record the amplitude and occurrence time of the abnormal vibration frequencies. Obtain the position information of the pressure sensors and the pipe segment connection relationship from the pipe network topology map, obtain the pressure data with a sampling frequency of 100 Hz from the pressure sensors, and align the data acquisition time bases of all sensors. Calculate the time difference between the arrival times of the pressure wave peaks at adjacent sensors, calculate the pressure wave propagation speed in combination with the pipe segment distance, and establish a directed graph of the pressure propagation path according to the pipe network connection relationship. Extract the node degree, path distance, and number of branches for the directed graph of the pressure propagation path, establish a node feature vector, and perform traceability analysis on the pressure propagation time series data using a Bayesian network. Combine the abnormal vibration frequency amplitude distribution and the pressure propagation time series relationship to construct a node priority scoring function, calculate the probability of each node as the impact source, and select the node with the highest probability as the impact source position.

[0007] Furthermore, the monitoring data of each sensor in the area where the impact source is located is processed by a multi-source data fusion algorithm, and the propagation speed and attenuation characteristics of the shock wave in different pipeline partitions are calculated to obtain the impact source location and the affected partition, including: obtaining real-time monitoring data with sampling frequencies of 100 Hz, 1000 Hz, and 500 Hz from the pressure sensor, vibration sensor, and acoustic sensor in the pipeline network, respectively, resampling the monitoring data to 100 Hz according to the sensor acquisition timestamp, and performing noise reduction processing on the resampled data by a mean filter with a sliding window length of 5 to obtain a filtered monitoring data sequence. The monitoring data includes pressure, vibration, and acoustic signals. The Kalman filter is used to perform state fusion on the filtered monitoring data sequence. The state variables include pressure value, vibration amplitude, and acoustic signal intensity. The observation noise covariance matrix is determined by the sensor measurement accuracy to obtain fused monitoring data. The sensor position association matrix is established according to the topological structure of the pipeline network partition. The matrix element value represents the pipe section distance between adjacent sensors. The time difference of the peak value of the fused monitoring data reaching the adjacent sensors is calculated to generate a time difference matrix. Based on the sensor position association matrix and the time difference matrix, a propagation path weighted graph is constructed. The Dijkstra algorithm is used to calculate the shortest path of the shock wave from the source point to each measuring point, and the actual propagation distance is recorded. The propagation velocity within the pipe network partition is calculated based on the actual propagation distance and the signal arrival time difference. The exponential function model is used to fit the peak attenuation curve of the monitoring data. The curve parameters include the initial peak value and the attenuation coefficient. The propagation velocity, peak initial value, and attenuation coefficient are input into the deep neural network. The network contains three hidden layers, with the number of hidden layer nodes being 64, 32, and 16 respectively, and the output layer is the spatial coordinates of the impact source. Based on the shock source position output by the deep neural network, the distance from each pipe network partition measuring point to the shock source is calculated. When the distance of the measuring point is less than the preset distance threshold, the partition where the measuring point is located is included in the affected partition set.

[0008] Further, perform pressure classification control with differences among partitions according to the affected partitions, including: dividing the pressure control priorities according to the distances of the affected partitions from the impact source. Those with distances less than 300 meters are marked as primary partitions, those between 300 meters and 500 meters are marked as secondary partitions, and those over 500 meters are marked as tertiary partitions. Calculate the partition control index using fuzzy rules. The rules include three input variables: distance weight, pressure overrun ratio, and affected area, and obtain the partition regulation priority sequence. Read the pressure values, flow values, and valve opening data within the affected partitions from the pipeline network monitoring database, and calculate the partition pressure adjustment amount according to the pressure change rate and the pressure fluctuation amplitude. The adjustment amount calculation formula consists of three terms: pressure deviation term, pressure change rate term, and fluctuation amplitude term. Optimize the pressure adjustment amount using an adaptive neural network. The input layer of the network includes three types of parameters: pipe segment flow rate, valve opening, and pressure fluctuation rate. The hidden layer adopts an adaptive number of nodes structure, and the output layer is the optimized adjustment amount. Establish the pipeline network hydraulic response characteristic curve based on the pipeline network test data, calculate the relationship between the pressure adjustment effect and the pipeline network response delay, and determine the partition adjustment step sequence through a hierarchical control algorithm. Calculate the mutual feedback influence coefficient of the pressure adjustment between adjacent partitions according to the pipeline network hydraulic characteristic curve, and correct the adjustment step of each partition using a time series attenuation function to generate a valve adjustment command considering the coupling effect between partitions. Calculate the partition pressure mean value and the fluctuation standard deviation during the pressure adjustment process using an exponential sliding window. When the fluctuation standard deviation is less than the upper limit value of the set interval within consecutive monitoring periods and the pressure mean value is within the upper and lower limit intervals of the rated working pressure, it is determined that the pressure adjustment is completed.

[0009] Further, obtain and analyze the steam flow rate fluctuation conditions of each affected partition, analyze the steam flow rate fluctuation conditions of each partition, determine the target flow rate and priority, and adjust the opening of the regulating valve, including: obtain the original flow rate data with a sampling frequency of 100 Hz from the flow meters in the affected partitions, eliminate outliers through data validity verification, perform three-layer decomposition and denoising on the original flow rate data through wavelet transform, calculate the flow rate mean value sequence and the standard deviation sequence using a sliding window with a length of 60 seconds, and obtain the flow rate fluctuation characteristic sequence. Mark the points where the standard deviation in the flow rate fluctuation characteristic sequence exceeds 20% of the mean value, calculate the flow rate fluctuation amplitude and the fluctuation duration of each partition, and record the corresponding pressure adjustment change amount during the fluctuation. Classify the flow rate fluctuation parameters using fuzzy inference rules. The input variables include flow rate fluctuation amplitude, fluctuation duration, and pressure adjustment amount, and output the partition flow rate control priority index. Sort the fluctuating flow rate data of each partition according to the partition flow rate control priority index, and use support vector regression to predict the flow rate change trend. The input features include the historical flow rate sequence, fluctuation frequency, and fluctuation amplitude. Obtain the pressure distribution state through the pipeline network online monitoring data, determine the target flow rate adjustment interval in combination with the flow rate prediction trend, and generate the partition flow rate adjustment index. Calculate the change amount of the regulating valve opening according to the pipeline network hydraulic characteristic curve, and divide the adjustment step using hierarchical control. The adjacent adjustment interval is not less than the response delay time.

[0010] Furthermore, the pipeline deformation data after adjusting the opening degree of the regulating valve is monitored in real time to identify the local deformation area and the degree of deformation. If the deformation amount exceeds the preset safety value, the pressure load in the local deformation area is reduced, including: obtaining the original deformation data with a sampling frequency of 10 Hz from the deformation sensors arranged every 50 meters along the pipeline, performing denoising processing on the deformation data using median filtering with a window length of 5, and calculating the deformation mean and standard deviation sequences through a 60-second sliding window. Calculate the deformation amount warning threshold according to the elastic modulus and strength limit of the pipeline material. When the deformation amount exceeds five-thousandths of the pipeline outer diameter, a warning is triggered to obtain the deformation monitoring reference value. Perform spatial clustering processing on the deformation data using the density-based clustering algorithm, set the clustering radius to 100 meters, identify the local deformation area of the pipeline, and record the deformation amount and the central position coordinates of the deformation area. Obtain the pipeline stress-strain curve from the material experiment database, and calculate the stress state of the local deformation area in combination with the measured deformation amount. When the stress exceeds 80% of the yield strength, it is determined to enter the dangerous state. For the dangerous state area, calculate the pressure adjustment amount using the steepest descent algorithm based on the pressure gradient. The algorithm inputs include the current pressure value, deformation amount, and stress state, and the output is the target pressure reduction value. Input the target pressure reduction value into a three-layer neural network controller with the number of hidden layer nodes being 32, 16, and 8 respectively, and output the regulating valve opening adjustment sequence. The controller updates the weights in an online learning manner. Implement pressure adjustment on the local deformation area according to the regulating valve opening adjustment sequence, and monitor the change trend of the deformation amount. When the deformation amount is reduced to 80% of the warning threshold and lasts for 300 seconds, it is determined that the adjustment is completed.

[0011] Furthermore, continuously monitor the real-time pressure and flow readings of the pipe network, and dynamically adjust the settings of the pressure regulation points at all levels and the opening degrees of the corresponding regulating valves to stably control steam transmission, including: collecting pressure and flow data with a sampling frequency of 100 Hz from each pressure regulation point, removing high-frequency interference through a digital low-pass filter with a cut-off frequency of 10 Hz, calculating the pressure-flow volatility using a 60-second sliding window, and generating a volatility feature sequence. Divide the pressure levels according to the pipe network structure, and successively divide them into three levels: high-pressure level, medium-pressure level, and low-pressure level from the source point to the end point. Calculate the deviation between the pressure-flow volatility of each level and the rated value, and generate a volatility deviation sequence. Mark the points exceeding the preset threshold in the volatility deviation sequence, and calculate the pressure regulation target value using the dynamic programming algorithm. The objective function of the algorithm includes a volatility deviation term, an adjustment amount term, and a response time term, and the constraint conditions include pressure-flow limits and adjustment step limits. Based on the regulation target values of each pressure level, construct a deep reinforcement learning trainer. The trainer adopts a four-layer network structure, with the input layer being the pressure-flow state vector, and the number of hidden layer nodes being 64, 32, and 16 respectively, and the output layer being the adjustment action vector. The trainer updates the network parameters through online learning, and uses a reward function to measure the control effect. The reward function includes a pressure deviation term, a flow deviation term, and a volatility term, and generates an opening degree action sequence of the regulating valve. Adjust the pressure points according to the opening degree action sequence of the regulating valve, and record the pressure-flow response curve. When the pressure volatility is less than 5% and the flow volatility is less than 10% within a continuous monitoring period, it is determined that the control is completed.

[0012] The present invention provides a steam transmission control system based on multi-level pressure regulation and dynamic feedback, mainly including: A data acquisition module for obtaining real-time data of multi-source sensors in the steam pipe network, including pressure, flow, and vibration signals, to form a multi-dimensional data set; A pressure analysis module for processing the real-time pressure data in the data set to obtain pressure time series data, determining the pressure change trend, and judging whether an instantaneous high-pressure peak appears. If the high-pressure peak exceeds the preset peak threshold, it is determined that a water hammer phenomenon occurs; A vibration analysis module for analyzing the vibration signal after determining the water hammer phenomenon, obtaining the abnormal vibration frequency in the pipe network, and combining with the pressure data to determine the propagation path of the shock wave and locate the area where the shock source is located; A shock source positioning module for processing the monitoring data of each sensor in the area where the shock source is located through a multi-source data fusion algorithm, calculating the propagation speed and attenuation characteristics of the shock wave in different pipe partitions, and obtaining the shock source position and the affected partitions; A pressure control module for performing zone-differentiated pressure grading control according to the affected partitions; The flow rate adjustment module is used to obtain and analyze the steam flow rate fluctuations in each affected partition, analyze the steam flow rate fluctuations in each partition, determine the target flow rate and priority, and adjust the opening degree of the regulating valve; The deformation monitoring module is used to monitor the pipeline deformation data in real time after adjusting the opening degree of the regulating valve, identify the local deformation area and the degree of deformation. If the deformation amount exceeds the preset safety value, the pressure load in the local deformation area is reduced; The dynamic regulation module is used to continuously monitor the real-time pressure and flow rate readings of the pipe network, and dynamically adjust the settings of the pressure regulation points at all levels and the opening degree of the corresponding regulating valves to stably control the steam transmission.

[0013] The technical solution provided by the embodiment of the present invention may include the following beneficial effects: The present invention discloses a steam transmission control method based on multi-level pressure regulation and dynamic feedback. This method obtains real-time data from multi-source sensors, analyzes the pressure time-series data, and determines whether a water hammer phenomenon occurs. When a water hammer is detected, the shock wave propagation path and the impact source location are determined by combining the analysis of vibration signals. Subsequently, the present invention uses a multi-source data fusion algorithm to calculate the shock wave propagation characteristics and determine the affected partitions. For the affected partitions, the present invention implements differential pressure grading control and adjusts the opening degree of the regulating valve in combination with the analysis of flow rate fluctuations. At the same time, the pipeline deformation is monitored in real time, and the local pressure load is reduced when necessary. Through continuous monitoring and dynamic adjustment, the present invention realizes the precise detection, rapid positioning and effective control of the water hammer phenomenon in the steam pipe network, ensuring the stability and safety of steam transmission. Description of the Drawings

[0014] Figure 1 It is a flowchart of the steam transmission control method based on multi-level pressure regulation and dynamic feedback of the present invention.

[0015] Figure 2 It is a structural diagram of the steam transmission control system based on multi-level pressure regulation and dynamic feedback of the present invention. Detailed Embodiments

[0016] To further understand the content of the present invention, the present invention will be described in detail in combination with the drawings and embodiments. The following further details the present application in combination with the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related invention, rather than limiting the invention. In addition, it should be noted that only the parts related to the invention are shown in the drawings for the convenience of description.

[0017] In the embodiment of the present invention, the steam pipe network system includes multi-source sensors and a control unit, aiming to realize the optimized control of steam transmission through dynamic feedback. The following details the specific implementation manners in combination with the drawings.

[0018] Such as Figure 1, the steam delivery control method based on multi - level pressure regulation and dynamic feedback in this embodiment may specifically include: S101. During the operation of the steam pipe network, obtain the real - time data of multi - source sensors, including pressure, flow, and vibration signals, to form a multi - dimensional data set.

[0019] In the embodiment of the present invention, a variety of sensors are deployed in the steam pipe network to monitor the system status in real time. Data acquisition first obtains the original signals from the pressure sensor, flow sensor, and vibration sensor through a preset period, and the data collector attaches a time stamp to ensure time - series consistency.

[0020] S1011. Collect data from the multi - source sensors of the steam pipe network according to a preset sampling period and perform normalization processing. Among them, the sampling period of the pressure sensor is 100 milliseconds, the flow sensor is 200 milliseconds, and the vibration sensor is 50 milliseconds. The data collector attaches a 13 - bit time stamp to each group of data, and then normalizes the collected data to the numerical range of 0 to 1 to obtain normalized sensor data.

[0021] S1012. Determine the normal value range of each sensor according to historical data and dynamically adjust the sampling frequency. The normal value range of pressure is 0.4 to 0.6 MPa, the flow is 50 to 150 cubic meters per hour, and the vibration is 0 to 2.5 millimeters per second. If the normalized data exceeds the normal range, such as the pressure suddenly increases to 0.7 MPa, the sampling frequency of the corresponding sensor is automatically increased to the preset maximum value of 20 milliseconds, and the original frequency is restored after the data returns to normal and lasts for 30 minutes.

[0022] S1013. Use the linear interpolation algorithm to repair the missing points of the normalized sensor data, perform fitting estimation based on the time - series information of the 3 data points before and after, and then eliminate the high - frequency noise through the moving window mean filter with a length of 5 to generate smooth sensor data to improve the subsequent analysis accuracy.

[0023] In practical applications, sensor data may be missing due to transmission interruption. The embodiment of the present invention ensures data integrity through linear interpolation and uses filtering technology to reduce noise interference, providing a reliable basis for subsequent analysis.

[0024] S1014. Store the smooth sensor data in the time - series database, perform hierarchical aggregation in the way of saving for 15 days at the minute level, 90 days at the hour level, and 365 days at the day level. The expired data is automatically archived to the historical database and compressed for storage. At the same time, extract data based on a 10 - minute time window, and use the Pearson correlation coefficient to analyze the multi - dimensional correlation of pressure, flow, and vibration. The features with a correlation coefficient exceeding 0.8 are listed in the feature association table.

[0025] S1015. Screen high-correlation feature combinations according to the feature correlation table. For example, when the correlation coefficient between pressure and flow rate reaches 0.85, select the pressure mean, pressure standard deviation, flow rate mean, and flow rate standard deviation as input features. Divide the training data set according to a 30-minute time window, and use the random forest regression method to construct a multi-dimensional sensor data prediction model. The model is configured with 50 decision trees, the maximum depth is 6, and the minimum number of samples in the leaf nodes is 5, to predict the pressure and flow rate values in the next 10 minutes.

[0026] In the embodiment of the present invention, the establishment of the multi-dimensional data prediction model provides data support for subsequent water hammer identification. The prediction results can be used to early warn of abnormal trends, thereby enhancing the active control ability of the system.

[0027] It can be understood that the sampling period and normal value range in this step can be flexibly adjusted according to the actual working conditions, and the specific parameters are not limited by the embodiments of the present invention. Through the above operations, the embodiments of the present invention can efficiently obtain and process multi-source data of the steam pipeline network, lay a foundation for subsequent water hammer phenomenon determination and dynamic regulation, and realize efficient control of steam transmission.

[0028] S102. Process the real-time pressure data collected from the steam pipeline network to generate pressure time series data, and judge whether there is an instantaneous high-pressure peak by analyzing the pressure change trend, and combine the feature classification technology to determine the occurrence of the water hammer phenomenon.

[0029] In the embodiment of the present invention, the processing of pressure data is a key link for identifying the water hammer phenomenon, aiming to extract effective information from multi-source data and perform anomaly detection. The pressure sensors in the steam pipeline network collect data at a fixed frequency, and then analyze the pressure fluctuation characteristics.

[0030] S1021. Obtain the original pressure sampling data sequence from the pressure sensors of the steam pipeline network and perform validity verification. The sensors continuously monitor the pipeline network status at a sampling frequency of 100 Hz, and remove outliers through preset data verification rules. For example, when the pressure value exceeds the range of 0 to 2 MPa or the change rate between adjacent sampling points exceeds 0.5 MPa per second, the corresponding data points are marked as invalid and removed, ensuring that subsequent analysis is based on reliable pressure valid data. In actual operation, the proportion of abnormal points is usually less than 0.1%.

[0031] S1022. Smooth the pressure valid data using median filtering in a sliding window. The window length is set to one-thousandth of the number of sampling points, that is, 1000 points. By traversing the data sequence and sorting the sampling values in each window and taking the median as the filtering result at that time point, this method can effectively suppress random noise while retaining the characteristics of pressure mutations. Compared with the original data, the smoothness of the filtered data can be improved by about 50%, laying a foundation for high-frequency feature extraction.

[0032] Through median filtering, the false fluctuations caused by external interference or sensor errors can be reduced, ensuring the authenticity of the pressure data.

[0033] S1023. Apply the wavelet decomposition algorithm to the filtered pressure data for three-layer scale analysis. Select the db4 wavelet basis function to decompose the data into low-frequency approximation coefficients and high-frequency detail coefficients. Among them, the high-frequency coefficients reflect the characteristics of rapid pressure changes. Calculate the mean and standard deviation of the data. For example, the mean is 0.6 MPa and the standard deviation is 0.05 MPa. Set the dynamic baseline threshold as the mean plus or minus three times the standard deviation, that is, 0.45 to 0.75 MPa. Based on this, generate normalized pressure data for further analysis of instantaneous fluctuations.

[0034] In the embodiment of the present invention, through the hierarchical analysis of the pressure signal by wavelet decomposition, the rapid change components caused by water hammer can be clearly separated from other low-frequency trends. The extraction of high-frequency coefficients provides an accurate basis for slope calculation.

[0035] S1024. Calculate the pressure change slope sequence between adjacent sampling points according to the normalized pressure data, and extract the set of peak points exceeding the dynamic baseline threshold. For example, when the pressure value exceeds 0.75 MPa and the rising rate is greater than 0.2 MPa per second, it is recorded as an over-limit point. At the same time, count the peak amplitude, duration, and pressure rising rate of each over-limit point. The peak duration of a typical water hammer event is usually between 100 and 300 milliseconds. This process aims to capture the significant characteristics of instantaneous high pressure.

[0036] S1025. Use the historical water hammer event data to construct a feature database and adopt a long short-term memory neural network for classification determination. The feature database records 100 groups of water hammer samples, including parameters such as the pressure peak amplitude is usually 1.5 to 2 times the normal pressure, the rising rate is between 0.3 and 0.8 MPa per second, and the median duration is 200 milliseconds. The neural network is configured with a two-layer structure, and the number of hidden layer nodes is 64. The input features include peak amplitude, duration, and rising rate, and the output is the probability of a water hammer event. When the probability exceeds 0.8, it is determined as a water hammer phenomenon, and the recognition accuracy can reach more than 95%, and the delay is less than 500 milliseconds.

[0037] In the embodiment of the present invention, the establishment of the water hammer feature database is based on the statistical analysis of long-term operation data, which can reflect the typical patterns of water hammer under different working conditions. The long short-term memory neural network enhances the adaptability to dynamic changes by memorizing the dependencies in the time series. For example, when the valve is quickly closed or the pump stops, the pressure may instantaneously rise to 3 times the normal value. If this mutation is not recognized in time, it may lead to pipeline rupture or equipment failure. Through the efficient processing of this step, the system can quickly trigger the response mechanism after an abnormality occurs.

[0038] It is worth mentioning that in the embodiments of the present invention, there are no strict limitations on the filtering window length or neural network parameters, and those skilled in the art can optimize and adjust them according to the actual pipe network scale and operating environment. In addition, the number of wavelet decomposition layers and the sample size of the feature database can also be expanded according to requirements to improve the determination accuracy.

[0039] Through the above steps, the time series analysis of pressure data can not only accurately capture the instantaneous high-pressure peak, but also combine intelligent classification technology to achieve rapid identification of water hammer phenomena, providing a reliable basis for dynamic control and ensuring the safe and stable operation of the steam pipe network.

[0040] S103. After determining that a water hammer phenomenon has occurred, analyze the vibration signal to extract the abnormal vibration frequency in the pipe network, and construct a propagation path in combination with the pressure data to locate the area where the impact source is located.

[0041] In the embodiments of the present invention, the positioning of the water hammer phenomenon relies on the collaborative analysis of vibration and pressure to ensure accurate tracing of the impact source from multi-source data. The steam pipe network collects dynamic signals in real time through sensors, and the analysis process aims to reveal abnormal features and determine their spatial distribution.

[0042] S1031. Obtain the original vibration data from the vibration sensors of the steam pipe network and perform filtering processing. Among them, a piezoelectric acceleration sensor is used to capture high-frequency vibration signals at a sampling frequency of 1000 Hz, with a measurement range of 0 to 100 m / s² and a sensitivity of 100 mV / (m / s²). A band-pass filter with a cut-off frequency of 10 to 200 Hz is used to remove low-frequency interference and high-frequency noise, and retain the core signal reflecting the mechanical vibration of the pipeline. Subsequently, the filtered data is converted into a vibration frequency spectrum using the fast Fourier transform, providing a basis for abnormal frequency extraction.

[0043] S1032. Perform anomaly detection on the vibration frequency spectrum and generate a frequency set. Among them, the background frequency distribution during the normal operation of the pipe network is statistically analyzed based on 30-day historical data, with an average value of 35 Hz and a standard deviation of 5 Hz. The anomaly determination threshold is set to 50 Hz, that is, the average value plus three times the standard deviation. When a point exceeding this threshold appears in the frequency spectrum, it is marked. For example, after detecting a high-frequency signal of 85 Hz, all the over-limit points are classified through a frequency clustering method, and their amplitudes and occurrence times are recorded to form an abnormal vibration frequency set to reflect the vibration characteristics caused by water hammer.

[0044] In the embodiments of the present invention, the application of band-pass filtering and frequency clustering can effectively isolate the instantaneous vibration caused by water hammer, avoid environmental noise or equipment operation interference, and ensure the accuracy of abnormal features. For example, the abnormal amplitude may be three times the normal value, indicating a strong mechanical impact.

[0045] S1033. Calculate the pressure wave propagation characteristics using the pressure sensor data and establish a path diagram. Among them, a piezoresistive sensor is used to collect pressure data at a sampling frequency of 100 Hz, with a measurement range of 0 to 2 MPa and an accuracy of 0.1%. Align the time bases of each measurement point through the pipe network topology information, calculate the time difference for the pressure wave peak to reach adjacent measurement points. For example, the propagation speed in a DN200 pipeline is about 1200 m / s. Combine the pipe section distance to generate a directed graph of the pressure propagation path. The nodes in the graph represent the measurement points, and the edges represent the propagation direction and time, which are used for subsequent traceability analysis.

[0046] S1034. Combine the abnormal vibration frequency set and the directed graph of the pressure propagation path for shock source localization. Among them, extract the degree, distance, and number of branches of the nodes in the path diagram as feature vectors, use a Bayesian network for probabilistic inference of the pressure time series data, and fuse the vibration amplitude distribution. For example, in the pipe network of a chemical plant, the 1st measurement point is connected to 4 branches, and the pressure wave propagation times to other measurement points are 8 ms, 15 ms, and 22 ms respectively, corresponding distances are 9.6 m, 18 m, and 26.4 m. The vibration frequency of 85 Hz is only significant at the 1st measurement point. Calculate the probability of each node being the shock source. The 1st measurement point scores 0.85, much higher than 0.45 of the 2nd measurement point and 0.32 of the 3rd measurement point, and determine that the shock source is located near the 1st measurement point.

[0047] In the embodiment of the present invention, the Bayesian network improves the positioning accuracy by integrating multi-dimensional data. For example, in a complex mesh pipe network, it includes 15 pressure measurement points and 12 vibration measurement points, with a total length of 3000 m and 25 pipe sections. When the pressure wave propagates from the 1st measurement point to the 2nd and 3rd measurement points, the differences in time and amplitude are used as the judgment basis. Locating the shock source helps to quickly take targeted measures to avoid systematic damage caused by the spread of water hammer.

[0048] It should be noted that the embodiment of the present invention does not strictly limit the filter cut-off frequency or the clustering method. Those skilled in the art can adjust the parameters according to the characteristics of the pipe network. For example, increase the sampling frequency to capture more subtle vibration changes. In addition, the pressure propagation speed varies with the pipe material and the medium state, and the calculation accuracy can be improved through field measurement and calibration.

[0049] Through the above steps, the embodiment of the present invention realizes a complete process from vibration frequency analysis to pressure path construction, can quickly locate the shock source after the occurrence of water hammer, provides accurate support for zonal control, and ensures the safety and stability of the steam pipe network operation.

[0050] S104. Process the sensor monitoring data in the shock source area through a multi-source data fusion algorithm, including pressure, vibration, and acoustic signals, to calculate the propagation speed and attenuation characteristics of the shock wave in the pipe section, so as to determine the shock source location and the affected area.

[0051] In the embodiments of the present invention, the fusion and analysis of multi-source data are the core steps for accurately locating the water hammer impact source, aiming to reveal the propagation law of shock waves by comprehensively processing different sensor signals. Multiple types of sensors in the steam pipe network work together to provide accurate basis for control.

[0052] S1041. Obtain the real-time monitoring data of pressure sensors, vibration sensors, and acoustic sensors from the steam pipe network and perform unified processing. Among them, the pressure sensor collects pressure signals at a sampling frequency of 100 Hz, the vibration sensor collects vibration signals at 1000 Hz, and the acoustic sensor collects acoustic signals at 500 Hz. All data are resampled to 100 Hz through timestamp alignment to ensure consistent time sequence. Subsequently, a sliding window mean filter with a length of 5 is used to denoise the resampled data, generating a filtered monitoring data sequence, effectively removing random noise and retaining the characteristics of shock waves.

[0053] S1042. Apply a Kalman filter to the filtered monitoring data sequence for state fusion. The state variables include pressure value, vibration amplitude, and acoustic signal intensity. The observation noise covariance matrix is determined according to the sensor accuracy. For example, the accuracy of the pressure sensor is 0.1%, the sensitivity of the vibration sensor is 100 mV / (m·s²), and the sensitivity of the acoustic sensor is 50 dB. The fusion result is optimized through the prediction and update stages to obtain the fused monitoring data, improving data reliability and consistency and laying a foundation for path analysis.

[0054] In the embodiments of the present invention, the Kalman filter reduces the measurement deviation between sensors through iterative estimation. For example, in a certain pipe section, the fused data can more accurately reflect the dynamic changes of shock waves, avoiding the interference of the instability of a single sensor signal on the analysis result.

[0055] S1043. Construct a propagation path weighted graph based on the pipe network topology and the fused monitoring data. Among them, a sensor position association matrix is generated according to the partition topology, and the matrix elements represent the pipe section distances between adjacent measurement points. For example, the distance between measurement point 1 and measurement point 2 is 120 m. Calculate the time difference between the arrival of the signal peak in the fused data at each measurement point. For example, the time difference between measurement point 1 and measurement point 2 is 100 ms, and the time difference between measurement point 2 and measurement point 3 is 80 ms. Combine the time difference matrix and calculate the shortest path from the impact source to each measurement point through the Dijkstra algorithm to obtain the actual propagation distance. For example, the distance from the source point to measurement point 1 is 50 m, to measurement point 2 is 170 m, and to measurement point 3 is 250 m.

[0056] S1044. Calculate the shock wave propagation speed based on the actual propagation distance and time difference and fit the attenuation characteristics. The propagation speed within the partition is obtained by dividing the distance by the time difference. For example, the speed in a certain section of the pipeline is 1200 meters per second. At the same time, the exponential function model is used to fit the peak attenuation curve. Taking the pressure data as an example, the initial peak is 2.5 MPa, it drops to 1.8 MPa after propagating 100 meters, 1.3 MPa after 200 meters, and 0.9 MPa after 300 meters. The fitted attenuation coefficient is 0.003 per meter, which reflects the energy loss law of the shock wave in the pipe network and provides a quantitative basis for positioning.

[0057] S1045. Use a deep neural network to process the propagation speed and attenuation parameters to determine the location of the impact source. The network input includes 9 features, namely the propagation speeds of 3 partitions, the initial peak value, and the attenuation coefficient. The network structure includes three hidden layers with the number of nodes being 64, 32, and 16 respectively. It is trained with 500 sets of historical water hammer event samples, and the three-dimensional space coordinates of the impact source are output. The distances between each measuring point and the impact source are calculated according to the coordinates. When the distance is less than the preset threshold of 300 meters, the corresponding partition is designated as the affected partition. For example, the 1st partition with a distance of 150 meters and the 2nd partition with a distance of 280 meters are included in the affected area, while the 3rd partition with a distance of 450 meters is not affected.

[0058] In the embodiment of the present invention, the deep neural network integrates multi-dimensional features through non-linear mapping and can meet the requirements of impact source positioning in complex pipe network environments. Compared with traditional methods, this method has significant advantages in terms of accuracy and robustness. For example, in a certain water hammer event, the coordinates output by the network can be accurate to the meter level, greatly improving the positioning efficiency.

[0059] It should be noted that the filter window length, the number of neural network layers, and the threshold parameters in the embodiment of the present invention can be flexibly adjusted according to the actual pipe network scale and operating conditions. Technical personnel can optimize these settings through experiments to adapt to different scenarios. In addition, the calculation of the propagation speed needs to consider the influence of pipe material and medium temperature, and further calibration can be carried out to improve the accuracy.

[0060] Through the above steps, the embodiment of the present invention realizes the effective fusion of multi-source data and the accurate calculation of shock wave characteristics. It can not only quickly locate the impact source but also clarify the affected partitions, providing reliable data support for subsequent pressure regulation and flow optimization, and ensuring the high efficiency and safety of the steam pipe network operation.

[0061] S105. Implement differential pressure grading control according to the affected partitions. Calculate the adjustment amount by analyzing the partition characteristics and optimize the valve command to achieve the effects of shock attenuation and pressure stability.

[0062] In the embodiments of the present invention, the goal of pressure control is to conduct zonal management for the influence range and intensity of water hammer shock waves to ensure the stable restoration of the pipeline network operation. Differential control combines multi-source data and intelligent algorithms to effectively cope with complex working condition changes.

[0063] S1051. Determine the pressure control priority according to the distance between the affected zone and the shock source and calculate the control index. Among them, the zones with a distance less than 300 meters are marked as the first-level priority, 300 to 500 meters as the second-level priority, and more than 500 meters as the third-level priority. Evaluate the control requirements of each zone through fuzzy rules. The input variables include distance weight, pressure overrun ratio, and influence area. For example, in the first-level zone, the distance is close, the pressure fluctuation exceeds the limit by 0.9 MPa, and the influence area accounts for more than 80%. Calculate the priority index of 0.9 and generate a regulation priority sequence to guide subsequent regulation.

[0064] S1052. Extract the real-time data of the affected zone from the pipeline network monitoring database and calculate the pressure adjustment amount, which includes pressure value, flow value, and valve opening. The pressure adjustment amount consists of three parts. First is the pressure deviation term, that is, the difference between the current pressure and the target value. For example, from 1.5 MPa to 0.6 MPa, the difference is 0.9 MPa. Second is the pressure change rate term, which reflects the pressure change speed, such as 0.2 MPa per second. Finally is the fluctuation amplitude term, which represents the fluctuation degree, such as 0.3 MPa. The three are combined to determine the initial adjustment target to ensure that the adjustment amount matches the actual demand.

[0065] Use an adaptive neural network to optimize the initial adjustment amount. The network input layer receives three types of parameters: pipe section flow, valve opening, and pressure volatility. For example, the flow is 120 cubic meters per hour, the valve opening is 65%, and the volatility is 0.33. The number of hidden layer nodes is dynamically adjusted from the initial 16 to 24. Through non-linear mapping, the optimized adjustment amount is output as 0.4 MPa. This method can adjust parameters according to real-time working conditions to avoid over-adjustment or under-adjustment problems under fixed rules.

[0066] S1053. Analyze the pressure adjustment effect based on the hydraulic response characteristic curve of the pipeline network and generate valve commands. Among them, establish a hydraulic response curve through experimental data to reveal the time lag relationship between valve action and pressure change. For example, when the valve opening is adjusted from 65% to 45%, the pressure starts to drop after 2 seconds and stabilizes after 5 seconds. Based on this, set the adjustment step size to 5% and the interval to 6 seconds. At the same time, calculate the mutual feedback influence coefficient between adjacent zones. For example, the influence coefficient of the first-level zone adjustment on the second-level zone is 0.3. Use a time series attenuation function to correct the step size. The first-level zone remains 5%, and the second-level zone is adjusted to 3.5% to generate a valve adjustment command sequence under the coupling effect.

[0067] In the embodiments of the present invention, the introduction of the hydraulic response curve makes the regulation process more scientific. For example, in the pipe network of a certain chemical plant, the amplitude of pressure fluctuation in the primary partition is as high as 1.2 MPa, and the secondary partition is less affected by the coupling effect. Through step-by-step regulation and real-time correction, the pressure fluctuation rapidly decays to the normal range, avoiding the occurrence of secondary impact.

[0068] The control effect is verified by monitoring the pressure regulation process. An exponential sliding window of 60 seconds is used to calculate the pressure mean and standard deviation of each partition. The upper limit of the standard deviation is set at 0.1 MPa, and the target pressure range is 0.5 to 0.7 MPa. When the standard deviation is lower than the upper limit and the mean is stable at 0.6 MPa for 5 consecutive cycles, it indicates that the regulation meets the expectation and the pipe network resumes stable operation.

[0069] It should be noted that the priority division, regulation step size, and neural network structure in the embodiments of the present invention can be optimized according to the pipe network scale and operating environment. For example, in the case of larger flow rate or different pipe materials, input features can be increased or the parameters of the attenuation function can be adjusted to improve the control accuracy and response speed.

[0070] Through the above steps, the embodiments of the present invention achieve pressure control based on partition differentiation, not only rapidly attenuating the water hammer impact, but also ensuring the smooth transition of pressure through intelligent optimization and mutual feedback analysis, providing technical support for the long-term stable operation of the steam pipe network.

[0071] S106. Obtain and analyze the steam flow rate fluctuation of the affected partition, and determine the target flow rate and priority by combining feature extraction and intelligent prediction with the pressure regulation result, and adjust the opening degree of the regulating valve to optimize the flow rate distribution.

[0072] In the embodiments of the present invention, flow rate regulation is an important link to ensure the stable operation of the steam pipe network, and precise control is achieved through the analysis and prediction of fluctuation characteristics. The processing of steam flow rate data is closely combined with pressure regulation to meet the requirements of different partitions.

[0073] S1061. Collect the original flow rate data from the flow meter of the affected partition and perform denoising processing. Among them, a vortex street flow meter is used to monitor the flow rate at a sampling frequency of 100 Hz, with a measurement range of 0 to 200 cubic meters per hour and an accuracy of 1%. Abnormal values are removed through validity verification. For example, when the flow rate exceeds the range or the change rate of adjacent sampling points exceeds 50%, the abnormal points are removed. The abnormal proportion is usually lower than 0.5%. Subsequently, the data is decomposed into three layers using the db4 wavelet basis function to separate the low-frequency trend and high-frequency noise, and the noise is suppressed by the threshold method to generate smooth flow rate data to improve the analysis accuracy.

[0074] S1062. Extract the fluctuation characteristics from the smoothed flow data and calculate the key parameters. Among them, calculate the flow mean sequence and the standard deviation sequence with a 60 - second sliding window. For example, the mean is 150 cubic meters per hour and the standard deviation is 30 cubic meters per hour. The points where the standard deviation exceeds 20% of the mean are marked as significant fluctuation points. Calculate the flow fluctuation amplitude and duration for these points. For example, the amplitude reaches 30 cubic meters per hour and the duration is 20 seconds. At the same time, record the pressure regulation change amount in the corresponding period to provide a basis for priority evaluation.

[0075] Adopt fuzzy inference rules to evaluate the priority of flow control in each zone. The rules take the flow fluctuation amplitude, duration, and pressure regulation amount as input variables. The amplitude is divided into three levels: low, medium, and high. Below 10% is low, 10% to 30% is medium, and above 30% is high. Combine other variables to calculate the priority index. For example, for Zone 1, due to high amplitude and long duration, the index reaches 0.85, and for Zone 2, it is 0.65. The zones with higher priority need to be adjusted first to quickly stabilize the flow.

[0076] S1063. Sort according to the priority and predict the flow trend to determine the adjustment target. Among them, sort each zone according to the priority index, and use the support vector regression model to predict the flow change. The model is based on the radial basis kernel function, inputting the historical flow data of the previous 10 minutes, the fluctuation frequency of 0.5 Hz, and the amplitude of 30%. Optimize the parameters through cross - validation. The prediction result shows that the flow will drop to 120 cubic meters per hour within 30 minutes. Combine the real - time pressure distribution data. For example, the average pressure in Zone 1 is 0.55 MPa and the fluctuation range is 0.05 MPa. Set the target flow range to 110 to 130 cubic meters per hour and generate an adjustment index of 0.8 to guide the valve adjustment.

[0077] In the embodiment of the present invention, the support vector regression can capture the time - dependence of the flow and can predict the change trend in advance. For example, after a water hammer, the flow may fluctuate violently due to a sudden pressure change. The prediction result helps the system take measures before the fluctuation intensifies to avoid steam supply imbalance.

[0078] S1064. Calculate the valve opening change based on the pipe network hydraulic characteristic curve and implement the adjustment. Among them, the hydraulic characteristic curve shows that when the valve opening is adjusted from 80% to 60%, the flow drops from 150 cubic meters per hour to 120 cubic meters per hour, and the response delay is 3 seconds. Set the adjustment step size to 5% and an interval of 4 seconds to ensure a smooth transition. Gradually adjust the valve through hierarchical control, and monitor the flow mean and standard deviation in real - time. When the mean reaches 125 cubic meters per hour, the standard deviation drops to 12 cubic meters per hour and is lower than 10% of the mean, and remains stable for 5 consecutive 10 - second cycles, it is determined that the adjustment is completed.

[0079] It is worth mentioning that the number of wavelet decomposition layers, window length, and prediction model parameters in the embodiments of the present invention can be adjusted according to the actual flow characteristics. For example, in high-load scenarios, the number of decomposition layers can be increased to more finely separate the noise, or the window can be extended to capture long-term trends, thereby improving the control effect.

[0080] Through the above steps, the embodiments of the present invention achieve precise analysis and regulation of flow fluctuations, not only quickly responding to abnormal changes after water hammer, but also optimizing steam distribution through priority management and trend prediction, ensuring the high efficiency and reliability of the pipe network operation.

[0081] S107. Real-time monitor the pipeline deformation data after the opening of the regulating valve is adjusted, identify the local deformation area and its degree by analyzing the deformation characteristics, and reduce the pressure load in the corresponding area when the deformation amount exceeds the standard to ensure the safety of the pipeline.

[0082] In the embodiments of the present invention, pipeline deformation monitoring is a key link to ensure the long-term stable operation of the steam pipe network, especially the local stress concentration that may be caused after a water hammer impact. Through multi-level data processing and intelligent control, it can quickly respond to anomalies and take protective measures.

[0083] S1071. Obtain the original deformation data from the deformation sensors of the steam pipe network and perform preprocessing. Among them, a resistive strain gauge sensor is used, with a measuring point arranged every 50 meters, covering a total of 80 measuring points on a 4000-meter pipeline. The sensitivity of the sensor is 2 micrometers per volt, the measurement range is 0 to 5 millimeters, and the data is collected at a sampling frequency of 10 hertz. Pulse interference is removed through median filtering with a window length of 5 points. Subsequently, the deformation mean and standard deviation sequences are calculated with a 60-second sliding window. For example, the mean is 0.8 millimeters and the standard deviation is 0.2 millimeters, providing a reliable data basis for anomaly detection.

[0084] S1072. Set the deformation warning threshold based on the pipeline material characteristics and perform spatial analysis. The pipeline material is 20# steel, with an elastic modulus of 210 GPa and a yield strength of 235 MPa. The warning threshold is set to five-thousandths of the pipeline outer diameter, that is, 2.5 millimeters. When the deformation amount exceeds this value, a warning is triggered. The density-based clustering algorithm is used to perform spatial analysis on the filtered data, and the clustering radius is set to 100 meters to identify the local deformation area. For example, an anomaly is detected within the range of measuring points 3 to 8, the center is located at measuring point 5, and the maximum deformation amount is 3.2 millimeters, exceeding the threshold by 28%. Record the area coordinates and deformation amount for further evaluation.

[0085] S1073. Evaluate the stress state of the deformed area in combination with the pipeline stress-strain curve and determine the risk. Among them, obtain the stress-strain curve of 20# steel through the material experiment database. The curve shows that when the strain is 0.2%, the corresponding yield stress is 235 MPa. The measured deformation of 3.2 mm is converted into a strain of 0.16%, and the corresponding stress is 188 MPa, reaching 80% of the yield strength, which is determined to be a dangerous state. This process clarifies the pipeline damage risk by quantifying the stress distribution and triggers the pressure regulation requirement to prevent permanent deformation or rupture.

[0086] In the embodiment of the present invention, the stress-strain analysis provides a scientific basis for pressure regulation. For example, the stress state close to the yield strength may lead to pipeline fatigue accumulation, and timely reducing the pressure load can effectively extend the pipeline life.

[0087] S1074. Calculate the pressure regulation amount for the dangerous state area and generate a valve control sequence. Among them, use the steepest descent algorithm based on the pressure gradient. Input the current pressure of 0.8 MPa, the deformation of 3.2 mm, and the stress state of 188 MPa. After 5 iterations, determine the target pressure to be 0.6 MPa, and the corresponding valve opening is adjusted from 75% to 55%. Subsequently, optimize the adjustment step size through a three-layer neural network controller. The number of hidden layer nodes in the network is 32, 16, and 8. Input the pressure, deformation, and stress, and output the opening sequence of 20 steps. Each step is adjusted by 5%, with an interval of 10 seconds to ensure smooth adjustment and avoid secondary impact.

[0088] Verify the control effectiveness by real-time monitoring the adjustment effect. After the pressure is reduced, the deformation gradually drops to 2 mm, and the corresponding stress is 150 MPa, which is lower than the safety threshold and lasts for 300 seconds, indicating that the pipeline has returned to the safe operating state. This method not only prevents pipeline damage but also optimizes the operating efficiency of the system through accurate identification and rapid response.

[0089] It should be noted that the clustering radius, filtering window, and neural network structure in the embodiment of the present invention can be adjusted according to the pipeline material and operating environment. For example, under high-temperature or high-pressure working conditions, the measurement point density can be increased or the sampling interval can be shortened to improve the monitoring sensitivity, thereby further improving the control accuracy. Through the above steps, the embodiment of the present invention realizes the closed-loop control from deformation monitoring to pressure regulation, fully combines the pipeline mechanical characteristics and intelligent algorithms, and ensures the safety and stability of the steam pipe network under abnormal working conditions.

[0090] S108. Continuously monitor the real-time pressure and flow readings of the steam pipe network, and dynamically adjust the set values of each pressure regulation point and the opening of the regulating valve through data processing and intelligent algorithms to ensure the stability and efficiency of steam delivery.

[0091] In the embodiments of the present invention, the real-time monitoring and dynamic regulation of the pipe network are key steps to maintain the system balance. Hierarchical management is carried out according to the fluctuation characteristics of pressure and flow to optimize the overall operation state.

[0092] S1081. Collect pressure and flow data from each pressure regulation point and perform filtering processing. Among them, a vortex flowmeter and a pressure transmitter are used to obtain real-time data at a sampling frequency of 100 Hz. High-frequency interferences such as mechanical vibrations are removed through a digital low-pass filter with a cut-off frequency of 10 Hz. Subsequently, a 60-second sliding window is used to calculate the volatility of pressure and flow. For example, the pressure volatility is 7% and the flow volatility is 12%, generating a fluctuation characteristic sequence to reflect the dynamic changes of the system and providing a reliable basis for analysis.

[0093] S1082. Divide the pressure levels according to the pipe network structure and calculate the fluctuation deviation. From the source point to the end point, it is divided into a high-pressure level, a medium-pressure level, and a low-pressure level in sequence. The operating pressure of the high-pressure level is 1.2 MPa and the rated flow is 200 cubic meters per hour. The medium-pressure level is 0.8 MPa and 150 cubic meters per hour. The low-pressure level is 0.4 MPa and 100 cubic meters per hour. The measured pressure deviation of the high-pressure level reaches 0.2 MPa, exceeding the preset threshold of 0.1 MPa, generating a fluctuation deviation sequence for each level and clarifying the key areas that need to be adjusted.

[0094] For the outlier points in the fluctuation deviation sequence, a dynamic programming algorithm is used to calculate the pressure regulation target value. The algorithm comprehensively evaluates the fluctuation deviation term, the regulation amount term, and the response time term through the objective function, and the weights are set to 0.5, 0.3, and 0.2 respectively. The regulation path is optimized under the constraints of the pressure limit and the step size. For example, the calculated target pressure of the high-pressure level is 1.1 MPa and the medium-pressure level is 0.7 MPa, ensuring that the regulation is both fast and stable.

[0095] S1083. Train a deep reinforcement learning network based on the regulation target value and output a control sequence. Among them, the network adopts a four-layer structure. The input layer contains 12 state variables, including the pressure value, flow value, volatility, and current regulation amount of each pressure level. The number of hidden layer nodes is 64, 32, and 16. Six regulation actions are output through non-linear mapping, corresponding to the opening changes of the valves at each level. The trainer updates the parameters through online learning. The reward function is composed of 40% of the pressure deviation term, 40% of the flow deviation term, and 20% of the volatility term. For example, when the pressure of the high-pressure level drops to 1.15 MPa and the volatility decreases to 4%, a positive reward of 0.8 is obtained, otherwise a negative feedback is given.

[0096] In the embodiments of the present invention, deep reinforcement learning continuously optimizes the control strategy by simulating the response of the pipe network. For example, when improper regulation causes the volatility to rise, the network will adjust the action direction to avoid a chain reaction and ensure that the system gradually tends to be stable.

[0097] S1084. Implement control according to the regulating valve opening action sequence and verify the effect. The high-pressure stage valve opening is adjusted from 80% to 65%, the medium-pressure stage is adjusted from 75% to 65%, and the low-pressure stage is maintained at 70%. After 300 seconds of adjustment, the high-pressure stage pressure fluctuation rate is reduced to 3%, and the flow fluctuation rate is reduced to 8%. The fluctuation rate during the continuous monitoring period is lower than the target values of 5% and 10%, indicating that the control has achieved the expected effect, steam transportation has returned to stability, the pressure fluctuation amplitude has been reduced by 50%, the flow fluctuation has been reduced by 40%, and the adjustment time has been shortened by about 30%.

[0098] It is worth mentioning that the filter frequency, window length and network parameters in the embodiment of the present invention can be flexibly adjusted according to the scale of the pipe network and the operating conditions. For example, when the flow demand changes drastically, the window can be shortened to improve the response speed, or the hidden layer nodes can be increased to improve the model complexity, thereby further optimizing the control accuracy and efficiency.

[0099] Through the above steps, the embodiment of the present invention realizes multi-level dynamic regulation of pressure and flow, which not only effectively suppresses local fluctuations, but also enhances the adaptability of the system through intelligent algorithms, ensuring the reliable operation of the steam pipeline network under complex working conditions.

[0100] like Figure 2 The present invention provides a steam delivery control system based on multi-stage pressure regulation and dynamic feedback, which mainly includes: Data acquisition module, used to obtain real-time data from multi-source sensors in the steam network, including pressure, flow and vibration signals, to form a multi-dimensional data set; The pressure analysis module is used to process the real-time pressure data in the data set, obtain the pressure time series data, determine the pressure change trend, and judge whether an instantaneous high pressure peak value occurs. If the high pressure peak value exceeds the preset peak value threshold, it is determined that water hammer phenomenon occurs; The vibration analysis module is used to analyze the vibration signal after determining the occurrence of water hammer phenomenon, obtain the abnormal vibration frequency in the pipe network, and determine the propagation path of the shock wave in combination with the pressure data to locate the area where the shock source is located; The shock source positioning module is used to process the monitoring data of each sensor in the area where the shock source is located through a multi-source data fusion algorithm, calculate the propagation speed and attenuation characteristics of the shock wave in different pipeline partitions, and obtain the shock source location and the affected partition; A pressure control module, used for performing zone-differentiated pressure graded control according to the affected zones; The flow control module is used to obtain and analyze the steam flow fluctuation of each affected zone, analyze the steam flow fluctuation of each zone, determine the target flow and priority, and adjust the opening of the control valve; The deformation monitoring module is used to monitor the pipeline deformation data in real time after adjusting the opening of the regulating valve, identify the local deformation area and the degree of deformation. If the deformation amount exceeds the preset safety value, the pressure load in the local deformation area is reduced. The dynamic regulation module is used to continuously monitor the real-time pressure and flow readings of the pipeline network, and dynamically adjust the settings of the pressure regulation points at all levels and the corresponding opening of the regulating valves to stably control the steam transmission.

[0101] Only some preferred embodiments of the present invention are listed above, but the present invention is not limited thereto, and many improvements and transformations can be made. As long as the improvements and transformations are made based on the basic principles of the present invention, they shall be regarded as falling within the protection scope of the present invention.

Claims

1. A steam delivery control method based on multi-level pressure regulation and dynamic feedback, characterized in that, The method includes: Obtaining real-time data of multi-source sensors in the steam pipe network to form a data set; Processing the real-time pressure data in the data set to obtain pressure time-series data, determining the pressure change trend, and judging whether an instantaneous high-pressure peak appears. If the high-pressure peak exceeds the preset peak threshold, it is determined that a water hammer phenomenon occurs; After determining that a water hammer phenomenon occurs, analyzing the vibration signal to obtain the abnormal vibration frequency in the pipe network, combining with the pressure data, determining the propagation path of the shock wave, and locating the area where the impact source is located; Processing the monitoring data in the area where the impact source is located through a multi-source data fusion algorithm, calculating the propagation speed and attenuation characteristics of the shock wave in different pipe partitions, and obtaining the impact source position and the affected partitions; Conducting partition-differentiated pressure classification control according to the affected partitions; Obtaining and analyzing the steam flow fluctuations in each affected partition, analyzing the steam flow fluctuations in each partition, determining the target flow rate and priority, and adjusting the opening degree of the regulating valve; Real-time monitoring the pipe deformation data after adjusting the opening degree of the regulating valve, identifying the local deformation area and the degree of deformation. If the deformation amount exceeds the preset safety value, reducing the pressure load in the local deformation area; Continuously monitoring the real-time pressure and flow readings of the pipe network, and dynamically adjusting the settings of each pressure regulation point and the corresponding opening degree of the regulating valve.

2. The method according to claim 1, wherein The obtaining of the real-time data of multi-source sensors in the steam pipe network to form a data set includes: Obtaining pressure sensor data, flow sensor data, and vibration sensor data from the multi-source sensors in the steam pipe network, and performing timestamp marking on the sensor data through a data collector to obtain normalized sensor data; Repairing the missing points according to the normalized sensor data using a linear interpolation algorithm, and performing noise reduction processing on the repaired data through a sliding window mean filter to obtain smooth sensor data; Storing the data in segments according to the time period for the smooth sensor data to form a data set.

3. The method according to claim 1, wherein The processing of the real-time pressure data in the data set to obtain pressure time-series data, determining the pressure change trend, and judging whether an instantaneous high-pressure peak appears. If the high-pressure peak exceeds the preset peak threshold, it is determined that a water hammer phenomenon occurs, includes: Obtaining a pressure sampling data sequence from the pipe network pressure sensor, and removing outliers from the pressure sampling data through the data verification rule to obtain valid pressure data; Performing median filtering operation on the valid pressure data using a sliding window, and performing wavelet decomposition on the data after median filtering to obtain high-frequency coefficient data; Calculating a pressure change slope sequence for the high-frequency coefficient data, extracting a set of pressure peak points, marking the points that exceed the dynamic baseline threshold in the peak point sequence, and calculating the amplitude, duration, and pressure rise rate of each over-limit peak point; Using a long short-term memory neural network to perform feature classification on the peak amplitude and pressure rise rate, and determining the water hammer phenomenon according to the matching result of the feature classification and the preset water hammer feature database.

4. The method according to claim 1, wherein After determining that a water hammer phenomenon occurs, analyzing the vibration signal to obtain the abnormal vibration frequency in the pipe network, combining with the pressure data, determining the propagation path of the shock wave, and locating the area where the impact source is located, includes: The original vibration data collected by the vibration sensor is filtered by a bandpass filter, and the spectrum of the filtered vibration data is analyzed by fast Fourier transform to obtain the vibration frequency spectrum; According to the comparison between the vibration frequency spectrum and a preset background frequency upper limit threshold, the points exceeding the threshold in the spectrum are marked and frequency clustered to obtain an abnormal vibration frequency set; Acquire pressure data from the pressure sensor, calculate the time difference between the pressure wave peaks reaching adjacent sensors, determine the pressure wave propagation speed according to the pipe section distance and the time difference, and establish a directed graph of the pressure propagation path; For the abnormal vibration frequency set and the directed graph of the pressure propagation path, a Bayesian network is used to perform source tracing analysis on the pressure propagation time series data, calculate the probability of a node being a shock source, and determine the shock source location.

5. The method according to claim 1, wherein The monitoring data of the area where the shock source is located is processed by the multi-source data fusion algorithm, the propagation speed and attenuation characteristics of the shock wave in different pipeline partitions are calculated, and the shock source position and the affected partition are obtained, including: Acquire sampling data from pressure sensors, vibration sensors, and acoustic sensors, resample the sampled data according to the sensor acquisition timestamp, and obtain a filtered monitoring data sequence through sliding window mean filtering; According to the filtered monitoring data sequence, a Kalman filter is used to perform state fusion on the pressure value, the vibration amplitude and the acoustic signal intensity to obtain fused monitoring data; According to the fused monitoring data, a sensor position association matrix and a time difference matrix are used to construct a propagation path weighted graph, and the actual propagation distance is calculated by the Dijkstra algorithm; The propagation velocity and the peak attenuation coefficient are calculated according to the actual propagation distance, and the propagation velocity and the peak attenuation coefficient are processed by a deep neural network to output the impact source position, and the distance from each pipe network partition measuring point to the impact source is calculated. When the measuring point distance is less than the preset distance threshold, the partition where the measuring point is located is classified into the affected partition set.

6. The method according to claim 1, wherein The pressure grading control based on the affected zones is differentiated by zones, including: Obtaining a pressure control priority sequence according to the distance value between the affected partition and the impact source, and calculating the control index value of the partition using fuzzy rules; Calculate the partition pressure adjustment amount according to the control index value, and the pressure adjustment amount is obtained by three parameters: pressure deviation term, pressure change rate term and fluctuation amplitude term; Adopting an adaptive neural network to optimize the pressure regulation amount, the adaptive neural network receives three types of input parameters: pipe flow, valve opening and pressure fluctuation rate, and outputs the optimized regulation amount through an adaptive node number structure; A hydraulic response characteristic curve of the pipe network is established according to the optimized adjustment amount, and the mutual feedback influence coefficient of the pressure adjustment of adjacent partitions is calculated through the hydraulic response characteristic curve to generate a valve adjustment instruction sequence.

7. The method according to claim 1, characterized in that, The obtaining and analyzing the fluctuation of steam flow in each affected zone, analyzing the fluctuation of steam flow in each zone, determining the target flow and priority, and adjusting the opening of the regulating valve include: The original flow data is decomposed and denoised by wavelet transform, and the flow fluctuation feature sequence is calculated by using a sliding window. Judgment is made according to the ratio of the standard deviation sequence to the mean sequence in the flow fluctuation feature sequence. If the value of the standard deviation sequence exceeds the value of the mean sequence, the flow fluctuation amplitude and the fluctuation duration are calculated; The fuzzy inference rule is used to process the flow fluctuation amplitude and the fluctuation duration to obtain the partition flow control priority index; According to the partition flow control priority index, each partition is sorted, and the support vector regression is used to predict the flow change trend, and the target flow regulation interval is determined through the flow change trend and the pressure distribution state.

8. The method according to claim 1, wherein The pipeline deformation data after adjusting the opening degree of the regulating valve is monitored in real time, and the local deformation area and the deformation degree are identified. If the deformation amount exceeds the preset safety value, the pressure load of the local deformation area is reduced, including: The original deformation data collected by the pipeline deformation sensor is obtained, and the filtered deformation data is obtained by using the median filtering algorithm according to the original deformation data, and the deformation mean and the standard deviation sequence are calculated by using a sliding window; The density-based clustering algorithm is used to perform spatial clustering processing on the filtered deformation data, and the position coordinates and the deformation amount of the local deformation area are identified according to the spatial clustering processing result; The stress state value is calculated according to the deformation amount of the local deformation area combined with the stress-strain curve. If the stress state value exceeds the preset yield strength threshold, it is determined that the dangerous state is entered; For the dangerous state area, the pressure adjustment amount is calculated by using the steepest descent algorithm based on the pressure gradient, and the opening degree sequence of the regulating valve is output through the neural network controller according to the pressure adjustment amount.

9. The method according to claim 1, characterized in that, The real-time pressure and flow readings of the pipe network are continuously monitored, and the settings of the pressure adjustment points at all levels and the corresponding opening degrees of the regulating valves are dynamically adjusted, including: The pressure and flow data collected at the pressure adjustment point are filtered, and the pressure-flow volatility sequence is calculated by using a sliding window according to the filtered data; According to the pressure-flow volatility sequence, three-level fluctuation deviation sequences of high pressure level, medium pressure level and low pressure level are sequentially divided from the source point to the end point according to the pipe network structure; For the fluctuation deviation sequence, the pressure adjustment target value is calculated by using the dynamic programming algorithm; The deep reinforcement learning network is trained according to the pressure adjustment target value, and the deep reinforcement learning network uses a four-layer network structure to output the opening degree action sequence of the regulating valve.

10. A steam delivery control system based on multi - level pressure regulation and dynamic feedback, characterized in that, The system includes: A data acquisition module for obtaining real-time data of multi-source sensors in the steam pipe network, including pressure, flow and vibration signals, and forming a data set; A pressure analysis module for processing the real-time pressure data in the data set to obtain pressure time series data, determining the pressure change trend, and judging whether an instantaneous high-pressure peak appears. If the high-pressure peak exceeds the preset peak threshold, it is determined that a water hammer phenomenon occurs; A vibration analysis module for analyzing the vibration signal after determining that a water hammer phenomenon occurs, obtaining the abnormal vibration frequency in the pipe network, combining the pressure data, determining the propagation path of the shock wave, and positioning the area where the shock source is located; The impact source positioning module is used to process the monitoring data of each sensor in the area where the impact source is located through a multi-source data fusion algorithm, calculate the propagation speed and attenuation characteristics of the shock wave in different pipeline partitions, and obtain the impact source position and the affected partitions; The pressure control module is used to perform pressure grading control with partition differentiation according to the affected partitions; The flow rate regulation module is used to obtain and analyze the steam flow rate fluctuations in each affected partition, analyze the steam flow rate fluctuations in each partition, determine the target flow rate and priority, and adjust the opening degree of the regulating valve; The deformation monitoring module is used to continuously monitor the pipeline deformation data after adjusting the opening degree of the regulating valve, identify the local deformation area and the degree of deformation. If the deformation amount exceeds the preset safety value, the pressure load in the local deformation area is reduced; The dynamic regulation module is used to continuously monitor the real-time pressure and flow rate readings of the pipe network, and dynamically adjust the settings of each level of pressure regulation points and the corresponding opening degrees of the regulating valves to stably control steam transportation.

Images