Distributed liquid terminal and centralized liquid supply source cooperative connection control method and system

Abstract

This application relates to the field of fluid pipeline control technology, specifically a method and system for the coordinated connection control of distributed liquid-using terminals and centralized liquid supply sources. The method includes: acquiring the pressure of the main supply pipe, the pressure of each terminal branch pipe, and pipeline vibration spectrum data to establish a dynamic monitoring dataset; extracting multi-dimensional features such as pressure fluctuation amplitude, propagation phase difference, and vibration spectrum peak value to identify the disturbance contribution of each terminal and generate a pressure disturbance weight coefficient vector; constructing a pressure coupling relationship graph based on this, inputting it into a graph attention neural network to output the optimal flow limit value for each terminal; designing a time-sharing peak-shifting scheduling strategy based on the limit value, and using a proportional reduction algorithm to limit the global total flow; monitoring pressure and vibration indicators in real time during execution, triggering rapid network retraining when limits are exceeded, and dynamically updating parameters until the fluctuations converge. This application achieves accurate flow limiting and adaptive coordinated scheduling under complex disturbances of multiple terminals, effectively ensuring the stability of the pipeline liquid supply.

Description

Technical Field

[0001] This application relates to the field of fluid pipeline control technology, specifically to a method and system for the coordinated connection and control of distributed liquid consumption terminals and centralized liquid supply sources. Background Technology

[0002] With the development of industrial and experimental fluid supply technology, pipeline systems that connect distributed fluid terminals and centralized fluid supply sources in parallel are widely used. In this multi-terminal parallel operation, the fluid extraction behavior of each terminal will propagate complex pressure pulses in the reverse direction through the pipeline medium to the centralized fluid supply source and other branches, accompanied by flow-induced vibrations in the pipeline.

[0003] Existing control technologies struggle to accurately isolate and quantify the implicit interference weight of a single terminal on the global pressure field when facing dynamic disturbances involving multi-node spatial coupling. This results in the system's inability to proactively formulate coordinated flow limiting and peak-shaving scheduling strategies that can both suppress global pressure surges and pipeline vibrations while accommodating the dynamic demands of each terminal when facing concurrent extraction demands from multiple terminals. Furthermore, when the pipeline network's operational status experiences dynamic drift, or pressure or vibration indicators exceed safety boundaries, existing methods are prone to falling into a state of continuous oscillation and over-limit due to the solidification of control parameters, failing to achieve adaptive and rapid convergence of the control strategy.

[0004] In view of this, this application proposes a method and system for coordinated connection and control between distributed liquid consumption terminals and centralized liquid supply sources. Summary of the Invention

[0005] To achieve the above objectives, this application provides a method and system for coordinated connection and control of distributed liquid consumption terminals and centralized liquid supply sources, the specific technical solution of which is as follows:

[0006] In a first aspect, this application provides a method for coordinated connection and control of distributed liquid consumption terminals and centralized liquid supply sources, including:

[0007] Acquire pressure data of the main outlet pipe of the centralized liquid supply source, pressure data of the inlet branch pipes of each distributed liquid consumption terminal, and vibration spectrum data of the connecting pipeline nodes to establish a dynamic monitoring dataset of pipeline pressure under multi-terminal parallel operation.

[0008] The pressure fluctuation amplitude characteristics, pressure propagation phase difference characteristics, and vibration spectrum peak characteristics are extracted from the pipeline pressure dynamic monitoring dataset. The multidimensional feature correlation analysis method is used to identify the disturbance contribution of each terminal to the pressure field of the liquid supply source when extracting fluid, and to generate the pressure interference weight coefficient vector of each terminal.

[0009] Based on the pressure interference weight coefficient vector, a pressure coupling relationship map between the terminal group and the centralized liquid supply source is constructed. The pressure coupling relationship map is input into the graph attention neural network for node feature aggregation and edge weight learning, and the optimal flow limit value of each terminal and liquid supply source connection branch is output at the current time.

[0010] Based on the optimal flow limit value, a time-sharing peak scheduling strategy is designed to stagger the fluid extraction time periods of each terminal. For terminals with overlapping extraction time periods, a proportional reduction algorithm is used to reduce the instantaneous extraction flow rate, and the total flow rate at the outlet of the centralized liquid supply source is limited to not exceeding the stable supply threshold at any time.

[0011] During the execution of the time-sharing peak-shaving scheduling strategy, the pressure stability index of the centralized liquid supply source and the pipeline vibration intensity index are monitored. When the index exceeds the safety boundary, the graph attention neural network is triggered to quickly retrain and dynamically update the collaborative connection scheduling parameters until the pressure fluctuation converges to the allowable range.

[0012] Preferably, a pressure transmitter is installed at the outlet main pipe of the centralized liquid supply source, and a pressure sensor is installed at the inlet branch pipe of each distributed liquid consumption terminal. Each pressure sensor and the pressure transmitter at the outlet main pipe of the liquid supply source are synchronously collected using the same sampling clock.

[0013] A piezoelectric accelerometer is installed at the node connecting the pipeline to collect vibration signals, and the vibration time-domain signal is subjected to a fast Fourier transform to obtain vibration spectrum data.

[0014] The pipeline pressure dynamic monitoring dataset uses the sampling timestamp as the primary index key and includes a pressure value field and a vibration spectrum vector field. It is associated with and stores the opening status of fluid extraction valves at each terminal, the start / stop status of circulating pumps, and the fluid load demand level. The collected data is validated, including range and rate of change validation for pressure data and spectral energy and fundamental frequency offset validation for vibration spectrum data. Abnormal data is compensated and filled using interpolation of adjacent valid data points.

[0015] Preferably, the pressure fluctuation amplitude characteristics are obtained by calculating the difference between the maximum and minimum pressure values ​​within the window after processing through a sliding window segmentation, and the pressure standard deviation is calculated as an auxiliary measure of the severity of pressure fluctuations.

[0016] The pressure propagation phase difference characteristics are calculated by cross-correlation analysis to determine the propagation delay between the pressure time series of the central liquid supply source outlet main pipe and the pressure time series of each terminal inlet branch pipe. The propagation delay is then combined with the main frequency of pressure fluctuation and converted into a phase difference representation.

[0017] The main frequency of pressure fluctuation is obtained by performing a fast Fourier transform on the pressure time series of the centralized liquid supply source outlet main pipe and selecting the frequency component with the largest amplitude.

[0018] Preferably, the peak features of the vibration spectrum include peak frequency and peak amplitude. Local peaks with amplitudes higher than adjacent frequency points in the vibration spectrum are identified by peak detection, and a preset number of peaks are selected in descending order of peak amplitude to form a multi-peak feature vector.

[0019] The pressure fluctuation amplitude, pressure propagation phase difference, and vibration spectrum peak amplitude are combined to form a multidimensional feature vector for each terminal. Multivariate regression analysis is used to determine the interference of each terminal feature on the pressure fluctuation of the centralized liquid supply source.

[0020] Partial correlation analysis was used to eliminate the mutual interference between terminals, the perturbation correlation strength of each terminal was calculated, and the partial correlation coefficient and standardized regression coefficient were weighted and combined to obtain the perturbation contribution score.

[0021] The disturbance contribution score is normalized to generate a pressure disturbance weight coefficient vector, and a moving average fusion strategy is used for dynamic updating.

[0022] Preferably, the pressure coupling relationship diagram adopts a directed weighted graph structure, with the centralized liquid supply source as the central node and each distributed liquid consumption terminal as the peripheral node. The direction of the directed edge from the liquid supply source node to each terminal node indicates the direction of fluid flow, and the coupling relationship between each terminal node is represented by the coupling association edge.

[0023] The feature vector of the centralized liquid supply source node includes the outlet main pipe pressure value, pressure fluctuation amplitude, fluid level height, and supply capacity margin. The feature vector of each liquid terminal node includes the inlet branch pipe pressure value, pressure fluctuation amplitude, fluid load demand, and fluid extraction valve opening. The node feature vector and edge weights are updated according to the latest data of the pipeline pressure dynamic monitoring dataset.

[0024] Preferably, the weight of the main supply edge from the centralized liquid supply source to each terminal node is assigned using the pressure interference weight coefficient of the corresponding terminal, and the weight of the coupling association edge between each terminal node is calculated based on the product of the pressure interference weight coefficients of the two terminals.

[0025] The graph attention neural network adopts a multi-layer stacked architecture, including an input layer, multiple graph attention layers, and an output layer. The graph attention layer uses an attention mechanism to weighted aggregate the features of neighboring nodes, and the edge weights are used as prior knowledge to participate in the calculation of attention coefficients. The output layer adopts a fully connected network structure to map the hidden feature vectors of each terminal node to the optimal traffic limit value. The output layer sets an activation function to constrain the range of the optimal traffic limit value.

[0026] Preferably, the time-slot staggered arrangement adopts a method that combines priority sorting and time slot allocation, and determines the terminal scheduling priority order based on the pressure interference weight coefficient;

[0027] Fluid extraction time slots are allocated to each terminal in descending order of priority. The terminal currently allocated a time slot selects its fluid extraction time slot according to the principle of maximizing the time slot interval, and prioritizes time slots that do not overlap with those of terminals that have already completed their time slot allocation.

[0028] The proportional reduction algorithm determines the reduction ratio allocation based on the pressure disturbance weight coefficient and the fluid load demand level. The flow reduction ratio of each terminal is positively correlated with the corresponding pressure disturbance weight coefficient. A reduction lower limit protection is set for terminals with high fluid load demand levels. The proportional reduction algorithm iteratively adjusts the reduction ratio until the total flow does not exceed the stable supply threshold.

[0029] Preferably, the liquid supply gap caused by the reduced flow rate at each terminal after the reduction is compensated by scheduling the lost fluid extraction volume into additional extraction time periods, which are then scheduled to be executed during non-overlapping or low-load periods.

[0030] The compensation period is selected based on the available time period with the shortest time interval from the original reduction period. Flow sensors are installed at the inlet branch pipes of each terminal to collect the instantaneous fluid extraction flow rate.

[0031] When the total flow rate at the outlet of the centralized liquid supply source is greater than or equal to the stable supply threshold, the flow limit control is triggered to adjust the opening of the fluid extraction valve. At the beginning of each scheduling period, the scheduling controller sends a scheduling instruction containing the allowed extraction status and the upper limit of the allowed extraction flow rate to the fluid extraction valve controller of each terminal.

[0032] An emergency dispatch response mechanism is set up to temporarily transfer the fluid extraction time of non-urgent terminals to terminals whose incremental demand exceeds the set incremental threshold when the incremental demand of terminal fluid exceeds the set incremental threshold.

[0033] Preferably, the pressure stability index of the centralized liquid supply source is constructed by weighted fusion of pressure fluctuation amplitude index, pressure change rate index and pressure deviation index, and the pipeline vibration intensity index is determined by combining frequency domain energy integral and time domain peak statistics and selecting the maximum value.

[0034] The graph attention neural network fast retraining adopts an incremental learning approach, which fine-tunes the parameters based on the existing network parameters using newly collected running data. When the index exceeds the safety boundary, the optimal traffic limit value label is adjusted downward and the adjustment magnitude is proportional to the degree of exceeding the limit. The loss function adds a safety boundary constraint penalty term. The collaborative connection scheduling parameter update adopts a gradual switching approach, gradually adjusting the upper limit of allowed extraction traffic during the transition time. A maximum retraining iteration limit is set, and when the limit is exceeded, it switches to a conservative running mode. A log recording mechanism is established to store retraining trigger information and index convergence status.

[0035] Secondly, this application provides a distributed liquid consumption terminal and centralized liquid supply source collaborative connection control system, which is used to implement the distributed liquid consumption terminal and centralized liquid supply source collaborative connection control method, including: pipeline pressure monitoring module, pressure interference characteristic analysis module, flow limiting intelligent module, peak-shaving scheduling control module and parameter update module;

[0036] The pipeline pressure monitoring module acquires the pressure data of the main outlet pipe of the centralized liquid supply source, the pressure data of the inlet branch pipe of each distributed liquid user terminal, and the vibration spectrum data of the connecting pipeline nodes, and establishes a dynamic monitoring dataset of pipeline pressure under the condition of multiple terminals in parallel operation.

[0037] The pressure interference feature analysis module extracts pressure fluctuation amplitude features, pressure propagation phase difference features, and vibration spectrum peak features from the pipeline pressure dynamic monitoring dataset. It uses a multi-dimensional feature correlation analysis method to identify the disturbance contribution of each terminal to the pressure field of the liquid supply source when extracting fluid, and generates a pressure interference weight coefficient vector for each terminal.

[0038] The intelligent flow limiting module constructs a pressure coupling relationship map between the terminal group and the centralized liquid supply source based on the pressure interference weight coefficient vector. It inputs the pressure coupling relationship map into a graph attention neural network for node feature aggregation and edge weight learning, and outputs the optimal flow limiting value of each terminal and liquid supply source connection branch at the current time.

[0039] The staggered peak scheduling control module designs a time-sharing staggered peak scheduling strategy based on the optimal flow limit value, staggers the fluid extraction time periods of each terminal, and uses a proportional reduction algorithm to reduce the instantaneous extraction flow for terminals with overlapping extraction time periods, limiting the total flow of the centralized liquid supply source outlet to not exceed the stable supply threshold at any time.

[0040] The parameter update module monitors the pressure stability index of the liquid supply source and the vibration intensity index of the pipeline during the execution of the time-sharing peak scheduling strategy. When the index exceeds the safety boundary, it triggers the graph attention neural network to quickly retrain and dynamically updates the collaborative connection scheduling parameters until the pressure fluctuation converges to the allowable range.

[0041] The beneficial effects of this application are as follows: By collecting pressure and vibration data in all directions, this application can capture the operating status of the pipeline network under multi-terminal parallel operation in real time; the establishment of a dynamic monitoring dataset provides an accurate data foundation for subsequent analysis, solves the problem of difficulty in tracing the source of pressure fluctuations in complex pipeline networks, and ensures the perception accuracy of the control process.

[0042] This application utilizes multidimensional feature correlation analysis to quantify and identify the specific degree of interference of each terminal to the pressure field of the centralized liquid supply source; it generates a pressure interference weight coefficient vector, realizing the accurate location and classification of the disturbance source, and avoiding the control error caused by ignoring individual differences in traditional methods.

[0043] This application introduces a graph attention neural network to process pressure coupling relationships, which can deeply explore the nonlinear interactions between terminal groups; by intelligently calculating and outputting the optimal flow limit value, it achieves global optimization of flow distribution, which minimizes the risk of fluid impact within the pipeline network while ensuring fluid load demand.

[0044] This application employs a time-sharing peak-shifting scheduling and proportional reduction algorithm, which effectively avoids water competition caused by multiple terminals simultaneously extracting large volumes of water; it strictly limits the total flow rate within a stable threshold, eliminating the hidden danger of sudden drops in liquid supply pressure, and significantly improving the continuity and operational stability of fluid supply.

[0045] This application establishes a closed-loop feedback and rapid retraining mechanism with adaptive adjustment capabilities. When sudden changes in operating conditions cause abnormal indicators, it can quickly and dynamically update scheduling parameters to ensure rapid convergence of pressure fluctuations. This effectively prevents pipeline fatigue damage caused by long-term vibration or pressure instability and extends the service life of the equipment.

[0046] This application realizes intelligent collaborative control of terminal groups, effectively solves the pressure oscillation problem when multiple terminals are connected in parallel, significantly improves fluid supply stability, suppresses pipeline vibration, reduces the risk of equipment fatigue damage, and ensures the safe and efficient operation of the system under complex working conditions. Attached Figure Description

[0047] Figure 1 Flowchart of the control method for coordinated connection between distributed liquid consumption terminals and centralized liquid supply sources provided in this application;

[0048] Figure 2 This is a schematic diagram of pipeline pressure dynamic monitoring data acquisition provided in this application;

[0049] Figure 3 The time-sharing peak-shifting and proportional reduction scheduling strategy diagram provided for this application;

[0050] Figure 4 This is a schematic diagram of the end-effector collaborative control system scenario provided in this application;

[0051] Figure 5 The structure diagram of the distributed liquid consumption terminal and centralized liquid supply source collaborative connection control system provided in this application. Detailed Implementation

[0052] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0053] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0054] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of this application. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0055] Example 1

[0056] Reference Figures 1 to 4 This is the first embodiment of the present application, such as Figure 1 As shown, a method for coordinated connection and control between distributed liquid consumption terminals and centralized liquid supply sources is provided.

[0057] Step 1: Obtain the pressure data of the main outlet pipe of the centralized liquid supply source, the pressure data of the inlet branch pipes of each distributed liquid consumption terminal, and the vibration spectrum data of the connecting pipeline nodes, and establish a dynamic monitoring dataset of pipeline pressure under the condition of multiple terminals in parallel operation.

[0058] Figure 2 This is a schematic diagram of pipeline pressure dynamic monitoring data acquisition. Figure 2 The hardware topology and signal flow characteristics are demonstrated. The fluid supply source connects to the distribution pipe via a main supply pipe, which then branches into three parallel branch pipes, connecting to the first, second, and third terminal fluid units, respectively. In terms of sensor arrangement, a main pipe pressure sensor P is installed at the main pipe, and pressure sensors P are also installed on each of the three branch pipes near the inlet of the terminal fluid units; vibration monitoring sensors V are installed on the distribution pipes. All pressure and vibration sensors are connected to a data acquisition controller, enabling real-time aggregation of pressure pulsations and mechanical vibration data from multiple nodes in the pipeline network to the central controller.

[0059] In the coordinated control of distributed liquid consumption terminals and centralized liquid supply sources, a comprehensive dynamic monitoring system for pipeline pressure is first required. A high-precision pressure transmitter is installed at the outlet main pipe of the centralized liquid supply source. This transmitter uses a diffused silicon piezoresistive sensing element, with a range covering the working pressure range and an accuracy class of no less than 0.25. The sampling period of the pressure transmitter is set according to the fluid flow characteristics, and the sampling period must meet the requirement of capturing transient responses to pressure pulsations. The pressure transmitter at the outlet main pipe of the centralized liquid supply source outputs a standard current signal, which is converted into a digital pressure value by an analog-to-digital converter and then transmitted to the data acquisition controller, forming a time-series data of the pressure at the outlet main pipe of the centralized liquid supply source. ,in This represents the absolute pressure value of the centralized liquid supply source outlet manifold at the nth sampling time. This indicates the nth sampling time. By installing a pressure transmitter on the main outlet pipe of the centralized liquid supply source and performing high-frequency sampling, it is possible to obtain the dynamic information of the pressure source when the centralized liquid supply source supplies fluid to each terminal in real time, laying a data foundation for subsequent analysis of the pressure fluctuation propagation law.

[0060] For multi-terminal parallel operation, pressure sensors are installed at the fluid inlet branch pipe of each distributed liquid-using terminal. Each branch pipe pressure sensor and the pressure transmitter on the central liquid supply source outlet main pipe are synchronously acquired using the same sampling clock, ensuring a unified time reference for pressure data from different measuring points. The pressure data collected by the pressure sensor on each terminal's inlet branch pipe forms an independent pressure time series. The pressure sequence of each terminal inlet branch pipe is represented as follows: ,in Indicates the first The absolute pressure value of the terminal inlet branch pipe at the nth sampling time. The value range corresponds to the actual number of terminals.

[0061] Pressure sensors at each terminal inlet branch pipe establish a communication connection with the data acquisition controller via an industrial fieldbus. Pressure data from each channel undergoes timestamp alignment within the data acquisition controller, generating a multi-channel synchronous pressure acquisition data matrix. By deploying pressure sensors at the inlet branch pipe locations of each terminal and achieving synchronous acquisition, the differences in pressure response at different locations during fluid extraction at each terminal can be accurately captured, providing distributed monitoring data support for identifying the disturbance impact of each terminal on the pipeline pressure field.

[0062] Vibration sensors are installed at several key locations on the connecting pipeline between the central liquid supply source outlet main pipe and the inlet branch pipes of each terminal. These locations include pipe branch tee joints, pipe bends, and pipe support fixing points. The vibration sensors are piezoelectric accelerometers, capable of sensing the mechanical vibration generated in the pipeline under the impact of fluid flow. The vibration sensors convert the pipeline vibration acceleration signal into a voltage signal output. After amplification and filtering by a signal conditioning circuit, the signal is digitally sampled by a high-speed analog-to-digital converter. The vibration sampling frequency is set to a multiple of the dominant frequency of the pipeline flow-induced vibration to ensure complete acquisition of the frequency components of the vibration signal. The acquired vibration time-domain signal undergoes Fast Fourier Transform processing within the data acquisition controller, converting the time-domain vibration acceleration sequence into frequency-domain vibration spectrum data. The vibration spectrum data of each pipeline node is represented as follows: ,in Indicates the first The vibration acceleration amplitude of a pipeline node at the m-th frequency component. This represents the m-th frequency component. By placing vibration sensors at key nodes in the connecting pipeline and performing spectral analysis, changes in the fluid flow state within the pipeline can be reflected from the perspective of vibration response. The vibration spectrum data corroborates the pressure data, enhancing the reliability and comprehensiveness of pipeline network operation status monitoring.

[0063] The collected pressure data from the centralized liquid supply source outlet main pipe, the pressure data from the inlet branch pipes of each terminal, and the vibration spectrum data of the connecting pipeline nodes are integrated to construct a dynamic monitoring dataset for pipeline network pressure. The dataset uses a structured storage format, with the sampling timestamp as the primary key for the data index. Each data record includes a field for the pressure value of the centralized liquid supply source outlet main pipe, a sequence of fields for the pressure values ​​of the inlet branch pipes of each terminal, and a sequence of fields for the vibration spectrum vector of each pipeline node. The data records in the dynamic monitoring dataset are stored sequentially in chronological order, forming a time-series database reflecting the dynamic evolution of pipeline network pressure under multi-terminal parallel operation. The data acquisition controller appends newly collected monitoring data to the dynamic monitoring dataset according to a set storage period. Simultaneously, based on data storage capacity limitations, a historical data sliding window cleanup strategy is executed to retain recently collected monitoring data for subsequent pressure feature extraction and analysis.

[0064] A data quality verification mechanism is introduced during the data acquisition process to verify the validity of the acquired pressure and vibration spectrum data. Pressure data validity verification includes range verification and rate of change verification. When the pressure value output by the pressure transmitter or pressure sensor exceeds the preset range (set according to the maximum working pressure of the pipeline network and sensor specifications, e.g., 0~1.0 MPa), the corresponding data is marked as abnormal. When the rate of change of pressure value between adjacent sampling times exceeds the maximum allowable rate of change based on physical characteristics (this maximum rate of change is set based on the fluid water hammer effect limit and the fastest valve action time, e.g., 0.1 MPa / s), the corresponding data is marked as abnormal. The validity verification of vibration spectrum data includes spectral energy verification and fundamental frequency offset verification. Data is considered invalid when the total vibration spectrum energy is lower than the sensor's noise floor level. The noise floor level is determined based on the sensor's inherent electrical noise and environmental background micro-vibration calibration, for example, set to 0.01g. When the fundamental frequency of the vibration spectrum significantly deviates from the pipeline's natural frequency, the corresponding data is marked as abnormal, triggering the sensor's status diagnosis process. Significant deviation is judged based on frequency drift limits caused by abnormal structural stiffness or sudden changes in fluid mass, for example, a deviation exceeding 5% of the pipeline's natural frequency. For data records marked as abnormal, compensation is performed using interpolation of adjacent valid data points to ensure the temporal continuity and data integrity of the pipeline pressure dynamic monitoring dataset.

[0065] The pipeline pressure dynamic monitoring dataset also records the operational status information of each terminal, including the opening status of the fluid extraction valves, the start / stop status of the circulating pumps, and the fluid load demand level of each terminal. This operational status information is obtained through the management system's communication interface and stored in association with pressure data and vibration spectrum data using the same timestamps. Incorporating the terminal operational status information into the pipeline pressure dynamic monitoring dataset provides background data for subsequent analysis of the causal relationship between the extraction actions of each terminal and the pipeline pressure response, enabling targeted analysis of pressure disturbance contribution based on specific operational scenarios.

[0066] The data acquisition and dataset construction method in this step forms a multi-source heterogeneous monitoring data system covering the centralized liquid supply source outlet, the inlets of each terminal, and key nodes of the connecting pipeline. The pipeline pressure dynamic monitoring dataset integrates pressure and vibration information with a unified time base, providing a high-quality raw data foundation for subsequent extraction of pressure fluctuation characteristics, analysis of pressure propagation laws, and identification of the contribution of terminal pressure disturbances. At the same time, the data quality verification mechanism and the operation status-related storage method enhance the reliability and analyzability of the monitoring dataset, ensuring that the distributed liquid terminal and centralized liquid supply source collaborative connection control method has robust data support capabilities under complex and ever-changing actual operating conditions.

[0067] Step 2: Extract pressure fluctuation amplitude features, pressure propagation phase difference features, and vibration spectrum peak features from the pipeline pressure dynamic monitoring dataset. Use multidimensional feature correlation analysis to identify the disturbance contribution of each terminal to the pressure field of the liquid supply source when extracting fluid, and generate the pressure interference weight coefficient vector of each terminal.

[0068] Read the time series data of the pressure of the main outlet pipe of the centralized liquid supply source and the pressure time series data of the inlet branch pipe of each terminal within a set time window from the dynamic monitoring data of pipeline pressure, and extract the pressure fluctuation amplitude characteristics from the pressure time series data.

[0069] Specifically, the pressure time series of the centralized liquid supply source outlet main pipe is first segmented into sliding windows, and statistical characteristics of the pressure series are calculated within each sliding window. The pressure fluctuation amplitude is defined as the difference between the maximum and minimum pressure values ​​within the sliding window. ,in This indicates the maximum pressure in the main outlet pipe of the centralized liquid supply source within the sliding window. This represents the minimum pressure in the main outlet pipe of the centralized liquid supply source within the sliding window. Simultaneously, the standard deviation of the pressure series within the sliding window is calculated as an auxiliary measure of the severity of pressure fluctuations; the pressure standard deviation reflects the degree of dispersion of the pressure value relative to the pressure mean within the window. The same sliding window segmentation method is used to calculate the pressure fluctuation amplitude for the pressure time series of each terminal inlet branch pipe. Pressure fluctuation amplitude of each terminal inlet branch pipe ,in Indicates the first [number]th ... The maximum value of the inlet branch pressure of each terminal. Indicates the first [number]th ... The minimum pressure of each terminal inlet branch pipe is obtained. By extracting the pressure fluctuation amplitude characteristics of the centralized liquid supply source outlet main pipe and each terminal inlet branch pipe, the pressure pulsation intensity of each measuring point in the pipeline network can be quantitatively characterized, providing basic characteristic data for subsequent analysis of the degree of disturbance of the pipeline network pressure field by the extraction action of each terminal.

[0070] Pressure propagation phase difference features were extracted from pressure time series data to reveal the time lag relationship in the propagation of pressure fluctuations from the centralized liquid supply source outlet main pipe to the inlet branch pipes of each terminal. Using the pressure time series of the centralized liquid supply source outlet main pipe as the reference signal and the pressure time series of each terminal inlet branch pipe as the response signal, cross-correlation analysis was used to calculate the phase difference between the reference signal and the response signal. Cross-correlation calculations were performed on the time series of pressures in the inlet branch pipes of each terminal to obtain the cross-correlation function curve. The time offset corresponding to the peak value of the cross-correlation function is the propagation of the pressure fluctuation of the centralized liquid supply source outlet main pipe to the [number missing] terminal. The propagation delay of each terminal inlet branch pipe. Converting the propagation delay into a phase difference representation, pressure propagation phase difference. ,in This represents the dominant frequency component of pressure fluctuations. This indicates that pressure fluctuations in the central liquid supply source outlet main pipe propagate to the first... The propagation delay of each terminal inlet branch pipe.

[0071] The dominant frequency component of pressure fluctuations was obtained by performing a Fast Fourier Transform on the pressure time series of the main outlet pipe of the centralized fluid supply source, and the frequency component with the largest amplitude in the spectrum was selected as the dominant frequency of pressure fluctuations. Simultaneously, the pressure propagation phase difference between the inlet branches of each terminal was extracted, reflecting the relative temporal relationship of pressure fluctuation propagation between different terminal branches. The pressure propagation phase difference characteristics can reveal the propagation path and speed information of pressure disturbances in the pipeline network. When a terminal's fluid extraction action triggers pressure fluctuations, the source location of the pressure disturbance can be traced through phase difference analysis, providing a causal relationship criterion for identifying the contribution of each terminal to pressure disturbances.

[0072] Vibration spectrum data of connected pipeline nodes were read from the pipeline pressure dynamic monitoring dataset, and peak feature extraction was performed. Peak detection was conducted on the vibration spectrum data of each pipeline node to identify local peaks in the vibration spectrum with amplitudes significantly higher than adjacent frequency points. The peak feature of the vibration spectrum includes two components: peak frequency and peak amplitude. The peak frequency reflects the main frequency component of the pipeline vibration, and the peak amplitude reflects the energy intensity of the pipeline vibration at the main frequency component. The peak frequency of the vibration spectrum of each pipeline node is expressed as: The peak amplitude of the vibration spectrum is expressed as When extracting multi-peak values ​​from vibration spectrum data, the first few significant peaks are selected in descending order of peak amplitude to form a multi-peak feature vector. This multi-peak feature vector can more comprehensively describe the frequency composition characteristics of pipeline vibration. Pipeline vibration is closely related to the fluid flow state. When the flow rate of the fluid extracted from the terminal changes, the fluid velocity in the pipeline changes accordingly, causing a corresponding change in the flow-induced vibration characteristics of the pipeline. The peak characteristics of the vibration spectrum can reflect the disturbance impact of the extraction actions of each terminal on the pipeline network from the perspective of vibration response.

[0073] The extracted pressure fluctuation amplitude features, pressure propagation phase difference features, and vibration spectrum peak value features are fused to construct a multi-dimensional feature vector for subsequent correlation analysis. For each terminal, the pressure fluctuation amplitude, pressure propagation phase difference, and vibration spectrum peak amplitude of the corresponding inlet branch pipe are combined to form a multi-dimensional feature vector. Multidimensional feature vectors of each terminal ;in Indicates the first The pressure fluctuation amplitude of each terminal inlet branch pipe This indicates that pressure fluctuations in the central liquid supply source outlet main pipe propagate to the first... Phase difference of each terminal inlet branch pipe, Indicates the relationship with the first The peak amplitude of the vibration spectrum of each terminal-associated pipeline node is given by T, which represents the transpose. Multidimensional feature vectors integrate feature data of different physical dimensions into a unified feature representation, laying the data structure foundation for subsequent identification of disturbance contribution using multidimensional feature correlation analysis methods.

[0074] Multidimensional feature correlation analysis was used to quantitatively analyze the correlation between the multidimensional feature vectors of each terminal and the pressure fluctuation of the central liquid supply source outlet manifold, identifying the contribution of each terminal to the pressure field of the liquid supply source when extracting fluid. A multiple regression analysis model was established between the pressure fluctuation amplitude of the central liquid supply source outlet manifold and the multidimensional feature vectors of each terminal, and the interference of each terminal's features on the pressure fluctuation of the central liquid supply source was determined by the regression coefficients. Partial correlation analysis was used to eliminate the mutual interference between terminals, and the partial correlation coefficient between the multidimensional feature vector of each terminal and the pressure fluctuation amplitude of the central liquid supply source outlet manifold was calculated. The partial correlation coefficient can independently reflect the correlation strength of the disturbance of a single terminal to the pressure field of the central liquid supply source.

[0075] The partial correlation coefficient and the standardized regression coefficients from the multiple regression analysis are weighted and combined to obtain a score for the disturbance contribution of each terminal to the pressure field of the centralized liquid supply source. A higher disturbance contribution score indicates a more significant disturbance impact on the pressure field of the liquid supply source when the corresponding terminal extracts fluid. The multidimensional feature correlation analysis method comprehensively considers multiple dimensions of feature information, such as pressure fluctuation amplitude, pressure propagation phase difference, and vibration spectrum peak value, avoiding the one-sidedness that may result from single feature analysis and enabling a more accurate identification of the pressure disturbance contribution of each terminal.

[0076] Based on the disturbance contribution scores of each terminal, a pressure interference weight coefficient vector is generated for each terminal. The disturbance contribution scores of all terminals are then normalized so that the sum of the disturbance contribution scores equals a unit value. The normalized disturbance contribution score is the pressure interference weight coefficient for the corresponding terminal. Pressure Interference Weight Coefficient Vector ,in Indicates the first Pressure interference weighting coefficient for each terminal This represents the total number of terminals, and the pressure interference weight coefficient vector satisfies the normalization constraint.

[0077] For example, if there are four terminals, and after multidimensional feature correlation analysis, the disturbance contribution scores for each terminal are 0.35, 0.25, 0.22, and 0.18, respectively, then the corresponding pressure disturbance weight coefficient vector is: This indicates that the first terminal contributes the most to the disturbance of the pressure field of the centralized liquid supply source, and the fluid extraction scheduling of the first terminal needs to be focused on in subsequent coordinated connection control. For the pressure disturbance weight coefficient vector, a dynamic update mechanism also needs to be constructed to adapt to changes in operating conditions. When the pipeline pressure dynamic monitoring dataset accumulates new monitoring data, the pressure fluctuation amplitude feature extraction, pressure propagation phase difference feature extraction, vibration spectrum peak feature extraction, and multi-dimensional feature correlation analysis processes are re-executed according to the set update cycle to generate an updated pressure disturbance weight coefficient vector. The dynamic update of the pressure disturbance weight coefficient vector adopts a moving average fusion strategy, which weights the newly calculated pressure disturbance weight coefficient with the historical pressure disturbance weight coefficient, avoiding the impact of random fluctuations in single calculation results on the stability of the pressure disturbance weight coefficient vector.

[0078] The multidimensional feature extraction and correlation analysis method constructed in this step extracts quantitative indicators reflecting the impact of each terminal on the pressure field disturbance of the centralized liquid supply source from the pipeline pressure dynamic monitoring dataset. The pressure disturbance weight coefficient vector expresses the disturbance contribution of each terminal in numerical form, providing the basis for assigning edge weights for the subsequent construction of the pressure coupling relationship map between the terminal group and the centralized liquid supply source. At the same time, the multidimensional feature fusion analysis method comprehensively utilizes pressure information and vibration information, improving the accuracy and reliability of disturbance contribution identification, enabling the distributed liquid consumption terminal and centralized liquid supply source collaborative connection control to design differentiated scheduling strategies for the disturbance characteristics of different terminals.

[0079] Step 3: Construct a pressure coupling relationship map between the terminal group and the centralized liquid supply source based on the pressure interference weight coefficient vector. Input the pressure coupling relationship map into the graph attention neural network for node feature aggregation and edge weight learning, and output the optimal flow limit value of each terminal and liquid supply source connection branch at the current time.

[0080] Based on the pressure interference weighting coefficient vector, a pressure coupling relationship graph is constructed between the terminal group and the centralized liquid supply source. This graph is represented using a directed weighted graph structure. In the graph, the centralized liquid supply source is the central node, and each distributed liquid-using terminal is an outer node. The central node and each outer node are connected by directed edges, with the direction of the edges pointing from the supply source node to each terminal node, representing the physical flow direction of fluid from the centralized supply source to each terminal. Directed edges are also established between the terminal nodes, representing the pressure coupling relationship between different terminals arising from sharing the pressure source of the centralized liquid supply source. The node set of the pressure coupling relationship graph is shown below. ,in Indicates the central node of the centralized liquid supply source. Indicates the first One terminal node, This represents the total number of terminals. The edge set of the pressure coupling relationship graph includes the main supply edges from the liquid supply source node to each terminal node, as well as the coupling relationship edges between each terminal node. Modeling the topological connection relationship between the terminal group and the centralized liquid supply source through a graph structure can intuitively express the characteristics of the pipeline network structure under multi-terminal parallel operation, providing structured input data for subsequent learning of pressure coupling relationships using graph neural networks. Each node in the pressure coupling relationship graph is assigned a node feature vector, which contains multi-dimensional attribute information reflecting the current operating status of the node. The feature vector of the central node of the centralized liquid supply source includes attributes such as the current pressure value of the liquid supply source outlet main pipe, the pressure fluctuation amplitude of the liquid supply source outlet main pipe, the fluid level height in the liquid supply source, and the supply capacity margin of the liquid supply source. ,in This indicates the current pressure value of the main outlet pipe of the centralized liquid supply source. This indicates the pressure fluctuation amplitude of the main outlet pipe of the centralized liquid supply source. Indicates the fluid level height within the centralized liquid supply source. This indicates the slack capacity of the centralized liquid supply source.

[0081] The feature vector of each terminal node includes attributes such as the current pressure value of the corresponding terminal inlet branch pipe, the pressure fluctuation amplitude of the inlet branch pipe, the current fluid load demand of the terminal, and the opening degree of the terminal fluid extraction valve. Feature vectors of terminal nodes ,in Indicates the first Current pressure value of each terminal inlet branch pipe. Indicates the first The pressure fluctuation amplitude of each terminal inlet branch pipe Indicates the first Current fluid load demand of each terminal Indicates the first The opening degree of the fluid extraction valve at each terminal. The node feature vector integrates the dynamic monitoring data of pipeline pressure with the terminal operating status data, so that the pressure coupling relationship map not only contains topological structure information, but also carries the real-time operating status information of each node.

[0082] The pressure interference weighting coefficient vector is used to assign weights to the edges in the pressure coupling relationship graph. The edge weights reflect the strength of the pressure coupling relationship between adjacent nodes. (The text then abruptly shifts to a seemingly unrelated topic: "From the central node of the centralized liquid supply source to the...") The main supply edge weight of each terminal node is directly adopted from the pressure interference weight coefficient of the corresponding terminal in the pressure interference weight coefficient vector. The pressure interference weight coefficient is assigned a value; a larger coefficient indicates a tighter pressure coupling between the centralized liquid supply source and the corresponding terminal. The coupling weight between each terminal node is calculated based on the product of the pressure interference weight coefficients of the two terminals. The terminal node and the first Weight of coupling edges between terminal nodes ,in Indicates the first Pressure interference weighting coefficient for each terminal Indicates the first The pressure interference weight coefficients of each terminal are considered. When the pressure interference weight coefficients of both terminals are large, the weight of the coupling relationship edge between the two terminals increases accordingly, indicating that the superposition effect of pressure interference when the two terminals simultaneously extract fluid is more significant. By assigning edge weights, the disturbance contribution information contained in the pressure interference weight coefficient vector is integrated into the edge attributes of the pressure coupling relationship graph, so that the graph structure can fully express the pressure coupling strength distribution characteristics between the terminal group and the centralized liquid supply source.

[0083] The constructed pressure coupling graph is input into a graph attention neural network for node feature aggregation and edge weight learning. The graph attention neural network employs a multi-layer stacked architecture, including an input layer, multiple graph attention layers, and an output layer. The input layer receives the feature vectors of each node in the pressure coupling graph and maps these feature vectors to the hidden feature space of the graph attention neural network. The graph attention layers use an attention mechanism to weighted aggregate the features of neighboring nodes, with the attention weights dynamically calculated based on the correlation between the features of the central node and the features of its neighbors. In the graph attention layers, for any node, the attention coefficient between the node and all its neighbors is first calculated. This attention coefficient is obtained by performing a linear transformation on the node's feature vector, calculating its inner product, and then normalizing it using softmax.

[0084] The node feature aggregation process in the graph attention layer comprehensively considers the feature information of neighboring nodes and edge weights. Edge weights, as prior knowledge, participate in the calculation of the attention coefficient, ensuring that terminal nodes with higher pressure interference weights receive greater attention during feature aggregation. After feature aggregation through multiple graph attention layers, the hidden feature vector of each node integrates pressure coupling information from neighboring nodes and even more distant nodes. The hidden feature vector of the central node of the centralized fluid supply source aggregates the feature information of all terminal nodes. The hidden feature vector of each terminal node contains comprehensive information about the pressure source state of the fluid supply source and the coupling effects of other terminals. The graph attention neural network adaptively learns the feature aggregation weights between different nodes through an attention mechanism, which, compared to fixed-weight graph convolution methods, can more flexibly capture the dynamic changes in pressure coupling relationships.

[0085] The output layer of the graph attention neural network generates the corresponding optimal flow limit value for each terminal node. The output layer employs a fully connected network structure, mapping the hidden feature vectors of each terminal node, aggregated by the graph attention layer, to a single numerical output. This single numerical output represents the optimal flow limit value for the branch connecting the corresponding terminal to the liquid supply source. The optimal flow limit value represents the maximum allowable flow rate of fluid extracted by the corresponding terminal from the centralized liquid supply source under the current pipeline pressure and terminal operating conditions. When the actual extracted flow rate exceeds the optimal flow limit value, it will cause excessive disturbance to the pressure field of the liquid supply source. The output layer also sets an activation function to constrain the range of the optimal flow limit value, ensuring that the optimal flow limit value lies between the terminal's minimum operating flow rate and the pipeline's maximum flow capacity. The optimal flow limit value for each terminal connected to the liquid supply source branch is expressed as: The optimal traffic limit values ​​of all terminals form an optimal traffic limit value vector. ,in Indicates the first The optimal flow limit value for each terminal connected to the liquid supply source branch. This indicates the total number of terminals.

[0086] The graph attention neural network is trained using supervised learning. The training dataset consists of a pressure coupling relationship graph under historical operating conditions and corresponding optimal flow limit labels. The optimal flow limit labels are assigned based on the pressure stability performance in the historical operating data. When the pressure fluctuation amplitude of the centralized liquid supply source remains within the allowable range during a certain period, the actual extraction flow rate of each terminal during that period is used as the optimal flow limit label. When the pressure fluctuation amplitude exceeds the allowable range, the actual extraction flow rate of each terminal is reduced and corrected according to the degree of pressure fluctuation exceeding the limit before being used as the optimal flow limit label. The loss function of the graph attention neural network uses mean squared error loss to measure the deviation between the optimal flow limit value output by the network and the label value. The training process uses the Adam optimizer to update network parameters, and the learning rate is dynamically adjusted according to the cosine annealing strategy. Training iterations continue until the loss function converges to below a set threshold. After training, the graph attention neural network can quickly infer and output the optimal flow limit value of each terminal's connection branch to the liquid supply source based on the input pressure coupling relationship graph.

[0087] The graph attention neural network (GNN) receives a real-time updated pressure coupling graph as input during the inference phase. The node feature vectors and edge weights in the pressure coupling graph are updated in real-time based on the latest data from the pipeline pressure dynamic monitoring dataset. When the pipeline pressure dynamic monitoring dataset is updated with new monitoring data, the feature vectors of each node and the weights of each edge are recalculated. The updated pressure coupling graph is then input into the GNN, which outputs an optimal flow limit vector reflecting the current pipeline pressure state. For example, if the pressure fluctuation amplitude of the centralized liquid supply source outlet main pipe increases at a certain moment, the GNN identifies a decrease in the stability of the centralized liquid supply source pressure during node feature aggregation. Correspondingly, it reduces the output optimal flow limit value for each terminal, mitigating the pressure fluctuation of the centralized liquid supply source by limiting the extraction flow of each terminal.

[0088] This step transforms the complex pressure coupling relationship between the terminal group and the centralized liquid supply source into a graph-structured data representation by constructing a pressure coupling relationship graph and using a graph attention neural network inference method. The graph attention neural network's node feature aggregation and edge weight learning capabilities are used to deeply model the pressure coupling relationship. The output optimal flow limit value comprehensively considers the supply capacity of the centralized liquid supply source, the fluid load demand of each terminal, and the pressure coupling influence between terminals. This provides a scientific and reasonable flow constraint basis for the subsequent design of time-sharing peak-shaving scheduling strategies. Simultaneously, the real-time inference capability of the graph attention neural network ensures that the optimal flow limit value can be dynamically adjusted according to changes in the pipeline network's operating conditions, improving the adaptability and intelligence level of the coordinated connection control between distributed liquid-using terminals and the centralized liquid supply source.

[0089] Step 4: Design a time-sharing peak scheduling strategy based on the optimal flow limit value, stagger the fluid extraction time periods of each terminal, and use a proportional reduction algorithm to reduce the instantaneous extraction flow for terminals with overlapping extraction time periods, so as to limit the total flow of the centralized liquid supply source outlet at any time to not exceed the stable supply threshold.

[0090] Figure 3 This is a diagram illustrating the time-sharing peak-shifting and proportional reduction scheduling strategies. Figure 3Using time as the horizontal axis and flow rate as the vertical axis, a global stable supply threshold Max is set, and three dynamic scheduling features are demonstrated: First, peak shifting, where the flow demands of the first and third end fluid units are staggered in time and do not overlap, so the total flow rate is naturally lower than the stable supply threshold; Second, proportional reduction, where when the original demands of the first and second end fluid units exceed the stable supply threshold, proportional reduction in the overlapping area is triggered, synchronously reducing the actual flow rates of the first and second end fluid units (reduction) to within the stable supply threshold limit; Third, compensation, where the flow rate restricted by the second end fluid unit during the reduction phase is compensated by the second end fluid unit in subsequent time periods, reflecting the flow conservation control logic of first reducing and then compensating on the time axis.

[0091] Based on the optimal flow limit vector output by the graph attention neural network and combined with the fluid load demand information of each terminal, a basic time period division scheme for the time-sharing peak-shaving scheduling strategy is designed. Specifically, the length of the scheduling cycle and the granularity of the time period division are first determined. The length of the scheduling cycle is set according to the typical operating cycle, and the granularity of the time period division is determined based on the response time constant of the terminal fluid extraction process. The granularity of the time period division must be smaller than the time required for the terminal fluid extraction process to go from start-up to stabilization. The scheduling cycle is divided into several consecutive scheduling time periods, and each scheduling time period serves as the basic time unit for terminal fluid extraction scheduling.

[0092] For each terminal, the number of fluid extraction time slots required by the terminal within the scheduling cycle is determined based on the terminal's fluid load demand level. Terminals with higher fluid load demand levels are allocated more fluid extraction time slots. The number of fluid extraction periods required for each terminal is expressed as follows: Number of fluid extraction periods According to the Fluid load requirements of each terminal The amount of fluid available to a terminal within a single time period is calculated and determined. By allocating the number of time periods based on fluid load demand, it is ensured that each terminal can obtain the total amount of fluid required within the scheduling cycle, laying the foundation for demand matching for subsequent time period staggered scheduling.

[0093] The fluid extraction time slots of each terminal are staggered to avoid pressure superposition disturbances caused by multiple terminals extracting fluid simultaneously in the same time period. The staggered arrangement uses a combination of priority ranking and time slot allocation. The scheduling priority order of terminals is determined based on the pressure interference weight coefficient of each terminal in the pressure interference weight coefficient vector. Terminals with larger pressure interference weight coefficients have higher scheduling priority, and higher-priority terminals are given priority in selecting scheduling time slots. The scheduling priority is defined according to the magnitude of the pressure interference weight coefficient of each terminal; terminals with larger pressure interference weight coefficients have higher scheduling priority. During time slot allocation, the highest-priority terminal first selects the required number of time slots as fluid extraction time slots from all available time slots within the scheduling period. The selection of time slots follows the principle of maximizing time slot intervals, ensuring that the fluid extraction time slots of the highest-priority terminals are evenly distributed within the scheduling period. The next highest-priority terminal selects fluid extraction time slots from the remaining available time slots, prioritizing time slots that do not overlap with the fluid extraction time slots of higher-priority terminals. This process continues, allocating fluid extraction time slots to each terminal in descending order of priority, with each terminal avoiding time slots already occupied by higher-priority terminals as much as possible during time slot allocation. By prioritizing and staggering time slots, terminals that contribute significantly to pressure disturbances can obtain independent extraction time slots, reducing the probability of highly disturbed terminals extracting fluid simultaneously with other terminals, thus reducing the occurrence of pressure superposition disturbances at the source.

[0094] When there are many terminals or the fluid load demand is concentrated, the fluid extraction periods of some terminals may overlap. A proportional reduction algorithm is used to reduce the instantaneous extraction flow rate for terminals with overlapping extraction periods. The algorithm detects terminal extraction overlap in each time period within the scheduling cycle and counts the set of terminals scheduled to extract fluid in each time period. For overlapping periods where multiple terminals extract fluid simultaneously, the total flow rate extracted by each terminal at its optimal flow limit during the overlapping period is calculated and compared with the stable supply threshold of the centralized fluid supply source. The stable supply threshold of the centralized fluid supply source represents the maximum fluid flow rate that the centralized fluid supply source can continuously output while maintaining stable pressure. The stable supply threshold of the centralized fluid supply source is determined based on parameters such as the volume, replenishment capacity, and cross-sectional area of ​​the outlet main pipe of the centralized fluid supply source, for example, set to 50 L / min. When the sum of the optimal flow limit values ​​of all terminals during the overlapping period exceeds the stable supply threshold of the centralized fluid supply source, the proportional reduction algorithm is triggered to reduce and adjust the instantaneous extraction flow rate of each terminal. The proportional reduction algorithm determines the reduction ratio allocation based on the pressure interference weight coefficient and fluid load demand level of each terminal. The reduction ratio allocation follows the principle of prioritizing pressure disturbance reduction while ensuring fluid load demand. Terminals with a high pressure disturbance weighting coefficient bear a larger proportion of flow reduction, while terminals with high fluid load demand levels are protected by a reduction lower limit. The reduction ratio of each terminal during the overlapping time period is expressed as follows: Reduction ratio According to the Pressure interference weighting coefficient of each terminal Weighted calculations are performed, with terminals having higher pressure interference weighting coefficients resulting in higher reduction ratios. After reduction, the... Actual traffic extracted by each terminal during the overlapping time period ,in Indicates the first The optimal flow limit value for each terminal connected to the liquid supply source branch. Indicates the first The reduction ratio of each terminal during overlapping time periods Indicates the number after reduction The actual extraction flow rate of each terminal. The proportional reduction algorithm iteratively adjusts the reduction ratio of each terminal until the total actual extraction flow rate of all terminals during the overlapping period does not exceed the stable supply threshold of the centralized fluid supply source. The proportional reduction algorithm assigns more reduction responsibility to terminals that contribute more to pressure disturbances through differentiated reduction allocation, thereby controlling the total flow rate while ensuring the fluid supply to terminals with urgent fluid load demands as much as possible.

[0095] Compensation scheduling is implemented to address the fluid supply gap caused by reduced flow rates at each terminal after flow reduction. The cumulative fluid extraction loss of each terminal during overlapping periods after proportional reduction is calculated, and this lost fluid extraction is converted into additional extraction time slots scheduled for non-overlapping or low-load periods. Compensation time slots are prioritized for available time slots closer to the original reduction period to minimize the impact of extraction time delays on terminal operation. For fluid supply gaps that cannot be fully compensated within the scheduling cycle, the compensation requirement is carried over to the next scheduling cycle, and an additional fluid extraction time slot quota is allocated to the corresponding terminal in the next scheduling cycle's time slot allocation. This compensation scheduling mechanism ensures that each terminal obtains the total fluid required to meet its fluid load within the scheduling cycle or adjacent scheduling cycles, preventing insufficient terminal operation due to flow reduction.

[0096] Establish a real-time monitoring and limiting mechanism for the total flow rate at the centralized liquid supply source outlet at any given time to ensure that the total flow rate does not exceed the stable supply threshold of the centralized liquid supply source. Flow sensors are installed at the inlet branch pipes of each terminal to collect the instantaneous fluid extraction flow rate at each terminal in real time. The total flow rate at the centralized liquid supply source outlet is obtained by summing the instantaneous extraction flow rates of each terminal. The total flow rate at the centralized liquid supply source outlet is expressed as... ,in This indicates the total flow rate at the centralized liquid supply source outlet. Indicates the first The actual traffic extracted by each terminal This indicates the total number of terminals. It also represents the total flow rate at the centralized liquid supply source outlet. Stable supply threshold of centralized liquid supply source Real-time comparison is performed. When the total flow rate at the outlet of the centralized liquid supply source approaches or exceeds the stable supply threshold of the centralized liquid supply source, the flow restriction control is triggered to adjust the opening of the fluid extraction valve. The opening of the fluid extraction valve at each terminal is reduced sequentially according to the reduction ratio calculated by the proportional reduction algorithm until the total flow rate at the outlet of the centralized liquid supply source falls back below the stable supply threshold of the centralized liquid supply source.

[0097] The time-sharing peak-shaving scheduling strategy is implemented through a linkage between scheduling command issuance and valve control. Based on the time-segmentation staggered arrangement scheme and proportional reduction algorithm, the scheduling controller issues scheduling commands to the fluid extraction valve controllers of each terminal at the start of each scheduling period. The scheduling command includes two parameters: the terminal's permitted extraction status for the current period and the maximum permitted extraction flow rate. The permitted extraction status parameter indicates whether the terminal is in a permitted or prohibited extraction state during the current period, while the maximum permitted extraction flow rate parameter indicates the maximum flow rate that the terminal must not exceed when extracting fluid during the current period. Upon receiving the scheduling command, the fluid extraction valve controllers of each terminal control the opening and closing of the valves according to the permitted extraction status parameter and adjust the valve opening according to the maximum permitted extraction flow rate parameter to ensure that the actual extraction flow rate of the terminal does not exceed the maximum flow rate specified in the scheduling command. For example, if the scheduling instruction of the first terminal during a certain scheduling period is to enable extraction and the maximum allowable extraction flow rate is 15 liters per minute, and the scheduling instruction of the second terminal is to disable extraction, then the first terminal can perform fluid extraction during the current period, but the extraction flow rate must not exceed 15 liters per minute, and the second terminal does not perform fluid extraction during the current period.

[0098] The time-sharing peak-shaving scheduling strategy also includes an emergency dispatch response mechanism to address sudden changes in fluid load demand that may occur during operation. When the fluid load demand of a terminal suddenly increases significantly within the scheduling cycle, the emergency dispatch response mechanism reassesses the terminal's fluid extraction time slot demand. Without affecting the normal operation of other terminals, it temporarily allocates some fluid extraction time slots from non-urgent terminals to the terminal with the sudden increase in load demand. The emergency dispatch response mechanism is triggered when the increase in terminal fluid load demand exceeds a set emergency response threshold, which is set based on capacity and allowable fluctuation range. The emergency dispatch response mechanism ensures that the time-sharing peak-shaving scheduling strategy can flexibly respond to dynamic changes in operating conditions, meeting the real-time needs of each terminal while ensuring the stability of the centralized fluid supply source pressure.

[0099] The time-sharing peak-shaving scheduling strategy constructed in this step reasonably staggers the fluid extraction periods of each terminal, uses a proportional reduction algorithm to differentiate the extraction flow of terminals in overlapping periods, establishes a real-time monitoring and limiting mechanism for the total flow of the centralized liquid supply source outlet, and controls the disturbance impact of multiple terminals extracting fluid in parallel on the pressure field of the centralized liquid supply source from both time and flow dimensions. The stable supply threshold constraint of the centralized liquid supply source ensures that the output capacity of the centralized liquid supply source matches the extraction demand of the terminal group at any time. The compensation scheduling mechanism and emergency response mechanism enhance the adaptability and robustness of the time-sharing peak-shaving scheduling strategy, providing a reliable scheduling and control means for the coordinated and stable operation of distributed liquid-using terminals and centralized liquid supply sources.

[0100] Step 5: During the execution of the time-sharing peak scheduling strategy, monitor the pressure stability index of the centralized liquid supply source and the pipeline vibration intensity index. When the index exceeds the safety boundary, trigger the graph attention neural network to quickly retrain and dynamically update the collaborative connection scheduling parameters until the pressure fluctuation converges to the allowable range.

[0101] A real-time monitoring mechanism for the pressure stability index of centralized liquid supply sources is established. Time-series pressure data from the outlet main pipe of the centralized liquid supply source is continuously read from the pipeline network pressure dynamic monitoring dataset, and the pressure stability of the centralized liquid supply source is evaluated on a rolling basis according to a set evaluation window period. The pressure stability index of the centralized liquid supply source is constructed using a multi-dimensional comprehensive evaluation method, including three sub-indicators: pressure fluctuation amplitude, pressure change rate, and pressure deviation. The pressure fluctuation amplitude index is obtained by calculating the difference between the maximum and minimum pressure values ​​of the outlet main pipe of the centralized liquid supply source within the evaluation window, reflecting the pulsation range of the centralized liquid supply source pressure within the evaluation window. The pressure change rate index is obtained by calculating the average absolute value of the pressure difference between adjacent sampling times within the evaluation window, reflecting the severity of pressure changes in the centralized liquid supply source. The pressure deviation index is obtained by calculating the deviation between the average pressure of the outlet main pipe of the centralized liquid supply source within the evaluation window and the rated operating pressure of the centralized liquid supply source, reflecting the degree of deviation of the centralized liquid supply source pressure from the design operating point.

[0102] By weighting and fusing the pressure fluctuation amplitude index, the pressure change rate index, and the pressure deviation index, a comprehensive index of pressure stability for centralized liquid supply sources is obtained. This comprehensive index of pressure stability for centralized liquid supply sources is expressed as follows: ,in This represents a comprehensive index of the pressure stability of a centralized liquid supply source. This represents the normalized value of the pressure fluctuation amplitude index. This represents the normalized value of the pressure change rate index. This represents the normalized value of the pressure deviation index. This indicates the fusion weight of the pressure fluctuation amplitude index. The weighting of the pressure change rate index indicates the fusion weighting. This indicates the fusion weight of the pressure deviation index. The weights are set based on the relative impact of each index on pipeline fatigue damage and fluid supply instability. For example, setting... , , A lower value for the comprehensive index of centralized liquid supply source pressure stability indicates a more stable pressure, while a higher value indicates more drastic pressure fluctuations. This multi-dimensional comprehensive evaluation method can comprehensively characterize the stability of the centralized liquid supply source pressure from different perspectives, avoiding the one-sidedness that may result from evaluating a single indicator.

[0103] A real-time monitoring mechanism for pipeline vibration intensity indicators is established simultaneously. Vibration spectrum data of connected pipeline nodes are continuously read from the pipeline network pressure dynamic monitoring dataset to perform rolling assessments of pipeline vibration intensity. The pipeline vibration intensity indicators are constructed using a combination of frequency domain energy integration and time domain peak statistics. The frequency domain energy integration indicator is obtained by integrating the vibration spectrum of each pipeline node within a range of interest. This range covers the main frequency components of flow-induced vibration and the frequency band near the resonant frequency of the pipeline structure. The frequency domain energy integration indicator reflects the total energy level of pipeline vibration within the range of interest. The time domain peak statistics indicator is obtained by peak detection and statistics of the time-domain vibration acceleration signals of each pipeline node. It statistically analyzes the number of times the vibration acceleration peak exceeds a set threshold and the average peak amplitude within the evaluation window. The time domain peak statistics indicator reflects the impact intensity and frequency of pipeline vibration. The frequency domain energy integral index and time domain peak statistical index of each pipeline node are summarized. A maximum value selection strategy is used to determine the comprehensive pipeline vibration intensity index. The index value of the node with the most severe vibration among all pipeline nodes is selected as the comprehensive pipeline vibration intensity index value. The comprehensive pipeline vibration intensity index is expressed as follows: Using the maximum value selection strategy can ensure that the comprehensive index of pipeline vibration intensity reflects the vibration state of the most severely vibrated location in the pipeline network, and promptly detect abnormal local pipeline vibration.

[0104] Safety boundary thresholds are set for the comprehensive pressure stability index of the centralized liquid supply source and the comprehensive pipeline vibration intensity index. These safety boundary thresholds are calibrated based on the yield strength, fatigue limit, and equipment safety operation specifications of the pipeline materials. For example, the safety boundary threshold for the comprehensive pressure stability index is set to 0.2 MPa, and the safety boundary threshold for the comprehensive pipeline vibration intensity index is set to 0.5 g. The safety boundary threshold for the comprehensive pressure stability index of the centralized liquid supply source is expressed as follows: When the comprehensive index of pressure stability of centralized liquid supply source Exceeding the safety boundary threshold When the pressure stability of the centralized liquid supply source exceeds the safety boundary, it is determined that the pressure stability exceeds the safety boundary. The safety boundary threshold for the comprehensive index of pipeline vibration intensity is expressed as follows: When the comprehensive index of pipeline vibration intensity Exceeding the safety boundary threshold When the vibration intensity exceeds the safety boundary, the pipeline is deemed to have exceeded the safety boundary. The calibration of the safety boundary threshold considers factors such as the material strength, fatigue life, and sealing reliability of the centralized liquid supply source and pipeline, reserving an appropriate safety margin while ensuring safe equipment operation. At the end of each evaluation window period, the calculated comprehensive index of centralized liquid supply source pressure stability and the comprehensive index of pipeline vibration intensity are compared with the corresponding safety boundary thresholds to determine whether the current pipeline network operation is within a safe range.

[0105] When the comprehensive index of centralized liquid supply source pressure stability or the comprehensive index of pipeline vibration intensity exceeds the corresponding safety boundary threshold, the rapid retraining process of the graph attention neural network is triggered. The rapid retraining process first diagnoses and analyzes the cause of the retraining trigger, determining the focus of retraining based on the extent to which the comprehensive index of centralized liquid supply source pressure stability and the comprehensive index of pipeline vibration intensity exceed the limits. When only the comprehensive index of centralized liquid supply source pressure stability exceeds the safety boundary, retraining focuses on optimizing the optimal flow limit value for each terminal to reduce pressure fluctuations in the centralized liquid supply source. When only the comprehensive index of pipeline vibration intensity exceeds the safety boundary, retraining focuses on adjusting the terminal extraction timing to avoid sensitive periods of pipeline resonance. When both the comprehensive index of centralized liquid supply source pressure stability and the comprehensive index of pipeline vibration intensity exceed the safety boundary, retraining simultaneously optimizes the optimal flow limit value and the extraction timing arrangement. The diagnostic analysis results guide the adjustment of the loss function weights during retraining, enabling the graph attention neural network retraining to specifically improve the control effect corresponding to the exceeding indicators.

[0106] The graph attention neural network (GNN) employs an incremental learning approach for rapid retraining. It fine-tunes parameters using newly acquired operational data, building upon existing network parameters and avoiding the time overhead of retraining from scratch. The training dataset for rapid retraining consists of a pressure coupling relationship graph and corresponding operational effect labels within a predetermined time period before retraining. Operational effect labels are assigned based on the actual performance of the comprehensive index of centralized liquid supply source pressure stability and the comprehensive index of pipeline vibration intensity. When an index exceeds the safety boundary, the optimal flow limit label for that moment is adjusted downwards, with the adjustment magnitude proportional to the degree of index exceedance. The loss function for rapid retraining adds a safety boundary constraint penalty term to the original mean squared error loss. This penalty term applies additional punishment to the network output that causes the index to exceed the limit, guiding the GNN to learn and output a more conservative optimal flow limit value. Rapid retraining uses a small learning rate for parameter updates to prevent drastic changes in network parameters during retraining that could affect control stability. The number of iterations in rapid retraining is dynamically determined based on the degree of index exceedance; the greater the degree of exceedance, the more iterations are needed to ensure sufficient adjustment of network parameters. Incremental learning enables graph attention neural networks to quickly adapt to new operating conditions while retaining historical operating experience, thus shortening the time required for retraining.

[0107] After rapid retraining, the optimal flow limit values ​​for each terminal's connection to the fluid supply source branch are recalculated using the retrained graph attention neural network, dynamically updating the collaborative connection scheduling parameters. The updated optimal flow limit value vector is input into the time-sharing peak-shaving scheduling strategy, and the time-segmentation staggered arrangement and proportional reduction algorithm are re-executed to generate updated scheduling instructions. The updated scheduling instructions are issued to the fluid extraction valve controllers of each terminal through the scheduling controller, and each terminal adjusts its fluid extraction behavior according to the updated scheduling instructions. The updating of the collaborative connection scheduling parameters is performed in a gradual switching manner. Within a set transition time, the allowable extraction flow limit of each terminal is gradually adjusted from the original parameter value to the updated parameter value to avoid sudden changes in scheduling parameters causing shock disturbances to the pipeline pressure.

[0108] After the collaborative connection scheduling parameters are updated, the changing trends of the comprehensive index of centralized liquid supply source pressure stability and the comprehensive index of pipeline vibration intensity are continuously monitored to determine whether the pressure fluctuation has converged to the allowable range. The allowable range is defined as the numerical interval obtained by multiplying the safety boundary threshold by the convergence coefficient. The convergence coefficient is less than a unit value, making the allowable range more stringent than the safety boundary threshold. When the comprehensive index of centralized liquid supply source pressure stability remains within the allowable range and the comprehensive index of pipeline vibration intensity remains within the allowable range for a set stabilization time, it is determined that the pressure fluctuation has converged to the allowable range, and the collaborative connection scheduling enters a steady-state operation mode. If the comprehensive index of centralized liquid supply source pressure stability or the comprehensive index of pipeline vibration intensity still exceeds the safety boundary after the collaborative connection scheduling parameters are updated, the graph attention neural network fast retraining process is triggered again to iteratively update the collaborative connection scheduling parameters until the indicators converge. To prevent retraining iterations from getting stuck in an infinite loop, a maximum limit on the number of retraining iterations is set. If the metrics still do not converge when the maximum number of retraining iterations is reached, an alarm is triggered and the system switches to conservative operation mode. In conservative operation mode, the maximum allowable traffic extraction limit for each terminal is uniformly reduced to a safe lower limit, such as 20% of the terminal's rated traffic or 10L / min.

[0109] A log recording mechanism for the retraining process and scheduling parameter updates is established, recording information such as the time of each retraining trigger, the triggering reason, changes in network parameters before and after retraining, the content of scheduling parameter updates, and the convergence status of indicators. The log information is stored in the operation log database for subsequent analysis of pressure fluctuation patterns and the optimization effect of scheduling strategies during operation. Statistical analysis of the log information can identify typical operating conditions that lead to frequent exceedances of indicators, providing data support for offline optimization training of graph attention neural networks. For example, if the log records show frequent retraining triggers during a fixed period each day, the terminal fluid load distribution characteristics and centralized liquid supply status during that period can be analyzed, allowing for targeted adjustments to the time-sharing peak-shifting scheduling strategy or supplementation of centralized liquid supply capacity.

[0110] Figure 4 This is a schematic diagram of a collaborative control system for end-point fluid units, comprehensively demonstrating the physical spatial layout and pipeline connection logic. The centralized supply station integrates a fluid supply source and pumps, with a collaborative controller positioned beside it. The system comprises three side-by-side cabinets, each housing four load devices, and each integrating an end-point fluid unit at its bottom, designated as the first, second, and third end-point fluid units. The centralized supply station is connected to the top and bottom of the three cabinets via two bidirectional circulation mains, forming a complete physical closed-loop system consisting of the supply station, circulation network, multi-cabinet load devices, and the underlying collaborative controller.

[0111] This step establishes a closed-loop feedback control system for the execution of the time-sharing peak-shaving scheduling strategy by constructing a real-time monitoring mechanism for the pressure stability index of the centralized liquid supply source and the pipeline vibration intensity index, a rapid retraining process for the graph attention neural network, and a dynamic update mechanism for the collaborative connection scheduling parameters. This system can promptly detect abnormal changes in the pipeline network's operating status and automatically correct the scheduling strategy. The incremental learning capability of the graph attention neural network enables the collaborative connection control method to continuously adapt to the dynamic evolution of operating conditions. The multi-index comprehensive monitoring and iterative convergence mechanism ensures that the pressure fluctuation of the centralized liquid supply source is always kept within a safe and permissible range, guaranteeing the long-term stable and reliable operation of the collaborative connection control mechanism between the distributed liquid consumption terminals and the centralized liquid supply source.

[0112] Example 2

[0113] The second embodiment of this application, as follows: Figure 5 As shown, a distributed liquid consumption terminal and a centralized liquid supply source collaborative connection control system is provided.

[0114] The system includes: a pipeline pressure monitoring module, a pressure interference characteristic analysis module, a flow limiting intelligent module, a peak-shaving scheduling control module, and a parameter update module.

[0115] The pipeline pressure monitoring module acquires the pressure data of the main outlet pipe of the centralized liquid supply source, the pressure data of the inlet branch pipes of each distributed liquid user terminal, and the vibration spectrum data of the connecting pipeline nodes, and establishes a dynamic monitoring dataset of pipeline pressure under the condition of multiple terminals in parallel operation.

[0116] The pressure interference feature analysis module extracts pressure fluctuation amplitude features, pressure propagation phase difference features, and vibration spectrum peak features from the pipeline pressure dynamic monitoring dataset. It uses a multi-dimensional feature correlation analysis method to identify the disturbance contribution of each terminal to the pressure field of the liquid supply source when extracting fluid, and generates a pressure interference weight coefficient vector for each terminal.

[0117] The intelligent flow limiting module constructs a pressure coupling relationship map between the terminal group and the centralized liquid supply source based on the pressure interference weight coefficient vector. It inputs the pressure coupling relationship map into a graph attention neural network for node feature aggregation and edge weight learning, and outputs the optimal flow limiting value of each terminal and liquid supply source connection branch at the current time.

[0118] The staggered peak scheduling control module designs a time-sharing staggered peak scheduling strategy based on the optimal flow limit value, staggers the fluid extraction time periods of each terminal, and uses a proportional reduction algorithm to reduce the instantaneous extraction flow for terminals with overlapping extraction time periods, limiting the total flow of the centralized liquid supply source outlet to not exceed the stable supply threshold at any time.

[0119] The parameter update module monitors the pressure stability index of the liquid supply source and the vibration intensity index of the pipeline during the execution of the time-sharing peak scheduling strategy. When the index exceeds the safety boundary, it triggers the graph attention neural network to quickly retrain and dynamically updates the collaborative connection scheduling parameters until the pressure fluctuation converges to the allowable range.

[0120] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is merely a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces. Indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms. The embodiments of this application have been described above in conjunction with specific embodiments, but this application is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art, under the guidance of this application, can make changes, modifications, substitutions, and variations to the above embodiments without departing from the spirit and scope of protection of the claims, and all of these fall within the protection scope of this application.

Claims

1. A method for coordinated connection and control of distributed liquid consumption terminals and centralized liquid supply sources, characterized in that, include: Acquire pressure data of the main outlet pipe of the centralized liquid supply source, pressure data of the inlet branch pipes of each distributed liquid consumption terminal, and vibration spectrum data of the connecting pipeline nodes to establish a dynamic monitoring dataset of pipeline pressure under multi-terminal parallel operation. The pressure fluctuation amplitude characteristics, pressure propagation phase difference characteristics, and vibration spectrum peak characteristics are extracted from the pipeline pressure dynamic monitoring dataset. The multidimensional feature correlation analysis method is used to identify the disturbance contribution of each terminal to the pressure field of the liquid supply source when extracting fluid, and to generate the pressure interference weight coefficient vector of each terminal. Based on the pressure interference weight coefficient vector, a pressure coupling relationship map between the terminal group and the centralized liquid supply source is constructed. The pressure coupling relationship map is input into the graph attention neural network for node feature aggregation and edge weight learning, and the optimal flow limit value of each terminal and liquid supply source connection branch is output at the current time. Based on the optimal flow limit value, a time-sharing peak scheduling strategy is designed to stagger the fluid extraction time periods of each terminal. For terminals with overlapping extraction time periods, a proportional reduction algorithm is used to reduce the instantaneous extraction flow rate, and the total flow rate at the outlet of the centralized liquid supply source is limited to not exceeding the stable supply threshold at any time. During the execution of the time-sharing peak-shaving scheduling strategy, the pressure stability index of the centralized liquid supply source and the pipeline vibration intensity index are monitored. When the index exceeds the safety boundary, the graph attention neural network is triggered to quickly retrain and dynamically update the collaborative connection scheduling parameters until the pressure fluctuation converges to the allowable range.

2. The method for coordinated connection and control of distributed liquid-using terminals and centralized liquid supply sources according to claim 1, characterized in that, A pressure transmitter is installed at the outlet main pipe of the centralized liquid supply source, and a pressure sensor is installed at the inlet branch pipe of each distributed liquid consumption terminal. Each pressure sensor and the pressure transmitter at the outlet main pipe of the liquid supply source are synchronized with each other using the same sampling clock. A piezoelectric accelerometer is installed at the node connecting the pipeline to collect vibration signals, and the vibration time-domain signal is subjected to a fast Fourier transform to obtain vibration spectrum data. The pipeline pressure dynamic monitoring dataset uses the sampling timestamp as the primary index key and includes a pressure value field and a vibration spectrum vector field. It is associated with and stores the opening status of fluid extraction valves at each terminal, the start / stop status of circulating pumps, and the fluid load demand level. The collected data is validated, including range and rate of change validation for pressure data and spectral energy and fundamental frequency offset validation for vibration spectrum data. Abnormal data is compensated and filled using interpolation of adjacent valid data points.

3. The method for coordinated connection and control of distributed liquid-using terminals and centralized liquid supply sources according to claim 2, characterized in that, The pressure fluctuation amplitude characteristics are obtained by calculating the difference between the maximum and minimum pressure values ​​within the window after processing by a sliding window segmentation. At the same time, the pressure standard deviation is calculated as an auxiliary measure of the severity of pressure fluctuations. The pressure propagation phase difference characteristics are calculated by cross-correlation analysis to determine the propagation delay between the pressure time series of the central liquid supply source outlet main pipe and the pressure time series of each terminal inlet branch pipe. The propagation delay is then combined with the main frequency of pressure fluctuation and converted into a phase difference representation. The main frequency of pressure fluctuation is obtained by performing a fast Fourier transform on the pressure time series of the centralized liquid supply source outlet main pipe and selecting the frequency component with the largest amplitude.

4. The method for coordinated connection and control of distributed liquid-using terminals and centralized liquid supply sources according to claim 3, characterized in that, The peak characteristics of the vibration spectrum include peak frequency and peak amplitude. By detecting peaks, local peaks in the vibration spectrum with amplitudes higher than adjacent frequency points are identified. A preset number of peaks are selected in descending order of peak amplitude to form a multi-peak feature vector. The pressure fluctuation amplitude, pressure propagation phase difference, and vibration spectrum peak amplitude are combined to form a multidimensional feature vector for each terminal. Multivariate regression analysis is used to determine the interference of each terminal feature on the pressure fluctuation of the centralized liquid supply source. Partial correlation analysis was used to eliminate the mutual interference between terminals, the perturbation correlation strength of each terminal was calculated, and the partial correlation coefficient and standardized regression coefficient were weighted and combined to obtain the perturbation contribution score. The disturbance contribution score is normalized to generate a pressure disturbance weight coefficient vector, and a moving average fusion strategy is used for dynamic updating.

5. The method for coordinated connection and control of distributed liquid-using terminals and centralized liquid supply sources according to claim 4, characterized in that, The pressure coupling relationship diagram adopts a directed weighted graph structure, with the centralized liquid supply source as the central node and each distributed liquid consumption terminal as the peripheral node. The direction of the directed edge from the liquid supply source node to each terminal node indicates the direction of fluid flow, and the coupling relationship between each terminal node is represented by the coupling association edge. The feature vector of the centralized liquid supply source node includes the outlet main pipe pressure value, pressure fluctuation amplitude, fluid level height, and supply capacity margin. The feature vector of each liquid terminal node includes the inlet branch pipe pressure value, pressure fluctuation amplitude, fluid load demand, and fluid extraction valve opening. The node feature vector and edge weights are updated according to the latest data of the pipeline pressure dynamic monitoring dataset.

6. The method for coordinated connection and control of distributed liquid-using terminals and centralized liquid supply sources according to claim 5, characterized in that, The weight of the main supply edge from the centralized liquid supply source to each terminal node is assigned using the pressure interference weight coefficient of the corresponding terminal, and the weight of the coupling association edge between each terminal node is calculated based on the product of the pressure interference weight coefficients of the two terminals. The graph attention neural network adopts a multi-layer stacked architecture, including an input layer, multiple graph attention layers, and an output layer. The graph attention layer uses an attention mechanism to weighted aggregate the features of neighboring nodes, and the edge weights are used as prior knowledge to participate in the calculation of attention coefficients. The output layer adopts a fully connected network structure to map the hidden feature vectors of each terminal node to the optimal traffic limit value. The output layer sets an activation function to constrain the range of the optimal traffic limit value.

7. The method for coordinated connection and control of distributed liquid-using terminals and centralized liquid supply sources according to claim 6, characterized in that, The time-slot staggered arrangement adopts a method that combines priority sorting and time slot allocation, and determines the terminal scheduling priority order based on the pressure interference weight coefficient; Fluid extraction time slots are allocated to each terminal in descending order of priority. The terminal currently allocated a time slot selects its fluid extraction time slot according to the principle of maximizing the time slot interval, and prioritizes time slots that do not overlap with those of terminals that have already completed their time slot allocation. The proportional reduction algorithm determines the reduction ratio allocation based on the pressure disturbance weight coefficient and the fluid load demand level. The flow reduction ratio of each terminal is positively correlated with the corresponding pressure disturbance weight coefficient. A reduction lower limit protection is set for terminals with high fluid load demand levels. The proportional reduction algorithm iteratively adjusts the reduction ratio until the total flow does not exceed the stable supply threshold.

8. The method for coordinated connection and control of distributed liquid-using terminals and centralized liquid supply sources according to claim 7, characterized in that, Compensation scheduling is carried out for the liquid supply gap caused by the reduced flow rate at each terminal after the reduction, and the lost fluid extraction volume is converted into additional extraction time periods and scheduled to be executed in non-overlapping or low-load periods. The compensation period is selected based on the available time period with the shortest time interval from the original reduction period. Flow sensors are installed at the inlet branch pipes of each terminal to collect the instantaneous fluid extraction flow rate. When the total flow rate at the outlet of the centralized liquid supply source is greater than or equal to the stable supply threshold, the flow limit control is triggered to adjust the opening of the fluid extraction valve. At the beginning of each scheduling period, the scheduling controller sends a scheduling instruction containing the allowed extraction status and the upper limit of the allowed extraction flow rate to the fluid extraction valve controller of each terminal. An emergency dispatch response mechanism is set up to temporarily transfer the fluid extraction time of non-urgent terminals to terminals whose incremental demand exceeds the set incremental threshold when the incremental demand of terminal fluid exceeds the set incremental threshold.

9. The method for coordinated connection and control of distributed liquid-using terminals and centralized liquid supply sources according to claim 8, characterized in that, The pressure stability index of the centralized liquid supply source is constructed by weighted fusion of pressure fluctuation amplitude index, pressure change rate index and pressure deviation index. The pipeline vibration intensity index is determined by combining frequency domain energy integral and time domain peak statistics and selecting the maximum value. The graph attention neural network fast retraining adopts an incremental learning approach, which fine-tunes the parameters based on the existing network parameters using newly collected running data. When the index exceeds the safety boundary, the optimal traffic limit value label is adjusted downward and the adjustment magnitude is proportional to the degree of exceeding the limit. The loss function adds a safety boundary constraint penalty term. The collaborative connection scheduling parameter update adopts a gradual switching approach, gradually adjusting the upper limit of allowed extraction traffic during the transition time. A maximum retraining iteration limit is set, and when the limit is exceeded, it switches to a conservative running mode. A log recording mechanism is established to store retraining trigger information and index convergence status.

10. A distributed liquid consumption terminal and centralized liquid supply source coordinated connection control system, used to implement the distributed liquid consumption terminal and centralized liquid supply source coordinated connection control method according to any one of claims 1 to 9, characterized in that, include: Pipeline pressure monitoring module, pressure interference characteristic analysis module, flow limiting intelligent module, peak-shaving scheduling control module, and parameter update module; The pipeline pressure monitoring module acquires the pressure data of the main outlet pipe of the centralized liquid supply source, the pressure data of the inlet branch pipe of each distributed liquid user terminal, and the vibration spectrum data of the connecting pipeline nodes, and establishes a dynamic monitoring dataset of pipeline pressure under the condition of multiple terminals in parallel operation. The pressure interference feature analysis module extracts pressure fluctuation amplitude features, pressure propagation phase difference features, and vibration spectrum peak features from the pipeline pressure dynamic monitoring dataset. It uses a multi-dimensional feature correlation analysis method to identify the disturbance contribution of each terminal to the pressure field of the liquid supply source when extracting fluid, and generates a pressure interference weight coefficient vector for each terminal. The intelligent flow limiting module constructs a pressure coupling relationship map between the terminal group and the centralized liquid supply source based on the pressure interference weight coefficient vector. It inputs the pressure coupling relationship map into a graph attention neural network for node feature aggregation and edge weight learning, and outputs the optimal flow limiting value of each terminal and liquid supply source connection branch at the current time. The staggered peak scheduling control module designs a time-sharing staggered peak scheduling strategy based on the optimal flow limit value, staggers the fluid extraction time periods of each terminal, and uses a proportional reduction algorithm to reduce the instantaneous extraction flow for terminals with overlapping extraction time periods, limiting the total flow of the centralized liquid supply source outlet to not exceed the stable supply threshold at any time. The parameter update module monitors the pressure stability index of the liquid supply source and the vibration intensity index of the pipeline during the execution of the time-sharing peak scheduling strategy. When the index exceeds the safety boundary, it triggers the graph attention neural network to quickly retrain and dynamically updates the collaborative connection scheduling parameters until the pressure fluctuation converges to the allowable range.

Images