Water supply network dynamic pressure control method and system based on multi-parameter fusion

Abstract

The invention discloses a water supply pipe network dynamic pressure control method and system based on multi-parameter fusion, and relates to the technical field of water supply pipe network control, and the method comprises the steps: carrying out the standardization processing and topological optimization of pipe network geographic information system data, and constructing an urban pipe network topological graph; identifying a pressure sensitive node and a distance sensitive node based on a hydraulic calculation result of the mechanism calculation model; performing association coding on historical monitoring data of the key nodes and a topological connection relationship, and training a machine learning model fusing pipe network topological constraints; collecting real-time monitoring data, inputting the real-time monitoring data into the mechanism calculation model and the machine learning model, and respectively obtaining offline adjustment parameters and online adjustment parameters; and determining a final adjusting instruction through consistency verification to control water supply pump set equipment. According to the method, the calculation precision of the mechanism model and the response speed of the machine learning model are fused, organic unification of accuracy and real-time performance is achieved, and the control stability and reliability under the abnormal working condition are improved.

Description

Technical Field

[0001] This invention relates to the field of water supply network control technology, specifically to a dynamic pressure control method and system for water supply networks based on multi-parameter fusion. Background Technology

[0002] In recent years, with the deepening of urban development and construction and the adjustment of regional industrial structure, the water demand of urban and industrial users has changed rapidly. As an important component of urban infrastructure, urban water supply systems were constructed before large-scale urban development, including water source intake projects, urban water treatment plants, and urban water supply networks. Conventional design follows relevant design manuals and uses recommended experience values ​​to estimate near-term and long-term plans, aiming to reduce water consumption discrepancies caused by inconsistencies over time. However, recent changes in industrial structure and residents' lifestyles (such as public emergencies, age structure adjustments due to population aging, and work patterns) have brought problems to the actual water supply system. Operating according to the design parameters of traditional water supply systems often results in insufficient water pressure during high-flow periods, leading to water restrictions for some end-users, and excessive water pressure during low-flow periods, causing leaks and pipe bursts, resulting in economic losses.

[0003] To address the aforementioned issues, the industry primarily employs theoretical mechanistic models for calculations and machine learning models for data prediction. The results are then manually verified and fed back to the mechanical pump units for adjustment. The drawbacks of this process are that mechanistic model calculations are time-consuming, hindering timely adjustments to operating conditions; and machine learning models, detached from the actual topology of the pipeline network, rely solely on numerical training and fitting to obtain mapping relationships, making it difficult to guarantee accuracy in predicting unforeseen events. Therefore, there is an urgent need for a dynamic pressure control method for water supply networks that integrates the advantages of both mechanistic and machine learning models, ensuring both computational accuracy and rapid response, and enabling the scientific selection of key monitoring nodes. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for dynamic pressure control of water supply networks based on multi-parameter fusion, aiming to solve at least one of the technical problems existing in the prior art.

[0005] The technical solution of this invention is: a dynamic pressure control method for water supply networks based on multi-parameter fusion, comprising the following steps: Data from the geographic information system of the water supply network is acquired and standardized to generate effective network data. The effective network data is then processed by pipe segment merging and end-point pruning to construct an urban network topology map. Input the urban pipeline network topology map into the mechanism calculation model to perform hydraulic calculations and identify pressure-sensitive nodes and distance-sensitive nodes. Historical monitoring data of pressure-sensitive nodes and distance-sensitive nodes are extracted, and the historical monitoring data is associated with the topological connection relationship of the corresponding nodes in the urban pipeline network topology map. A machine learning model is trained to establish a predictive mapping relationship that integrates pipeline network topology constraints. Real-time monitoring data of pressure-sensitive nodes and distance-sensitive nodes are collected, and the real-time monitoring data is input in parallel into the mechanism calculation model and the machine learning model to obtain the pump set offline adjustment parameters and the pump set online adjustment parameters, respectively. The consistency of the pump set offline adjustment parameters and the pump set online adjustment parameters is verified. Based on the verification results, the pump set adjustment command is generated, and the water supply pump set equipment is adjusted according to the pump set adjustment command.

[0006] The process involves acquiring and standardizing geographic information system (GIS) data of the water supply area's pipeline network to generate effective pipeline network data. This effective network data is then subjected to pipe segment merging and end-point pruning to construct an urban pipeline network topology map. The pipeline geographic information system data includes pipeline node coordinate information, pipe segment connection relationship information, and pipe diameter parameter information; Based on the pipe segment connection rules and hydraulic physical constraint rules, the connectivity index of each pipe segment in the geographic information system data of the water supply area is calculated, and the pipe segment data that meets the hydraulic calculation requirements is selected to form effective pipe network data; Based on the topological connection relationship of pipe segments in the effective pipe network data, a traversal operation is performed to calculate the difference in pipe diameter parameters between adjacent pipe segments. Pipe segment groups with zero difference in pipe diameter parameters and physical continuity are marked as pipe segment groups to be merged. Accumulate the length parameters of each pipe segment in the pipe segment group to be merged while keeping the positions of the first and last nodes unchanged to complete the pipe segment merging process; Hydraulic distribution calculations are performed on the terminal pipe segments in the effective pipe network data after pipe segment merging. The flow rate change and pressure change of each terminal pipe segment during the overall water supply process are obtained. The weighted sum of the flow rate change and pressure change is used as the impact index of the terminal pipe segment. Terminal pipe segments with impact index less than a preset threshold are deleted to complete the terminal pruning process and construct the urban pipe network topology map.

[0007] The urban pipeline network topology map is input into the mechanism calculation model for hydraulic calculations to identify pressure-sensitive and distance-sensitive nodes, including: A mechanism calculation model is constructed, which includes a node flow balance equation and a pipe section head loss equation. The node flow balance equation is used to calculate the node flow conservation value, and the pipe section head loss equation is used to calculate the pipe section pressure loss value. Extract the network structure information from the urban network topology map, and substitute the network structure information into the node flow balance equation and the pipe section head loss equation to generate a set of network hydraulic calculation equations. Solve the set of hydraulic calculation equations for the pipeline network to obtain the node pressure sequence. Calculate the node pressure change amplitude based on the node pressure sequence and mark nodes whose pressure change amplitude exceeds a preset range as pressure-sensitive nodes. Calculate the path length from each node to the pressure-sensitive node in the urban pipeline network topology map, and mark the nodes whose path length is less than a preset length value as distance-sensitive nodes.

[0008] Historical monitoring data of pressure-sensitive and distance-sensitive nodes are extracted. This historical monitoring data is then correlated with the topological connections of corresponding nodes in the urban pipeline network topology map. A machine learning model is trained to train this model, establishing a predictive mapping relationship that integrates pipeline network topology constraints. Extract historical pressure data and historical flow data from pressure-sensitive nodes and distance-sensitive nodes, and align them according to timestamps to form historical monitoring data; The topological connection relationship between nodes is extracted from the urban pipeline network topology map, the shortest connection distance between nodes is calculated, the shortest connection distance is converted into a location code, and the location code is combined with historical monitoring data to generate training data. Based on the topological connection relationship, the network nodes are traversed to identify node pairs with direct pipe segment connections, the pressure difference sequence between node pairs is calculated, and the node pressure attenuation function is constructed. The node pressure decay function is added as a loss term to the machine learning model. The prediction error of the training dataset is calculated. Training is stopped when the prediction error is less than a preset error threshold, and the trained machine learning model is obtained. The location code is input into the trained machine learning model, and the output is the node pressure prediction value. A mapping relationship from the location code to the pressure prediction value is established. The mapping relationship is used to predict the pressure value of pressure-sensitive nodes and distance-sensitive nodes.

[0009] Real-time monitoring data of pressure-sensitive nodes and distance-sensitive nodes are collected, and the real-time monitoring data is input in parallel into the mechanism calculation model and the machine learning model to obtain the pump set offline adjustment parameters and the pump set online adjustment parameters, respectively: Real-time pressure data and real-time flow data of pressure-sensitive nodes and distance-sensitive nodes are collected, and the real-time pressure data and real-time flow data are time-series aligned and numerically verified to generate real-time monitoring data. The water load characteristics of the water supply network at different times are obtained. Based on the water load characteristics, the real-time monitoring data is divided into steady-state operation data and fluctuating operation data. The steady-state operation data is input into the mechanism calculation model for offline calculation to obtain the pressure distribution of the network nodes and the flow distribution of the pipe segments. Based on the pressure distribution and flow distribution, the pump set offline adjustment parameters are generated. The fluctuating operating data is input into a machine learning model for online prediction to obtain node pressure prediction values ​​and pipe segment flow prediction values. Based on the pressure prediction values ​​and flow prediction values, online adjustment parameters for the pump set are generated.

[0010] The consistency verification of the pump set's offline and online adjustment parameters includes: Read the pump set offline adjustment parameters from the offline control command set, read the pump set online adjustment parameters from the online control command set, calculate the pressure adjustment difference and flow adjustment difference between the pump set offline adjustment parameters and the pump set online adjustment parameters, and generate adjustment difference data; The pressure and flow rate changes of the water supply pump set before and after adjustment are calculated based on the adjustment difference data, and the pressure and flow rate changes are used as water supply status change data. Determine whether the changes in water supply status will lead to insufficient water pressure or excessive pressure in the water supply system, and generate a water supply status verification result. The water supply status verification results are compared with the historical operation data of the water supply network to determine whether the water supply status verification results meet the water demand change pattern, and the demand matching results are generated as the consistency verification results.

[0011] The consistency of the pump set's offline and online adjustment parameters is verified, and pump set adjustment commands are generated based on the verification results, including: A verification pass flag is generated when the consistency verification result meets the operating requirements of the water supply system; a verification failure flag is generated when the consistency verification result does not meet the operating requirements of the water supply system. When the verification is successful, the pump set online adjustment parameters are written into the control command cache, and the pump set adjustment command is generated and sent to the water supply pump set based on the pressure adjustment and flow adjustment in the pump set online adjustment parameters. When verification fails, the pump set offline adjustment parameters are written into the control command cache. Based on the pressure adjustment and flow adjustment in the pump set offline adjustment parameters, a pump set adjustment command is generated and sent to the water supply pump set. At the same time, the real-time monitoring data when verification fails is used as training samples to input into the machine learning model for incremental training.

[0012] Adjusting the water supply pump set equipment according to the pump set adjustment command includes: Parse the pump set adjustment command, extract the pressure adjustment target value and flow adjustment target value from the pump set adjustment command, and convert them into pump set operation control parameters; The required output pressure and flow rate of the water supply pump set are calculated based on the pump set operation control parameters. The number of pumps to be put into operation and the speed of each pump are determined according to the output pressure and flow rate. Pump adjustment control commands are generated and sent to the water supply pump set to perform pump set adjustment operations. Collect the pipeline pressure and flow values ​​after the pump set adjustment operation is performed, compare the pipeline pressure and flow values ​​with the pressure adjustment target values ​​and flow adjustment target values, and generate adjustment effect evaluation data; The aforementioned adjustment effect evaluation data will be used as new real-time monitoring data for the next round of dynamic pressure control of the water supply network.

[0013] This invention provides a dynamic pressure control system for water supply networks based on multi-parameter fusion, the system comprising: The data acquisition module is used to acquire geographic information system data of the water supply area's pipe network and perform standardized processing to generate effective pipe network data. The effective pipe network data is then processed by pipe segment merging and end-point pruning to construct an urban pipe network topology map. The node identification module is used to input the urban pipeline network topology map into the mechanism calculation model for hydraulic calculation and to identify pressure-sensitive nodes and distance-sensitive nodes. The model training module is used to extract historical monitoring data of pressure-sensitive nodes and distance-sensitive nodes, associate the historical monitoring data with the topological connection relationship of the corresponding nodes in the urban pipeline network topology map, train the machine learning model, and establish a predictive mapping relationship that integrates pipeline network topology constraints. The parameter calculation module is used to collect real-time monitoring data of pressure-sensitive nodes and distance-sensitive nodes, and input the real-time monitoring data into the mechanism calculation model and the machine learning model in parallel to obtain the pump set offline adjustment parameters and the pump set online adjustment parameters, respectively. The adjustment and control module is used to verify the consistency between the offline and online adjustment parameters of the pump set, generate pump set adjustment commands based on the verification results, and adjust the water supply pump set equipment according to the pump set adjustment commands.

[0014] One technical solution provided in this embodiment of the invention is an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0015] This invention achieves rapid and accurate regulation of water supply pressure by integrating a machine learning model based on pipeline topology constraints with a mechanistic calculation model, balancing computational accuracy and response speed. Key monitoring nodes are scientifically identified based on hydraulic calculations, improving the representativeness of monitoring data and control effectiveness. A dual-model consistency verification mechanism enhances control stability under abnormal operating conditions; when the machine learning model's predictions deviate, it automatically switches to the mechanistic model and triggers model optimization, ensuring system reliability and adaptability. Pipeline merging and end-point pruning optimization reduce computational complexity and improve the real-time control performance of large-scale pipeline networks. Attached Figure Description

[0016] Figure 1 A flowchart of a dynamic pressure control method for water supply networks based on multi-parameter fusion provided in an embodiment of the present invention; Figure 2 This is a diagram of the urban pipeline network topology according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a dynamic pressure control system for water supply networks based on multi-parameter fusion, provided in an embodiment of the present invention. Detailed Implementation

[0017] like Figure 1 As shown, Figure 1 A flowchart of a dynamic pressure control method for water supply networks based on multi-parameter fusion provided in an embodiment of the present invention is shown. The method includes the following steps: Data from the geographic information system of the water supply network is acquired and standardized to generate effective network data. The effective network data is then processed by pipe segment merging and end-point pruning to construct an urban network topology map. Input the urban pipeline network topology map into the mechanism calculation model to perform hydraulic calculations and identify pressure-sensitive nodes and distance-sensitive nodes. Historical monitoring data of pressure-sensitive nodes and distance-sensitive nodes are extracted, and the historical monitoring data is associated with the topological connection relationship of the corresponding nodes in the urban pipeline network topology map. A machine learning model is trained to establish a predictive mapping relationship that integrates pipeline network topology constraints. Real-time monitoring data of pressure-sensitive nodes and distance-sensitive nodes are collected, and the real-time monitoring data is input in parallel into the mechanism calculation model and the machine learning model to obtain the pump set offline adjustment parameters and the pump set online adjustment parameters, respectively. The consistency of the pump set offline adjustment parameters and the pump set online adjustment parameters is verified. Based on the verification results, the pump set adjustment command is generated, and the water supply pump set equipment is adjusted according to the pump set adjustment command.

[0018] The process involves acquiring and standardizing geographic information system (GIS) data of the water supply area's pipeline network to generate effective pipeline network data. This effective network data is then subjected to pipe segment merging and end-point pruning to construct an urban pipeline network topology map. The pipeline geographic information system data includes pipeline node coordinate information, pipe segment connection relationship information, and pipe diameter parameter information; Based on the pipe segment connection rules and hydraulic physical constraint rules, the connectivity index of each pipe segment in the geographic information system data of the water supply area is calculated, and the pipe segment data that meets the hydraulic calculation requirements is selected to form effective pipe network data; Based on the topological connection relationship of pipe segments in the effective pipe network data, a traversal operation is performed to calculate the difference in pipe diameter parameters between adjacent pipe segments. Pipe segment groups with zero difference in pipe diameter parameters and physical continuity are marked as pipe segment groups to be merged. Accumulate the length parameters of each pipe segment in the pipe segment group to be merged while keeping the positions of the first and last nodes unchanged to complete the pipe segment merging process; Hydraulic distribution calculations are performed on the terminal pipe segments in the effective pipe network data after pipe segment merging. The flow rate change and pressure change of each terminal pipe segment during the overall water supply process are obtained. The weighted sum of the flow rate change and pressure change is used as the impact index of the terminal pipe segment. Terminal pipe segments with impact index less than a preset threshold are deleted to complete the terminal pruning process and construct the urban pipe network topology map.

[0019] Acquire geographic information system (GIS) data of the water supply network in the area. This data includes network node coordinates, pipe segment connection relationships, and pipe diameter parameters. Network node coordinates include the Cartesian coordinates of each node; pipe segment connection relationships include the start and end node numbers of each segment; and pipe diameter parameters include the inner diameter of each segment. Standardize the acquired raw network data, unifying the coordinates to a single coordinate system. Check and correct outliers and missing values ​​to ensure data integrity and consistency.

[0020] The connectivity index of each pipe segment in the geographic information system data of the water supply area network is calculated based on the pipe segment connection rules and hydraulic physical constraint rules. The pipe segment connection rules stipulate that pipe segments must be connected through nodes, and suspended pipe segments are not allowed. The hydraulic physical constraint rules stipulate that water flow in the network must follow the principles of mass and energy conservation. The connectivity index calculation considers the accessibility of the pipe segment's path, the importance of the pipe segment's location in the network, and the rationality of the pipe segment's hydraulic parameters. If the upstream node of a pipe segment cannot be traced from the water source, or if the pipe diameter of the segment is too small to meet the minimum flow requirement, the connectivity index of that pipe segment is zero; otherwise, the connectivity index ranges from 0 to 1, with a larger value indicating higher connectivity importance of the pipe segment in the network.

[0021] Pipe segment data that meets the connectivity requirements for hydraulic calculations are selected to form valid pipe network data. A connectivity threshold of 0.3 is set; pipe segments with a connectivity index greater than or equal to 0.3 are retained, while those less than 0.3 are discarded. This process excludes isolated, dead-end, and hydraulically incompatible pipe segments, ensuring that the resulting valid pipe network data can support subsequent hydraulic calculations and topology analysis.

[0022] Based on the topological connections of pipe segments in the effective pipe network data, a traversal operation is performed to calculate the diameter parameter difference between adjacent pipe segments. A depth-first search algorithm is used to traverse the pipe network and identify all adjacent connected pipe segment pairs. For each pair of adjacent pipe segments, their diameter difference is calculated. Pipe segment groups with zero diameter parameter difference and physical continuity are marked as pipe segment groups to be merged. Physical continuity means that the two pipe segments share a node, and that node has no other connected pipe segments besides these two segments. For example, in a water supply area, there are pipe segments A and B, which are connected through node N, and both segments have a diameter of 200 mm. There are no other connected pipe segments at node N. In this case, pipe segments A and B are marked as a pipe segment group to be merged.

[0023] The pipe segment merging process is performed on the pipe segment groups to be merged. The length parameters of each pipe segment in the group to be merged are accumulated, while keeping the positions of the first and last nodes unchanged, thus completing the pipe segment merging process. The length of the merged pipe segment is the sum of the lengths of all pipe segments in the original pipe segment group, the pipe diameter remains unchanged, the starting node is the starting node of the original pipe segment group, the ending node is the ending node of the original pipe segment group, and intermediate nodes are deleted. For example, in the previous example, pipe segment A is 100 meters long, pipe segment B is 150 meters long, and after merging, a new pipe segment C is formed with a length of 250 meters and a diameter of 200 millimeters. The starting node is the starting node of pipe segment A, and the ending node is the ending node of pipe segment B.

[0024] Hydraulic distribution calculations are performed on the terminal pipe segments in the effective pipe network data after segment merging. A terminal pipe segment is defined as a segment connected to other pipe segments at only one end. The hydraulic distribution calculation includes simulation analysis of the flow and pressure distributions within the pipe network. Based on the nodal hydraulic calculation principle, nodal flow balance equations and loop pressure balance equations are constructed. Iterative calculations are then used to obtain the pressure values ​​at each node and the flow values ​​for each pipe segment in the network. The flow and pressure changes of each terminal pipe segment during the overall water supply process are obtained and denoted as the flow influence coefficient and pressure influence coefficient, respectively.

[0025] The weighted sum of flow rate and pressure changes is used as the impact index for terminal pipe sections. The flow rate impact weight is set to 0.6, and the pressure impact weight is set to 0.4. The impact index of a certain terminal pipe section is equal to its flow rate impact coefficient × 0.6 + pressure impact coefficient × 0.4. An impact index threshold of 0.25 is set, and terminal pipe sections with an impact index less than 0.25 are deleted, completing the terminal pruning process. Terminal pruning removes terminal pipe sections with minimal impact on the overall water supply, simplifying the pipe network structure and reducing computational complexity.

[0026] Final construction as Figure 2The diagram shows the urban pipeline network topology. The pipeline data, after connectivity filtering, segment merging, and end-point pruning, is converted into a topology graph structure. Nodes in the graph represent pipeline intersections or user access points, and edges represent pipeline segments. The weight of each edge can be set as the segment length, pipe diameter, or hydraulic loss coefficient. The completed topology graph visually demonstrates the connection structure and hydraulic characteristics of the water supply network, providing a foundation for subsequent pressure control strategies.

[0027] This invention simplifies and optimizes complex water supply networks by acquiring and standardizing data from a geographic information system (GIS) of the pipeline network, selecting effective pipe segments based on connectivity indicators, optimizing the network structure through segment merging and end-point pruning, and ultimately constructing a topology map of the urban pipeline network. This effectively reduces computational burden, improves the accuracy and efficiency of hydraulic simulation, and provides a reliable topological foundation for dynamic pressure control of the water supply network. By properly processing the pipeline network data, singularity issues in hydraulic calculations can be avoided, enhancing the stability of the water supply system.

[0028] The urban pipeline network topology map is input into the mechanism calculation model for hydraulic calculations to identify pressure-sensitive and distance-sensitive nodes, including: A mechanism calculation model is constructed, which includes a node flow balance equation and a pipe section head loss equation. The node flow balance equation is used to calculate the node flow conservation value, and the pipe section head loss equation is used to calculate the pipe section pressure loss value. Extract the network structure information from the urban network topology map, and substitute the network structure information into the node flow balance equation and the pipe section head loss equation to generate a set of network hydraulic calculation equations. Solve the set of hydraulic calculation equations for the pipeline network to obtain the node pressure sequence. Calculate the node pressure change amplitude based on the node pressure sequence and mark nodes whose pressure change amplitude exceeds a preset range as pressure-sensitive nodes. Calculate the path length from each node to the pressure-sensitive node in the urban pipeline network topology map, and mark the nodes whose path length is less than a preset length value as distance-sensitive nodes.

[0029] A mechanistic calculation model is constructed, which includes nodal flow balance equations and pipe segment head loss equations. The nodal flow balance equations, based on the principle of mass conservation, state that the total flow into each node equals the total flow out of each node. For any non-source node in the water supply network, the algebraic sum of the flow rates of all pipe segments connecting that node equals the water demand of that node; the flow rates into the node are positive, and the flow rates out of the node are negative. The pipe segment head loss equations, based on the principle of energy conservation, state that the head difference between the nodes at both ends of a pipe segment equals the head loss of that segment. Head loss is related to flow rate, pipe segment length, pipe diameter, and roughness coefficient; the Heize-Williams formula is generally used to calculate head loss.

[0030] The network structure information, including the node connection matrix and the pipe segment characteristic matrix, is extracted from the urban pipeline network topology map. The node connection matrix records the connection status between nodes; if node i and node j are directly connected via a pipe segment, the corresponding element in the matrix has a value of 1, otherwise it has a value of 0. The pipe segment characteristic matrix records the physical parameters of each pipe segment, such as pipe length, pipe diameter, and roughness coefficient. The extracted network structure information is then substituted into the node flow balance equation and the pipe segment head loss equation to establish corresponding equations for each node and each pipe segment in the network, generating a set of hydraulic calculation equations for the pipeline network.

[0031] For a water supply network consisting of 100 nodes and 120 pipe segments, 100 node flow balance equations and 120 pipe segment head loss equations can be established. Since the number of unknowns in the equation system equals the total number of node pressures and pipe segment flow rates, some known conditions are required for the solution. Typically, the pressure of the source nodes is assumed to be known, such as a source node pressure of 60 meters of water column for a given water supply network. The water demand of the demand nodes in the network is also considered known, such as node 1 demanding 2 L / s and node 2 demanding 1.5 L / s.

[0032] A global gradient method is used to solve the hydraulic equations for the pipe network. Initial pressure values ​​for each node are assumed, such as 50 meters of water column. Then, the node pressures and pipe flow rates are iteratively adjusted until convergence conditions are met. Each iteration includes three steps: calculating the pipe flow rate based on the current node pressure; checking the flow balance at each node based on the flow rate; and adjusting the node pressure based on the flow imbalance. The equations are considered converged when the flow imbalance at all nodes is less than a preset threshold, such as 0.001 L / s. The solved equations yield the pressure values ​​at each node and the flow rates for each pipe segment, forming a node pressure sequence and a pipe flow rate sequence.

[0033] The amplitude of node pressure change is calculated based on node pressure sequences. The amplitude of node pressure change is defined as the difference between the maximum and minimum pressure values ​​of the node under different water demand conditions. Water demand data for different times of a typical day are collected, such as 3:00 AM, 7:00 AM, 12:00 PM, 6:00 PM, and 10:00 PM. Hydraulic calculations are performed to obtain the pressure values ​​of each node at different times. The maximum and minimum pressure values ​​for each node during these times are calculated, and the difference between the two is the amplitude of the node's pressure change. A preset amplitude range of 10 meters of water column is set, and nodes with pressure change amplitudes exceeding 10 meters of water column are marked as pressure-sensitive nodes. For example, in a pipeline network, the pressure change amplitude of node 10 is 15 meters of water column, and the pressure change amplitude of node 20 is 8 meters of water column. Node 10 is marked as a pressure-sensitive node, while node 20 is not marked.

[0034] Calculate the path length from each node to a pressure-sensitive node in the urban pipeline network topology map. Path length is defined as the total length of the pipe segments traversed from one node to another. Dijkstra's algorithm is used to calculate the shortest path length. This algorithm first initializes the distances of all nodes to infinity, sets the distance of the starting node to 0, and then finds the shortest path length from the starting node to all other nodes by continuously updating and comparing. For each pressure-sensitive node, calculate the shortest path length from all other nodes in the pipeline network to that pressure-sensitive node. Set a preset length value of 500 meters, and mark nodes with path lengths less than 500 meters as distance-sensitive nodes. For example, if the path length from a node to a pressure-sensitive node is 450 meters, then that node is marked as a distance-sensitive node.

[0035] This invention performs hydraulic calculations by inputting the urban pipeline network topology map into a mechanistic calculation model, identifying pressure-sensitive and distance-sensitive nodes, and providing key reference points for dynamic pressure control of the water supply network. It fully considers the topology and hydraulic characteristics of the water supply network, accurately identifying areas with significant pressure fluctuations and wide-ranging influence within the network. By analyzing the pressure change patterns of nodes and the connections between nodes, precise location of key control nodes is achieved. The distribution of sensitive nodes can be dynamically adjusted according to changes in the pipeline network, laying the foundation for intelligent operation and management of the water supply network.

[0036] Historical monitoring data of pressure-sensitive and distance-sensitive nodes are extracted. This historical monitoring data is then correlated with the topological connections of corresponding nodes in the urban pipeline network topology map. A machine learning model is trained to train this model, establishing a predictive mapping relationship that integrates pipeline network topology constraints. Extract historical pressure data and historical flow data from pressure-sensitive nodes and distance-sensitive nodes, and align them according to timestamps to form historical monitoring data; The topological connection relationship between nodes is extracted from the urban pipeline network topology map, the shortest connection distance between nodes is calculated, the shortest connection distance is converted into a location code, and the location code is combined with historical monitoring data to generate training data. Based on the topological connection relationship, the network nodes are traversed to identify node pairs with direct pipe segment connections, the pressure difference sequence between node pairs is calculated, and the node pressure attenuation function is constructed. The node pressure decay function is added as a loss term to the machine learning model. The prediction error of the training dataset is calculated. Training is stopped when the prediction error is less than a preset error threshold, and the trained machine learning model is obtained. The location code is input into the trained machine learning model, and the output is the node pressure prediction value. A mapping relationship from the location code to the pressure prediction value is established. The mapping relationship is used to predict the pressure value of pressure-sensitive nodes and distance-sensitive nodes.

[0037] Historical pressure and flow data for pressure-sensitive and distance-sensitive nodes are extracted. The historical data is obtained from the pipeline monitoring system database and includes node number, timestamp, pressure value, and flow rate value. The unit for historical pressure data is meters of water column, and the unit for historical flow data is liters per second. The data extraction period is the past three months, and the sampling frequency is once per hour. The historical pressure and flow data are aligned according to the timestamp to form historical monitoring data. The format of historical monitoring data is: node number, timestamp, pressure value, and flow rate value. For a specific pressure-sensitive node, its historical monitoring data might include: Node 25, May 1, 2023, 08:00:00, 45.6 meters of water column, 2.3 L / s.

[0038] The topological connections between nodes are extracted from the urban pipeline network topology map. The urban pipeline network topology map contains node information and pipeline segment information. The topological connections between nodes can be obtained through the starting and ending nodes of the pipeline segments. An adjacency matrix is ​​constructed to represent the connection status between nodes. The rows and columns of the matrix represent node numbers. If there is a direct pipeline connection between node i and node j, the element at the corresponding position in the matrix has a value of 1; otherwise, it has a value of 0. The shortest connection distance between nodes is calculated. A breadth-first search algorithm is used to traverse the adjacency matrix, calculating the minimum number of pipeline segments traversed between any two nodes, resulting in the topological distance matrix between nodes.

[0039] The shortest connection distance is converted into a location code. Location coding is a method that converts topological distance information into a vector representation, facilitating processing by machine learning models. A sine-cosine coding method is used to convert the topological distance *d* into a location code vector of length 10. The coding process is as follows: For the topological distance *d*, calculate the sine and cosine values ​​of *d* at different frequencies, with frequencies of 1 / 100, 1 / 1000, 1 / 100000, 1 / 1000000, and 1 / 1000000. The obtained sine and cosine values ​​are arranged in order to form the location code vector. The location code is combined with historical monitoring data to generate training data. The training data includes: node number, timestamp, pressure value, flow value, and location code vector.

[0040] The network nodes are traversed based on topological connections to identify node pairs with direct pipe segment connections. The adjacency matrix is ​​traversed to find all positions where each element is 1; the corresponding row and column indices represent the directly connected node pairs. The pressure difference sequence between node pairs is calculated. For each directly connected node i and node j, their pressure values ​​at the same timestamp are extracted, and the pressure difference is calculated to form a pressure difference sequence. The pressure difference equals the pressure value of node i minus the pressure value of node j. Statistical analysis is performed on the pressure difference sequence for each node pair, calculating the mean, standard deviation, and distribution characteristics. A node pressure decay function is constructed. The node pressure decay function describes the pressure relationship between adjacent nodes and can be represented by a linear model. The pressure decay parameters for each pair of adjacent nodes are obtained by fitting the pressure difference sequence using the least squares method.

[0041] The node pressure decay function is incorporated as a loss term into the machine learning model. The machine learning model employs a Long Short-Term Memory (LSTM) network structure, comprising one input layer, two LSM layers, and one output layer. The input layer receives location encoding and historical pressure-flow data, the LSM layers process temporal features, and the output layer predicts future pressure values. The model's loss function is defined, consisting of a prediction error term and a topology constraint term. The prediction error term is the squared difference between the predicted and actual pressure values, and the topology constraint term is the squared difference between the difference in predicted pressure values ​​between adjacent nodes and the value calculated by the pressure decay function. The loss function is expressed as: total loss equals prediction error term + topology constraint term × weight factor, with the weight factor set to 0.5. The model parameters are updated by minimizing the loss function using the stochastic gradient descent algorithm.

[0042] The training process employs batch training, with each batch containing 32 data points and 100 training epochs. After each training epoch, the model performance is evaluated using a validation set, and the prediction error on the validation set is calculated. Training stops when the relative rate of change of the prediction error on the validation set is less than 0.001 after 5 consecutive training epochs, or when the maximum number of training epochs is reached. The prediction error is calculated as the mean absolute error between the predicted and actual values, with a preset error threshold of 0.5 meters of water column. Training stops when the prediction error falls below the preset error threshold, resulting in the completed machine learning model.

[0043] The location code is input into the trained machine learning model, which outputs a node stress prediction value. For any node, based on its position in the topology graph, the shortest connection distance from that node to all stress-sensitive nodes and distance-sensitive nodes is calculated, generating a location code. This location code is then input into the trained machine learning model to obtain the node's stress prediction value at future times. A mapping relationship from location code to stress prediction value is established, which is used to predict the stress values ​​of stress-sensitive nodes and distance-sensitive nodes. This mapping relationship can be expressed as: given a node's location code, the trained machine learning model can predict the node's stress value at any future time.

[0044] This invention achieves high-precision prediction of water supply network node pressure by extracting historical monitoring data from pressure-sensitive and distance-sensitive nodes and combining it with urban pipeline network topology information. Through location coding technology, it successfully captures the spatial correlation between network nodes, solving the problem of traditional prediction methods struggling to handle complex topologies. Based on a deep learning architecture using long short-term memory networks, it effectively mines the temporal characteristics of pressure and flow data, achieving accurate characterization of dynamic changes in the pipeline network. It is not only applicable to pressure prediction of monitored nodes but can also infer the pressure status of unmonitored nodes through topological relationships, significantly reducing the deployment cost of monitoring equipment and providing reliable data support for intelligent dynamic pressure control of water supply networks.

[0045] Real-time monitoring data of pressure-sensitive nodes and distance-sensitive nodes are collected, and the real-time monitoring data is input in parallel into the mechanism calculation model and the machine learning model to obtain the pump set offline adjustment parameters and the pump set online adjustment parameters, respectively: Real-time pressure data and real-time flow data of pressure-sensitive nodes and distance-sensitive nodes are collected, and the real-time pressure data and real-time flow data are time-series aligned and numerically verified to generate real-time monitoring data. The water load characteristics of the water supply network at different times are obtained. Based on the water load characteristics, the real-time monitoring data is divided into steady-state operation data and fluctuating operation data. The steady-state operation data is input into the mechanism calculation model for offline calculation to obtain the pressure distribution of the network nodes and the flow distribution of the pipe segments. Based on the pressure distribution and flow distribution, the pump set offline adjustment parameters are generated. The fluctuating operating data is input into a machine learning model for online prediction to obtain node pressure prediction values ​​and pipe segment flow prediction values. Based on the pressure prediction values ​​and flow prediction values, online adjustment parameters for the pump set are generated.

[0046] Real-time pressure and flow data are collected from pressure-sensitive and distance-sensitive nodes. Real-time pressure data is collected via pressure sensors installed at the nodes, at a frequency of once per minute, with the unit being meters of water column. Real-time flow data is collected via flow meters installed on adjacent pipe sections, also at a frequency of once per minute. The collected real-time pressure and flow data undergo time-series alignment and numerical verification. Time-series alignment matches data from different sensors according to timestamps to ensure consistency in data analysis. Numerical verification filters out outliers, including values ​​outside the measurement range and values ​​that significantly deviate from historical patterns. Numerical verification uses the three-standard-deviation method, marking data deviating from the average by more than three standard deviations as outliers and removing them. Missing data is supplemented using linear interpolation. After time-series alignment and numerical verification, a real-time monitoring dataset is formed, containing node numbers, timestamps, and verified pressure and flow values.

[0047] Acquire the water load characteristics of the water supply network at different times. Water load characteristics refer to the pattern of water consumption changes at different times, typically exhibiting intraday and weekly fluctuations. Intraday fluctuations show a bimodal characteristic, with peak water consumption periods from 6:00 AM to 8:00 AM and 6:00 PM to 8:00 PM, and a low water consumption period from 12:00 AM to 5:00 AM. Weekly fluctuations show that water consumption on weekdays is higher than on weekends. By analyzing historical water consumption data, water load curves are extracted, and the average water consumption and standard deviation for each time period are calculated. Based on water load characteristics, real-time monitoring data is divided into steady-state operation data and fluctuating operation data. When the rate of change of real-time water consumption is lower than a preset threshold, the system is considered to be in a steady-state operation state; when the rate of change of real-time water consumption is higher than the preset threshold, the system is considered to be in a fluctuating operation state. The rate of change is calculated by dividing the absolute value of the difference in water consumption between two adjacent time points by the water consumption value at the previous time point, with the preset threshold set at 5%. Steady-state operation usually occurs during low water consumption periods or stable periods, while fluctuating operation usually occurs during peak water consumption periods or periods of rapid water consumption changes.

[0048] Steady-state operating data is input into the mechanistic calculation model for offline calculation. The mechanistic calculation model is based on fluid mechanics principles and includes nodal flow balance equations and pipe section head loss equations. The nodal flow balance equation states that the flow rate into a node equals the flow rate out of the node. The pipe section head loss equation states that the head difference between the two ends of a pipe section equals the head loss of that pipe section. Head loss is related to flow rate, pipe diameter, pipe length, and roughness coefficient. Using the nodal water demand from the steady-state operating data as known conditions, combined with the pipe network topology and pipe section parameters, the equations of the mechanistic calculation model are solved using the global gradient method. The global gradient method is a numerical method for iteratively solving nonlinear equations, continuously adjusting the nodal head and pipe section flow rate until a convergence condition is met. The convergence condition is set as a nodal flow balance error of less than 0.001 L / s. The pressure distribution at the pipe network nodes and the flow distribution in the pipe sections are obtained through the solution.

[0049] Offline pump set control parameters are generated based on pressure and flow distribution. These parameters include pump head and pump set combination. The pump head is determined by the minimum pressure requirement of the pressure-sensitive node. This is achieved by adding the minimum pressure requirement to the network head loss and subtracting the source water level. For example, if the minimum pressure requirement is 35 meters of water column, the network head loss is 15 meters of water column, and the source water level is 10 meters of water column, the required pump head is 40 meters of water column. The pump set combination is determined based on the total network flow demand. By consulting the pump set characteristic curves, the combination that best matches the required flow rate and has the highest energy efficiency is selected. When the total network flow demand is 100 L / s, two pumps with a rated flow rate of 60 L / s can be operated in parallel.

[0050] Fluctuation data is input into a machine learning model for online prediction. The machine learning model employs a Long Short-Term Memory (LSTM) network structure, which is well-suited for processing time-series data and capturing long-term dependencies. The input layer receives the fluctuation data, the hidden layer contains 64 LSM units, and the output layer predicts future node pressure and pipe flow rates. Model training uses historical fluctuation data, employing mean squared error as the loss function, and updates network parameters through backpropagation. The prediction time window is set to 30 minutes, predicting changes in node pressure and pipe flow rates over the next 30 minutes based on current and past data. The prediction results are output as predicted node pressure and pipe flow rates, including minute-by-minute predictions for the next 30 minutes.

[0051] Online pump control parameters are generated based on predicted pressure and flow rates. These parameters include the variable frequency pump speed and the number of pumps in operation. The variable frequency pump speed is dynamically adjusted based on the predicted pressure change trend; the speed is increased when the predicted pressure is lower than the set value and decreased when it is higher. The speed adjustment is proportional to the pressure deviation, with the proportionality coefficient determined based on actual operating experience. The number of pumps in operation is determined based on the predicted flow demand; a standby pump is started when the predicted flow exceeds the rated flow of a single pump, and one pump is shut down when the predicted flow is lower than 80% of the total flow of the minimum number of operating pumps. The online pump control parameters are updated every 5 minutes to ensure the system can respond promptly to changes in pipeline demand.

[0052] This invention achieves precise dynamic control of water supply network pressure by collecting real-time monitoring data from key nodes and combining the advantages of mechanistic calculation models and machine learning models. It treats steady-state and fluctuating operation separately, fully utilizing the accuracy of the mechanistic model under steady-state conditions and the rapid response capability of the machine learning model to dynamic changes. Through the synergistic effect of offline and online pump unit regulation, the accuracy and stability of network pressure control are significantly improved, effectively reducing network pressure fluctuations and energy waste.

[0053] The consistency verification of the pump set's offline and online adjustment parameters includes: Read the pump set offline adjustment parameters from the offline control command set, read the pump set online adjustment parameters from the online control command set, calculate the pressure adjustment difference and flow adjustment difference between the pump set offline adjustment parameters and the pump set online adjustment parameters, and generate adjustment difference data; The pressure and flow rate changes of the water supply pump set before and after adjustment are calculated based on the adjustment difference data, and the pressure and flow rate changes are used as water supply status change data. Determine whether the changes in water supply status will lead to insufficient water pressure or excessive pressure in the water supply system, and generate a water supply status verification result. The water supply status verification results are compared with the historical operation data of the water supply network to determine whether the water supply status verification results meet the water demand change pattern, and the demand matching results are generated as the consistency verification results.

[0054] The pump set offline adjustment parameters are read from the offline control instruction set. The offline control instruction set is stored in the control system's parameter library, updated every 24 hours, and the data format is key-value pairs of parameter names and values. The pump set offline adjustment parameters include the base head value and the pump set combination mode. The former indicates the base head that the pump set needs to provide, in meters of water column; the latter indicates the combination of pump numbers that need to be activated. A typical set of pump set offline adjustment parameters might be: a base head value of 45 meters of water column, and a pump set combination mode of simultaneous operation of pumps 1 and 3. The pump set online adjustment parameters are read from the online control instruction set. The online control instruction set is maintained by the real-time control module, updated every 5 minutes, and the data format is also key-value pairs of parameter names and values. The pump set online adjustment parameters include the speed adjustment value and the dynamic pump on / off strategy. The former represents a percentage adjustment relative to the rated speed, and the latter represents the pump on / off operation that needs to be performed at a preset time point. A typical set of pump set online adjustment parameters might be: a speed adjustment value increased by 5%, and a dynamic pump on / off strategy to start standby pump 2 after 10 minutes.

[0055] The pressure and flow rate regulation differences between the offline and online adjustment parameters of the pump set are calculated to generate regulation difference data. The pressure regulation difference is calculated by comparing the offline head value with the actual head value after speed adjustment. Subtracting the offline head value from the actual head value results in a positive value indicating increased pressure and a negative value indicating decreased pressure. When the base head value is 45 meters of water column and the speed adjustment value increases by 5%, since the pump head is proportional to the square of the speed, the actual head value is approximately 49.6 meters of water column, and the pressure regulation difference is 4.6 meters of water column. The flow rate regulation difference is calculated by subtracting the offline flow rate value from the actual flow rate value. A positive value indicates increased flow rate, and a negative value indicates decreased flow rate. When the flow rate in the offline combination mode is 100 L / s, and the speed is increased by 5% with the standby pump starting after 10 minutes, based on the proportional relationship between pump flow rate and speed, the actual flow rate value is expected to increase from the initial 105 L / s to 150 L / s after the standby pump starts. The flow rate regulation difference ranges from 5 to 50 L / s.

[0056] The pressure and flow rate changes of the water supply pump set before and after adjustment are calculated based on the adjustment difference data, and these changes are used as data on water supply status changes. The pressure change calculation considers the transmission effect of the adjustment difference in the pipe network, and the pressure changes at each key node are obtained through hydraulic model simulation. The simulation calculation uses the relationship between node flow balance and pipe section head loss, and iterative calculations are performed until convergence. For the aforementioned pressure adjustment difference of 4.6 meters of water column, the calculated pressure changes at each pressure-sensitive node are between 2.8 and 4.3 meters of water column, and the pressure changes at each distance-sensitive node are between 3.1 and 4.5 meters of water column. The flow rate change considers the water demand characteristics of the water supply area, and the change in regional water demand caused by the flow rate change is calculated using the water demand elasticity coefficient. The water demand elasticity coefficient represents the percentage change in water demand caused by a unit pressure change, and is generally taken as 0.2 to 0.5. For the aforementioned flow rate adjustment difference, the calculated increase in water demand in the water supply area is approximately 3% to 8%.

[0057] The system determines whether changes in water supply status data will lead to insufficient or excessive water pressure in the water supply system, generating a water supply status verification result. The criteria for insufficient water pressure are: the adjusted pressure at any pressure-sensitive node is lower than the minimum service pressure requirement, typically 28 meters of water column. The criteria for excessive pressure are: the adjusted pressure at any node exceeds the upper limit of the pipeline network design pressure, typically 60 meters of water column, or the pressure change rate at adjacent time points exceeds 2 meters of water column per minute. For the aforementioned pressure change scenarios, inspection revealed that the adjusted pressure at each node was between 32 and 55 meters of water column, with a maximum pressure change rate of 0.8 meters of water column per minute. No insufficient or excessive water pressure issues were observed; therefore, the water supply status verification result is "passed."

[0058] The water supply status verification results are compared with historical operating data of the water supply network to determine whether the verification results meet the water demand change pattern, generating a demand matching result. Historical operating data includes records of water consumption and pressure changes for the same dates and time periods over the past three months. The water demand change pattern is extracted through time series decomposition, including trend, seasonal, and random components. The trend component reflects long-term trends, the seasonal component reflects periodic changes, and the random component reflects unpredictable random fluctuations. The criterion for meeting the water demand change pattern is: the deviation between the adjusted water supply status and historical data for the same period is within a reasonable range, and the change trend is consistent with the predicted trend. The reasonable deviation range is defined as within twice the standard deviation of the historical data for the same period. For the aforementioned water supply status, after comparing with historical data for the same period, the pressure change deviation is 0.5 meters of water column, and the flow rate change deviation is 2%, both within a reasonable range, and the change trend is consistent with the predicted trend. Therefore, the demand matching result is "matched". The demand matching result is used as the consistency verification result. In this example, the consistency verification result is "passed and matched", which means that the pump set offline adjustment parameters and the pump set online adjustment parameters have good consistency and the corresponding control commands can be executed.

[0059] This invention achieves dual assurance of safety and adaptability in water supply network pressure control by verifying the consistency of offline and online pump set adjustment parameters. Through simulation calculations and comparison with historical data, it ensures that control commands meet both water supply safety requirements and adapt to changes in water demand. The consistency verification mechanism effectively avoids conflicts between offline and online parameters, preventing potential problems such as insufficient or excessive water pressure during control command execution. By comparing the verification results with historical data, the adaptability of the control strategy to water usage patterns is enhanced, achieving intelligent and refined management of the water supply process.

[0060] The consistency of the pump set's offline and online adjustment parameters is verified, and pump set adjustment commands are generated based on the verification results, including: A verification pass flag is generated when the consistency verification result meets the operating requirements of the water supply system; a verification failure flag is generated when the consistency verification result does not meet the operating requirements of the water supply system. When the verification is successful, the pump set online adjustment parameters are written into the control command cache, and the pump set adjustment command is generated and sent to the water supply pump set based on the pressure adjustment and flow adjustment in the pump set online adjustment parameters. When verification fails, the pump set offline adjustment parameters are written into the control command cache. Based on the pressure adjustment and flow adjustment in the pump set offline adjustment parameters, a pump set adjustment command is generated and sent to the water supply pump set. At the same time, the real-time monitoring data when verification fails is used as training samples to input into the machine learning model for incremental training.

[0061] A verification pass flag is generated when the consistency verification result meets the operational requirements of the water supply system; a verification failure flag is generated when the result does not. The operational requirements of the water supply system mainly include two aspects: pressure safety and demand matching. Pressure safety requires that the predicted pressure at pressure-sensitive nodes is not lower than the minimum service pressure, typically set at 28 meters of water column, and not higher than the maximum pressure-bearing capacity of the pipe network, typically set at 60 meters of water column. Demand matching requires that the deviation between the predicted flow rate and the actual water demand be within an acceptable range, with a deviation threshold typically set at 10%. A verification pass flag indicates that the online adjustment parameters of the pump set meet the pressure safety and demand matching requirements of the water supply system and can be used as a control basis; a verification failure flag indicates that the online adjustment parameters of the pump set do not meet the above requirements, and more reliable offline adjustment parameters need to be used. The verification process is executed every 5 minutes, consistent with the update frequency of the online adjustment parameters. The verification result is stored as a binary flag, with 0 indicating failure and 1 indicating success. For example, the predicted minimum pressure of the pressure-sensitive node is 30.5 meters of water column and the maximum pressure is 52.8 meters of water column. The deviation between the predicted flow rate and the actual demand is 6.5%, which meets the operational requirements. Therefore, the verification pass mark 1 is generated.

[0062] Upon successful verification, the pump set online adjustment parameters are written to the control command cache. Based on the pressure and flow rate adjustments in the online adjustment parameters, pump set adjustment commands are generated and sent to the water supply pump set. The control command cache is a temporary storage area for upcoming control commands, using a first-in-first-out queue structure with a capacity of 10 commands. The content written to the pump set online adjustment parameters cache includes parameter type identifier, parameter value, effective time, and validity period. The parameter type identifier distinguishes different types of control parameters; for example, "SP" indicates a speed adjustment value, and "VP" indicates a pump on / off command. The effective time indicates the time when the parameter takes effect, and the validity period indicates the duration of the parameter's validity. Pressure adjustments are converted to pump set speed adjustment values; the conversion relationship is based on the pump performance curve, typically with the pump head proportional to the square of the speed. Flow rate adjustments are converted to pump on / off commands, determining the number of pumps to be activated based on the rated flow rate of a single pump and the current demand flow rate. For example, when the pressure adjustment is to increase the water column by 3.5 meters and the flow adjustment is to increase the flow rate by 25 L / s, the generated pump set adjustment command includes: increasing the speed of pumps 1 and 2 to 104.2% of their rated speed, and simultaneously starting standby pump 3 after 30 seconds. The command is sent to the frequency converter of the water supply pump set via an industrial communication protocol. The communication message includes the device address, function code, data area, and verification area. The device address identifies the target controller, the function code indicates the operation type, the data area contains specific parameters, and the verification area is used for communication reliability verification.

[0063] When verification fails, the offline adjustment parameters of the pump set are written into the control command cache. Based on the pressure and flow rate adjustments in the offline adjustment parameters, a pump set adjustment command is generated and sent to the water supply pump set. The real-time monitoring data at the time of verification failure is used as training samples to input into the machine learning model for incremental training. The offline adjustment parameters of the pump set are derived from the mechanistic calculation model and have high reliability, making them suitable as a backup plan when the online model prediction is inaccurate. The content structure of the offline adjustment parameters written into the cache is the same as that of the online adjustment parameters, but the parameter type identifier is prefixed with "BK_" to indicate that this is a set of backup parameters. The generated pump set adjustment commands also include speed adjustment and pump on / off operation, but the adjustment range is usually more conservative to ensure water supply safety. The real-time monitoring data at the time of verification failure records the scenario that led to the prediction inaccuracy, including timestamps, pressure values, flow rates at each monitoring point, and deviations from the predicted values. This data constitutes the incremental training sample set for optimizing the machine learning model. The machine learning model adopts a long short-term memory network structure, and incremental training uses a mini-batch gradient descent algorithm with a learning rate of 0.001 and 20 training epochs. For example, if the predicted pressure at a pressure-sensitive node is 25.8 meters of water column, which is lower than the minimum service pressure requirement of 28 meters of water column, the system generates a verification failure flag of 0. It then extracts the offline adjustment parameters for the pump group from the offline parameter library: a base head of 48 meters of water column, and starts pumps 1 and 4. Based on this, it generates adjustment instructions: set the speed of pump 1 to 105% of its rated speed, start pump 4, and shut down pump 2. The deviation data between the current real-time monitoring data (including pressure values ​​from 10 nodes and flow values ​​from 8 pipe segments) and the predicted values ​​are used to form a training sample, which is then input into the machine learning model for incremental training to improve the model's prediction accuracy in similar scenarios.

[0064] The verification result processing mechanism has priority control. When three consecutive verifications fail, the system will be temporarily locked in offline adjustment mode for one hour. Even if a verification passes during this period, it will not immediately switch back to online adjustment mode to avoid water supply fluctuations caused by frequent switching. While the offline adjustment mode is locked, the machine learning model will continue to perform incremental training until it passes the verification test more than five times consecutively. Only then will the lock be lifted, and the normal verification and switching mechanism be restored.

[0065] This invention achieves intelligent and safe pressure control of water supply networks by establishing a consistency verification mechanism for pump set adjustment parameters and a control strategy selection method based on verification results. Through incremental learning of verification failure data, the machine learning model is continuously optimized, enabling the control system to self-improve. The differentiated handling strategy between successful and failed verifications ensures that the system can select the most suitable control scheme under various operating conditions, effectively avoiding insufficient water pressure or pressure fluctuations. Through an instruction caching mechanism and priority control, frequent switching of control instructions is prevented, improving the stability and reliability of the water supply system.

[0066] Adjusting the water supply pump set equipment according to the pump set adjustment command includes: Parse the pump set adjustment command, extract the pressure adjustment target value and flow adjustment target value from the pump set adjustment command, and convert them into pump set operation control parameters; The required output pressure and flow rate of the water supply pump set are calculated based on the pump set operation control parameters. The number of pumps to be put into operation and the speed of each pump are determined according to the output pressure and flow rate. Pump adjustment control commands are generated and sent to the water supply pump set to perform pump set adjustment operations. Collect the pipeline pressure and flow values ​​after the pump set adjustment operation is performed, compare the pipeline pressure and flow values ​​with the pressure adjustment target values ​​and flow adjustment target values, and generate adjustment effect evaluation data; The aforementioned adjustment effect evaluation data will be used as new real-time monitoring data for the next round of dynamic pressure control of the water supply network.

[0067] The pump set regulation commands are parsed to extract the pressure and flow regulation target values, which are then converted into pump set operation control parameters. The pump set regulation commands use a structured data format, comprising a command header, command content, and a checksum. The command header defines the command type, priority, and timestamp; the command content includes the pressure and flow regulation target values ​​and additional control flags; the checksum is used to verify the integrity of the command. The pressure regulation target value represents the desired factory outlet pressure or critical node pressure, measured in meters of water column; the flow regulation target value represents the desired total water supply flow rate, measured in liters per second. The parsing process first verifies the correctness of the command checksum, then extracts the pressure and flow regulation target values ​​from the command content. The extraction method uses positional parsing, with pressure values ​​located in bytes 1-4 of the content field and flow values ​​in bytes 5-8. The pump set operation control parameters include the total pump head target value and the total pump flow target value. The conversion process takes into account pipeline losses and elevation differences. Taking a certain instruction as an example, when the pressure regulation target value is 38 meters of water column (representing the expected pressure at the critical node A), by calculating the elevation difference of 15 meters from the critical node A to the water plant and the head loss of 8 meters through the pipeline, the total head target value of the pump set is obtained as 61 meters of water column; when the flow regulation target value is 250 L / s, considering the pipeline leakage rate of 5%, the total flow target value of the pump set is obtained as 263 L / s.

[0068] Based on the pump set operation control parameters, the required output pressure and flow rate of the water supply pump set are calculated. The number of pumps to be put into operation and the speed of each pump are determined according to the output pressure and flow rate. Pump regulation control commands are generated and sent to the water supply pump set for execution. The required output pressure and flow rate of the water supply pump set are calculated using a correction coefficient method, considering changes in the dynamic resistance of the pipeline network. The output pressure equals the target total head of the pump set multiplied by a dynamic correction coefficient, which is determined based on the current pipeline network status, typically between 1.02 and 1.05. The output flow rate equals the target total flow rate of the pump set multiplied by a flow correction coefficient, which is adjusted according to changes in pipeline network pressure, typically between 0.98 and 1.03. The number of pumps to be put into operation is determined using an optimal combination search algorithm, comprehensively considering equipment efficiency and switching costs. All possible pump combinations are traversed, and the operating efficiency and switching cost of each combination under the target operating conditions are calculated. The combination with the highest comprehensive score is selected. The speed of each pump is determined by solving a set of nonlinear equations, which represents the balance between the pump characteristic curve and the system resistance curve. The solution employs a modified Newton's method, with an iteration accuracy set to 0.1%. Pump regulation control commands include two categories: equipment start / stop commands and variable frequency speed control commands. Taking a specific regulation as an example, when the output pressure is 62.5 meters of water column and the output flow rate is 270 L / s, the optimal combination search determines that pumps 1, 3, and 4 need to be activated, with speeds of 1450 rpm, 1480 rpm, and 1420 rpm respectively. The generated start / stop command format is "equipment number + operation type + time stamp," such as "3-START-1608456780" indicating the start of pump 3. The variable frequency speed control command format is "equipment number + speed value + time stamp," such as "1-1450-1608456780" indicating the speed of pump 1 is adjusted to 1450 rpm. Commands are sent to the water supply pump group controller via fieldbus to execute the pump group regulation operation. During regulation, the speed change rate is limited to within 10 revolutions per second to avoid water hammer effects.

[0069] The system collects pipeline pressure and flow rates after pump unit regulation operations, compares these values ​​with the target pressure and flow rates, and generates regulation effect evaluation data. Pipeline pressure data is collected at the factory outlet pressure metering point and key monitoring points, with a sampling frequency of 10 seconds per measurement using pressure transmitters with an accuracy of 0.1 meters of water column. Flow rates are collected at the factory outlet total flow meter and zone flow meters, also with a sampling frequency of 10 seconds per measurement using electromagnetic flow meters with an accuracy of 0.5%. Data collection lasts for 10 minutes after regulation to ensure the system reaches a stable state. The comparison method uses a 5-minute time window average comparison. The average value of pipeline pressure and flow rates within the window is calculated, and the deviation is calculated from the target regulation value. The deviation is calculated by subtracting the target value from the actual value; a positive result indicates overshoot, and a negative result indicates undershoot. The regulation effect evaluation data includes four indicators: pressure regulation accuracy, flow regulation accuracy, regulation response time, and stability. Pressure regulation accuracy is the absolute value of the pressure deviation divided by the target pressure value; flow regulation accuracy is the absolute value of the flow deviation divided by the target flow rate value; regulation response time is the time required from sending the command to the actual value first entering the ±5% range of the target value; stability is the standard deviation of the actual value after reaching a stable state. Taking a certain regulation as an example, when the target pressure regulation value is 38 meters of water column, the actual measured average pressure is 39.2 meters of water column, the pressure deviation is 1.2 meters of water column, and the pressure regulation accuracy is 3.2%; when the target flow regulation value is 250 L / s, the actual measured average flow rate is 247 L / s, the flow deviation is -3 L / s, and the flow regulation accuracy is 1.2%; the response time is 85 seconds; the pressure stability is 0.8 meters of water column, and the flow stability is 2.5 L / s.

[0070] The regulation effect evaluation data will be used as new real-time monitoring data for the next round of dynamic pressure control in the water supply network. The regulation effect evaluation data, after standardization, will be stored in a real-time database. Standardization includes outlier filtering and timestamp normalization. Outlier filtering uses the 3σ principle to remove data points exceeding the mean ± 3 standard deviations. Timestamp normalization adjusts all data timestamps to whole-minute values ​​for easier subsequent analysis. Data storage uses a time-series database structure, supporting efficient time-range queries and aggregation analysis. Stored fields include measurement point identifier, timestamp, measured value, quality label, and source marker. The quality label indicates data reliability, and the source marker distinguishes between raw and processed data. This data will serve as input for the next round of dynamic pressure control, used for model prediction, parameter optimization, and control decisions, forming a closed-loop control mechanism.

[0071] This invention achieves precise dynamic control of water supply network pressure by establishing a complete conversion mechanism from pump group adjustment commands to specific equipment control. Through refined command parsing and parameter conversion, it ensures accurate transmission of control intentions; through intelligent pump group combination optimization and speed calculation, it minimizes water supply energy consumption; and through comprehensive effect evaluation and data feedback, it constructs a self-learning and continuously optimizing control system. This dynamic pressure control method based on multi-parameter fusion significantly improves the response speed, control accuracy, and stability of the water supply system, effectively solving the problem that traditional control methods struggle to cope with complex changes in water demand.

[0072] like Figure 3 As shown, Figure 3 This is a schematic diagram of a dynamic pressure control system for water supply networks based on multi-parameter fusion, provided in an embodiment of the present invention. The system includes: The data acquisition module is used to acquire geographic information system data of the water supply area's pipe network and perform standardized processing to generate effective pipe network data. The effective pipe network data is then processed by pipe segment merging and end-point pruning to construct an urban pipe network topology map. The node identification module is used to input the urban pipeline network topology map into the mechanism calculation model for hydraulic calculation and to identify pressure-sensitive nodes and distance-sensitive nodes. The model training module is used to extract historical monitoring data of pressure-sensitive nodes and distance-sensitive nodes, associate the historical monitoring data with the topological connection relationship of the corresponding nodes in the urban pipeline network topology map, train the machine learning model, and establish a predictive mapping relationship that integrates pipeline network topology constraints. The parameter calculation module is used to collect real-time monitoring data of pressure-sensitive nodes and distance-sensitive nodes, and input the real-time monitoring data into the mechanism calculation model and the machine learning model in parallel to obtain the pump set offline adjustment parameters and the pump set online adjustment parameters, respectively. The adjustment and control module is used to verify the consistency between the offline and online adjustment parameters of the pump set, generate pump set adjustment commands based on the verification results, and adjust the water supply pump set equipment according to the pump set adjustment commands.

[0073] One technical solution provided in this embodiment of the invention is an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0074] The specific embodiments described above are preferred embodiments of the present invention and are not intended to limit the specific scope of the present invention. The scope of the present invention includes, but is not limited to, these specific embodiments. All equivalent changes made in accordance with the shape and structure of the present invention are within the protection scope of the present invention.

Claims

1. A dynamic pressure control method for water supply networks based on multi-parameter fusion, characterized in that, Includes the following steps: Data from the geographic information system of the water supply network is acquired and standardized to generate effective network data. The effective network data is then processed by pipe segment merging and end-point pruning to construct an urban network topology map. Input the urban pipeline network topology map into the mechanism calculation model to perform hydraulic calculations and identify pressure-sensitive nodes and distance-sensitive nodes. Historical monitoring data of pressure-sensitive nodes and distance-sensitive nodes are extracted, and the historical monitoring data is associated with the topological connection relationship of the corresponding nodes in the urban pipeline network topology map. A machine learning model is trained to establish a predictive mapping relationship that integrates pipeline network topology constraints. Real-time monitoring data of pressure-sensitive nodes and distance-sensitive nodes are collected, and the real-time monitoring data is input in parallel into the mechanism calculation model and the machine learning model to obtain the pump set offline adjustment parameters and the pump set online adjustment parameters, respectively. The consistency of the pump set offline adjustment parameters and the pump set online adjustment parameters is verified. Based on the verification results, the pump set adjustment command is generated, and the water supply pump set equipment is adjusted according to the pump set adjustment command.

2. The method according to claim 1, characterized in that, The process involves acquiring and standardizing geographic information system (GIS) data of the water supply area's pipeline network to generate effective pipeline network data. This effective network data is then subjected to pipe segment merging and end-point pruning to construct an urban pipeline network topology map. The pipeline geographic information system data includes pipeline node coordinate information, pipe segment connection relationship information, and pipe diameter parameter information; Based on the pipe segment connection rules and hydraulic physical constraint rules, the connectivity index of each pipe segment in the geographic information system data of the water supply area is calculated, and the pipe segment data that meets the hydraulic calculation requirements is selected to form effective pipe network data; Based on the topological connection relationship of pipe segments in the effective pipe network data, a traversal operation is performed to calculate the difference in pipe diameter parameters between adjacent pipe segments. Pipe segment groups with zero difference in pipe diameter parameters and physical continuity are marked as pipe segment groups to be merged. Accumulate the length parameters of each pipe segment in the pipe segment group to be merged while keeping the positions of the first and last nodes unchanged to complete the pipe segment merging process; Hydraulic distribution calculations are performed on the terminal pipe segments in the effective pipe network data after pipe segment merging. The flow rate change and pressure change of each terminal pipe segment during the overall water supply process are obtained. The weighted sum of the flow rate change and pressure change is used as the impact index of the terminal pipe segment. Terminal pipe segments with impact index less than a preset threshold are deleted to complete the terminal pruning process and construct the urban pipe network topology map.

3. The method according to claim 1, characterized in that, The urban pipeline network topology map is input into the mechanism calculation model for hydraulic calculations to identify pressure-sensitive and distance-sensitive nodes, including: A mechanism calculation model is constructed, which includes a node flow balance equation and a pipe section head loss equation. The node flow balance equation is used to calculate the node flow conservation value, and the pipe section head loss equation is used to calculate the pipe section pressure loss value. Extract the network structure information from the urban network topology map, and substitute the network structure information into the node flow balance equation and the pipe section head loss equation to generate a set of network hydraulic calculation equations. Solve the set of hydraulic calculation equations for the pipeline network to obtain the node pressure sequence. Calculate the node pressure change amplitude based on the node pressure sequence and mark nodes whose pressure change amplitude exceeds a preset range as pressure-sensitive nodes. Calculate the path length from each node to the pressure-sensitive node in the urban pipeline network topology map, and mark the nodes whose path length is less than a preset length value as distance-sensitive nodes.

4. The method according to claim 1, characterized in that, Historical monitoring data of pressure-sensitive and distance-sensitive nodes are extracted. This historical monitoring data is then correlated with the topological connections of corresponding nodes in the urban pipeline network topology map. A machine learning model is trained to train this model, establishing a predictive mapping relationship that integrates pipeline network topology constraints. Extract historical pressure data and historical flow data from pressure-sensitive nodes and distance-sensitive nodes, and align them according to timestamps to form historical monitoring data; The topological connection relationship between nodes is extracted from the urban pipeline network topology map, the shortest connection distance between nodes is calculated, the shortest connection distance is converted into a location code, and the location code is combined with historical monitoring data to generate training data. Based on the topological connection relationship, the network nodes are traversed to identify node pairs with direct pipe segment connections, the pressure difference sequence between node pairs is calculated, and the node pressure attenuation function is constructed. The node pressure decay function is added as a loss term to the machine learning model. The prediction error of the training dataset is calculated. Training is stopped when the prediction error is less than a preset error threshold, and the trained machine learning model is obtained. The location code is input into the trained machine learning model, and the output is the node pressure prediction value. A mapping relationship from the location code to the pressure prediction value is established. The mapping relationship is used to predict the pressure value of pressure-sensitive nodes and distance-sensitive nodes.

5. The method according to claim 1, characterized in that, Real-time monitoring data of pressure-sensitive nodes and distance-sensitive nodes are collected, and the real-time monitoring data is input in parallel into the mechanism calculation model and the machine learning model to obtain the pump set offline adjustment parameters and the pump set online adjustment parameters, respectively: Real-time pressure data and real-time flow data of pressure-sensitive nodes and distance-sensitive nodes are collected, and the real-time pressure data and real-time flow data are time-series aligned and numerically verified to generate real-time monitoring data. The water load characteristics of the water supply network at different times are obtained. Based on the water load characteristics, the real-time monitoring data is divided into steady-state operation data and fluctuating operation data. The steady-state operation data is input into the mechanism calculation model for offline calculation to obtain the pressure distribution of the network nodes and the flow distribution of the pipe segments. Based on the pressure distribution and flow distribution, the pump set offline adjustment parameters are generated. The fluctuating operating data is input into a machine learning model for online prediction to obtain node pressure prediction values ​​and pipe segment flow prediction values. Based on the pressure prediction values ​​and flow prediction values, online adjustment parameters for the pump set are generated.

6. The method according to claim 1, characterized in that, The consistency verification of the pump set's offline and online adjustment parameters includes: Read the pump set offline adjustment parameters from the offline control command set, read the pump set online adjustment parameters from the online control command set, calculate the pressure adjustment difference and flow adjustment difference between the pump set offline adjustment parameters and the pump set online adjustment parameters, and generate adjustment difference data; The pressure and flow rate changes of the water supply pump set before and after adjustment are calculated based on the adjustment difference data, and the pressure and flow rate changes are used as water supply status change data. Determine whether the changes in water supply status will lead to insufficient water pressure or excessive pressure in the water supply system, and generate a water supply status verification result. The water supply status verification results are compared with the historical operation data of the water supply network to determine whether the water supply status verification results meet the water demand change pattern, and the demand matching results are generated as the consistency verification results.

7. The method according to claim 1, characterized in that, The consistency of the pump set's offline and online adjustment parameters is verified, and pump set adjustment commands are generated based on the verification results, including: A verification pass flag is generated when the consistency verification result meets the operating requirements of the water supply system; a verification failure flag is generated when the consistency verification result does not meet the operating requirements of the water supply system. When the verification is successful, the pump set online adjustment parameters are written into the control command cache, and the pump set adjustment command is generated and sent to the water supply pump set based on the pressure adjustment and flow adjustment in the pump set online adjustment parameters. When verification fails, the pump set offline adjustment parameters are written into the control command cache. Based on the pressure adjustment and flow adjustment in the pump set offline adjustment parameters, a pump set adjustment command is generated and sent to the water supply pump set. At the same time, the real-time monitoring data when verification fails is used as training samples to input into the machine learning model for incremental training.

8. The method according to claim 1, characterized in that, Adjusting the water supply pump set equipment according to the pump set adjustment command includes: Parse the pump set adjustment command, extract the pressure adjustment target value and flow adjustment target value from the pump set adjustment command, and convert them into pump set operation control parameters; The required output pressure and flow rate of the water supply pump set are calculated based on the pump set operation control parameters. The number of pumps to be put into operation and the speed of each pump are determined according to the output pressure and flow rate. Pump adjustment control commands are generated and sent to the water supply pump set to perform pump set adjustment operations. Collect the pipeline pressure and flow values ​​after the pump set adjustment operation is performed, compare the pipeline pressure and flow values ​​with the pressure adjustment target values ​​and flow adjustment target values, and generate adjustment effect evaluation data; The aforementioned adjustment effect evaluation data will be used as new real-time monitoring data for the next round of dynamic pressure control of the water supply network.

9. A dynamic pressure control system for water supply networks based on multi-parameter fusion, used to implement the method described in any one of claims 1-8, characterized in that, The system includes: The data acquisition module is used to acquire geographic information system data of the water supply area's pipe network and perform standardized processing to generate effective pipe network data. The effective pipe network data is then processed by pipe segment merging and end-point pruning to construct an urban pipe network topology map. The node identification module is used to input the urban pipeline network topology map into the mechanism calculation model for hydraulic calculation and to identify pressure-sensitive nodes and distance-sensitive nodes. The model training module is used to extract historical monitoring data of pressure-sensitive nodes and distance-sensitive nodes, associate the historical monitoring data with the topological connection relationship of the corresponding nodes in the urban pipeline network topology map, train the machine learning model, and establish a predictive mapping relationship that integrates pipeline network topology constraints. The parameter calculation module is used to collect real-time monitoring data of pressure-sensitive nodes and distance-sensitive nodes, and input the real-time monitoring data into the mechanism calculation model and the machine learning model in parallel to obtain the pump set offline adjustment parameters and the pump set online adjustment parameters, respectively. The adjustment and control module is used to verify the consistency between the offline and online adjustment parameters of the pump set, generate pump set adjustment commands based on the verification results, and adjust the water supply pump set equipment according to the pump set adjustment commands.

10. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the method as described in any one of claims 1 to 8.

Images