A digital pump intelligent control method and system

Abstract

The present invention relates to the technical field of digital pump control, and particularly to an intelligent control method and system for digital pumps. The present invention proposes the following solutions. By acquiring the real-time parameter data of digital pumps collected by sensors and inputting it into a data analysis model including a first analysis network and a second prediction network, the operation characteristics are extracted and the future state is predicted. According to the analysis results, the operation strategy of the digital pumps is dynamically adjusted; regional modeling is carried out according to the topological structure and operation requirements of the water supply network, the historical data of associated digital pumps is screened, and reference data is generated by combining sliding processing of time windows to support operation strategies such as pressure regulation, flow optimization, and overpressure protection; the present invention solves the problems of poor real-time performance, insufficient risk identification, and high energy consumption in traditional water supply networks, and improves the safety, stability, and operation efficiency of the water supply system.

Description

Technical Field

[0001] The present invention relates to the technical field of digital pump control, and particularly to an intelligent control method and system for digital pumps. Background Art

[0002] In the prior art, the operation and management of urban water supply networks mainly rely on digital pumps to adjust pressure and flow. However, traditional methods usually rely on static parameters or simple threshold judgments, and it is difficult to adapt to the dynamic changes of water supply demand. During peak water usage periods, traditional systems often suffer from insufficient pressure or overpressure problems due to the inability to predict water supply trends in real time. Under abnormal operating conditions (such as leaks or equipment failures), there is a lack of intelligent analysis means, and it is impossible to quickly identify risks and dynamically adjust strategies. In addition, the layout and configuration of digital pumps mostly rely on experience and are difficult to accurately match the actual operating requirements of the water supply network, resulting in energy waste and equipment overload. Existing methods have significant deficiencies in whole-network collaborative regulation, risk identification, and trend prediction, and it is difficult to meet the requirements of modern water supply systems for efficient, safe, and intelligent management.

[0003] To solve the above problems, this application designs an intelligent control method and system for digital pumps. Summary of the Invention

[0004] In view of the deficiencies of the prior art, the present invention provides an intelligent control method and system for digital pumps, which obtain the real-time parameter data of digital pumps collected by sensors, input them into a data analysis model including a first analysis network and a second prediction network, extract operation characteristics and predict future states, and dynamically adjust the operation strategy of digital pumps according to the analysis results. Regional modeling is carried out according to the topological structure and operation requirements of the water supply network, historical data of associated digital pumps is screened, and reference data is generated by combining time-window sliding processing to support operation strategies such as pressure regulation, flow optimization, and overpressure protection.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] An intelligent control method for digital pumps, which is applied to an urban water supply network. The urban water supply network includes a network node model, and each node in the network node model is configured with a digital pump and a sensor assembly. The method includes:

[0007] Obtain the real-time parameter data of the digital pump collected by the sensor assembly;

[0008] Input the real-time parameter data into a preset data analysis model, and output a data analysis result through the data analysis model. The data analysis model includes a first analysis network and a second prediction network. The first analysis network is used to process the real-time parameter data, extract operation characteristics and output the current operation state, and the second prediction network outputs a predicted state according to the operation characteristics;

[0009] Adjust the operation strategy of the digital pump according to the data analysis result.

[0010] Obtaining the real-time parameter data of the digital pump collected by the sensor component includes:

[0011] Adjust the acquisition sequence priority of the sensor component according to the operation state of the water supply network;

[0012] Perform consistency verification on the data collected by the sensor component, adjust the data processing order according to the verification result, and obtain real-time parameter data, where the real-time parameter data includes pressure data and flow data.

[0013] The first analysis network includes:

[0014] Input layer, receive real-time parameter data according to the data processing order, perform weighted assignment on the real-time parameter data according to the reception order, and perform channel splitting on the weighted real-time parameter data;

[0015] Feature extraction layer, used to process pressure and flow data respectively in different channels and output channel features;

[0016] Feature fusion layer, merge the channel features according to the self-attention mechanism, perform non-linear mapping through the fully connected layer, and output operation features, where the weight of the input layer is used as the initial weight matrix of the self-attention mechanism;

[0017] Output layer, map the operation features to a high-dimensional space, process them layer by layer through the pooling kernel in the high-dimensional space, and classify the current operation state according to the Softmax classifier.

[0018] The feature extraction layer includes:

[0019] Pressure feature extraction branch, extract features from pressure data through a sparse convolutional network and calculate pressure channel features;

[0020] Flow feature extraction branch, decompose flow data through a grouped convolutional network and calculate flow channel features.

[0021] The second prediction network includes:

[0022] Input layer, used to receive the operation features output by the first analysis network and historical reference data;

[0023] Trend extraction layer, used to extract the pressure change trend and the flow change trend, where the trend extraction layer includes a pressure trend path and a flow trend path;

[0024] The output layer includes a pressure prediction branch and a flow prediction branch, and outputs the pressure prediction value and the flow prediction value for the future time step according to the pressure change trend and the flow change trend, wherein the future time step is calculated based on the current time.

[0025] The historical reference data is parameter data screened from the historical parameter data of all digital pumps in the water supply network, and the screening includes:

[0026] According to the historical parameter data of all digital pumps, screen the historical parameter data of the digital pumps that have a direct operation association with the current digital pump, wherein the direct operation association is calculated based on the pressure transmission path, the flow interaction relationship, and the correlation of the historical operation data between digital pumps;

[0027] Perform a time window sliding process on the historical parameter data of the digital pumps that have a direct operation association, and the size of the time window is adjusted according to the current time period and the area where the current digital pump is located;

[0028] Use the historical parameter data within the time window as the historical reference data.

[0029] Adjusting the operation strategy of the digital pump according to the data analysis result includes:

[0030] According to the current operation state output by the first analysis network and the predicted state output by the second prediction network, judge whether there are risks for the digital pump at the current and future time steps, wherein the risks include pressure risk, flow risk, and overload risk;

[0031] If there is a risk, adjust the operation strategy according to the risk type.

[0032] The adjusted operation strategy includes pressure regulation, flow optimization, and overpressure protection.

[0033] The method further includes performing a modeling analysis on the water supply network and configuring digital pumps at the model nodes of the water supply network, specifically including:

[0034] Collect the topological structure, regional characteristic data, and operation requirements of the water supply network;

[0035] Perform regional segmentation on the water supply network according to the topological structure and regional characteristic data, and output the regional boundary conditions;

[0036] Analyze the local operation characteristics according to the regional boundary conditions and operation requirements, and output the digital pump layout position and configuration.

[0037] A digital pump intelligent control system, the system includes a data acquisition module, a data analysis module, and an operation adjustment module;

[0038] The data acquisition module is used to collect the real-time parameter data of digital pumps in the water supply network, including pressure data and flow rate data, and perform preliminary processing;

[0039] The data analysis module is used to receive the real-time parameter data provided by the data acquisition module, extract the operation characteristics through the first analysis network and output the current operation state, and output the predicted values of pressure and flow rate through the second prediction network;

[0040] The operation adjustment module adjusts the operation strategy of the digital pump according to the operation state and risk prediction provided by the data analysis module, including pressure regulation, flow rate optimization and overpressure protection.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] The present invention combines real-time parameter data with historical reference data to accurately analyze the operation state of digital pumps and predict trends, providing forward-looking operation strategy adjustments, thereby ensuring the stability and efficiency of the water supply network; the present invention combines regional characteristic data and operation requirements to optimize the layout position and operation parameters of digital pumps, and also supports dynamic regulation of pressure and flow rate distribution, avoiding problems such as excessive energy consumption, insufficient water supply or equipment overload caused by unreasonable initial configuration or changes in operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] By reading the detailed description of the non-restrictive embodiments with reference to the following drawings, other features, objects and advantages of the present invention will become more apparent:

[0044] Figure 1 It is a schematic flow chart of a digital pump intelligent control method according to Embodiment 1 of the present invention;

[0045] Figure 2 It is a schematic structural diagram of a water supply network node model according to Embodiment 1 of the present invention;

[0046] Figure 3 It is a structural diagram of a data analysis model according to Embodiment 1 of the present invention;

[0047] Figure 4 It is a structural diagram of the first analysis network according to Embodiment 1 of the present invention;

[0048] Figure 5 It is a structural diagram of a feature extraction layer according to Embodiment 1 of the present invention;

[0049] Figure 6 It is a structural diagram of the second prediction network according to Embodiment 1 of the present invention;

[0050] Figure 7 It is a schematic flow chart of a modeling analysis method for a water supply network according to Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0051] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments.

[0052] Embodiment 1

[0053] Please refer to Figure 1 , an embodiment provided by the present invention: a digital pump intelligent control method, and the specific steps of the method are as follows:

[0054] S1: Obtain the real-time parameter data of the digital pump collected by the sensor assembly;

[0055] In this embodiment, through the sensor assembly configured at the node of the water supply network, the pressure data and flow data of the digital pump are collected in real time, and the collected data is preprocessed, including dynamically adjusting the collection frequency and priority of the sensor according to the operating state of the water supply network, and performing consistency verification to ensure the accuracy and integrity of the data. For example, during peak water usage periods, the real-time parameter data of key nodes is preferentially collected, and the data is weighted so that high-priority data has a greater influence weight in subsequent analysis.

[0056] S2: Input the real-time parameter data into a preset data analysis model, and output a data analysis result through the data analysis model;

[0057] In this embodiment, the collected real-time parameter data is input into a data analysis model including a first analysis network and a second prediction network. The first analysis network processes the real-time data, extracts the local characteristics of the pressure data through a sparse convolutional network, combines a grouped convolutional network to perform multi-dimensional feature decomposition on the flow data, generates the operating characteristics of the current digital pump, and uses a Softmax classifier to output the current operating state, such as "normal", "overpressure warning", or "low pressure anomaly" and other state results. At the same time, the second prediction network extracts the pressure change trend and flow change trend based on the operating characteristics output by the first analysis network and the historical reference data screened from the historical parameter data of the digital pumps across the network, and outputs the pressure and flow prediction results for multiple future time steps. It realizes the accurate evaluation of the current operating state of the digital pump and the forward-looking prediction of the future operating trend.

[0058] S3: Adjust the operating strategy of the digital pump according to the data analysis result;

[0059] In this embodiment, based on the current operating state of the first analysis network and the prediction result of the second prediction network, it is determined whether there are pressure risks, flow risks, or overload risks in the current digital pump. If there are risks, the operating strategy of the digital pump is dynamically adjusted according to the risk type. For example, when the pressure exceeds the safety threshold, the system will automatically reduce the operating speed of the digital pump, activate the pressure relief device, and at the same time notify the associated digital pumps to jointly share the pressure load in the high-pressure area; when a flow shortage or leakage risk is detected, the water supply flow at key nodes is preferentially increased, and an abnormal alarm message is sent to the operation and maintenance personnel. Through multi-level response mechanisms such as dynamic pressure regulation, flow optimization, and overpressure protection, the stable operation of the water supply network is ensured, while reducing the operating risks of the equipment and extending its service life.

[0060] Please refer to Figure 2 , which is a schematic diagram of the node model structure of the urban water supply network in the embodiment of the present invention, showing a node model of an urban water supply network. In this model, the digital acquisition and sensor components are located in different areas (Area 1, Area 2, and Area 3). These areas are connected through a control terminal, which collects and transmits data from each area. Each area represents a different water supply area, and sensors in each area monitor relevant data. This information is summarized through the control terminal to facilitate the effective management and control of the entire urban water supply system. In this embodiment, taking the urban water supply network as an example, a digital pump intelligent control method is proposed. By collecting the pressure and flow data of the digital pump in real time and combining the collaborative work of the first analysis network and the second prediction network, the operating state and trend of the digital pump are accurately analyzed and predicted, so as to dynamically adjust the operating strategy of the digital pump. Although digital pumps have been widely used in the water supply network in the prior art for pressure and flow regulation, the prior art usually relies on single real-time parameter data, lacks the ability to verify data quality and extract features, and is difficult to accurately identify abnormal states and predict future operating trends. For example, in the peak water supply scenario, due to the sudden increase in user demand, the pressure in a certain area drops sharply. The prior art may only trigger the pump speed increase through simple threshold judgment, unable to comprehensively consider future trends and network coordination, easily leading to overpressure in other areas or even equipment overload. In addition, in complex scenarios with large flow fluctuations, the prior art lacks accuracy in judging abnormal states (such as leakage or insufficient water supply), often requiring manual intervention, which not only increases the maintenance cost but also makes it difficult to respond to dynamic demands in a timely manner.

[0061] Exemplarily, in a water supply network of a new urban area, when the pressure in a certain area drops below the set threshold due to concentrated water use by users, the traditional method usually directly increases the operating speed of the digital pump in this area, which may cause overpressure problems in other areas due to unbalanced pressure distribution. In this embodiment, the first analysis network is used to identify the low-pressure state of the current area, and the second prediction network is used to predict the pressure recovery trend; at the same time, the water supply path is dynamically adjusted in combination with the historical parameter data of adjacent digital pumps, moderately increasing the pressure output of the low-pressure area and transferring the excess load to the standby digital pump to participate in the operation. This not only successfully solves the low-pressure problem but also avoids the overpressure risk in other areas.

[0062] Preferably, in this embodiment, the digital pump intelligent control method can also be applied to multiple practical application fields of digital pumps.

[0063] Specifically, in the field of agricultural intelligent irrigation, the method of this embodiment can intelligently control the digital pumps in the irrigation network by real-time monitoring of soil humidity and water pressure changes in different areas. For example, in arid areas, when the flow demand at a certain irrigation node surges, the system can judge the water supply status of the current node according to the operation characteristics of the first analysis network and dynamically adjust the pump speed in combination with the trend analysis of the second prediction network to ensure the reasonable distribution of water resources. At the same time, the overpressure protection function is used to avoid equipment damage caused by too high pipe network pressure, thereby achieving the goals of precise irrigation and water conservation.

[0064] Specifically, in the field of industrial cooling water circulation, the pressure and flow of water are crucial for the stable operation of equipment. The method of this embodiment can optimize the operation strategy of the cooling water pump by real-time collecting the operation parameters (such as flow rate, pressure) of the cooling water pump and combining the trend analysis of the prediction network. For example, when the cooling demand decreases, the system can automatically reduce the operation power of the cooling water pump to reduce energy consumption; when abnormal pressure or too high equipment load is detected in the pipe network, the system can immediately execute the dynamic adjustment strategy to share the pressure load and ensure the safe operation of the equipment.

[0065] The specific steps of S1 are as follows:

[0066] S1.1: Adjust the acquisition sequence priority of the sensor assembly according to the operation state of the water supply pipe network;

[0067] Specifically, by preferentially collecting data at key nodes, potential low-pressure problems can be identified in a timely manner and corresponding measures can be taken. The acquisition sequence priority of the sensor assembly is adjusted by a dynamic weight algorithm, and the specific steps of the dynamic weight algorithm are as follows:

[0068] S1.1.1: Calculate the real-time priority weight of each sensor node based on the operation state parameters of the water supply pipe network (such as regional flow fluctuation, pressure gradient distribution, and historical anomaly records).

[0069] The operating state is provided by the real-time monitoring data of the water supply network. For example, during the peak period of the water supply network, there is usually a sudden increase in regional flow and a rapid decrease in pressure. Therefore, during the peak period, priority is given to monitoring the pressure and flow data of the nodes at the end of the water supply and in high-demand areas, while the data collection frequency in low-demand areas is temporarily reduced.

[0070] S1.1.2: Optimize the collection order according to historical anomaly records;

[0071] For example, for areas with frequent leaks or anomalies in the past, the priority of their sensors will be automatically increased to ensure that abnormal situations can be detected more quickly.

[0072] Furthermore, this dynamic adjustment is achieved through the control terminal of the water supply network. After receiving the status data of the water supply network, the control terminal updates the data collection plan of the sensors in real time according to the preset priority algorithm, enabling each sensor to complete the data collection task in the order of priority.

[0073] S1.2: Perform consistency verification on the collected data, adjust the data processing order according to the verification results, and obtain real-time parameter data, where the real-time parameter data includes pressure data and flow data;

[0074] In this embodiment, the purpose of consistency verification is to quickly screen out sensitive data and dynamically adjust the data processing order according to the sensitivity level. High-sensitivity data will be preferentially input into the data analysis model to accelerate the analysis and response to potential problems; low-sensitivity data will be processed later according to the situation. This consistency verification strategy can not only improve the response speed of key node data but also optimize the allocation of data processing resources, ensuring that the data analysis model focuses on important data, thereby improving the overall intelligent level and abnormal response efficiency of the water supply network operation. Especially in scenarios of peak water supply or complex flow fluctuations, this mechanism for quickly screening sensitive data significantly enhances the real-time performance and accuracy of the system.

[0075] Specifically, the consistency verification adopts a dynamic sensitivity analysis method, which mainly verifies and screens the data quickly from three aspects: the time dimension, spatial correlation, and historical anomaly correlation.

[0076] In this embodiment, in the time dimension, the data is verified for consistency through time series comparison analysis to determine whether the volatility of the sensor data exceeds the preset range. For example, when the pressure data of a certain sensor shows continuous sudden increases or decreases within a short period, and the change amplitude is significantly higher than the historical average fluctuation range, this data is marked as "high-sensitivity data" and given a higher priority. This can quickly capture possible pressure anomalies in the water supply network, such as sudden overpressure or low-pressure risks.

[0077] In this embodiment, in the spatial correlation verification, based on the water supply network topology model, the pressure and flow rate data of adjacent nodes are compared to determine whether the difference between them conforms to the normal hydrodynamic distribution law. For example, if the flow rate of the upstream node increases sharply while the flow rate of the downstream node remains unchanged, it may indicate a blockage or leakage in the pipeline, and the relevant data of this node will be marked and thus given priority for processing.

[0078] In this embodiment, the historical anomaly correlation verification is based on the data anomaly historical records and trend analysis, and preferentially marks the data of the sensor nodes that have had frequent anomaly records in the past. For example, if a certain node has frequently had low pressure records during the peak water supply period, the latest data of this node will be focused on.

[0079] Please refer to Figure 3 , the data analysis model structure diagram of the embodiment of the present invention. The data analysis model includes a first analysis network and a second prediction network. The first analysis network is used to process the real-time parameter data, extract operation characteristics and output the current operation state, and the second prediction network outputs a predicted state according to the operation characteristics.

[0080] Please refer to Figure 4 , the structure diagram of the first analysis network of the embodiment of the present invention. Through the cooperation of the input layer, feature extraction layer, feature fusion layer and output layer, the real-time parameter data of the digital pumps in the water supply network are deeply analyzed, operation characteristics are extracted and the current operation state is evaluated. The first analysis network includes:

[0081] The input layer receives the real-time parameter data according to the data processing sequence, assigns weights to the real-time parameter data according to the reception sequence, and performs channel splitting on the weighted real-time parameter data;

[0082] In this embodiment, the input layer obtains the real-time parameter data of the digital pumps from the sensor assembly, including pressure data and flow rate data. Each input data stream will be arranged according to a pre-arranged sequence. For example, among the multiple nodes of the water supply network, the data of real-time input parameter 1 is sorted first and will be input into the first analysis network first, and so on, which is consistent with the sequence logic of the S1.2 data processing.

[0083] In this embodiment, a unique timestamp identifier is assigned to each group of input data to record the order of their reception. After the data is received, a time-ordered data matrix is formed. Each column corresponds to the real-time pressure and flow parameters of a node, and each row represents the data at different time points. The initial weight of the data is directly related to its reception order. The weight assignment follows the following principle: the earlier the input order, the higher the weight; the later the input order, the weight gradually decreases. After the weighted assignment is completed, all weights are normalized to ensure that the sum of the weights is 1. The calculation formula for the weight is:

[0084] ;

[0085] Among them, represents the weight of the i-th received data, N represents the total number of data groups, i represents the order of data reception, represents the normalization process, and j represents any value from 1 to N.

[0086] Specifically, based on the weighted assignment according to the input order, the input layer can automatically amplify the influence of the data that arrives first, thereby improving the efficiency and priority of data processing. In the dynamic operation scenario of the water supply network, it can ensure that the real-time data with abnormal conditions is quickly responded to.

[0087] In this embodiment, the weighted data is divided into a pressure channel and a flow channel according to the data type to ensure that the pressure and flow data can independently enter the subsequent feature extraction path. The weighted pressure data is organized in a matrix form, with each column corresponding to the pressure values of different nodes and each row corresponding to the pressure features at different time points. Similarly, the flow data is also organized in a matrix form, with each column corresponding to the flow values of different nodes, effectively reducing the interference between the pressure and flow data, making the feature extraction more targeted, and at the same time optimizing the calculation efficiency of the network.

[0088] Feature extraction layer, used to process the pressure and flow data in different channels respectively and output channel features;

[0089] In this embodiment, the feature extraction layer adopts a branched architecture and independently processes the characteristics of the pressure and flow data respectively, including a pressure feature extraction branch and a flow feature extraction branch. Each branch uses a dedicated network structure to process its unique data characteristics.

[0090] Feature fusion layer, merges the channel features according to the self-attention mechanism, performs non-linear mapping through a fully connected layer, and outputs operation features, where the weight of the input layer is used as the initial weight matrix of the self-attention mechanism;

[0091] In this embodiment, the pressure channel features and flow channel features are converted into corresponding feature matrices with the same dimension through matrix transformation. At the same time, the initial weight matrix is set as the dynamic weighted assignment result from the input layer, representing the initial importance of each feature. The feature matrices are merged through the initial weight matrix to calculate the joint feature matrix. Then, the query, key, and value matrices are constructed through the self-attention mechanism to reassign each feature in the joint feature matrix. The weight update matrix is calculated based on the assignment result. The basis for reallocation is to calculate the attention score of each feature, where the weights of highly similar features are amplified and the weights of low-similarity features are weakened. The joint feature matrix is updated through the weight update matrix.

[0092] In this embodiment, the joint feature matrix enhanced by the self-attention mechanism is input into the fully connected network for non-linear mapping to further improve the feature expression ability. Specifically, the fully connected layer expands the low-dimensional joint features to a high-dimensional space through multiple mapping operations to improve the discrimination of the features. Batch normalization is added after the output of each layer to ensure the stable distribution of the high-dimensional features and reduce the risk of gradient disappearance or explosion. The last fully connected network compresses the high-dimensional features and outputs the running feature representation, which is directly used as the input for classification in the output layer.

[0093] The output layer processes the running features layer by layer through the pooling kernel and classifies the current running state according to the Softmax classifier.

[0094] Specifically, the pooling kernel is used to process the running features belonging to the high-dimensional features layer by layer to extract the important information in the features and filter out the redundant features. The adaptive pooling technology is adopted during the pooling process, and the pooling strategy is dynamically adjusted according to the complexity of different types of features. For example, for the high-dimensional feature space, a larger pooling kernel is used to quickly reduce the dimension, while for the low-dimensional feature space, a smaller pooling kernel is used to retain more detailed information, so as to retain the key features and enhance the generalization ability of the model. Finally, the running features after pooling processing are input into the Softmax classifier. By calculating the probability distribution of each running state, the category with the highest probability is preferentially selected as the final state. If the probabilities of multiple categories are similar, multiple state categories can be output simultaneously and the confidence score of each category can be generated. The confidence score is calculated based on the probability output of the classifier, and a comprehensive judgment report is generated according to the confidence score to accurately classify and output the current running state.

[0095] Please refer to Figure 5 , the structural diagram of the feature extraction layer in the embodiment of the present invention. The feature extraction layer includes:

[0096] The pressure feature extraction branch extracts features from the pressure data through the sparse convolutional network and calculates the pressure channel features;

[0097] Specifically, the sparse convolutional network can effectively capture the non-uniform distribution characteristics in the data and is particularly suitable for dealing with local anomalies existing in the pressure data, such as local overpressure or low-pressure phenomena caused by pipeline characteristics or node layouts in the water supply network.

[0098] In this embodiment, the pressure data is input in the form of a two-dimensional matrix, and each element corresponds to the pressure value of a node in the water supply network. The data is sparsified according to the pressure value and the set upper and lower thresholds, and only the regions with significant pressure anomalies (such as overpressure and low-pressure regions) are retained, and the rest are set to zero to form a sparse matrix. The sparse matrix is convolved by a sparse convolutional kernel. The design of the convolutional kernel specifically considers the spatial distribution characteristics of the pressure data. Since most of the values in the sparse matrix are zero, the sparse convolution operation skips the zero-value positions and only calculates the non-zero elements, thereby significantly reducing the computational complexity and improving the feature extraction efficiency. At the same time, a multi-layer sparse convolution stacking structure is adopted, and the perception ability of local pressure anomalies and regional pressure gradients is gradually enhanced through layer-by-layer feature extraction. After each layer, the ReLU activation function is used to perform a non-linear mapping on the extracted features to retain the key features and filter out the redundancy. The finally output pressure channel features are a multi-dimensional feature vector, which includes: the significance of local pressure anomalies, the eigenvalue of the regional pressure gradient distribution, and the overall pattern of the pressure changing with space.

[0099] The flow feature extraction branch decomposes the flow data through a grouped convolutional network to calculate the flow channel features;

[0100] Specifically, the flow data has multi-dimensional attributes, usually including spatial distribution (such as the flow difference between nodes), temporal dynamics (such as the fluctuation law of the flow), and abnormal patterns (such as sudden increase or decrease). These characteristics need to be modeled independently to avoid feature aliasing. At the same time, the flow characteristics in different regions of the water supply network vary significantly. For example, there may be large fluctuations in industrial areas, while the flow in residential areas tends to be stable. After grouping the flow data, each group can focus on specific regional characteristics for analysis. The core idea of grouped convolution is to divide the input tensor into multiple groups, and the convolution operation is independently performed within each group, thereby achieving efficient feature separation and parallel processing.

[0101] In this embodiment, the flow data is input in the form of a three-dimensional tensor, and the dimensions are , where H and W represent the spatial distribution, C represents the multi-dimensional characteristics of the flow rate. The input tensor is divided into multiple groups along the channel dimension C, and each group contains several channels, representing different dimensions of flow rate characteristics. The convolution operation is applied separately to each group. The convolution kernel is used to extract features within a specific dimension, and the size of the convolution kernel is dynamically planned according to the sizes of H and W in each group. By independently processing different dimensional characteristics, the grouped convolution operation can avoid interference between features of different channels and significantly reduce the computational overhead. The output of the grouped convolution passes through a fully connected layer for feature fusion, integrating the feature vectors of independent groups into a high-dimensional feature representation. The finally output flow channel features are a multi-dimensional feature vector, including the spatial distribution characteristics of the flow rate amplitude, the temporal correlation of the flow rate fluctuation, and the potential patterns of flow rate anomalies.

[0102] Please refer to Figure 6 , the second prediction network structure diagram of the embodiment of the present invention. The second prediction network includes:

[0103] An input layer for receiving the operation characteristics and historical reference data output by the first analysis network;

[0104] Specifically, the operation characteristics include the pressure change rate, flow rate fluctuation amplitude, and multi-dimensional feature representation extracted by the first analysis network, which are used to describe the current operation state of the digital pump; the historical reference data is obtained by screening the historical parameter data of all digital pumps in the water supply network for data directly related to the current digital pump.

[0105] In this embodiment, the screening of historical reference data includes:

[0106] According to the historical parameter data of all digital pumps, screen the historical parameter data of digital pumps that have a direct operation association with the current digital pump, where the direct operation association is calculated based on the pressure transmission path, flow rate interaction relationship, and correlation of historical operation data between digital pumps;

[0107] In this embodiment, a graph model based on the topological structure of the water supply network is constructed to calculate the direct operation association between digital pumps. The digital pump nodes in the water supply network are represented as vertices in the graph model, and the pipelines are used as edges to describe the physical connection relationship between digital pumps. At the same time, combined with the pressure transmission path and flow rate interaction relationship in the historical operation data, the association weight between digital pumps is calculated. The correlation analysis of historical parameter data is based on the following dimensions:

[0108] Pressure transmission path: By analyzing the pressure distribution characteristics between nodes in the water supply network, identify other nodes that are highly correlated with the pressure change of the current digital pump. For example, when a certain digital pump has overpressure or low pressure, the pressure changes of the upstream or downstream nodes connected to it are directly related.

[0109] Flow interaction relationship: Based on historical flow data, calculate the flow transfer ratio and fluctuation correlation between adjacent nodes to identify the nodes that have a greater impact on the current digital pump flow change.

[0110] Historical data correlation: Through the correlation analysis of historical time series, screen the nodes that have a significant association with the current digital pump in the operation mode, such as the co-variation trend of the water supply pressure during peak periods.

[0111] Perform a time window sliding process on the historical parameter data of the digital pumps with direct operation association, and the size of the time window is adjusted according to the current time period and the area where the current digital pump is located;

[0112] For example, during the peak water supply period, expand the time window to capture more historical data and analyze the long-term operation trend; while during the stable operation or low-demand period, narrow the time window and only focus on short-term changes to improve real-time performance. The adjustment logic of the time window also combines the water supply characteristics of the area where the digital pump is located. For example, for the high-rise building area at the end of the water supply, the time window will pay more attention to capturing high-frequency fluctuation data; while for the main water supply trunk pipe area, the window will focus more on the long-term pressure stability.

[0113] Use the historical parameter data within the time window as historical reference data.

[0114] Furthermore, the input layer not only standardizes the above data to make its distribution more adaptable to the processing requirements of the subsequent network. Since there are significant differences in the physical properties, temporal characteristics, and spatial correlation of pressure and flow data, through the method of data channel division, the pressure characteristics and flow characteristics are divided into independent input paths to ensure the characteristics and independence of the two types of characteristics in subsequent processing.

[0115] Trend extraction layer, used to extract the pressure change trend and flow change trend, where the trend extraction layer includes a pressure trend path and a flow trend path;

[0116] In this embodiment, the pressure trend path is designed by combining a recursive graph neural network (R-GNN) and a temporal convolutional network (TCN). The recursive graph neural network is used to construct the spatial correlation of pressure characteristics and dynamically model the pressure transfer relationship between associated digital pumps based on the topological structure of the water supply network, while the temporal convolutional network captures the long-term and short-term temporal variation laws of pressure characteristics through multi-layer causal convolutions. Because pressure data not only has dynamic changes in time but is also affected by the spatial distribution of associated nodes, the combination of the recursive graph neural network and the temporal convolutional network can model spatio-temporal characteristics simultaneously, significantly improving the recognition accuracy of the pressure change trend.

[0117] In this embodiment, the flow trend path adopts a combination of a dynamic time-series graph attention network (D-TGAT) and a multi-layer residual convolutional network. The dynamic time-series graph attention network is used to capture the spatio-temporal dynamic characteristics of flow data, dynamically adjust the flow feature weights at different historical time points through a time perception mechanism, and analyze the spatial correlation of the flow based on the node interaction relationship of the water supply network. The residual convolutional network further performs deep modeling on the flow features, focusing on the flow mutation characteristics and abnormal patterns. This structure is adopted because the flow features may mutate in a short period of time (such as leakage or peak demand). The dynamic time-series graph attention network can enhance the sensitivity to key time points, and the introduction of the residual convolutional network improves the robustness to complex flow changes.

[0118] The output layer, including a pressure prediction branch and a flow prediction branch, outputs the pressure prediction value and the flow prediction value for future time steps according to the pressure change trend and the flow change trend, wherein the future time steps are calculated based on the current time.

[0119] In this embodiment, the pressure prediction branch performs a non-linear mapping on the features of the pressure trend path through a multi-layer fully connected network (FCN), converts the multi-dimensional features into the pressure prediction results for multiple future time steps, and attaches a confidence score to each time step to quantify the reliability of the prediction results. The flow prediction branch performs time-series modeling on the features of the flow trend path through a recurrent neural network (RNN), generates the time-series prediction values of future flow demands, and outputs the flow fluctuation range to evaluate the uncertainty of the flow changes. The independent branches can be optimized more precisely according to their respective characteristics. The pressure prediction branch emphasizes stability and global trends, while the flow prediction branch focuses on dynamic changes and short-term fluctuations.

[0120] Exemplarily, common alternative networks include long short-term memory networks (LSTM) and gated recurrent units (GRU). LSTM and GRU perform excellently in processing time-series data, especially suitable for capturing long-term and short-term dependencies in the time dimension. However, these architectures mainly focus on time-series data, and their limitation for a system with strong spatial correlation such as a water supply network is that they cannot effectively handle the spatial relationship between digital pumps. It is easier to ignore the pressure transmission path and flow interaction characteristics between nodes, so there is a deficiency in capturing spatial dynamic correlations. This deficiency may lead to a weak detection ability for abnormal trends (such as local overpressure or leakage) caused by spatial correlations in a complex water supply network.

[0121] Furthermore, in the processing of the input data of the second prediction network, the historical parameter data is screened, and only the data that has a direct operational association with the current digital pump is retained. The screening is based on the pressure transmission path, flow interaction relationship, and correlation of historical operation data among digital pumps, reducing the interference of irrelevant data, lowering the complexity of network calculations, and ensuring the high correlation and quality of the input data. For example, in a partitioned water supply network, only analyzing the data of adjacent nodes that directly affect the current operation state of the digital pump can effectively improve the modeling accuracy and efficiency of future trends. This screening process avoids the noise interference brought by the traditional global data input method, making the prediction results more targeted and practical.

[0122] Furthermore, the second prediction network specifically designs a dynamic calculation mechanism for the prediction time step, enabling the prediction time step to be dynamically adjusted according to the current time. For example, in a water supply network, the water supply demand changes are usually more frequent and complex during peak periods, and the prediction time step needs to be appropriately shortened to provide more real-time control support; while during off-peak periods, the water supply demand is relatively stable, and the time step can be appropriately extended to reduce the computational burden of the network. This dynamic adjustment strategy optimized for specific scenarios is one of the important innovations in the practical application of this application, making its performance in complex water supply scenarios superior to traditional fixed time step prediction methods.

[0123] The specific steps of S3 are as follows:

[0124] S3.1: According to the current operation state output by the first analysis network and the predicted state output by the second prediction network, determine whether there are risks for the digital pump at the current and future time steps, where the risks include pressure risk, flow risk, and overload risk;

[0125] In this embodiment, by comprehensively analyzing the actual operation parameters (such as pressure, flow, and pump speed) of the digital pump at the current time step and future trends, determine whether there are pressure risk, flow risk, and overload risk for the digital pump.

[0126] Specifically, the first analysis network uses real-time parameter data to extract the local characteristics and sparse distribution of pressure data through a sparse convolutional network, and deconstructs the multi-dimensional characteristics of flow data through a grouped convolutional network, maps it into multi-dimensional operation features, and classifies the features into states such as "normal", "warning", and "abnormal" using a Softmax classifier. In this way, the first analysis network can quickly and accurately capture the operation state of the current digital pump and provide real-time operation state analysis.

[0127] Specifically, the second prediction network receives the output features of the first analysis network and historical reference data, and models the dynamic trends of pressure and flow using the pressure trend path and the flow trend path respectively. The pressure trend path adopts a recurrent graph neural network to capture the pressure transfer relationship between the digital pump node and the associated nodes, and combines a temporal convolutional network to extract the long-term and short-term variation trends of pressure; the flow trend path extracts the dynamic change patterns of flow features through a temporal graph attention network, which is particularly suitable for capturing complex flow fluctuation characteristics. According to the predicted future pressure values and flow values, the possible risk types are comprehensively judged. For example, if it is predicted that the pressure in a certain period will exceed the threshold, an "overpressure risk" signal will be triggered, or if the predicted flow will significantly decrease, it will be marked as a "flow shortage risk".

[0128] S3.2: If there is a risk, adjust the operation strategy according to the risk type. Adjusting the operation strategy includes pressure regulation, flow optimization, and overpressure protection.

[0129] Specifically, when the overpressure risk is detected, the system reduces the pressure output by decreasing the operating speed of the digital pump, and at the same time starts the cooperation of the associated digital pumps to share the pressure. To achieve this function, the system dynamically optimizes the pressure distribution in the water supply area based on the predicted pressure change trend. For example, it models the topological relationship of the water supply network through graph convolution, adjusts the pressure transfer path between the network nodes, and ensures that the load in the high-pressure area is transferred to the low-pressure area to avoid overloading of single-node equipment. In addition, the system starts the pressure relief device when necessary to release the excess pressure in the water supply network, and at the same time transmits the alarm signal of the overpressure risk to the operation and maintenance end through the Internet of Things platform to further enhance the security of the water supply network.

[0130] For the flow risk, when the system determines that the flow shortage or leakage trend is significant, first, it dynamically optimizes the flow output capacity of the digital pump. For example, a multi-scale residual module is added to the flow prediction path to strengthen the response ability to short-term flow anomalies and increase the pump speed to increase the flow; second, it reallocates the water supply path in combination with the time series prediction results, and gives priority to ensuring the water supply flow in the high-demand area. For example, the water volume in the low-flow area is supplemented by adjusting the rotational speed of the digital pumps in adjacent areas; in addition, in the case of leakage risk, the possible leakage area is located by combining the pressure change data, the water supply pressure in the leakage area is reduced to reduce losses, and an alarm signal is sent to the operation and maintenance end.

[0131] For the overload risk, when the system determines that the digital pump is under long-term high load or the operating parameters have approached the equipment safety threshold, it first reduces the operating load of the digital pump, and takes over part of the tasks by the standby digital pump to reduce the pressure of the main pump. In addition, the system adjusts the target operating parameters of the digital pump in combination with the future trend results of the second prediction network to prevent problems such as equipment wear and efficiency decline caused by future high-load operation in advance.

[0132] Please refer to Figure 7, a schematic flow chart of a method for modeling and analyzing a water supply network according to an embodiment of the present invention. This embodiment also provides a method for modeling and analyzing a water supply network, and the specific steps of the method are as follows:

[0133] S401: Collect the topological structure, regional characteristic data, and operation requirements of the water supply network;

[0134] In this embodiment, the topological structure of the water supply network includes the node distribution of the pipeline, the pipeline length, diameter, material, and the connection relationship with the nodes. These data are collected in real time through the geographic information system (GIS) and sensor network deployed in the network. The regional characteristic data includes the terrain height difference, user distribution density, water demand type (such as industrial area, residential area, commercial area), and these data are obtained through terrain modeling and water demand surveys. At the same time, the operation requirements are determined by the flow rate, pressure changes, and seasonal fluctuation data monitored in real time. The collection and fusion of such multi-source data provide a comprehensive basic input for subsequent analysis. By introducing a dynamic data collection module, it is possible to preferentially collect data of key nodes according to the importance of the nodes. For example, in areas with large height differences, the system preferentially collects the pressure and flow rate data of high points and end nodes to capture possible water supply bottlenecks. Through this collection method, it can be ensured that the dynamic changes of the water supply network can be accurately reflected, laying a data foundation for water supply optimization.

[0135] S402: Perform regional segmentation on the water supply network according to the topological structure and regional characteristic data, and output regional boundary conditions;

[0136] The core of regional segmentation is to divide the water supply network into several sub-regions with independent characteristics. Each region is relatively independent in operation but collaborates as a whole. To achieve this goal, this embodiment adopts a regional division method based on the graph segmentation algorithm. By constructing a graph model of the water supply network, regarding the nodes of the network as vertices and the pipelines as edges, and on this basis, performing partitioning through the minimum cut algorithm or the K-means clustering method. The regional boundary conditions are determined by calculating the pressure gradient and flow transfer relationship between the internal and external nodes of the region, and the inlet pressure, outlet pressure, and key nodes of the region are output. Through regional segmentation, it is possible to effectively isolate different characteristic regions such as height difference regions, flow fluctuation regions, and end regions, thereby providing a more targeted basis for the layout of digital pumps. For example, in a complex terrain water supply network including mountains and plains, the segmented height difference region can be independently provided with digital pumps to adjust the pressure, while reducing the interference of high pressure on low pressure regions. In this way, the water supply demand within the region can be independently met, and the global coordination is ensured by the boundary conditions.

[0137] S403: Analyze the local operation characteristics according to the regional boundary conditions and operation requirements, and output the layout position and configuration of the digital pumps;

[0138] After completing the area segmentation and determining the boundary conditions, further analyze the local operating characteristics according to the operating requirements of the area, and finally determine the installation location and configuration parameters of the digital pump. In this embodiment, a pressure-flow coupling analysis technology is adopted. According to the average pressure demand and peak flow demand in the area, calculate the pump head and flow range required for each key node. The key nodes include the inlet node of the high-demand area, the end water supply node, and the nodes with significant pressure attenuation in the pipe network. For example, by comparing the flow demand fluctuations and pressure stability of each node in the area, preferentially select the node with the lowest pressure as the installation location of the digital pump. In addition, this embodiment adopts a pump configuration optimization algorithm based on historical operation data. According to the water consumption fluctuation curve of the node, dynamically adjust the operating parameters of the digital pump, such as head, speed, and power, to ensure that the digital pump can operate in the efficient range. In this way, the installation of the digital pump can not only meet the operating requirements in the area, but also adapt to the coordination goals of the entire network, thus achieving global energy conservation and stable regional water supply.

[0139] In this embodiment, the modeling and analysis method of the water supply pipe network is used as a precondition means of this application, aiming to provide an accurate optimization basis for the subsequent optimization of the digital pump operation strategy.

[0140] Specifically, by collecting the topological structure, regional characteristic data, and operating requirements of the water supply pipe network, accurately divide the pipe network area and determine the boundary conditions, and further analyze the installation location and configuration parameters of the digital pump in combination with the operating characteristics of the area, ensuring that the initial layout and operation strategy of the digital pump can highly match the dynamic requirements of the actual pipe network. It can achieve refined zoning and efficient resource allocation of the water supply pipe network, thus avoiding problems such as unbalanced pressure distribution, flow waste, and equipment overload caused by unreasonable layout or inaccurate initial conditions, and improving the overall efficiency and stability of the water supply system.

[0141] Furthermore, the pipe network node model in this embodiment can be applied to the dynamic monitoring and optimal control of key nodes in the water supply pipe network. In the peak water consumption scenario, the node model can real-time identify the water supply bottleneck points and potential pressure imbalance areas, serving as the basis for dynamically adjusting the digital pump operation strategy; while in the low-demand period, the node model can support the power optimization of the digital pump to reduce energy consumption.

[0142] Embodiment 2

[0143] The present invention provides an embodiment: a digital pump intelligent control system, the system includes a data acquisition module, a data analysis module, and an operation adjustment module;

[0144] The data acquisition module is used to collect the real-time parameter data of the digital pump in the water supply pipe network, including pressure data and flow data, and perform preliminary processing;

[0145] The data analysis module is used to receive the real-time parameter data provided by the data acquisition module, extract the operation characteristics through the first analysis network and output the current operation status, and output the predicted values of pressure and flow through the second prediction network.

[0146] The operation adjustment module adjusts the operation strategy of the digital pump according to the operation status and risk prediction provided by the data analysis module, including pressure regulation, flow optimization, and overpressure protection.

[0147] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A digital pump intelligent control method is applied to a city water supply network, wherein the city water supply network is configured with a network node model, and each node in the network node model is configured with a digital pump and a sensor component, characterized in that: The method comprises: Acquire real-time parameter data of the digital pump collected by the sensor assembly, wherein acquiring the real-time parameter data comprises: According to the operation status of the water supply network, adjusting the acquisition sequence priority of the sensor components; Performing consistency verification on the data collected by the sensor assembly, adjusting the data processing sequence according to the verification result, and acquiring real-time parameter data, wherein the real-time parameter data includes pressure data and flow data; Input the real-time parameter data into a preset data analysis model, and output the data analysis results through the data analysis model, wherein the data analysis model includes a first analysis network and a second prediction network, the first analysis network is used to receive the real-time parameter data according to a data processing order, and weight the real-time parameter data according to the receiving order, extract the operation characteristics and output the current operation status, and the second prediction network outputs the prediction status according to the operation characteristics; The operation strategy of the digital pump is adjusted according to the data analysis results.

2. According to claim 1, a digital pump intelligent control method is characterized in that: The first analysis network comprises: An input layer receives the real-time parameter data according to a data processing order, performs weighted assignment on the real-time parameter data according to the receiving order, and performs channel segmentation on the weighted real-time parameter data; The feature extraction layer is used to process pressure and flow data in different channels respectively and output channel features; A feature fusion layer, which merges the channel features according to the self-attention mechanism, performs nonlinear mapping through a fully connected layer, and outputs running features, wherein the weight of the input layer is used as the initial weight matrix of the self-attention mechanism; The output layer maps the operation features to a high-dimensional space, processes them layer by layer through a pooling kernel in the high-dimensional space, and classifies the current operation status according to a Softmax classifier.

3. According to claim 2, a digital pump intelligent control method is characterized in that: The feature extraction layer comprises: The pressure feature extraction branch extracts features from pressure data through a sparse convolutional network and calculates pressure channel features; The traffic feature extraction branch performs feature decomposition on traffic data through a group convolutional network and calculates traffic channel features.

4. According to claim 1, a digital pump intelligent control method is characterized in that: The second prediction network comprises: An input layer, used to receive the operation characteristics and historical reference data output by the first analysis network; A trend extraction layer, used to extract pressure change trends and flow change trends, wherein the trend extraction layer includes a pressure trend path and a flow trend path; The output layer includes a pressure prediction branch and a flow prediction branch, which outputs a pressure prediction value and a flow prediction value of a future time step according to the pressure change trend and the flow change trend, wherein the future time step is calculated according to the current time.

5. According to claim 4, a digital pump intelligent control method is characterized in that: The historical reference data is parameter data obtained by screening the historical parameter data of all digital pumps in the water supply network, and the screening includes: According to the historical parameter data of all digital pumps, the historical parameter data of digital pumps that are directly associated with the operation of the current digital pump are screened, wherein the direct operation association is calculated according to the pressure transmission path, flow interaction relationship and correlation of the historical operation data between the digital pumps; Performing time window sliding processing on the digital pump historical parameter data directly related to the operation, wherein the size of the time window is adjusted according to the current time period and the area where the current digital pump is located; The historical parameter data within the time window is used as historical reference data.

6. A digital pump intelligent control method according to claim 1, characterized in that: The step of adjusting the operation strategy of the digital pump according to the data analysis result includes: According to the current operating state output by the first analysis network and the predicted state output by the second prediction network, determine whether there are risks in the digital pump at the current and future time steps, wherein the risks include pressure risk, flow risk and overload risk; If there is a risk, adjust the operating strategy according to the risk type.

7. A digital pump intelligent control method according to claim 6, characterized in that: The adjustment operation strategy includes pressure regulation, flow optimization and overpressure protection.

8. A digital pump intelligent control method according to claim 1, characterized in that: The method further includes modeling and analyzing the water supply network, configuring a digital pump at a node of the network node model, and modeling and analyzing the water supply network includes: Collect water supply network topology, regional characteristics data and operation requirements; Performing regional segmentation on the water supply network according to the topological structure and regional characteristic data, and outputting regional boundary conditions; The local operation characteristics are analyzed according to the boundary conditions and operation requirements of the area, and the layout position and configuration of the digital pump are output.

9. A digital pump intelligent control system, used to implement a digital pump intelligent control method as claimed in any one of claims 1 to 8, characterized in that: The system includes a data acquisition module, a data analysis module and an operation adjustment module; The data acquisition module is used to collect real-time parameter data of the digital pump in the water supply network, including pressure data and flow data, and perform preliminary processing; The data analysis module is used to receive the real-time parameter data provided by the data acquisition module, extract the operation characteristics through the first analysis network and output the current operation status, and output the predicted values ​​of pressure and flow through the second prediction network; The operation adjustment module adjusts the operation strategy of the digital pump according to the operation status and risk prediction provided by the data analysis module, including pressure regulation, flow optimization and overpressure protection.

Images