A method and system for water meter collection management in water affairs

Abstract

The application relates to the technical field of intelligent water affairs, and discloses a water meter group copy management method and system for water affairs, wherein the water meter group copy management method comprises the following steps: constructing a multi-level space-time graph model, abstracting a water meter group copy network into a multi-level space-time graph structure; applying a graph neural network to extract network topology features, including applying a graph convolution network to process spatial dependence and applying a time convolution network to process time dependence; calculating attention weights between nodes to model mutual dependence between the nodes; performing multi-scale network state prediction to predict future network states on multiple time scales; and realizing an abnormal event response mechanism to detect and respond to emergencies in the water meter group copy network; the multi-level space-time graph model can realize adaptive adjustment of network topology, can dynamically optimize the network structure according to actual operation environment changes and network load conditions, and can effectively reduce a data packet loss rate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart water management, more specifically, it relates to a water meter collection and copy management method and system for water management. BACKGROUND

[0002] With the rapid development of smart city construction, the intelligent level of water management is continuously improving. As an important part of smart water management, the water meter collection and copy system realizes real-time monitoring and management of water consumption through automatic data collection and transmission. However, the existing water meter collection and copy management technology still has many technical defects and deficiencies in actual application.

[0003] In the prior art, the water meter collection and copy network usually adopts a fixed topology structure and a static routing strategy. This traditional method is difficult to dynamically adjust according to the changes of the actual running environment and the network load after the network deployment is completed. Specifically, the existing technology has the following main problems:

[0004] Firstly, the network topology structure lacks self-adaptive ability. Once the traditional water meter collection and copy system is deployed, its network topology structure is basically fixed and unchanged. During the water consumption peak period, some network nodes bear excessive data transmission pressure, resulting in a high data packet loss rate of 15%-25%, which seriously affects the accuracy and integrity of data collection. At the same time, due to the use of fixed communication paths, the energy consumption of some nodes in the network is too fast, while the energy utilization rate of other nodes is relatively low, causing serious imbalance in energy consumption between network nodes and reducing the service life of the entire network. Secondly, it lacks the ability to predict the evolution trend of the network topology. In the urban water environment, the network status will change due to various factors, including seasonal water consumption mode changes, changes in communication environment caused by urban construction, performance degradation caused by equipment aging, etc. The existing technology can only passively adapt to these changes and cannot predict in advance and make appropriate preparations, resulting in obvious lag in network adaptability and affecting the reliability and stability of data collection. Thirdly, the spatiotemporal dependence relationship modeling is not accurate enough. The existing technology usually adopts a simplified processing method with uniform weights when dealing with the spatiotemporal dependence relationship in the water meter collection and copy network, without fully considering the importance difference of different spatiotemporal scale characteristics. This processing method ignores the change characteristics of the network topology structure in different time scales and cannot accurately capture and model complex spatiotemporal dependence relationships, resulting in limited network optimization effect. Finally, the response ability to sudden events is insufficient. For sudden changes in network topology caused by sudden events (such as pipe rupture, equipment failure, communication interference, etc.), the existing technology reacts slowly and cannot adjust the network configuration and routing strategy in time. This lag causes the network to be unable to quickly recover to normal operation state when a sudden event occurs, affecting the continuity and reliability of data collection.

[0005] In addition, although some existing technical solutions attempt to introduce machine learning methods to optimize network management, they are mostly limited to single-time-scale analysis and lack comprehensive consideration of multi-scale spatio-temporal features. These methods often fail to fully utilize the spatio-temporal evolution rules of network topology when dealing with complex urban water management environments, resulting in suboptimal optimization results.

[0006] Therefore, there is an urgent need for a water meter cluster reading management method that can dynamically adapt to changes in network environment, has multi-scale prediction capability, accurately models spatio-temporal dependencies, and quickly responds to emergencies, in order to improve the intelligent level and operational efficiency of water management. SUMMARY

[0007] The present application provides a water meter cluster reading management method and system for water management, which solves the technical problems of fixed water meter cluster reading network topology structure, lack of prediction capability, inaccurate spatio-temporal dependency modeling, and slow response to emergencies in related technologies.

[0008] The present application provides a water meter cluster reading management method for water management, comprising:

[0009] A multi-level spatio-temporal graph model is constructed to abstract the water meter cluster reading network as a multi-level spatio-temporal graph structure.

[0010] Based on the multi-level spatio-temporal graph structure, graph neural networks are applied to extract network topology features, including applying graph convolution networks to process spatial dependencies and time convolution networks to process temporal dependencies.

[0011] Based on the network topology features, the attention weights between nodes are calculated to model the mutual dependencies between nodes.

[0012] Based on the attention weights between nodes, multi-scale network state prediction is performed to predict the future state of the network at multiple time scales.

[0013] Based on the output of the multi-scale network state prediction, an abnormal event response mechanism is implemented to detect and respond to emergencies in the water meter cluster reading network.

[0014] Further, the step of constructing a multi-level spatio-temporal graph model comprises:

[0015] All nodes in the water meter cluster reading network are numbered to form a node set.

[0016] Based on the communication connection relationship between nodes, a time-varying edge set is constructed.

[0017] The edge set is converted into a time-varying adjacency matrix.

[0018] A multi-level spatio-temporal graph is constructed, with each level representing network topology features at different spatio-temporal scales.

[0019] Establish inter-layer connection, connect different levels of the same node through cross-layer edge set.

[0020] Further, the step of applying a graph neural network to extract network topology features comprises:

[0021] For each layer in the multi-level spatio-temporal graph, a graph convolution network is applied to extract topology features;

[0022] Combine the time convolution network to capture the evolution pattern of network topology over time;

[0023] For the cross-layer connection in the multi-level spatio-temporal graph, a cross-layer feature fusion algorithm is applied to integrate information of different spatio-temporal scales;

[0024] Generate a node feature matrix reflecting the complete spatio-temporal characteristics of the water meter collection network.

[0025] Further, the step of calculating the attention weight between nodes comprises:

[0026] Based on the extracted node feature matrix, a graph attention network is applied to calculate the attention weight between nodes;

[0027] Apply a multi-head attention mechanism to simultaneously calculate multiple independent sets of attention weights, and then combine the results;

[0028] For different levels in the multi-level spatio-temporal graph, the attention weight is calculated respectively to form a hierarchical set of attention weights;

[0029] According to the attention weight, a weighted adjacency matrix is generated.

[0030] Further, the step of performing multi-scale network state prediction comprises:

[0031] Construct a multi-scale prediction model to simultaneously predict network state changes on four time scales: hour, day, week, and month;

[0032] For prediction tasks of different time scales, a hierarchical time series prediction architecture is adopted;

[0033] Introduce a multi-task learning framework to jointly optimize the prediction tasks of the four time scales;

[0034] Based on the prediction results, a prediction-driven network parameter optimization strategy is calculated.

[0035] Further, the step of implementing an abnormal event response mechanism comprises:

[0036] Construct an abnormal event detector to identify abnormal events by monitoring mutations in network topology;

[0037] For detected abnormal events, rapid positioning and impact range assessment are performed;

[0038] Based on the abnormal influence area, a quick response strategy is designed to preferentially adjust the network parameters of the affected area;

[0039] For different types of abnormal events, a special processing strategy is applied.

[0040] Further, the graph convolution network comprises:

[0041] An input layer receives initial features of nodes;

[0042] A plurality of graph convolution layers update node representations by aggregating neighbor node information;

[0043] An output layer generates final node feature representations;

[0044] The time convolution network comprises a causal convolution layer, a dilated convolution layer and a residual connection.

[0045] Further, the graph attention network comprises:

[0046] A feature transformation layer performs linear transformation on input node features;

[0047] An attention calculation layer calculates attention coefficients between node pairs;

[0048] A feature aggregation layer aggregates neighbor node information according to the attention coefficients;

[0049] The multi-head attention mechanism simultaneously calculates 8 to 16 attention heads, and each attention head focuses on a different feature subspace.

[0050] Further, the hierarchical time series prediction architecture comprises:

[0051] A short-term prediction layer predicts hour-level network state changes using a gated recurrent unit;

[0052] A medium-term prediction layer predicts day-level and week-level network state changes using a long short-term memory network;

[0053] A long-term prediction layer predicts month-level network state changes using an attention-enhanced time convolution network;

[0054] The prediction-driven network parameter optimization strategy comprises route strategy optimization, energy management optimization and communication frequency optimization.

[0055] The present application provides a water meter collection management system for water affairs for executing the above-mentioned water meter collection management method for water affairs, comprising:

[0056] A multi-level spatio-temporal graph construction module is used to abstract the water meter collection network into a multi-level spatio-temporal graph structure;

[0057] a graph neural network feature extraction module for extracting network topology features;

[0058] an attention weight calculation module for modeling the mutual dependence between nodes;

[0059] a multi-scale network state prediction module for predicting future network states at multiple time scales;

[0060] an abnormal event response module for detecting and responding to sudden events in the water meter collection network.

[0061] The beneficial effects of the present application are: through the multi-level spatiotemporal graph model, the adaptive adjustment of the network topology is realized, the network structure can be dynamically optimized according to the actual running environment changes and network load conditions, and the data packet loss rate is effectively reduced;

[0062] With multi-scale network state prediction capability, it can predict network state changes at different time scales, and through prediction-driven optimization strategies, it can respond to various changes in advance, avoiding performance degradation caused by passive adaptation;

[0063] The introduction of the attention mechanism can automatically learn and quantify the influence weight between network nodes, accurately capturing the importance differences of features at different spatiotemporal scales;

[0064] An event-driven rapid response system is constructed, which can quickly detect abnormal events, accurately locate the impact range, and take targeted measures;

[0065] Through intelligent routing strategies and load balancing mechanisms, the problem of uneven energy consumption between nodes is improved, the network service life is prolonged, the overall throughput of the system is improved, the data transmission efficiency is improved, and the operation and maintenance cost is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0066] Figure 1 is a flowchart of a water meter collection management method in water affairs in the present application;

[0067] Figure 2 is a line chart of the data packet loss rate comparison of the water meter collection network before and after applying the method of the present application at different time points;

[0068] Figure 3 is a bar chart of the data transmission success rate comparison before and after applying the method of the present application under different network load conditions;

[0069] Figure 4 is a radar chart of the performance comparison of the present application scheme and the traditional water meter collection management scheme;

[0070] Figure 5 is a pie chart of the proportion of nodes with different energy consumption levels in the network before and after applying the method of the present application.

[0071] Figure 6 is a broken line graph of the accuracy of the prediction of the network state at different time scales. DETAILED DESCRIPTION

[0072] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that discussions of these implementations are merely provided to enable those skilled in the art to better understand and utilize the subject matter described herein, and do not limit the scope of protection or application for the claimed subject matter. Various processes and components can be omitted, substituted, or added as desired. Additionally, features described with respect to one example can be combined in any manner with features described with respect to another example.

[0073] In at least one embodiment of the present disclosure, a water meter cluster reading management method for water affairs is disclosed, as shown in the formula (I): Figure 1 The formula (I) comprises the following steps:

[0074] Step 1, a multi-level space-time graph model is constructed, and the water meter cluster reading network is abstracted into a multi-level space-time graph structure;

[0075] In this step, a multi-level space-time graph construction algorithm is used to abstract the water meter cluster reading network into a multi-level space-time graph structure, which is used to comprehensively capture the space-time characteristics of the network topology.

[0076] Step 1.1, a node set is constructed;

[0077] According to one embodiment of the present application, all nodes (including smart water meters, data collectors, relay nodes, and data concentrators, etc.) in the water meter cluster reading network are numbered first to form a node set Each node has attribute information such as location coordinates, function type, and energy state.

[0078] Step 1.2, a time-varying edge set is constructed;

[0079] In addition, based on the communication connection relationship between nodes, a time-varying edge set is constructed. When there is a communication connection between two nodes, an edge is created at the corresponding time , and edge attributes such as communication quality and data transmission rate are recorded.

[0080] Step 1.3, a time-varying adjacency matrix is formed;

[0081] The edge set is converted into a time-varying adjacency matrix , where the matrix element represents that the node is connected to the node at the time The connection strength can be a binary value (0 or 1) indicating whether the connection exists or not, or a continuous value indicating the connection strength.

[0082] In some embodiments, the adjacency matrix can be extended to a multi-dimensional tensor, containing multiple types of connection relationships. For example, multiple connection types such as physical connection, logical connection, and data flow connection can be considered simultaneously to form a multi-relationship adjacency tensor where represents the relationship type.

[0083] Step 1.4, constructing a multi-level spatio-temporal graph;

[0084] The expression for constructing a multi-level spatio-temporal graph is:

[0085] ;

[0086] where, represents a set of multi-level spatio-temporal graphs, , , represent the first layer, the second layer, the layer spatio-temporal graph respectively, is the total number of layers of the spatio-temporal graph.

[0087] Each layer of the spatio-temporal graph represents the network topology characteristics at different spatio-temporal scales:

[0088] : Hour-level spatio-temporal graph, capturing short-term network topology changes;

[0089] : Daily spatio-temporal graph, capturing network topology change patterns within a day;

[0090] : Weekly spatio-temporal graph, capturing periodic network topology change rules;

[0091] : Monthly spatio-temporal graph, capturing long-term network topology evolution trends.

[0092] Optionally, according to the requirements of specific application scenarios, the division of spatio-temporal scales can be adjusted, for example, in areas with dramatic changes in water consumption, minute-level spatio-temporal graphs can be added, or in areas with obvious seasonal changes, quarterly spatio-temporal graphs can be added.

[0093] Step 1.5, forming a complete multi-level spatio-temporal graph;

[0094] Establishing inter-layer connections to connect the same nodes at different levels to form a complete multi-level spatio-temporal graph where the time-varying graph structure is used to model the characteristics of network topology changes over time:

[0095] ;

[0096] wherein denotes time varying graph structure, denotes a set of nodes, denotes a set of edges, denotes a set of edges, denotes an adjacency matrix.

[0097] In some embodiments, the cross-layer connection can adopt a non-uniform connection strategy, and different cross-layer connection weights are allocated to different nodes according to the importance or activity level of the nodes, so that the key nodes can more effectively transmit information at different time scales.

[0098] Through the above sub-steps, the complex water meter collection network is abstracted into a mathematical multi-level spatio-temporal graph model, laying a foundation for subsequent analysis and optimization based on graph neural network. It should be noted that this multi-level spatio-temporal graph model can comprehensively capture the evolution characteristics of the network topology at different time scales, providing a solid theoretical basis for realizing accurate network state prediction and intelligent management.

[0099] Step 2, based on the multi-level spatio-temporal graph structure, applying a graph neural network to extract network topology features, including applying a graph convolutional network to process spatial dependence and a time convolutional network to process temporal dependence;

[0100] This step applies a graph neural network to the multi-level spatio-temporal graph constructed in step 1 to extract the topology features of the water meter collection network for subsequent network state analysis and optimization.

[0101] Step 2.1, extracting topology features;

[0102] According to one embodiment of the present application, for each layer in the multi-level spatio-temporal graph, a graph convolutional network (GCN) is applied to extract topology features. The mathematical expression of graph convolution operation is:

[0103] ;

[0104] wherein denotes the node feature matrix of the layer; denotes the node feature matrix of the layer, i.e. the output feature after one graph convolution operation; is the degree matrix of , i.e. a diagonal matrix whose diagonal elements are the degrees of the corresponding nodes; denotes the degree matrix​ the negative square root of the degree of the node for normalization operation; is the trainable weight matrix of the layer; is the activation function;

[0105] is the adjacency matrix with self-connections added, denoted as:

[0106] ;

[0107] where is the original adjacency matrix, is the identity matrix;

[0108] Optionally, in some embodiments, different graph convolution variants such as GraphSAGE or Graph Isomorphism Network (GIN) can be used to adapt to different types of network topologies. For example, when the node degree distribution in the network is extremely uneven, the sampling aggregation method in GraphSAGE can be adopted to avoid the excessive influence of highly connected nodes on the convolution operation.

[0109] The graph convolution network used in the present application includes an input layer, multiple graph convolution layers and an output layer. Among them:

[0110] The input layer receives the initial features of the nodes, such as node type, location, communication ability and other attributes;

[0111] The graph convolution layer updates the node representation by aggregating neighbor node information, and each layer contains trainable weight parameters;

[0112] The output layer generates the final node feature representation.

[0113] In the water meter collection network scenario, the graph convolution network is particularly suitable for capturing the spatial dependence between water meter nodes, for example, it can identify that some water meter nodes may have similar data transmission patterns due to their close geographical location.

[0114] Step 2.2, capture the evolution pattern of network topology over time;

[0115] In addition, the time convolution network (Temporal Convolutional Network, TCN) is combined to capture the evolution pattern of network topology over time. The time convolution operation applies one-dimensional convolution in the time dimension to process the node feature sequence containing historical information, and generates a feature representation considering time dependence.

[0116] The time convolution network structure used in the present application includes:

[0117] Causal convolutional layer: ensures the model does not use "future" information;

[0118] Dilated convolutional layer: captures long-range temporal dependencies by increasing the receptive field;

[0119] Residual connection: helps train deeper networks and alleviates the vanishing gradient problem.

[0120] In some embodiments, different time convolution parameters can be configured for different levels of spatiotemporal graphs according to the characteristics of different time scales. For example, for hourly spatiotemporal graphs, smaller convolution kernels and smaller dilation rates can be used to capture fine-grained temporal patterns; while for monthly spatiotemporal graphs, larger convolution kernels and larger dilation rates can be used to capture long-term temporal dependencies.

[0121] In practical applications, the time convolution network can learn seasonal patterns (such as water usage peak) and periodic changes (such as the difference between weekdays and weekends) from historical data, thereby predicting possible future network load changes.

[0122] Step 2.3, integrate information of different spatiotemporal scales;

[0123] In addition, for cross-layer connections in multi-level spatiotemporal graphs, a cross-layer feature fusion algorithm is applied to integrate information of different spatiotemporal scales. The specific implementation is to design a cross-layer aggregation function to weight and fuse node features of different levels.

[0124] Optionally, the cross-layer aggregation function can take different forms of implementation, such as weighted average, attention weighting or gating mechanism, etc. In some embodiments, an adaptive weight learning mechanism can be introduced to dynamically adjust the fusion weights of features of different levels according to the current network state, so that the system can better adapt to the dynamic changes of the network environment.

[0125] Step 2.4, generate node feature matrix;

[0126] Therefore, through the above processing, a node feature matrix reflecting the complete spatiotemporal characteristics of the water meter collection network is generated , where each row corresponds to a network node and each column corresponds to a feature dimension. These features include node communication patterns, load characteristics, energy consumption characteristics, and other key information.

[0127] It should be understood that through this step, the system of the present application can learn the communication patterns and load characteristics between network nodes, providing data support for subsequent network optimization. It should be noted that the application of graph neural networks enables the system to automatically extract key features contained in complex network topologies without the need for manual design of feature extraction rules.

[0128] Step 3, based on the network topology features, calculate the attention weight between nodes, model the interdependence between nodes;

[0129] This step uses the attention mechanism to calculate the influence weight between each node in the water meter collection network, which is used to accurately model the interdependence between nodes.

[0130] Step 3.1, calculate the attention weight between nodes;

[0131] According to an embodiment of the present application, based on the node feature matrix extracted in step 2 , the Graph Attention Network (GAT) is applied to calculate the attention weight between nodes. For each pair of connected nodes in the network and , the attention weight calculation formula is:

[0132] ;

[0133] where represents the attention weight of node to node , and represent the source node and the target node respectively; , and represent the feature vectors of nodes , and respectively; is a trainable weight matrix used for linear transformation of node features; is an attention vector used to calculate the similarity score between node pairs; represents the vector concatenation operation, which connects two vectors into a longer vector; represents the neighbor node set of node , represents the center node; represents the LeakyReLU activation function, which is used to introduce nonlinearity and avoid gradient vanishing; represents the exponential function, which is used to convert the score to a positive value; represents the sum of all neighbor nodes of node , which is used for normalization; represents the vector transpose operation;

[0134] Optionally, in some embodiments, the attention calculation can consider the feature information of the edges, modified to include an edge feature vector containing attributes such as communication quality, bandwidth, stability, etc. This edge-enhanced attention mechanism can more comprehensively consider the communication characteristics between nodes, further improving the accuracy of the attention weights.

[0135] The graph attention network structure adopted in the present application mainly includes the following components:

[0136] Feature transformation layer: linearly transforms the input node features to enhance expression ability;

[0137] Attention calculation layer: calculates the attention coefficients between node pairs;

[0138] Feature aggregation layer: aggregates neighbor node information according to the attention coefficients.

[0139] In the water meter collection network scenario, the graph attention network can identify the asymmetry of the influence between different nodes. For example, the relay nodes located on the critical path have a much greater impact on the entire network than the edge nodes, and through the attention mechanism, this asymmetric influence relationship can be automatically learned and quantified.

[0140] In addition, in order to capture different types of dependency relationships, a multi-head attention mechanism is applied, i.e. to calculate independent sets of attention weights, and then combine the results. This mechanism fuses multiple sets of results by averaging and concatenating the attention weights, allowing the model to focus on dependency relationships in different feature subspaces at the same time.

[0141] The multi-head attention mechanism adopted in the present application usually sets 8 to 16 attention heads (i.e. 8-16), each of which focuses on a different feature subspace, thereby being able to capture multiple types of inter-node relationships. For example, in the water meter collection network, different attention heads may focus on different aspects such as the geographical distance relationship between nodes, the energy consumption state relationship, the data transmission volume relationship, etc.

[0142] Step 3.2, form a hierarchical set of attention weights;

[0143] Next, for different levels in the multi-level spatio-temporal graph, the attention weights are calculated respectively to form a hierarchical set of attention weights.

[0144] Optionally, in some embodiments, an inter-layer attention mechanism can be introduced to calculate the attention weights between different levels. The inter-layer attention calculation measures the degree of association between nodes in different levels through a similarity function, and standardizes the result into a probability distribution. This inter-layer attention mechanism can more finely model the dependency relationships between different time scales, improving the system's ability to capture multi-scale time patterns.

[0145] Step 3.3: Generate the weighted adjacency matrix;

[0146] Therefore, a weighted adjacency matrix is ​​generated based on the attention weights. , where the matrix elements are:

[0147] ;

[0148] in This represents the adjacency matrix weighted by attention weights; This represents the adjacency matrix after attention weighting. Line number The elements of the column, corresponding to the nodes To the node The weights; Represents a node For nodes Attention weights;

[0149] It should be noted that through this step, the system of this application can automatically learn the influence weights between network nodes, accurately capture the interdependencies between different nodes, and provide an important reference for subsequent network optimization. It should be understood that, compared with traditional methods using fixed weights, the attention mechanism adopted in this application can dynamically adjust the influence weights between nodes according to the actual network state, greatly improving the model's ability to express complex dependencies in network topology.

[0150] Step 4: Based on the attention weights between nodes, perform multi-scale network state prediction to predict the future state of the network at multiple time scales.

[0151] Based on the features extracted and the calculated attention weights in the previous steps, this step predicts the future state of the water meter collection network at multiple time scales, providing a basis for the proactive optimization of network parameters.

[0152] Step 4.1: Construct a multi-scale prediction model;

[0153] According to one embodiment of this application, a multi-scale prediction model is constructed to simultaneously predict network state changes at four time scales: hourly, daily, weekly, and monthly. For each time scale... Construct the corresponding prediction model :

[0154] ;

[0155] in Time scale Up to the future Predicting network state; Indicates the current time; is the time scale is the corresponding prediction step, denotes a specific time scale (1 hour for hourly scale, 1 day for daily scale, 1 week for weekly scale, 1 month for monthly scale); denotes the time scale is the corresponding prediction function, denotes the prediction model for a specific time scale; denotes the historical network state sequence from time to denotes the historical time window length.

[0156] Optionally, in some embodiments, the prediction model can employ a probabilistic prediction method, outputting the distribution of future network states instead of a single point prediction. Such probabilistic prediction method is able to estimate the uncertainty of prediction, providing more comprehensive information for subsequent decision making, which is particularly suitable for highly dynamic and uncertain network environment.

[0157] The multi-scale prediction model of the present application employs a hierarchical design, including the following key components:

[0158] Feature extractor: extracting spatio-temporal features from original network states;

[0159] Time scale encoder: encoding information of different time scales into a unified representation;

[0160] Prediction decoder: decoding prediction results for different time scales.

[0161] In a water meter collection network, the multi-scale prediction model is able to capture both short-term fluctuations (such as morning and evening peak) and long-term trends (such as seasonal changes), thus comprehensively predicting network state changes. For example, the model can predict the possible water usage peak in a certain area within the next few hours, while also predicting the overall water usage pattern shift due to seasonal changes within the next few weeks.

[0162] Step 4.2, hierarchical time series prediction architecture;

[0163] For prediction tasks of different time scales, a hierarchical time series prediction architecture is employed, including:

[0164] Short-term prediction layer: using Gated Recurrent Unit (GRU) to predict hourly network state changes;

[0165] Medium-term prediction layer: using Long Short-Term Memory (LSTM) to predict daily and weekly network state changes;

[0166] ​Long-term prediction layer: Predict monthly network state changes using attention-enhanced temporal convolutional networks.

[0167] In some embodiments, external factors such as weather data, holiday information, city activities, etc. can be introduced as auxiliary input to further improve the accuracy of the prediction. For example, an external factor encoder can be designed to encode this information into a feature vector, which is then fused with the network state features and input into the prediction model together.

[0168] The Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) structures in this application are specially optimized to adapt to the characteristics of water meter collection network data:

[0169] GRU unit: uses update gate and reset gate to control information flow, suitable for processing short-term fluctuation data;

[0170] LSTM unit: contains input gate, forget gate and output gate, which can better preserve long-term dependency information;

[0171] Temporal attention layer: Introduce attention mechanism in long sequence prediction, focusing on key patterns in historical data.

[0172] Step 4.3, jointly optimize the prediction tasks of four time scales;

[0173] In addition, in order to improve the prediction accuracy, a multi-task learning framework is introduced to jointly optimize the prediction tasks of four time scales. The loss function is designed as the weighted sum of the prediction loss of each time scale, and the weight coefficient of each time scale can be pre-set or automatically adjusted through learning.

[0174] Optionally, in some embodiments, an adaptive weight allocation strategy can be used to dynamically adjust the weight coefficients in the loss function according to the learning difficulty and current performance of different time scale prediction tasks. This adaptive weight allocation mechanism can automatically balance the optimization difficulty of different prediction tasks during the training process, avoiding the dominance of certain tasks in the learning process of the entire model.

[0175] The multi-task learning framework of the present application uses the complementary information between different time scale tasks by sharing the underlying feature representation, which improves the overall prediction accuracy. For example, daily scale prediction can learn fine-grained fluctuation patterns from hourly scale prediction, while hourly scale prediction can obtain long-term trend information from monthly scale prediction.

[0176] Step 4.4, network parameter optimization strategy;

[0177] Therefore, based on the prediction results , compute prediction-driven network parameter optimization strategies, including:

[0178] Route strategy optimization: adjust data transmission paths according to predicted network load;

[0179] Energy management optimization: adjust node sleep and wake-up strategies according to predicted node energy consumption status;

[0180] Communication frequency optimization: adjust data collection and transmission frequency according to predicted data importance.

[0181] It should be understood that through this step, the system of the present application can predict the state change of the network at multiple time scales, and based on the prediction result, the network parameters are adjusted in advance to realize prediction-driven proactive optimization, effectively responding to changes in the network environment and load fluctuations. It should be noted that compared with the traditional passive response optimization method, the prediction-driven optimization method of the present application can respond to network changes in advance, greatly reducing the performance loss caused by network adaptation lag.

[0182] Step 5, based on the output of the multi-scale network state prediction, an abnormal event response mechanism is realized to detect and respond to sudden events in the water meter collection network;

[0183] This step builds an event-driven rapid response system for detecting and responding to sudden events in the water meter collection network, realizing real-time optimization of the network topology.

[0184] Step 5.1, build an abnormal event detector;

[0185] According to an embodiment of the present application, an abnormal event detector is built to identify abnormal events by monitoring the mutation of the network topology. The detection function is defined as:

[0186] ;

[0187] Wherein represents the detection function, which is used to judge whether the network state at the current time and the previous time is abnormal; represents the network graph state at the current time , represents the current time; represents the network graph state at the previous time , represents the previous time; and are the adjacency matrices at the current time and the previous time, respectively, indicating the corresponding time; represents the Frobenius norm of the matrix, F represents the Frobenius norm, which is used to measure the difference between the adjacency matrices at the two times; is a preset threshold value, when the difference exceeds this threshold value, an abnormal event response is triggered.

[0188] Optionally, in some embodiments, anomaly detection can consider a longer historical window, using time series anomaly detection methods. This method detects anomalies by comparing the difference between the current observed network state and the normal network state predicted based on historical data. When the difference exceeds a preset threshold value, it is determined as an abnormal event. This prediction-based anomaly detection method can capture more complex anomaly patterns, such as gradually evolving anomalies or anomalies in periodic patterns.

[0189] The anomaly event detection system of the present application includes the following key components:

[0190] Real-time monitoring module: continuously monitor network topology state changes;

[0191] Difference calculation module: calculate the topology difference measure within the continuous time window;

[0192] Threshold adaptive adjustment module: dynamically adjust the anomaly detection threshold according to historical data;

[0193] Event classification module: classify the detected abnormal events into different types (such as node failure, link interruption, etc.).

[0194] In the water meter collection network scenario, the anomaly event detector can quickly identify various sudden situations, such as network topology changes caused by pipe bursts, node failures caused by device failures, etc., thereby triggering the corresponding processing flow.

[0195] Step 5.2, rapid positioning and impact range assessment;

[0196] For the detected abnormal events, rapid positioning and impact range assessment are carried out. First, calculate the abnormal impact area, i.e. the set of nodes in the network directly affected by the abnormal event. By calculating the change degree of each node's connection state, determine which nodes are affected. When the connection change amplitude of a node exceeds the node level threshold value, the node is marked as an affected node.

[0197] In some embodiments, impact range assessment can use a graph diffusion model to predict the area that may be affected by the chain effect by simulating the propagation process of the anomaly in the network. This method calculates the probability of other nodes being affected based on the known set of affected nodes, and when the probability exceeds a preset threshold value, these nodes are also included in the potential impact range. This impact range assessment based on the propagation model can identify potential risk areas in advance, enabling more forward-looking anomaly response.

[0198] The influence range evaluation algorithm of the present application is based on a graph propagation model, which determines the potential affected range by calculating the multi-hop neighbors of the abnormal node. For different severity of anomalies, the system will adopt different radius of influence range evaluation to ensure that the spread effect of abnormal events can be accurately captured.

[0199] Step 5.3, rapid response strategy;

[0200] Then, based on the abnormal influence area, a rapid response strategy is designed to adjust the network parameters of the affected area preferentially. The response process includes:

[0201] Emergency route reconstruction: constructing a temporary communication path for the affected area;

[0202] Emergency resource scheduling: temporarily deploying communication resources from surrounding areas to the affected area;

[0203] Data collection strategy adjustment: adjusting the data collection frequency and priority according to the abnormal situation.

[0204] Optionally, in some embodiments, a reinforcement learning-based response strategy generation method can be used to learn the optimal coping strategy by interacting with the environment. This method takes the current network state as the state input and the possible response actions as the action space, and selects the optimal strategy by evaluating the long-term benefits of different actions. This reinforcement learning method can learn the optimal coping strategy for different types of anomalies through continuous trial and improvement.

[0205] The rapid response system of the present application adopts a priority scheduling mechanism to ensure that the most critical affected nodes can obtain resource support preferentially. At the same time, the system maintains a standby resource pool for quickly deploying additional resources in the event of an abnormal event, improving the recovery ability of the network.

[0206] Step 5.4, application processing strategy;

[0207] For different types of abnormal events, special processing strategies are applied:

[0208] Node failure event: ensure data transmission through redundant paths and load balancing strategy;

[0209] Communication interference event: avoid interference by dynamically adjusting communication frequency and power;

[0210] Network congestion event: relieve congestion through flow control and priority queuing mechanism;

[0211] Security threat event: ensure network security through secure communication channels and abnormal traffic isolation.

[0212] In some embodiments, the system can establish an abnormal event knowledge base to record historical abnormal events and their effective treatment schemes, so that when similar abnormal situations are encountered, the proven effective coping strategies can be quickly retrieved and applied to achieve knowledge-driven abnormal processing.

[0213] The abnormal processing strategy library of the present application contains processing templates for various types of abnormal events. The system will select the most suitable processing scheme from the strategy library according to the type of abnormality detected in real time, and adjust the parameters according to the specific circumstances to achieve intelligent processing of abnormal events.

[0214] It should be noted that through this step, the system of the present application can trigger a rapid re-optimization process when a network topology mutation is detected, and preferentially adjust the affected area to achieve real-time response to sudden events and ensure the stable operation of the water meter collection network under various abnormal conditions. It should be understood that compared with traditional systems, the abnormal event response mechanism of the present application not only can quickly detect abnormalities, but also can accurately locate the affected area and take targeted measures, greatly improving the system's ability to respond to sudden events.

[0215] A water meter collection management system for water affairs for performing the above-mentioned water meter collection management method for water affairs, comprising:

[0216] A multi-level spatiotemporal graph construction module for abstracting the water meter collection network into a multi-level spatiotemporal graph structure;

[0217] A graph neural network feature extraction module for extracting network topology features;

[0218] An attention weight calculation module for modeling the interdependence between nodes;

[0219] A multi-scale network state prediction module for predicting the future state of the network at multiple time scales;

[0220] An abnormal event response module for detecting and responding to sudden events in the water meter collection network.

[0221] As shown in Figures 2 to 6 The line graphs show the comparison of data packet loss rates before and after applying the method of the present application at different time points; the bar graphs show the comparison of data transmission success rates before and after applying the method of the present application under different network load conditions; the radar graphs show the performance comparison between the present application and traditional water meter collection management schemes; the pie charts show the proportion of nodes with different energy consumption levels in the network before and after applying the method of the present application (the inner circle shows the proportion of nodes with different energy consumption levels in the network before applying the method of the present application, and the outer circle shows the proportion of nodes with different energy consumption levels in the network after applying the method of the present application); and the line graphs show the accuracy of network state prediction at different time scales.

[0222] In this embodiment, the application provides an implementation example:

[0223] The application scenario of this embodiment is a water meter collection network in a central area of a city, covering an area of about 20 square kilometers, containing more than 15,000 smart water meter nodes, 200 data collectors, 50 relay nodes and 10 data concentrators. This area has the following characteristics:

[0224] High building density, complex communication environment, and multiple interference sources;

[0225] Various water usage patterns, including residential areas, commercial areas and some industrial areas;

[0226] Seasonal water usage changes significantly, with water usage increasing by about 40% in summer than in winter;

[0227] Frequent municipal construction in the area, leading to frequent changes in the communication environment;

[0228] Before applying the method of the present application, the water meter collection network in this area faces problems such as high data packet loss rate, unbalanced energy consumption, poor network adaptability, etc. Especially during the peak water usage period, the data transmission success rate decreases significantly, affecting the accuracy of water management decisions.

[0229] In actual deployment, the system first collects the historical data of the water meter collection network in this area for three months, including:

[0230] Node location information: geographic coordinates of each water meter, collector, relay node and concentrator;

[0231] Communication connection record: connection status, signal strength, transmission rate, etc. between nodes;

[0232] Energy status information: remaining battery percentage of battery-powered nodes;

[0233] Data collection record: data collection volume, transmission success rate, etc. at each time period.

[0234] Based on these data, the system constructs a four-layer spatiotemporal graph model:

[0235] Hourly spatiotemporal graph: captures short-term changes within 24 hours, such as network load differences during morning and evening peak hours;

[0236] Daily spatiotemporal graph: captures differences in usage patterns between weekdays and weekends;

[0237] Weekly spatiotemporal graph: captures periodic changes within a week;

[0238] Monthly spatiotemporal graph: captures seasonal change trends.

[0239] For each smart water meter node, the system records its basic attributes such as functional type (e.g. normal water meter, key monitoring water meter), installation location characteristics (indoor / outdoor), communication capability parameters, etc. The time-varying adjacency matrix adopts continuous value representation, where the matrix element represents the communication quality between node and node at time t, ranging from 0 to 1, and the larger the value, the better the communication quality.

[0240] In practical applications, the system selects three types of relationships to construct the multi-relationship adjacency tensor: physical connection relationship (based on geographical distance), logical connection relationship (based on network topology), and data flow relationship (based on actual data transmission path), so that the system can comprehensively capture the multi-dimensional characteristics of the network.

[0241] For the water meter collection network in this area, the system configures a graph convolutional network with three layers of graph convolutional layers, each containing 64 hidden units. The input features include node type (one-hot encoding), geographical location coordinates, remaining power ratio, average data transmission volume, and other 10-dimensional features.

[0242] In the actual configuration of the graph convolutional network, the hour-level spatio-temporal graph adopts an attention-weighted GraphSAGE variant to handle the extremely uneven node degree distribution; the day-level and week-level spatio-temporal graphs adopt standard GCN; and the month-level spatio-temporal graph uses GIN with a larger receptive field to capture long-term evolution patterns.

[0243] For the time convolutional network part, the system adopts different configurations for different levels of spatio-temporal graphs:

[0244] Hour level: causal convolution with kernel size 3 and dilation rate 1;

[0245] Day level: causal convolution with kernel size 3 and dilation rate 2;

[0246] Week level: causal convolution with kernel size 3 and dilation rate 4;

[0247] Month level: causal convolution with kernel size 3 and dilation rate 8.

[0248] Cross-layer feature fusion adopts an attention-weighted mechanism, and the system dynamically adjusts the fusion weight by calculating the importance of different levels of features for the current prediction task, so that in short-term prediction tasks, the hour-level features obtain higher weight, while in long-term prediction tasks, the month-level features obtain higher weight.

[0249] In practical applications, the system configures a graph attention network with 16 attention heads, each with an output dimension of 32, collectively generating a 512-dimensional node representation. The attention calculation not only considers node features but also incorporates edge features, including communication quality, stability, and bandwidth attributes.

[0250] During a peak network operation, the system automatically identified several relay nodes located in commercial areas that assumed excessive forwarding burdens. Through the attention mechanism, it quantified the influence of these nodes on the entire network. Specifically, the attention weights of these key relay nodes averaged 0.18, much higher than the 0.03 of ordinary nodes, indicating their crucial role in the network.

[0251] Each head of the multi-head attention mechanism indeed focuses on different types of node relationships: some attention heads mainly focus on nodes with similar geographical distances; some focus on nodes with similar energy consumption states; and some focus on nodes with similar data transmission modes. This multi-dimensional attention enables the system to comprehensively understand the complex dependency relationships in the network.

[0252] The inter-layer attention mechanism performs well in practical applications, especially in capturing correlation patterns at different time scales. For example, the system found that seasonal variation patterns in the monthly spatiotemporal graph influenced daily predictions, so it assigned higher inter-layer attention weights in related prediction tasks.

[0253] In the practical application of the city water meter collection network, the system adopted specially optimized prediction architectures for different prediction tasks:

[0254] Short-term prediction uses a 2-layer GRU network with a hidden layer size of 128 and an input window length of 24 (corresponding to 24 hours of historical data); medium-term prediction uses a 2-layer LSTM network with a hidden layer size of 256 and input window lengths of 7 (daily prediction) and 4 (weekly prediction); long-term prediction uses a convolutional network with temporal attention, with an input window length of 12 (corresponding to 12 months of historical data).

[0255] The system also integrates external factors as auxiliary inputs, including temperature data, precipitation forecasts, holiday information, etc. These information is converted into feature vectors by a special encoder and fused with network state features to improve prediction accuracy. For example, by incorporating weather forecast data, the system can predict changes in water usage patterns in advance when the rainy season arrives and adjust network parameters accordingly.

[0256] The adaptive weight allocation strategy in the multitask learning framework performs well in practice. In the initial stage, the system allocates the same weight to the prediction tasks of the four time scales, but as the training progresses, the system finds that the daily prediction task is more difficult, so it automatically increases the weight of the daily prediction task (from 0.25 to 0.35), so that the training loss of each task tends to be balanced.

[0257] In actual operation, the system successfully detects and responds to multiple abnormal events. Taking a pipeline burst event as an example:

[0258] At 3:14 one morning, the system detected a mutation in the adjacency matrix in the region, and the Frobenius norm change exceeded the preset threshold (0.15). The anomaly detector was triggered immediately, and the affected area was located to include 12 water meter nodes and 2 relay nodes. The system immediately implemented the following response measures:

[0259] Emergency routing reconstruction: a temporary communication path was constructed for the water meter nodes in the affected area, bypassing the damaged relay nodes;

[0260] Resource scheduling: temporarily allocate the communication resources of the 3 nearby low-load relay nodes to the affected area;

[0261] Data acquisition adjustment: increase the acquisition frequency of the water meters in the affected area from once an hour to once every 10 minutes to closely monitor possible water quantity anomalies;

[0262] These measures were deployed within 30 seconds, so that during the pipeline repair process (about 4 hours), the data acquisition success rate of the region remained above 92%, much higher than the 60% or so of traditional systems in similar situations. At the same time, the system accurately identified the location of the water leakage point through real-time analysis of the collected data, providing accurate guidance for the repair work.

[0263] Comparison of data packet loss rates before and after applying the method of the present application:

[0264] Before application: average packet loss rate 12.3% in ordinary period, average packet loss rate 20.5% in peak period, highest 26.8%;

[0265] After application: average packet loss rate 2.1% in ordinary period, average packet loss rate 2.8% in peak period, highest not more than 3.5%;

[0266] At the same time, the real-time performance of data acquisition has also been improved, and the average delay from data collection to aggregation has been reduced from 8.5 minutes to 1.2 minutes, an increase of about 85.9%.

[0267] Comparison of node energy consumption before and after applying the method of the present application:

[0268] Before application: the power consumption ratio between the 10% nodes with the highest energy consumption and the 10% nodes with the lowest energy consumption in the network is 4.7:1, resulting in that the battery replacement frequency of some nodes is more than 4 times that of other nodes;

[0269] After application: the ratio is reduced to 1.9:1, the energy consumption difference between nodes is reduced by about 60%, and the overall service life of the network is prolonged;

[0270] The average service life of the battery-powered nodes is prolonged from 10.5 months to 17.8 months, reducing the maintenance and replacement frequency and reducing the operation and maintenance cost.

[0271] As can be seen from the above practical application examples, the water meter group copy management method provided by the present application can effectively solve the problems existing in the prior art, improve the reliability, stability and efficiency of the water meter group copy network, and provide strong technical support for smart water management.

[0272] The embodiments of the present application are described above, but the embodiments are not limited to the specific implementation described above, and the specific implementation described above is only illustrative and not limiting, and those skilled in the art can make more forms of equivalent embodiments under the inspiration of the embodiments, which are all within the protection of the embodiments.

Claims

1. A method for water meter collection management in water affairs, characterized by, The method comprises the following steps: a multi-level spatio-temporal graph model is constructed to abstract the water meter collection network into a multi-level spatio-temporal graph structure; the step of constructing the multi-level spatio-temporal graph model comprises: all nodes in the water meter collection network are numbered to form a node set, and each node has position coordinates, a functional type, and energy state attribute information; a time-varying edge set is constructed based on the communication connection relationship between the nodes, and the communication quality and data transmission rate edge attributes are recorded; the edge set is converted into a time-varying adjacency matrix; a multi-level spatio-temporal graph is constructed, and each level represents the network topological characteristics at different spatio-temporal scales; inter-layer connections are established to connect the same nodes at different levels through a cross-layer edge set; based on the multi-level spatio-temporal graph structure, a graph neural network is applied to extract network topological characteristics, including applying a graph convolution network to process spatial dependence and a time convolution network to process temporal dependence; based on the network topological characteristics, the attention weights between nodes are calculated to model the mutual dependence between nodes; based on the attention weights between nodes, multi-scale network state prediction is performed to predict the future state of the network at multiple time scales; the step of performing multi-scale network state prediction comprises: a multi-scale prediction model is constructed to simultaneously predict network state changes at four time scales: hour, day, week, and month; a hierarchical time series prediction architecture is adopted for prediction tasks at different time scales; a multi-task learning framework is introduced to jointly optimize the prediction tasks at the four time scales; based on the prediction results, a prediction-driven network parameter optimization strategy is calculated; the hierarchical time series prediction architecture comprises: a short-term prediction layer that uses a gated recurrent unit to predict hour-level network state changes; a medium-term prediction layer that uses a long short-term memory network to predict day-level and week-level network state changes; a long-term prediction layer that uses an attention-enhanced time convolution network to predict month-level network state changes; the prediction-driven network parameter optimization strategy comprises route strategy optimization, energy management optimization, and communication frequency optimization; based on the output of the multi-scale network state prediction, an abnormal event response mechanism is implemented to detect and respond to sudden events in the water meter collection network.

2. The method for water meter collection management in water affairs according to claim 1, characterized in that, The step of applying a graph neural network to extract network topological characteristics comprises: for each layer in the multi-level spatio-temporal graph, a graph convolution network is applied to extract topological characteristics; a time convolution network is combined to capture the evolution pattern of network topology over time; for cross-layer connections in the multi-level spatio-temporal graph, a cross-layer feature fusion algorithm is applied to integrate information at different spatio-temporal scales; a node feature matrix reflecting the complete spatio-temporal characteristics of the water meter collection network is generated.

3. The method for water meter collection management in water affairs according to claim 1, characterized in that, The step of calculating the attention weights between nodes comprises: based on the extracted node feature matrix, a graph attention network is applied to calculate the attention weights between nodes; a multi-head attention mechanism is applied to simultaneously calculate multiple independent sets of attention weights, and then the results are combined; for different levels in the multi-level spatio-temporal graph, the attention weights are calculated separately to form a hierarchical set of attention weights; a weighted adjacency matrix is generated based on the attention weights.

4. The method for water meter collection management in water affairs according to claim 1, characterized in that, The step of implementing an abnormal event response mechanism comprises: an abnormal event detector is constructed to identify abnormal events by monitoring mutations in the network topology; For detected abnormal events, rapid positioning and impact assessment are performed; Based on the abnormal impact area, a rapid response strategy is designed to prioritize network parameter adjustment in the affected area; For different types of abnormal events, special processing strategies are applied.

5. The method for water meter collection management in water affairs according to claim 1, characterized in that, The graph convolution network includes: An input layer that receives initial features of nodes; Multiple graph convolution layers that update node representations by aggregating neighbor node information; An output layer that generates final node feature representations; The time convolution network includes causal convolution layers, dilated convolution layers, and residual connections.

6. The method for water meter collection management in water affairs according to claim 3, characterized in that, The graph attention network includes: A feature transformation layer that performs linear transformation on input node features; An attention calculation layer that calculates attention coefficients between node pairs; A feature aggregation layer that aggregates neighbor node information based on attention coefficients; The multi-head attention mechanism simultaneously calculates 8 to 16 attention heads, each focusing on a different feature subspace.

7. A water meter collection management system for use in water utilities, characterized in that, A method for water meter collection management according to any one of claims 1-6, comprising: A multi-level spatio-temporal graph construction module for abstracting the water meter collection network into a multi-level spatio-temporal graph structure; A graph neural network feature extraction module for extracting network topology features; An attention weight calculation module for modeling the interdependence between nodes; A multi-scale network state prediction module for predicting future network states at multiple time scales; An abnormal event response module for detecting and responding to sudden events in the water meter collection network.

Images