Method and system for predicting water demand trend of urban and rural integrated water supply network

Abstract

The invention discloses a water demand trend prediction method and system for an urban and rural integrated water supply pipe network, and the method comprises the steps: defining a water pipe network terminal node, obtaining a flow matrix of the water consumption of each node in nearly 7 days, and carrying out the water balance processing according to the water yield of all water plants in nearly 7 days; training the graph convolutional network according to the node flow and the node adjacency relation matrix of nearly 7 days after water balance and the selected activation function; forming a seven-day node traffic matrix by the node traffic of the nearly 6 days and the 24-hour traffic of the day, inputting the seven-day node traffic matrix into the trained graph convolutional network, and obtaining a node water demand predicted value in the next 24 hours; and calculating the daily water demand and the water demand mode according to the predicted values of the water demand of the community and village nodes. According to the invention, the graph convolutional network is applied to the water demand prediction of the integrated water supply network, the water demand trend prediction precision is higher, and the urban and rural water supply difference and relation are better embodied.

Description

Technical Field

[0001] The present invention relates to the technical field of water demand prediction, and in particular to a method and system for predicting water demand trend of an urban-rural integrated water supply network. Background Art

[0002] Accurate prediction of water demand plays an important role in improving the quality of water supply in rural areas and optimizing the daily scheduling of agricultural drinking water plants. However, in the process of promoting the integration of urban and rural water supply, due to changes in living habits, the changes in water demand in rural areas not only show a daily cycle and seasonal change cycle, but also show an annual increase trend. Due to the large changes in water consumption during rural festivals and customs, and the large proportion of population mobility, the changes in village water demand are complex and less regular than those in urban areas.

[0003] At present, relevant research at home and abroad is mostly focused on urban water demand prediction. Although existing studies have proposed mechanism models, neural networks and other related methods for rural water demand prediction, the degree of informatization of water supply projects in many rural areas of my country lags far behind that in urban areas. Informatized water meters are difficult to maintain and have low efficiency. The rural population density is low, the proportion of floating population is large, and the statistical characteristics are poor. There are great differences in rural living habits, and water consumption is greatly affected by traditional water use habits. Therefore, compared with the water demand prediction problem of urban water supply, there are new challenges in water demand prediction in rural areas.

[0004] Professor Fu Guangtao of the University of Exeter in the UK made an important review on deep learning in urban water management in Water Research in 2022 [Fu Guangtao, Jin Yiwen, Sun Siao, Yuan Zhiguo, etc. The role of deep learning in urban water management: A critical review. Water Research 223 (2022) 118973.] He summarized the application level of machine learning methods such as Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), Autoencoder (Autoencoder), Graph Neural Network (GNN), Deep Reinforcement Learning (DRL) in urban water supply volume forecasting, water supply leakage detection, etc., and believed that the current application of machine learning in this area is still in its infancy and cannot be substantially applied to production practice. In terms of water supply volume forecasting, the most successful and widely used are long short-term memory LSTM and hybrid convolutional neural network-long short-term memory (hybrid CNN-LSTM). Auto Regressive Integrated Moving Average (ARIMA) and Support Vector Machine (SVM) have more advantages in predicting peak water demand. In terms of water demand prediction for 2-20 households, Long Short-Term Memory (LSTM) exceeds Support Vector Machine (SVM), but both fail in predicting peak water supply. This shows that in rural areas of my country where the population is dispersed, the proportion of mobility is high, and customs are highly influenced, predicting water demand is still very challenging. In addition, in the context of integrated urban and rural water supply, the mobility of urban and rural populations is correlated and intertwined, which is more complicated.

[0005] Water demand changes are a key factor in scientific decision-making for secondary chlorination in water supply networks. In order to improve the scientificity and accuracy of secondary chlorination, the water demand trend method for urban-rural integrated water supply networks has become an important key technology in the process of promoting urban-rural integrated water supply. Summary of the invention

[0006] Purpose of the invention: In view of the shortcomings of the prior art, the present invention provides a method and system for predicting the water demand trend of an urban-rural integrated water supply network, so as to solve the problem that in rural areas with a dispersed population, a high mobility ratio, and a high degree of influence of customs, the water demand is highly random, and various methods have limitations in predicting water demand. It is difficult to grasp the correlation between urban and rural population mobility in the context of urban-rural integrated water supply, and it is difficult to reflect the hierarchical relationship between regions, villages, and households in the water supply system, as well as the relationship between water balance and water supply.

[0007] Technical solution: A method for predicting water demand trend of urban and rural integrated water supply network, comprising the following steps:

[0008] Define the terminal nodes of the water pipe network and obtain the flow matrix of the water consumption of each node in the past 7 days, and perform water balance processing based on the water output of all water plants in the past 7 days;

[0009] The graph convolutional network is trained by the node flow and node adjacency matrix of the last 7 days after water balance and the selected activation function, wherein the graph convolutional network is obtained according to the water supply network, and the layout of the water supply nodes and the connection relationship in the water supply network is embedded in the graph convolutional network, wherein the nodes of the graph convolutional network represent the flow of the water supply nodes in the water supply network, and the connection relationship between the water supply nodes is extracted as the node adjacency matrix of the graph convolutional network;

[0010] The 7-day node flow matrix formed by the node flow in the past 6 days and the 24-hour flow on the day is input into the trained graph convolutional network to obtain the predicted value of node water demand in the next 24 hours;

[0011] The daily water demand and water demand pattern are calculated based on the predicted water demand values ​​at the community and village nodes.

[0012] Preferably, the flow prediction range of the water supply network node is defined as the village, community and above pipeline networks, and the node water supply flow is defined as the total flow leaving the node. If the node is connected to a water supply pipe, the flow is equal to the monitored value of the pipe flow, otherwise it is calculated from the flow monitoring values ​​of all connected pipes. If the node has no monitoring value, no flow prediction is made for the node.

[0013] Preferably, water balance treatment is performed based on the water output of all water plants in the past 7 days, including:

[0014] Based on the water output of all water plants in the past seven days and the water consumption of all village and community-level nodes, a hierarchical tree of information-based water meters is established, and the water volume is balanced step by step according to the water volume monitored by the superior until the water consumption of all village and community-level nodes is equal to the total water output of all water plants.

[0015] Preferably, establishing a hierarchical tree of an information-based water meter includes:

[0016] First, determine the top level of the informationized water meter, i.e., the factory meter;

[0017] Starting from the factory table, it is divided into several regional tables, and the total flow of these regional tables is equal to the flow of the factory table;

[0018] For each regional table, it is further divided into several community tables and village tables. The total flow of these community tables and village tables is equal to the flow of the corresponding regional table.

[0019] Preferably, the water volume is balanced step by step according to the water volume monitored by the upper level, including:

[0020] The water volume of the root node is balanced to the water volume of the next layer. The sum of the water volume of the next layer should be equal to the comprehensive water volume of the previous layer. The balance method adopts the difference geometric balance. The calculation formula is as follows:

[0021]

[0022] Where: Q i is the traffic of a node at this level; ∑Q i is the total traffic of all nodes at this level; Q is the total traffic of the upper level nodes; It is the corrected flow of a node at this level.

[0023] Preferably, the selected activation function is a Swish function.

[0024] Preferably, the daily water demand and water demand pattern are calculated based on the predicted water demand values ​​of the community and village nodes, including:

[0025] Obtain water demand forecast data for community and village nodes, and aggregate the forecast data for all community and village nodes into a unified data set;

[0026] The predicted daily water demand values ​​of each community and village-level node are summed up; the predicted water demand values ​​of all village-level nodes in the same community are added together to obtain the predicted daily water demand value of the community;

[0027] Add up the predicted daily water demand values ​​of all communities and all village-level nodes;

[0028] The total daily water demand of the entire system is obtained by adding the sum of the daily water demand of the community to the sum of the daily water demand of the village level;

[0029] Analyze the temporal distribution of water demand at each community and village-level node, determine peak and off-peak periods, draw a daily water demand curve to show the changes in water demand during different time periods of the day, and analyze the cyclical changes in water demand, including daily cycles, weekly cycles, and annual cycles.

[0030] A water demand trend prediction system for an urban-rural integrated water supply network, comprising:

[0031] The water consumption acquisition module is used to define the terminal nodes of the water pipe network and obtain the flow matrix of the water consumption of each node in the past 7 days, and perform water balance processing based on the water output of all water plants in the past 7 days;

[0032] A network training module, used for training a graph convolutional network according to the node flow and node adjacency matrix of the last 7 days after water balance and a selected activation function, wherein the graph convolutional network is obtained according to a water supply network, and the layout of water supply nodes and connection relationships in the water supply network is embedded in the graph convolutional network, wherein the nodes of the graph convolutional network represent the flow of water supply nodes in the water supply network, and the connection relationships between the water supply nodes are extracted as the node adjacency matrix of the graph convolutional network;

[0033] The network prediction module is used to input the 7-day node flow matrix formed by the node flow of the past 6 days and the 24-hour flow of the day into the trained graph convolutional network to obtain the predicted value of node water demand in the next 24 hours;

[0034] The water demand prediction module is used to calculate the daily water demand and water demand pattern based on the predicted water demand values ​​of community and village nodes.

[0035] The present invention also provides a computer device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the method for predicting water demand trends in an urban and rural integrated water supply network are implemented as described above.

[0036] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method for predicting water demand trend of an urban-rural integrated water supply network as described above are implemented.

[0037] Beneficial effects: Compared with the prior art, the present invention has the following beneficial effects: The present invention proposes a method for balancing the water supply of urban and rural integrated water supply by using a balanced tree algorithm, which makes up for the water balance error and metering error, and basically reflects the hierarchical relationship between the region, village, and household of the water supply system, and reflects the relationship between water balance and water supply. The method of learning a graph neural network composed of a water supply network reflects the network topology and the mutual influence and relationship between different nodes, and optimizes the activation function according to the characteristics of water demand changes in the urban and rural integrated water supply network, thereby having a higher water demand trend prediction accuracy and better reflecting the differences and connections between urban and rural water supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The schematic diagram of the water demand trend prediction method for the urban-rural integrated water supply network;

[0039] Figure 2A flow chart of the water demand trend method for the integrated urban and rural water supply network;

[0040] Figure 3 It is a water consumption balance tree for the integrated urban and rural water supply network. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings.

[0042] The present invention is based on the information-based water meter monitoring of the integrated urban and rural water supply network, integrates the Internet of Things technology, intelligent learning algorithm and other technologies, and uses the water consumption of the past 7 days to learn with graph convolutional networks (Graph Conversion Nets, GCN) to achieve the prediction of water demand and water demand pattern of the terminal nodes (village and community water meters) in the next 24 hours. Its implementation principle is as follows Figure 1 As shown in the figure, first, the input network converts the original data such as the water volume at the factory and the water consumption of the village (community) water meter to form a data form suitable for graph convolution processing, and then extracts the node trend and correlation features through the graph convolution operation. Next, the extracted node features (trends and correlations) are nonlinearly transformed using the activation function to enhance the representation ability of the model. After the activation function, the data is normalized to standardize the signal strength of the information, making the training process more stable and improving the learning efficiency. Finally, the model output prediction result is obtained through the causal relationship of the graph convolution nonlinear operation. By repeating this process training with a large number of data sets, the coefficient matrix of the graph convolution nonlinear operation is optimized, thereby obtaining the optimized model. The optimized model can predict the results corresponding to the input layer data outside of a certain non-training set, including node flow, new connection relationships, and the new graph structure of the entire network.

[0043] Specifically, refer to Figure 2 The method for predicting water demand trend of urban and rural integrated water supply network provided by the present invention comprises the following steps:

[0044] Step S1, obtain the node flow matrix of the past 7 days, perform water balance, and use it as the input training set data of the graph convolutional network;

[0045] The water supply network node flow prediction range is defined as the village, community and above pipeline network, and the node water supply flow is defined as the total flow leaving the node. All village and community entrance water supply pipes are required to be monitored by information technology. If the node is connected to a water supply pipe (usually the village or community water supply main water meter installation point), the flow is equal to the monitored value of the pipeline flow. Otherwise, it is calculated from the monitored values ​​of all connected pipeline flows. If there is no monitored value, it means that the node is not important and is not used as a flow prediction node. Figure 2, Q1 to Qn represent the hourly traffic from node 1 to node n in the past 7 days. Xi is a set represented by a two-dimensional matrix composed of Qi, one dimension is the node and the other dimension is the time, that is, Xi represents the hourly traffic of all nodes in the past 7 days.

[0046] Since the measurement error of the informationized water meter is inevitable, and some of them may not be able to be measured and reported to the information platform due to battery shortage, the factory water volume measurement value and all village and community water consumption measurement values ​​need to be balanced to reduce the impact of informationization errors. By establishing a hierarchical tree of informationized water meters, the subordinate relationship between the meters is represented, and then the balancing process is performed according to the hierarchical tree. This hierarchical tree is also called the water supply and water consumption balance tree.

[0047] like Figure 3 As shown, first determine the top layer of the informationized water meter, that is, the factory meter; then start from the factory meter and divide it downward into several regional meters, the total flow of these regional meters is equal to the flow of the factory meter; for each regional meter, continue to divide it downward into several community meters and village meters, the total flow of these community meters and village meters is equal to the flow of the corresponding regional meter.

[0048] According to one embodiment of the present invention, a method for establishing a water supply and water consumption balance tree is as follows:

[0049] (1) The total water supply is composed of the root node (virtual node, there is no corresponding facility physically), and all water plants under the water supply network. The total water supply is obtained by the sum of the water supply of each water plant. Figure 3 As shown in, the total water supply is represented as the factory table 1, which serves as the root node.

[0050] (2) Starting from the water supply plant, the water supply plant is the parent node, and the first-level branch pipes connected to the water supply plant main pipe and the second-level branch pipes connected to the first-level branch pipes are child nodes to establish a second-level hierarchical tree. Figure 3 As shown in , the secondary hierarchy tree contains multiple region tables.

[0051] (3) A three-level hierarchical tree is established with the starting node of the secondary branch as the parent node and the nodes connected by the secondary branch and the tertiary branch as child nodes. Figure 3 As shown in , the three-level hierarchy tree contains two region tables.

[0052] (4) And so on, until the community or village node location. Figure 3 As shown in , the four-level hierarchical tree in the case contains multiple community tables and village tables.

[0053] The specific water balance algorithm process is as follows: balance the water volume of the root node to the next layer, the sum of the water volume of the next layer should be equal to the comprehensive water volume of the previous layer, and the balance method adopts the difference geometric balance, and the calculation formula is as follows:

[0054]

[0055] Where: Q i is the traffic of a node at this level; ∑Q i is the total traffic of all nodes at this level; Q is the total traffic of the upper level nodes; It is the corrected flow of a node at this level.

[0056] Until the water consumption of all village and community-level nodes is equal to the total output of all water plants.

[0057] Step S2, using the node flow in the past 7 days after water balance and the node adjacency matrix and the selected activation function to train the hourly node flow in the past 24 hours using a graph convolutional network;

[0058] Xi* represents the node flow matrix of the past 7 days after water balance, Gi represents the node adjacency matrix, and σ represents the activation function. The node flow matrix Yi* of the day obtained by the graph convolutional network is expressed as:

[0059] Yi*=σ(Σ(Gi⊙Xi*))

[0060] ⊙ represents the Hadamard product, which is the element-by-element multiplication of the corresponding positions of the elements. The above expression means that the node flow matrix Xi* of the last 7 days after water balance is element-by-element multiplied with the node adjacency matrix Gi, and then the results of all nodes are summed up, and finally the final node flow prediction matrix Yi* of the day is obtained through the activation function σ.

[0061] It should be noted that the above expression is a simplified representation. The actual graph convolution operation also includes the weight matrix W, bias b, and a multi-layer network structure. A more complete expression is as follows:

[0062] Yi*(l+1)=σ(Σ(W(l)⊙(Gi⊙Yi*(l))+b(l)))

[0063] Here Yi*(l) represents the output of layer l, W(l) is the weight matrix of layer l, b(l) is the bias vector of layer l, and Yi*(l+1) is the output of layer l+1. Through such a multi-layer structure, GCN can learn more complex node representations.

[0064] An activation function is a function that runs on the neurons of an artificial neural network. It is responsible for mapping the input of the neuron to the output, and is designed to help the network learn complex patterns in the data.

[0065] An activation function is a nonlinear transformation in a neural network. It is defined on each neuron and converts the neuron input signal into an output signal. In deep learning, activation functions are very important because they enable neural networks to capture nonlinear relationships, thereby better approximating complex functions or mappings. Activation functions can be divided into two categories, namely saturated activation functions and unsaturated activation functions. Among them, saturated activation functions include: Sigmoid, Tanh..., etc., and unsaturated activation functions include: ReLU, Leaky Relu, ELU, PReLU, RReLU..., etc. In neural network learning, activation functions are very important. They are an important guarantee for the convergence of neural networks and are also a key influencing factor for the learning efficiency of neural networks. In the water demand trend forecasting of urban and rural integrated water supply pipelines, the scale of network output points is basically between hundreds and thousands, which is a medium-sized neural network. Since the frequency of water demand forecasting is not high, at most once an hour, efficiency is not a limiting factor.

[0066] In this solution, the input layer data set is a large amount of 7-day node traffic data, and the known output layer data is the 24-hour traffic data corresponding to the 7-day traffic data. Since the training process of the neural network is universal, it is based on the input layer of a large amount of known data and the output layer data set corresponding to the input layer, and is trained according to error statistics. The training process can be automatically performed by the software to optimize the parameters and connection relationships in the neural network model. Therefore, the specific training process will not be repeated.

[0067] However, due to the large number of neural network training functions and the variety of neural network types, it is challenging to realize a neural network and activation function combination that is very suitable for the differences and connections between urban and rural water supply. This invention applies graph neural networks to the water supply forecast of water supply networks for the first time. The graph neural network reflects the correlation, and strengthening the effective information through appropriate activation functions is actually a process of strengthening and selecting information.

[0068] The present invention adopts a graph convolutional neural network GCN composed of a water supply network, which is a deep learning model specially used to process graph data. The network topology is shown as follows:

[0069] Node representation: In a water supply network, each node represents a water plant, reservoir, or pumping station. Each node in a GCN will have a feature vector to represent its state, such as flow.

[0070] Edge representation: Edges represent physical connections between nodes, such as pipes. In GCN, edges can be represented by adjacency matrices, which can also be called node adjacency matrices or node adjacency matrices. The elements in the adjacency matrix indicate whether there is a connection between two nodes and the weight of the connection. The node adjacency matrix is ​​the input data for training the network, which is directly read from the pipe network diagram and written into the data file.

[0071] Graph structure: GCN learns directly on the graph structure, which means that the topological structure in the network (i.e., the layout of nodes and edges) is directly embedded into the model.

[0072] GCN can not only reflect the network topology, but also express the mutual influence and relationship between different nodes.

[0073] Graph convolution operation: GCN updates the feature representation of nodes through the graph convolution layer. This operation takes into account the neighbor nodes of each node and updates the state of the current node by aggregating the information of the neighbor nodes. This reflects the mutual influence between nodes.

[0074] Weight sharing: In GCN, each node uses the same convolution kernel to process the information of its neighbor nodes, which ensures that the learned features are globally consistent, that is, nodes at different positions in the network but with similar functions will be given similar feature representations.

[0075] Message passing mechanism: The message passing mechanism in GCN allows nodes to exchange information. For example, traffic changes in one node may affect the status of other nodes connected to it, and this impact is reflected in the model through message passing.

[0076] The role of the adjacency matrix: The adjacency matrix defines the connection relationship between nodes. GCN uses the adjacency matrix to weight the information of neighbor nodes in the convolution operation, which directly reflects the relationship between nodes.

[0077] Consideration of multi-hop neighbors: By stacking multiple graph convolutional layers, GCN is able to consider one-hop or multi-hop neighbors of a node, thereby capturing more distant interactions between nodes.

[0078] By learning the topology and flow patterns in the water supply network, GCN can predict the flow at a certain node in the future, helping with resource allocation and water pressure management.

[0079] This method uses the records of one year of water meters in the urban-rural integrated water supply network in a city in northern Jiangsu to train and evaluate different activation functions. The initial activation functions include: Tanh, ReLU, Leaky Relu, ELU, Swish, Mish. The evaluation results are shown in Table 1, which shows that all activation functions have good training convergence, and the Nash efficiency coefficient (NSE) is greater than 0.9, and the average error is less than 3%. In terms of performance, in the intelligent learning and identification of water demand trends and patterns in urban-rural integrated water supply networks, the Swish function as an activation function has the highest convergence accuracy and high efficiency.

[0080] Table 1 Activation function efficiency and training convergence error statistics

[0081]

[0082] After training, the trained node adjacency matrix Gi, the trained node weight matrix Wi, and the daily node traffic training output matrix Hi will be obtained.

[0083] Step S3, the node flow in the past 6 days and the 24-hour flow on the current day form a 7-day node flow matrix and input it into the trained graph convolutional network to obtain the predicted value of node water demand in the next 24 hours;

[0084] According to the trained graph convolutional network, using the trained node adjacency matrix Gi' and the trained node weight matrix Wi', the predicted value of node water demand in the next 24 hours is obtained through the calculation of the neural network.

[0085] Step S4, calculating the daily water demand and water demand pattern based on the community and village node water demand forecast values.

[0086] Obtain water demand forecast data for community and village-level nodes, and aggregate the forecast data for all community and village-level nodes into a unified data set; sum up the daily water demand forecast values ​​for each community and village-level node; add up the water demand forecast values ​​for all village-level nodes in the same community to obtain the daily water demand forecast value for the community; add up the daily water demand forecast values ​​for all communities and the daily water demand forecast values ​​for all village-level nodes; add the sum of the community's daily water demand and the sum of the village-level daily water demand to obtain the total daily water demand for the entire system; analyze the time distribution of water demand for each community and village-level node, determine the peak and off-peak periods, draw a daily water demand curve, show the changes in water demand in different time periods of the day, and analyze the periodic changes in water demand, including daily cycles, weekly cycles, and annual cycles.

[0087] The predicted daily water demand is the sum of the predicted water demand values ​​of the nodes in the next 24 hours, and the 24-hour average water demand is calculated The water demand pattern is a standardized dimensionless water demand curve, from hourly water demand to 24-hour average water demand. The specific calculation formula is as follows:

[0088]

[0089] Where: Q i For the predicted hourly water demand for the next 24 hours, is the average water demand in 24 hours, The water demand pattern has good stability and continuity, reflecting the habitual changes in the daily water consumption of the village or community.

[0090] The present invention also provides a water demand trend prediction system for an urban-rural integrated water supply network, comprising:

[0091] The water consumption acquisition module is used to define the terminal nodes of the water pipe network and obtain the flow matrix of the water consumption of each node in the past 7 days, and perform water balance processing based on the water output of all water plants in the past 7 days;

[0092] A network training module, used for training a graph convolutional network according to the node flow and node adjacency matrix of the last 7 days after water balance and a selected activation function, wherein the graph convolutional network is obtained according to a water supply network, and the layout of water supply nodes and connection relationships in the water supply network is embedded in the graph convolutional network, wherein the nodes of the graph convolutional network represent the flow of water supply nodes in the water supply network, and the connection relationships between the water supply nodes are extracted as the node adjacency matrix of the graph convolutional network, and the node adjacency matrix is ​​obtained by directly reading the pipe network graph;

[0093] The network prediction module is used to input the 7-day node flow matrix formed by the node flow of the past 6 days and the 24-hour flow of the day into the trained graph convolutional network to obtain the predicted value of node water demand in the next 24 hours;

[0094] The water demand prediction module is used to calculate the daily water demand and water demand pattern based on the predicted water demand values ​​of community and village nodes.

[0095] It should be understood that the water demand trend prediction system for the urban-rural integrated water supply network in the embodiment of the present invention can implement all the technical solutions in the above-mentioned method embodiment, and the functions of its various functional modules can be specifically implemented according to the methods in the above-mentioned method embodiments. The specific implementation process can refer to the relevant description in the above-mentioned embodiments, which will not be repeated here.

[0096] The present invention also provides a computer device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the method for predicting water demand trends in an urban and rural integrated water supply network are implemented as described above.

[0097] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method for predicting water demand trend of an urban-rural integrated water supply network as described above are implemented.

[0098] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices (systems), computer equipment or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0099] The present invention is described with reference to a flowchart of a method according to an embodiment of the present invention. It should be understood that each process in the flowchart and a combination of processes in the flowchart can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A device that specifies functions in a process or multiple processes.

[0100] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A function specified in a process or multiple processes.

[0101] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 The steps of a specified function in a process or multiple processes.

Claims

1. A method for predicting water demand trend of urban and rural integrated water supply network, characterized in that: The following steps are involved: Define the terminal nodes of the water pipe network and obtain the flow matrix of the water consumption of each node in the past 7 days, and perform water balance processing based on the water output of all water plants in the past 7 days; The graph convolutional network is trained by the node flow and node adjacency matrix of the last 7 days after water balance and the selected activation function, wherein the graph convolutional network is obtained according to the water supply network, and the layout of the water supply nodes and the connection relationship in the water supply network is embedded in the graph convolutional network, wherein the nodes of the graph convolutional network represent the flow of the water supply nodes in the water supply network, and the connection relationship between the water supply nodes is extracted as the node adjacency matrix of the graph convolutional network; The 7-day node flow matrix formed by the node flow in the past 6 days and the 24-hour flow on the day is input into the trained graph convolutional network to obtain the predicted value of node water demand in the next 24 hours; The daily water demand and water demand pattern are calculated based on the predicted water demand values ​​at the community and village nodes.

2. The method according to claim 1, characterized in that The flow prediction range of the water supply network node is defined as the village, community and above pipeline networks. The node water supply flow is defined as the total flow leaving the node. If the node is connected to a water supply pipe, the flow is equal to the monitored value of the pipe flow. Otherwise, it is calculated from the monitored values ​​of the flow of all connected pipes. If the node has no monitored value, no flow prediction is made for the node.

3. The method according to claim 1, characterized in that Water balance processing is carried out based on the water output of all water plants in the past 7 days, including: Based on the water output of all water plants in the past seven days and the water consumption of all village and community-level nodes, a hierarchical tree of information-based water meters is established, and the water volume is balanced step by step according to the water volume monitored by the superior until the water consumption of all village and community-level nodes is equal to the total water output of all water plants.

4. The method according to claim 3, characterized in that: The hierarchical tree for establishing an information-based water meter includes: First, determine the top level of the informationized water meter, i.e., the factory meter; Starting from the factory table, it is divided into several regional tables, and the total flow of these regional tables is equal to the flow of the factory table; For each regional table, it is further divided into several community tables and village tables. The total flow of these community tables and village tables is equal to the flow of the corresponding regional table.

5. The method according to claim 3, characterized in that: Balance the water volume step by step according to the water volume monitored by the superior, including: The water volume of the root node is balanced to the water volume of the next layer. The sum of the water volume of the next layer should be equal to the comprehensive water volume of the previous layer. The balance method adopts the difference geometric balance. The calculation formula is as follows: Where: Q i is the traffic of a node at this level; ∑Q i is the total traffic of all nodes at this level; Q is the total traffic of the upper level nodes; It is the corrected flow of a node at this level.

6. The method according to claim 1, characterized in that The selected activation function is the Swish function.

7. The method according to claim 1, characterized in that Calculate daily water demand and water demand patterns based on community and village node water demand forecasts, including: Obtain water demand forecast data for community and village nodes, and aggregate the forecast data for all community and village nodes into a unified data set; The predicted daily water demand values ​​of each community and village-level node are summed up; the predicted water demand values ​​of all village-level nodes in the same community are added together to obtain the predicted daily water demand value of the community; Add up the predicted daily water demand values ​​of all communities and all village-level nodes; The total daily water demand of the entire system is obtained by adding the sum of the daily water demand of the community to the sum of the daily water demand of the village level; Analyze the temporal distribution of water demand at each community and village-level node, determine peak and off-peak periods, draw a daily water demand curve to show the changes in water demand during different time periods of the day, and analyze the cyclical changes in water demand, including daily cycles, weekly cycles, and annual cycles.

8. A water demand trend prediction system for urban and rural integrated water supply network, characterized in that: include: The water consumption acquisition module is used to define the terminal nodes of the water pipe network and obtain the flow matrix of the water consumption of each node in the past 7 days, and perform water balance processing based on the water output of all water plants in the past 7 days; A network training module, used for training a graph convolutional network according to the node flow and node adjacency matrix of the last 7 days after water balance and a selected activation function, wherein the graph convolutional network is obtained according to a water supply network, and the layout of water supply nodes and connection relationships in the water supply network is embedded in the graph convolutional network, wherein the nodes of the graph convolutional network represent the flow of water supply nodes in the water supply network, and the connection relationships between the water supply nodes are extracted as the node adjacency matrix of the graph convolutional network; The network prediction module is used to input the 7-day node flow matrix formed by the node flow of the past 6 days and the 24-hour flow of the day into the trained graph convolutional network to obtain the predicted value of node water demand in the next 24 hours; The water demand prediction module is used to calculate the daily water demand and water demand pattern based on the predicted water demand values ​​of community and village nodes.

9. A computer device, characterized in that: include: one or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the method for predicting water demand trend of urban and rural integrated water supply network as described in any one of claims 1-7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of the method for predicting water demand trend of urban and rural integrated water supply network as described in any one of claims 1-7 are implemented.

Images