Water plant water volume prediction method based on time-varying characteristics and intelligent scheduling system

Abstract

The invention provides a water plant water volume prediction method based on time-varying characteristics. The method comprises the following steps: generating a digital topology model based on key physical elements and connection relations of a water plant physical water circulation network; wherein the digital topology model comprises topology nodes and directed edges; defining a water volume index for a topological node in the digital topological model based on the available water volume definition, collecting topological node time sequence data, and processing the topological node time sequence data and external feature data to obtain a multi-dimensional time sequence data set; based on the digital topology model and the multi-dimensional time sequence data set, constructing a node spatial-temporal feature matrix fused with the spatial-temporal dimension; and establishing a prediction model for simultaneously capturing time dependence and space dependence by using the node features fused with the space-time dimensions. The method effectively solves the common problem that the traditional method is difficult to model the complex spatial dependency relationship and the time dynamic characteristic at the same time, thereby achieving the more accurate and more reliable prediction of the future water consumption.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water supply scheduling optimization, in particular to a water plant water quantity prediction method based on time-varying characteristics and an intelligent scheduling system. BACKGROUND

[0002] The traditional water supply scheduling mode seriously depends on the personal experience of the scheduling personnel, and the judgment is made through rough grasping of the historical water use law (such as season, day type) and limited real-time monitoring data (such as clear water pool water level). The water pump or valve is manually started and stopped. This way is highly subjective and has a lagging response, and it is difficult to cope with the suddenness and randomness of water use, which can easily lead to low system operation efficiency. The specific performance is as follows: the conservative scheduling strategy taken to ensure water supply safety often causes frequent start and stop of water pumps, high pressure of pipe network at night, which not only wastes a lot of electric energy and accelerates equipment wear and tear, but also may affect water quality safety due to water stagnation.

[0003] In recent years, although some researches have tried to introduce deep learning models (such as LSTM) to better capture the nonlinear time sequence law of water use, they have not fundamentally solved the problem of spatial dependence. Although the data of different nodes can be simply spliced as model input, the physical topology structure composed of pipes, valves and pump stations cannot be explicitly expressed and learned, and the prediction results lack consistency in the spatial dimension, which is difficult to meet the demand of collaborative optimization scheduling of the whole system.

[0004] In terms of scheduling, the existing optimization methods are mostly based on static working conditions or historical average data, and cannot use high-precision future water quantity prediction results as pre-input. The scheduling scheme generated by them cannot predict the changes in water demand in the next few hours to tens of hours, cannot store water in advance before the peak of electricity consumption, and cannot run smoothly during the low water consumption period, so the optimization effect is limited, and it is essentially a "post-response" rather than a "pre-scheduling".

[0005] Therefore, there is an urgent need in the art for a new prediction method that can deeply integrate the physical topology structure of the water supply system and accurately capture the time and space variation characteristics of water use. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a water plant water quantity prediction method based on time-varying characteristics and an intelligent scheduling system to solve the problems raised in the background art.

[0007] To achieve the above purpose, the present application realizes the following technical scheme: a water plant water quantity prediction method based on time-varying characteristics, comprising the following steps: S1: generating a digital topology model based on the key physical elements and connection relationships of the physical water circulation network of the water plant; In the digital topology model, there are topology nodes and directed edges. S2: defining a water quantity index based on the available water quantity for the topological nodes in the digitalized topological model, collecting time series data of the topological nodes and external characteristic data to obtain a multi-dimensional time series data set after processing; S3: based on the digitalized topological model and the multi-dimensional time series data set, constructing a node space-time feature matrix fusing space-time dimensions; S4: using the node features fusing space-time dimensions, establishing a prediction model capturing time dependence and space dependence simultaneously; S5: training the prediction model using historical data, and inputting data collected in real time and processed through S2 and S3 into the trained prediction model for rolling prediction to obtain available predicted water quantity of each node in a future period; S6: taking the available predicted water quantity as input, solving by an optimization algorithm to generate an optimal device scheduling instruction set.

[0008] As a further optimization, the S1 specifically comprises: all key physical elements in the physical water circulation network of the water plant are listed, each physical element is abstracted as a node, a unique identifier is assigned to each node, and its static attributes are recorded; According to the actual water flow direction and the pipe connection relationship, the connection between all nodes is defined, each pipe connection is abstracted as a directed edge, a unique identifier is assigned to each edge, and the static attributes are recorded; Using graph theory knowledge, all the above nodes and edges are combined to construct a directed graph G = (V, E), where V represents the set of all nodes, E represents the set of all edges, and the directed graph is the digitalized topological model of the physical water circulation network of the water plant.

[0009] As a further optimization, the specific process of S2 comprises: S21: defining the available water quantity for different types of nodes; S22: acquiring time series data related to the available water quantity of each topological node through a sensor array and a SCADA system, and collecting external characteristic data closely related to water consumption; S23: processing all collected data to obtain a multi-dimensional time series data set.

[0010] As a further optimization, the steps of S3 specifically comprise: Extracting historical time series features of each topological node from the multi-dimensional time series data set, including sliding window statistics, periodicity and trend features; Based on the adjacency relationship of the digitalized topological model, the historical data of the upstream and downstream adjacent nodes of each node are aggregated to form spatial correlation features; The weather forecast and the date type external feature are associated and fused with the corresponding node or global network; A spatio-temporal feature matrix of each node at each time step is formed, which fuses the node's own history, neighbor state and external factors; As a further optimization, the prediction model comprises a spatial dimension modeling module and a temporal dimension modeling module; The spatial dimension modeling module is a graph convolution network or a graph attention network, which is used to capture the spatial correlation between nodes; The temporal dimension modeling module is a gated recurrent unit, a temporal convolution network or a Transformer architecture, which is used to capture the variation law of node features over time.

[0011] As a further optimization, the specific process of S6 comprises: Taking the available predicted water quantity as the core input, a multi-objective optimization function is established with the water plant physical water cycle network topology and the hydraulic model as constraints; Wherein, the optimization objectives include: minimizing the total energy consumption of water pumps, ensuring that the pipe network pressure and the clear water pool water level are in the safety interval, reducing the equipment start-stop frequency, and optimizing the water quality residence time; A mathematical programming or intelligent optimization algorithm is used to solve the optimization function to generate an optimal scheduling plan for the future period; The optimal scheduling plan is output to the scheduling personnel or directly issued to the production control system for execution.

[0012] As a further optimization, a water plant water quantity intelligent scheduling system based on time-varying features is used to implement the above-mentioned water plant water quantity prediction method based on time-varying features, comprising: A digital topology model construction module generates a digital topology model based on the key physical elements and connection relationships of the water plant physical water cycle network; Wherein, the digital topology model contains topology nodes and directed edges, and the digital topology model is a directed graph G = (V, E), wherein V represents the set of all nodes, and E represents the set of all edges; A data acquisition and processing module defines water quantity indicators for the topology nodes in the digital topology model based on available water quantity, and acquires multi-dimensional time series data sets after processing topology node time series data and external feature data; A node feature construction module constructs a node spatio-temporal feature matrix that fuses the spatial and temporal dimensions based on the digital topology model and the multi-dimensional time series data set; A model construction and prediction output module is configured to use the node features in the fused space-time dimension to establish a prediction model that captures time dependence and space dependence, train the prediction model using historical data, input data collected in real time and processed by the data acquisition and processing module and the node feature construction module into the trained prediction model for rolling prediction, and obtain the available predicted water volume of each node in a future period; An instruction generation module is configured to use the available predicted water volume as input, solve by an optimization algorithm, and generate an optimal device scheduling instruction set.

[0013] As a further optimization, the specific process of the data acquisition and processing module acquiring the multi-dimensional time series data set includes: The available water volume of each node is defined according to the type of the node; Time series data related to the available water volume of each topology node is acquired by a sensor array and a SCADA system, and external feature data closely related to the water consumption is collected; All the collected data is processed to obtain a multi-dimensional time series data set.

[0014] As a further optimization, the node feature construction module constructing the node space-time feature matrix in the fused space-time dimension includes: The historical time series features of each topology node itself are extracted from the multi-dimensional time series data set, including sliding window statistics, periodicity and trend features; Based on the adjacency relationship of the digital topology model, the historical data of the upstream and downstream adjacent nodes of each node is aggregated to form a spatial correlation feature; The weather forecast and date type external features are associated and fused with the corresponding node or global network; A space-time feature matrix of each node is formed at each time step, which integrates the historical data, neighbor state and external factors of each node; As a further optimization, the prediction model includes a space dimension modeling module and a time dimension modeling module; The space dimension modeling module is a graph convolution network or a graph attention network, which is configured to capture the spatial correlation between nodes; The time dimension modeling module is a gated recurrent unit, a time convolution network or a Transformer architecture, which is configured to capture the change rule of the node features over time.

[0015] The application provides a water plant water quantity prediction method and intelligent scheduling system based on time-varying characteristics, which has the following beneficial effects: the physical water cycle network is accurately abstracted into a digital topological model, laying a solid spatial foundation for subsequent data fusion and model construction, and ensuring that all calculations and analyses conform to the actual physical constraints. By systematically defining node water quantity indicators and fusing multi-source time-varying data, high-dimensional features that can capture the spatial and temporal variation of water consumption are constructed, greatly improving the depth and breadth of data utilization. The spatiotemporal graph neural network prediction model used innovatively couples the graph convolution network with the recurrent neural unit, effectively solving the common problem that traditional methods cannot simultaneously model complex spatial dependence relationships and temporal dynamic characteristics, thereby achieving more accurate and reliable prediction of future available water quantity.

[0016] And based on the high-precision prediction result, a multi-objective optimization solution is obtained with the hydraulic model as a strict constraint, which can generate a scheduling scheme that achieves comprehensive optimization in economic benefits, safe operation, equipment life, and water quality safety. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A flowchart of the water plant water quantity prediction method based on time-varying characteristics of the application; Figure 2 A block diagram of the interface element dynamic adaptive regulation and control system based on user real-time behavior perception of the application. DETAILED DESCRIPTION

[0018] The embodiments of the application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the application, and cannot be understood as limiting the application.

[0019] The following disclosure provides many different embodiments or examples for implementing different structures of the application. In order to simplify the disclosure of the application, the components and settings of specific examples are described below. Of course, they are only examples and the purpose is not to limit the application. In addition, the application can repeatedly refer to numerals and / or letters in different examples, and such repetition is for the purpose of simplification and clarity, which itself does not indicate the relationship between the various embodiments and / or settings discussed. In addition, the application provides examples of various specific processes and materials, but those skilled in the art can realize the application of other processes and / or the use of other materials.

[0020] As Figure 1 shown, the embodiment of the application provides a water plant water quantity prediction method based on time-varying characteristics, which includes the following steps: S1: generating a digital topology model based on key physical elements and connection relationships of a physical water circulation network of a water plant; In the digital topology model, there are topology nodes and directed edges. Specifically, all key physical elements in the physical water circulation network of the water plant are identified and listed, each physical element is abstracted as a node, a unique identifier (ID) is assigned to each node, and its static attributes are recorded. The key physical elements include water sources (such as rivers, reservoirs, deep wells), water intake pumping stations, water transmission pipelines, water treatment plants (including various treatment structures such as sedimentation tanks, filter tanks, and clear water tanks), water distribution pumping stations, booster pumping stations, water user areas / large user nodes, water towers / high-level water tanks, valves, and monitoring points. The static attributes of the nodes are specifically types, capacities, elevations, design flow rates, etc.

[0021] According to the actual water flow direction and pipeline connection relationship, the connection between all nodes is defined, each pipeline connection is abstracted as a directed edge, a unique identifier is assigned to each edge, and the static attributes are recorded. The static attributes of the directed edge are specifically the starting node ID and the ending node ID, the pipeline length, the pipe diameter, the roughness, and other static attributes.

[0022] Using graph theory knowledge, all the above nodes and edges are combined to construct a directed graph G = (V, E), where V represents the set of all nodes, E represents the set of all edges, and the directed graph is the digital topology model of the physical water circulation network of the water plant.

[0023] It can be understood that the directed graph should accurately reflect the physical flow path of water and the overall structure of the system.

[0024] S2: defining water quantity indicators for the topology nodes in the digital topology model based on available water quantity, collecting time series data of the topology nodes and processing external feature data to obtain a multi-dimensional time series data set; Specifically, S21: for different types of nodes, define their available water quantity. In the digital topology model, each topology node defines the available water quantity with specific indicators. The available water quantity with specific indicators corresponds to specific and measurable physical quantities one-to-one.

[0025] For example, for water sources and clear water tanks, the real-time water level or water storage capacity is used; for water plants and pumping stations, the instantaneous water inflow / outflow is used; for water user nodes, the inlet pressure (which indirectly reflects water quantity demand and availability) is used.

[0026] S22: Obtain time series data related to the available water quantity of each topology node through the sensor array and the SCADA system, and collect external feature data closely related to water consumption; The time series data includes but is not limited to water level, flow, pressure, water pump start-stop state, valve opening degree, etc. These data are recorded at fixed time intervals (such as every 5 minutes) and accurately bound to the nodes in the topology graph; the external feature data includes but is not limited to future weather forecast (temperature, precipitation, holidays), historical same-period water consumption, social and economic activity index (such as epidemic control level), etc.

[0027] S23: Process all collected data to obtain a multi-dimensional time series data set.

[0028] Specifically, the collected time series data and external feature data are cleaned, aligned, normalized, etc. to form a multi-dimensional time series data set, and each data point in the multi-dimensional time series data set corresponds to a specific topology node and time on the digital topology model.

[0029] S3: Based on the digital topology model and the multi-dimensional time series data set, a node spatio-temporal feature matrix integrating space and time dimensions is constructed; In this embodiment, the step of S3 specifically includes: From the multi-dimensional time series data set, the historical time series features of each topology node itself are extracted, including sliding window statistics, periodicity and trend features; Specifically, for each topology node, features are extracted from its historical time series data, such as: sliding window statistics (mean, variance, maximum) of the past 1 hour, 3 hours, 24 hours, time series trend, periodicity features (daily cycle, weekly cycle).

[0030] Based on the adjacency relationship of the digital topology model, the historical data of the upstream and downstream adjacent nodes of each node are aggregated to form spatial correlation features; Specifically, based on the adjacency relationship of the digital topology model, the historical water quantity / flow data of the upstream or downstream adjacent nodes of each topology node is aggregated. For example, the available water quantity of a water distribution node is largely affected by the current flow of the water supply pump station upstream. Using techniques such as graph embedding, the position and structural importance of the node in the entire network are converted into a feature vector.

[0031] Correlate and integrate weather forecasts and date type external features with the corresponding nodes or the global network; Form a spatio-temporal feature matrix of each node that integrates its own history, neighbor status and external factors at each time step; S4: Use the node features integrating space and time dimensions to establish a prediction model that captures both time dependence and spatial dependence; Specifically, the prediction model includes a spatial dimension modeling module and a temporal dimension modeling module; the spatial dimension modeling module is a graph convolutional network or a graph attention network, used to capture the spatial correlation between nodes; the temporal dimension modeling module is a gated recurrent unit, a temporal convolutional network, or a Transformer architecture, used to capture the changing patterns of node features over time.

[0032] It should be noted that the input of the prediction model is the spatiotemporal feature matrix of the past T consecutive time steps, and the output is the predicted value of available water volume for all nodes or key nodes in the next K time steps.

[0033] In this embodiment, the prediction model specifically couples a graph convolutional network and a gated recurrent unit in a reasonable order to construct a complete graph convolutional-GRU model. The input of this model is the feature matrix of all nodes and the graph structure at multiple historical time steps, and the output is the predicted available water volume of each node at multiple future time steps (e.g., the next 24 hours). The graph convolution-GRU calculation formula is as follows: ; ; ; ; in, , This represents the update gate and reset gate matrix for all nodes; This represents the input feature matrix of all nodes at time step t; Let A represent the hidden state matrix of all nodes at time step t-1; let A represent the adjacency matrix of the graph. A graph convolution operation that takes matrix M as input and A as adjacency relation (where the subscripts are...). This means that each gate uses independent, trainable GCN weight parameters. , Indicates the activation function; This represents the candidate hidden state matrix for all nodes; This represents the updated hidden state matrix of all nodes at time step t. This represents the Hadamard product.

[0034] The final prediction formula is: ; in, This represents the predicted water volume values ​​for all nodes at the next time step (or K steps in the future). This indicates the output function.

[0035] S5: Use historical data to train the prediction model, and input the data collected in real time and processed by S2 and S3 into the trained prediction model for rolling prediction to obtain the available predicted water volume of each node in the future period. Specifically, the prediction model is trained using historical time-series data in a supervised learning manner, with the previous T steps of data as input and the next K steps of real data as labels, and the model parameters are optimized by minimizing the prediction error loss function. It receives the latest sensor data in real time, automatically executes S2 and S3, and then inputs it into the model to predict the available water volume of each key node in the entire water cycle network over a period of time in the future, thus forming the prediction results.

[0036] Among them, optimizing model parameters by minimizing the prediction error loss function is an existing technique, which will not be elaborated on here.

[0037] S6: Using the available predicted water volume as input, the optimal equipment scheduling instruction set is generated through optimization algorithms.

[0038] Specifically, using the available predicted water volume as the core input, a multi-objective optimization function is established with constraints of the water plant's physical water cycle network topology and hydraulic model; The optimization objectives include: minimizing the total energy consumption of water pumps, ensuring that the pipeline pressure and the water level in the clear water tank are within a safe range, reducing the frequency of equipment start-up and shutdown, and optimizing the water retention time. The optimization function is solved using mathematical programming or intelligent optimization algorithms to generate the optimal scheduling plan for the future time period; The optimal scheduling plan is output to the scheduling personnel or directly issued to the production control system for execution.

[0039] Understandably, the scheduling plan includes: the start-up and shutdown schedule of each water pump, speed settings, valve opening adjustment plan, etc.

[0040] The water plant water volume prediction method based on time-varying features provided in this embodiment lays a solid spatial foundation for subsequent data fusion and model construction by accurately abstracting the physical water cycle network into a digital topology model, ensuring that all calculations and analyses conform to actual physical constraints. By systematically defining node water volume indicators and fusing multi-source time-varying data, a high-dimensional feature capable of simultaneously capturing the spatiotemporal variation patterns of water consumption is constructed, greatly improving the depth and breadth of data utilization. The spatiotemporal graph neural network prediction model adopted innovatively couples graph convolutional networks with recurrent neural units, effectively solving the common problem of traditional methods being unable to simultaneously model complex spatial dependencies (such as pipeline topology, upstream and downstream influences) and temporal dynamic characteristics (such as daily / weekly cycles, trends), thereby achieving more accurate and reliable predictions of future available water volume. Furthermore, based on the high-precision prediction results, multi-objective optimization is performed with the hydraulic model as a strict constraint, generating a scheduling scheme that achieves comprehensive optimality in terms of economic benefits, safe operation, equipment lifespan, and water quality safety.

[0041] like Figure 2 As shown, this embodiment also provides a water plant water volume intelligent scheduling system based on time-varying characteristics, used to implement the above-mentioned water plant water volume prediction method based on time-varying characteristics, including: The digital topology model building module generates a digital topology model based on the key physical components and connections of the water plant's physical water circulation network. The digital topology model includes topological nodes and directed edges. Specifically, the digital topology model is a directed graph G = (V, E), where V represents the set of all nodes and E represents the set of all edges. The data acquisition and processing module defines the available water volume as the water volume index of the topological nodes in the digital topology model, and obtains a multidimensional time series dataset by collecting and processing the time series data of the topological nodes and external feature data. The node feature construction module constructs a node spatiotemporal feature matrix that integrates spatiotemporal dimensions based on a digital topology model and a multidimensional time series dataset. The model building and prediction output module is used to establish a prediction model that simultaneously captures time dependence and spatial dependence by utilizing the node features that integrate spatiotemporal dimensions. At the same time, the prediction model is trained using historical data, and the data collected in real time and processed by the data acquisition and processing module and the node feature building module is input into the trained prediction model for rolling prediction to obtain the available predicted water volume of each node in the future period. Instruction generation module: Taking the available predicted water volume as input, it uses an optimization algorithm to generate the optimal set of equipment scheduling instructions.

[0042] As a further preferred embodiment, the specific process by which the data acquisition and processing module acquires the multidimensional time series dataset includes: Define the available water volume for different types of nodes; Time-series data related to the available water volume of each topology node are acquired through sensor arrays and SCADA systems, and external characteristic data closely related to water consumption are collected. All collected data are processed to obtain a multidimensional time series dataset.

[0043] As a further preferred embodiment, the node feature construction module constructs a node spatiotemporal feature matrix that integrates spatiotemporal dimensions, specifically including: Extract the historical time series features of each topological node from the multidimensional time series dataset, including sliding window statistics, periodicity and trend features; Based on the adjacency relationship of the digital topology model, historical data of upstream and downstream adjacent nodes of each node are aggregated to form spatial association features; Associate and integrate weather forecasts and date type external features with corresponding nodes or the global network; Form a spatiotemporal feature matrix for each node at each time step, which integrates its own history, neighbor states, and external factors; As a further preferred embodiment, the prediction model includes a spatial dimension modeling module and a temporal dimension modeling module; The spatial dimension modeling module is a graph convolutional network or a graph attention network, used to capture the spatial correlation between nodes; The time dimension modeling module is a gated recurrent unit, a temporal convolutional network, or a Transformer architecture, used to capture the changing patterns of node features over time.

[0044] It should be noted that the input of the prediction model is the spatiotemporal feature matrix of the past T consecutive time steps, and the output is the predicted value of available water volume for all nodes or key nodes in the next K time steps.

[0045] In this embodiment, the prediction model specifically couples a graph convolutional network and a gated recurrent unit in a reasonable order to construct a complete graph convolutional-GRU model. The input of this model is the feature matrix of all nodes and the graph structure at multiple historical time steps, and the output is the predicted available water volume of each node at multiple future time steps (e.g., the next 24 hours). The graph convolution-GRU calculation formula is as follows: ; ; ; ; in, , This represents the update gate and reset gate matrix for all nodes; This represents the input feature matrix of all nodes at time step t; Let A represent the hidden state matrix of all nodes at time step t-1; let A represent the adjacency matrix of the graph. A graph convolution operation that takes matrix M as input and A as adjacency relation (where the subscripts are...). This means that each gate uses independent, trainable GCN weight parameters. , Indicates the activation function; This represents the candidate hidden state matrix for all nodes; This represents the updated hidden state matrix of all nodes at time step t. This represents the Hadamard product.

[0046] The final prediction formula is: ; in, This represents the predicted water volume values ​​for all nodes at the next time step (or K steps in the future). This indicates the output function.

[0047] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting water volume in water plants based on time-varying characteristics, characterized in that, Includes the following steps: S1: Generate a digital topology model based on the key physical components and connections of the water plant's physical water circulation network; The digital topology model includes topological nodes and directed edges; S2: Based on the definition of available water volume, the topological nodes in the digital topological model are defined as water volume indicators. After collecting and processing the time series data of the topological nodes and external feature data, a multidimensional time series dataset is obtained. S3: Based on the digital topology model and multidimensional time series dataset, construct a node spatiotemporal feature matrix that integrates spatiotemporal dimensions; S4: Utilize the node features that fuse the spatiotemporal dimensions to establish a prediction model that simultaneously captures temporal and spatial dependencies; S5: Use historical data to train the prediction model, and input the data collected in real time and processed by S2 and S3 into the trained prediction model for rolling prediction to obtain the available predicted water volume of each node in the future period. S6: Using the available predicted water volume as input, the optimal equipment scheduling instruction set is generated through an optimization algorithm.

2. The water plant water volume prediction method based on time-varying characteristics according to claim 1, characterized in that, The interior of S1 specifically includes: It lists all the key physical components in the physical water circulation network of the water plant, abstracts each physical component into a node, assigns a unique identifier to each node, and records its static attributes; Based on the actual water flow direction and pipe connection relationship, define the connection between all nodes, abstract each pipe connection as a directed edge, assign a unique identifier to each edge, and record static attributes; Using graph theory, all the nodes and edges mentioned above are combined to construct a directed graph G = (V, E), where V represents the set of all nodes and E represents the set of all edges. The directed graph is the digital topology model of the physical water circulation network of the water plant.

3. The water plant water volume prediction method based on time-varying characteristics according to claim 1, characterized in that, The specific process of S2 includes: S21: Define the available water volume for different types of nodes; S22: Acquire time-series data related to the available water volume of each topology node through sensor arrays and SCADA systems, and collect external characteristic data closely related to water consumption; S23: Process all collected data to obtain a multidimensional time series dataset.

4. The water plant water volume prediction method based on time-varying characteristics according to claim 3, characterized in that, The steps in S3 specifically include: Extract the historical time series features of each topological node from the multidimensional time series dataset, including sliding window statistics, periodicity and trend features; Based on the adjacency relationship of the digital topology model, historical data of upstream and downstream adjacent nodes of each node are aggregated to form spatial association features; Associate and integrate weather forecasts and date type external features with corresponding nodes or the global network; This forms a spatiotemporal feature matrix for each node at each time step, which integrates its own history, neighbor states, and external factors.

5. The water plant water volume prediction method based on time-varying characteristics according to claim 1, characterized in that, The prediction model includes a spatial dimension modeling module and a temporal dimension modeling module; The spatial dimension modeling module is a graph convolutional network or a graph attention network, used to capture the spatial correlation between nodes; The time dimension modeling module is a gated recurrent unit, a temporal convolutional network, or a Transformer architecture, used to capture the changing patterns of node features over time.

6. The water plant water volume prediction method based on time-varying characteristics according to claim 5, characterized in that, The specific process of S6 includes: Using the available predicted water volume as the core input, a multi-objective optimization function is established with constraints of the water plant's physical water cycle network topology and hydraulic model. The optimization objectives include: minimizing the total energy consumption of water pumps, ensuring that the pipeline pressure and the water level in the clear water tank are within a safe range, reducing the frequency of equipment start-up and shutdown, and optimizing the water retention time. The optimization function is solved using mathematical programming or intelligent optimization algorithms to generate the optimal scheduling plan for the future time period; The optimal scheduling plan is output to the scheduler or directly sent to the production control system for execution.

7. A water plant water volume intelligent scheduling system based on time-varying characteristics, used to implement the water plant water volume prediction method based on time-varying characteristics as described in any one of claims 1-6, characterized in that, include: The digital topology model building module generates a digital topology model based on the key physical components and connections of the water plant's physical water circulation network. The digital topology model includes topological nodes and directed edges. Specifically, the digital topology model is a directed graph G = (V, E), where V represents the set of all nodes and E represents the set of all edges. The data acquisition and processing module defines the available water volume as the water volume index of the topological nodes in the digital topology model, and obtains a multidimensional time series dataset by collecting and processing the time series data of the topological nodes and external feature data. The node feature construction module constructs a node spatiotemporal feature matrix that integrates spatiotemporal dimensions based on a digital topology model and a multidimensional time series dataset. The model building and prediction output module is used to establish a prediction model that simultaneously captures time dependence and spatial dependence by utilizing the node features that integrate spatiotemporal dimensions. At the same time, the prediction model is trained using historical data, and the data collected in real time and processed by the data acquisition and processing module and the node feature building module is input into the trained prediction model for rolling prediction to obtain the available predicted water volume of each node in the future period. Instruction generation module: Taking the available predicted water volume as input, it uses an optimization algorithm to generate the optimal set of equipment scheduling instructions.

8. A water plant intelligent water volume scheduling system based on time-varying characteristics according to claim 7, characterized in that, The specific process by which the data acquisition and processing module acquires a multidimensional time series dataset includes: Define the available water volume for different types of nodes; Time-series data related to the available water volume of each topology node are acquired through sensor arrays and SCADA systems, and external characteristic data closely related to water consumption are collected. All collected data are processed to obtain a multidimensional time series dataset.

9. A water plant intelligent water volume scheduling system based on time-varying characteristics according to claim 8, characterized in that, The node feature construction module constructs a node spatiotemporal feature matrix that integrates spatiotemporal dimensions, specifically including: Extract the historical time series features of each topological node from the multidimensional time series dataset, including sliding window statistics, periodicity and trend features; Based on the adjacency relationship of the digital topology model, historical data of upstream and downstream adjacent nodes of each node are aggregated to form spatial association features; Associate and integrate weather forecasts and date type external features with corresponding nodes or the global network; This forms a spatiotemporal feature matrix for each node at each time step, which integrates its own history, neighbor states, and external factors.

10. A water plant intelligent water volume scheduling system based on time-varying characteristics according to claim 7, characterized in that, The prediction model includes a spatial dimension modeling module and a temporal dimension modeling module; The spatial dimension modeling module is a graph convolutional network or a graph attention network, used to capture the spatial correlation between nodes; The time dimension modeling module is a gated recurrent unit, a temporal convolutional network, or a Transformer architecture, used to capture the changing patterns of node features over time.

Images