Intelligent early warning and adaptive regulation platform driven by digital twin of water system

Abstract

The application relates to the technical field of intelligent water conservancy engineering, and discloses an intelligent early warning and self-adaptive regulation and control platform driven by a digital twin of a water system, which comprises the following steps: collecting multi-source data related to the operation of the water system through Internet of Things equipment to generate a data set in a unified format; integrating the multi-source data through a cross-domain data fusion engine to generate a fusion feature set; analyzing the causal relationship between variables in the multi-source data through a time series causal reasoning engine to generate a dynamic causal diagram; optimizing the system operation mode through an inter-node information interaction system to generate distributed cognitive data; generating a state atlas of the water system based on a topological network and the dynamic causal diagram through a synchronous state atlas visualization engine; and generating regulation and control instructions according to a multi-target collaborative regulation and control strategy to update the data set. In the application, the cross-domain data fusion engine and the time series causal reasoning engine can accurately identify abnormal events in the water system.

Description

Technical Field

[0001] This invention relates to the field of intelligent water conservancy engineering technology, and in particular to an intelligent early warning and adaptive control platform driven by digital twins for water systems. Background Technology

[0002] As a key component of urban infrastructure, the water system undertakes core tasks such as water resource collection, transportation, treatment, and water supply and drainage scheduling. Its stable operation is directly related to the safety of residents' drinking water, the guarantee of industrial production, and the sustainability of the ecological environment.

[0003] Currently, the identification of abnormal events in water systems mainly adopts traditional data analysis methods: these methods take local data collected by a single type of sensor as the core analysis object, such as water quality pH data at a certain monitoring point or flow data of a certain pipe section, and make anomaly judgments by setting fixed thresholds or using simple time-series data statistical methods.

[0004] However, in the process of implementing the technical solution of this application, the inventors of this application discovered that the above-mentioned technology has at least the following technical problems: the existing technology cannot combine multi-dimensional data correlation and causal logic for comprehensive judgment, which ultimately results in low accuracy of abnormal event identification and easy misjudgment or omission. The root cause of the problem is that the existing solution lacks a unified technical framework that can integrate multi-source heterogeneous data, reveal deep causal relationships, and realize distributed collaboration and closed-loop optimization. Summary of the Invention

[0005] To overcome the above shortcomings, this invention provides a digital twin-driven intelligent early warning and adaptive control platform for water systems. It aims to improve the existing technology, which cannot combine multi-dimensional data correlation and causal logic for comprehensive judgment, resulting in low accuracy in identifying abnormal events and easy misjudgment or omission.

[0006] This invention provides the following technical solution: a digital twin-driven intelligent early warning and adaptive control platform for water systems, comprising a processor and a memory, wherein the memory stores instructions, characterized in that the instructions are executed by the processor to achieve the following steps:

[0007] The system collects multi-source data related to the operation of the water system through IoT devices, including sensor data, SCADA system data and external data, and preprocesses the data to generate a dataset in a unified format.

[0008] By integrating multi-source data through a cross-domain data fusion engine, the spatiotemporal correlation patterns between the water system's operational status and the external environment are mined to generate a fusion feature set. The parameters of the model used to generate the fusion feature set are based on joint feedback provided by temporal causal reasoning and distributed cognitive enhancement, and are updated online through an objective function that includes causal constraint terms and cognitive consistency terms.

[0009] The time-series causal inference engine analyzes the causal relationships between variables in multi-source data, generates a dynamic causal graph, and dynamically adjusts the probability distribution of the causal graph through an online incremental update algorithm.

[0010] By using a distributed cognitive enhancement module, the nodes of the water system are given node-level state perception and action decision-making capabilities. The system operation mode is optimized through information interaction between nodes, and distributed cognitive data is generated.

[0011] Based on complex network theory, a self-organizing topology mapping method is used to dynamically construct the topology network of the water system by combining fused feature sets, dynamic causal graphs, and distributed cognitive data. The topology edge weights are determined by a weighted combination of feature similarity, causal strength, and cognitive association, and are updated online incrementally within a predetermined period.

[0012] By optimizing the multi-agent game theory through dynamic game theory, and combining topological networks, dynamic causal graphs, and distributed cognitive data, a multi-objective collaborative regulation strategy is generated. The multi-objective collaborative regulation introduces dynamic weights formed by the topology, causal strength, and distributed cognitive data, and uses the alternating direction multiplier method to decompose and solve the global problem in parallel to meet real-time constraints.

[0013] A state graph of the water system is generated based on topology network and dynamic cause-effect graph using a synchronous state graph visualization engine.

[0014] Control commands are generated based on a multi-objective collaborative control strategy, executed through IoT devices, and monitored and fed back in real time to update the dataset. The system state after execution serves as feedback, which is fed back to jointly update the model parameters used to generate the fused feature set, the causal strength metric of the dynamic causal graph, and the edge weights of the topological network, thereby forming an adaptive optimization closed loop based on joint feedback that integrates perception, analysis, decision-making, and execution.

[0015] Preferably, the objective function comprising the causal constraint term and the cognitive consistency term is specifically: ,in, For embedding loss, The causal constraint term is constructed by calculating the difference between the similarity in the feature space and the causal strength in the dynamic causal graph; The cognitive consistency term is constructed by calculating the distance between the fused features and the node state vectors in the distributed cognitive data. and These are dynamically adjusted hyperparameters.

[0016] Preferably, the probability distribution of the causal graph is dynamically adjusted through an online incremental update algorithm, specifically employing an incremental Bayesian learning method. Each time a new fused feature set is received, the conditional probability distribution is updated according to the following formula: in, For learning rate, For the new feature set The probability increment.

[0017] Preferably, the step of generating the state map includes:

[0018] The edge weights dynamically updated by the joint feedback mechanism in the topological network, and the causal strength updated in the dynamic causal graph, are mapped to the visual thickness attribute of the edges in the graph; and the node state data integrated from the distributed cognitive enhancement module are labeled as the visual attributes of the corresponding nodes, thereby rendering a state graph reflecting the latest operational risks of the system in real time.

[0019] Preferably, the collection of multi-source data related to the operation of the water system via IoT devices includes:

[0020] Deploy a distributed sensor network to collect water pressure, flow rate, and water quality data; acquire pump station power and valve opening data through a SCADA system; integrate external weather forecast data; and clean and standardize the data on edge computing devices to generate a dataset in a unified format.

[0021] Preferably, the step of generating control commands based on a multi-objective collaborative control strategy and monitoring feedback in real time includes:

[0022] The strategy is converted into equipment instructions for adjusting pump station power and valve opening, and sent to IoT devices for execution via the MQTT protocol; and system status data after execution is collected in real time as feedback for the joint update process.

[0023] Preferably, the platform achieves the adaptive optimization closed loop through the synergistic effect of a cross-domain data fusion engine, a temporal causal reasoning engine, a distributed cognitive enhancement module, a self-organizing topology mapping, dynamic game optimization, and a synchronous state graph visualization engine, wherein:

[0024] The dynamic causal graph generated by the temporal causal reasoning engine and the distributed cognitive data generated by the distributed cognitive enhancement module constitute the joint feedback, which is used to optimize the model parameters of the cross-domain data fusion engine.

[0025] The self-organizing topology mapping module dynamically constructs a topology network using an optimized fusion feature set, an updated dynamic causal graph, and distributed cognitive data.

[0026] The dynamic game optimization module uses the topological network, dynamic causal graph and distributed cognitive data to generate multi-objective collaborative regulation strategies.

[0027] The system status feedback data after the control command is executed is synchronously fed back to trigger a new round of joint updates.

[0028] The present invention has the following beneficial effects:

[0029] 1. This invention introduces a cross-domain data fusion engine driven by a joint feedback mechanism, which not only integrates multi-source data but, more importantly, dynamically guides the feature extraction process using causal reasoning and distributed cognitive results. This deep fusion enables the generated fused feature set to fundamentally capture the key causal chains leading to system anomalies. Compared to traditional anomaly identification methods based on thresholds or statistical features, this significantly improves the accuracy of early warnings and substantially reduces false alarms and false negatives.

[0030] 2. This invention optimizes the operating mode through asynchronous communication between nodes. The self-organizing topology mapping is based on complex network theory, combining fused feature sets, dynamic causal graphs, and distributed cognitive data to dynamically construct the topology network. This combination of dynamic modeling and node optimization enables the system to adapt to complex dynamic environments.

[0031] 3. In this invention, a multi-objective collaborative control strategy is generated through a dynamic weight adjustment mechanism to balance the operational stability of subsystems and the overall efficiency of the system; the synchronous state graph visualization engine uses WebGL real-time rendering technology to generate a state graph based on the topology network and dynamic causal graph, providing an intuitive display of key nodes and risk paths to assist users in making control decisions. Attached Figure Description

[0032] Figure 1 This is a flowchart of the method for a digital twin-driven intelligent early warning and adaptive control platform for water systems proposed in this invention. Detailed Implementation

[0033] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Reference Figure 1In the first embodiment of the present invention, the present invention provides a digital twin-driven intelligent early warning and adaptive control platform for water systems, comprising the following steps:

[0035] The system collects multi-source data related to the operation of the water system through IoT devices, including sensor data, SCADA system data and external data, and preprocesses the data to generate a dataset in a unified format.

[0036] By integrating multi-source data through a cross-domain data fusion engine, we can uncover the spatiotemporal correlation patterns between the water system's operational status and the external environment, and generate a fusion feature set.

[0037] The engine analyzes the causal relationships between variables in multi-source data using a time-series causal inference engine, generating a dynamic causal graph.

[0038] By using a distributed cognitive enhancement module, the nodes of the water system are given node-level state perception and action decision-making capabilities. The system operation mode is optimized through information interaction between nodes, and distributed cognitive data is generated.

[0039] Based on complex network theory, the topological network of the water system is dynamically constructed by using self-organizing topology mapping, combined with fusion feature sets, dynamic causal graphs and distributed cognitive data.

[0040] By optimizing dynamic game theory based on multi-agent game theory, and combining topological networks, dynamic causal graphs and distributed cognitive data, multi-objective collaborative regulation strategies are generated.

[0041] A state graph of the water system is generated based on topology network and dynamic cause-effect graph using a synchronous state graph visualization engine.

[0042] Control commands are generated based on a multi-objective collaborative control strategy, executed through IoT devices, and monitored and fed back in real time to update the dataset.

[0043] Specifically, the platform consists of the following modules: multi-source data acquisition and preprocessing module, cross-domain data fusion engine, time-series causal reasoning engine, distributed cognitive enhancement module, self-organizing topology mapping module, dynamic game optimization module, synchronous state graph visualization engine, and control command generation and execution module.

[0044] The multi-source data acquisition and preprocessing module collects sensor data, SCADA system data, and external data related to the operation of the water system through IoT devices. It uses edge computing devices to clean and standardize the data, generating a dataset in a unified format to provide input for subsequent analysis.

[0045] The cross-domain data fusion engine integrates multi-source data and uses graph embedding technology, time series analysis, and spatial analysis to uncover the spatiotemporal correlation patterns between the water system's operating status and the external environment, generating a fusion feature set to provide data support for causal reasoning and topology modeling.

[0046] The temporal causal reasoning engine analyzes the causal relationships between variables in multi-source data through dynamic Bayesian networks, generating dynamic causal graphs to provide explanatory basis for early warning and regulation.

[0047] The distributed cognitive enhancement module embeds a lightweight intelligent agent into the nodes of the water system. Through asynchronous communication between nodes and reinforcement learning optimization, it endows the nodes with state perception and action decision-making capabilities, generates distributed cognitive data, and optimizes the system operation mode.

[0048] The self-organizing topology mapping module, based on complex network theory, combines fusion feature sets, dynamic causal graphs, and distributed cognitive data to dynamically construct the topology network of the water system, reflecting the spatiotemporal evolution characteristics of its operational status.

[0049] The dynamic game optimization module is based on multi-agent game theory. It utilizes topological networks, dynamic causal graphs, and distributed cognitive data to generate multi-objective collaborative control strategies through Nash equilibrium, thereby balancing the operational stability of subsystems and the overall efficiency of the system.

[0050] The synchronous state graph visualization engine generates a state graph of the water system based on topological networks and dynamic causal graphs, using WebGL real-time rendering technology to display key nodes and risk paths, supporting decision-making.

[0051] The control command generation and execution module generates device commands based on a multi-objective collaborative control strategy, sends them to IoT devices for execution via the MQTT protocol, and collects feedback data in real time to update the dataset.

[0052] The platform forms a closed-loop system through the collaborative operation of its modules. A unified-format dataset from the multi-source data acquisition and preprocessing module is input into the cross-domain data fusion engine to generate a fused feature set. This fused feature set is then transmitted to the temporal causal inference engine and the self-organizing topology mapping module, generating a dynamic causal graph and a topology network, respectively. Distributed cognitive data generated by the distributed cognitive enhancement module is fed back to the cross-domain data fusion engine and the self-organizing topology mapping module to optimize feature extraction and topology modeling. The dynamic game optimization module utilizes the topology network, dynamic causal graph, and distributed cognitive data to generate a multi-objective collaborative control strategy, which is then transmitted to the control instruction generation and execution module. The synchronous state graph visualization engine generates a state graph based on the topology network and dynamic causal graph, providing decision support. Feedback data from the control instruction generation and execution module updates the dataset, forming a closed-loop operation.

[0053] More specifically, the platform achieves continuous improvement in overall system performance and closed-loop operation through a refined cross-module joint feedback and adaptive optimization mechanism. The core of this mechanism lies in the feature extraction process of the cross-domain data fusion engine. The parameter optimization of the graph embedding model does not rely solely on traditional data fitting, but is performed online by minimizing a composite objective function, defined as: ,in, To ensure the accuracy of feature reconstruction, a traditional embedding loss based on feature spatial distance is used. As a causal constraint term, it can be constructed by calculating the KL divergence between the similarity of the feature space and the causal strength between corresponding variables in the dynamic causal graph. It aims to drive the fused features to encode and strengthen the identified causal relationships in the vector space, so that subsequent analysis will focus more on the abnormal precursors with causal associations. The cognitive consistency term is constructed by calculating the cosine similarity between the fused features and the node state vectors in the distributed cognitive data. It is used to ensure that the global data fusion results and the intelligent decisions made by each node based on local perception are mutually verified and coordinated, thus avoiding decision conflicts between the global model and local behavior. and The hyperparameters used to balance the weights of each component are described above. The innovation of this optimization mechanism lies in its ability to tightly couple the three previously independent technical layers of causal reasoning, distributed decision-making, and data fusion through the aforementioned joint feedback, forming a mutually reinforcing closed loop. The dynamic causal graph of the temporal causal reasoning engine provides directional guidance for feature fusion, while the real-time decision data from the distributed cognitive enhancement module provides local state verification for feature fusion. This allows the system to learn from the feedback of each regulatory execution, dynamically optimizing its internal model, thereby achieving increasingly accurate perception, early warning, and adaptive regulation of the entire water system.

[0054] Simultaneously, the system status feedback data collected after the execution of control commands is used to synchronously update multiple core modules online, forming a tightly coupled optimization loop: the dynamic causal graph uses feedback data to update the probability distribution and causal strength measure of its causal edges through an incremental Bayesian learning algorithm, while the topology network recalculates the feature similarity, causal strength, and cognitive correlation between nodes based on the latest fused feature set, updated causal strength, and cognitive data, thereby dynamically adjusting the topology edge weights. This cross-module joint feedback and update mechanism enables the system to learn from each decision execution, continuously refining its internal model, thereby achieving increasingly accurate perception, early warning, and control of the water system.

[0055] The steps involved in integrating multi-source data through a cross-domain data fusion engine to uncover spatiotemporal correlation patterns between the operational status of the water system and the external environment, and generating a fused feature set, include:

[0056] Feature extraction is performed on sensor data, SCADA system data, and external data, and graph embedding technology is used to map multi-source data to a unified feature space;

[0057] Using time series analysis and spatial analysis methods, we extract the spatiotemporal correlation characteristics between the operating status of the water system and the external environment;

[0058] The similarity between features is calculated using kernel functions to generate a fused feature set;

[0059] By combining the dynamic causal graph of the temporal causal reasoning engine and the distributed cognitive data of the distributed cognitive enhancement module, the graph embedding model parameters are dynamically adjusted through a feedback optimization mechanism to optimize the fusion feature set.

[0060] Specifically, the cross-domain data fusion engine is used to integrate multi-source data, mine the spatiotemporal correlation patterns between the water system's operating status and the external environment, generate a fusion feature set, and provide high-quality input data for subsequent causal inference, topology modeling, and regulation optimization. The cross-domain data fusion engine is deployed on a distributed computing framework, receives a unified format dataset output by the multi-source data acquisition and preprocessing module through a data interface, and outputs the fusion feature set to the temporal causal inference engine and the self-organizing topology mapping module.

[0061] Sensor data, SCADA system data, and external data are provided by the multi-source data acquisition and preprocessing module. The sensor data includes water pressure, flow rate, and pH value, and is in time series format. ,in, Represents a timestamp. Represents multidimensional measurement values. Represents a time set, Representing an n-dimensional real space, SCADA system data includes pump station power and valve opening, etc., in the following format: ,in, Indicates device status. This represents a set of states. External data includes weather forecasts, geographic information, and user water usage behavior data, in the following format: ,in, Represents environment variables, This represents a set of environment variables.

[0062] The feature extraction process uses graph embedding technology to map multi-source data to a unified feature space. The GraphSAGE algorithm is employed to construct a data association graph for the water system, where nodes represent data points and edges represent the relationships between data points. GraphSAGE generates low-dimensional feature vectors by aggregating node neighbor information. ,in This represents the dimension of the feature space. The mapping function is defined as:

[0063] ;

[0064] in, Represents the fused feature set, This represents the parameters of the graph embedding model.

[0065] Time series analysis uses the ARIMA model to analyze the time dependencies of multi-source data and extract the temporal characteristics of the water system's operating status, such as the periodic changes in water pressure over time. Spatial analysis uses Moran's I statistical method to calculate the spatial correlation between data points and extract the spatial impact of the external environment on the operating status, such as the effect of rainfall on pipeline pressure. Spatiotemporal correlation features are generated by combining the results of time series and spatial analysis, reflecting the dynamic interaction between the water system's operating status and the external environment.

[0066] Feature similarity is calculated using the radial basis function kernel method, and the formula is as follows:

[0067] ;

[0068] in Representing the eigenvector and Similarity, This represents the kernel function bandwidth parameter. Through similarity calculation, a fused feature set z is generated, containing feature vectors reflecting spatiotemporal correlation patterns.

[0069] The feedback optimization mechanism combines the dynamic causal graph output by the temporal causal inference engine and the distributed cognitive data output by the distributed cognitive enhancement module to dynamically adjust the graph embedding model parameters. Specifically, it optimizes feature extraction by minimizing the loss function.

[0070] ;

[0071] in, Represents the spatial distance of features. This represents the correlation between the original data, based on the causal strength of the dynamic causal graph and the node state adjustment of the distributed cognitive data. The optimization process uses the gradient descent algorithm to iteratively update the graph embedding model parameters, ensuring that the fused feature set accurately reflects the operating status of the water system and its correlation with the external environment. The cross-domain data fusion engine is deployed on the distributed computing framework Apache Spark, achieving large-scale data processing through parallel processing. Input data is received from the multi-source data acquisition and preprocessing module through a data interface, and the output fused feature set is transmitted to the time-series causal inference engine and the self-organizing topology mapping module through the data interface.

[0072] The steps involved in analyzing causal relationships between variables in multi-source data and generating a dynamic causal graph using a time-series causal inference engine include:

[0073] A time-series causal graph is constructed using a dynamic Bayesian network, which includes variable nodes, causal edges, and conditional probability distributions.

[0074] Initialize the causal graph structure using a structure learning algorithm;

[0075] Based on intervention analysis, assess the causal impact of regulatory actions on the state of the water system;

[0076] By utilizing the fusion feature set provided by the cross-domain data fusion engine, the probability distribution of the causal graph is dynamically adjusted through an online incremental update algorithm.

[0077] Specifically, this time-series causal reasoning engine analyzes the causal relationships between variables in multi-source data to generate a dynamic causal graph, providing explanatory evidence for intelligent early warning and control of water systems. Deployed on a distributed computing cluster, the engine receives the fused feature set output by the cross-domain data fusion engine through a data interface, and outputs the dynamic causal graph to the self-organizing topology mapping module and the dynamic game optimization module, supporting subsequent topology modeling and control strategy generation.

[0078] The fusion feature set is provided by the cross-domain data fusion engine, and the format is as follows: ,in, Indicates the first Each feature vector contains the spatiotemporal correlation features between the water system's operating status and the external environment. Represents the dimension of the feature space. This indicates the number of data points. The time-series causal inference engine analyzes the causal relationships between variables based on a fused feature set, such as the impact of water pressure changes on pipe leaks or the causal effect of rainfall on overflow.

[0079] The steps involved in analyzing causal relationships between variables in multi-source data and generating a dynamic causal graph using a time-series causal inference engine include:

[0080] A time-series causal graph is constructed using a dynamic Bayesian network, which includes variable nodes, causal edges, and conditional probability distributions.

[0081] Initialize the causal graph structure using a structure learning algorithm;

[0082] Based on intervention analysis, assess the causal impact of regulatory actions on the state of the water system;

[0083] By utilizing the fusion feature set provided by the cross-domain data fusion engine, the probability distribution of the causal graph is dynamically adjusted through an online incremental update algorithm.

[0084] Dynamic Bayesian networks are used to construct temporal causal graphs, defined as follows: ,in Represents variable nodes, Indicates a causal edge. Represents the conditional probability distribution. Dynamic Bayesian networks model the temporal dependencies between variables through time slices, as shown in the formula:

[0085] ;

[0086] in Indicates time The variable state, Indicates time The fusion feature set, This represents the Bayesian update function, which updates the conditional probability distribution by fusing feature sets.

[0087] Structural learning employs the PC algorithm to initialize the causal graph structure. The PC algorithm identifies direct causal relationships between variables through a conditional independence test, generating the initial causal graph. Specific steps include: first, calculating the correlation matrix between variables based on the fused feature set Z; then, constructing a skeleton graph by progressively removing insignificant edges; and finally, determining the direction of causal edges through orientation rules to generate the initial causal graph. The computational complexity is O(n). .

[0088] Intervention analysis is used to assess the causal impact of regulatory actions on the state of the water system. The causal effect is calculated based on do-calculus, using the following formula: ,in, Represents the target variable. Indicates regulatory action, Indicates the expected value. Indicates intervention operation. This represents a dynamic causal graph. Intervention analysis quantifies the impact of regulatory actions on target variables by simulating regulatory actions, such as the causal effect of adjusting pump station power on water pressure.

[0089] Online incremental update algorithms are used to dynamically adjust the probability distribution of causal graphs, based on fused feature sets. Specifically, an incremental Bayesian learning method is used, which updates the conditional probability distribution each time a new feature set is received:

[0090] ;

[0091] in, Indicates the learning rate. This represents the probability increment based on the new feature set. The update process is implemented in parallel through a distributed message passing interface (MPI) to ensure real-time performance.

[0092] The temporal causal inference engine is deployed on a distributed computing cluster and uses Apache Flink for streaming data processing. The input fused feature set is received from the cross-domain data fusion engine via a data interface, and the output dynamic causal graph is transmitted to the self-organizing topology mapping module and the dynamic game optimization module via the same data interface. The dynamic causal graph is also fed back to the cross-domain data fusion engine to optimize the feature extraction process.

[0093] The steps involved in generating distributed cognitive data include: Empowering water system nodes with node-level state awareness and action decision-making capabilities through a distributed cognitive enhancement module, optimizing system operation modes through inter-node information exchange, and generating distributed cognitive data.

[0094] A lightweight intelligent agent is embedded in each water system node to record local states and actions;

[0095] A distributed cognitive network is constructed by exchanging state information between nodes through an asynchronous communication protocol.

[0096] Based on the idea of ​​reinforcement learning, node actions are optimized through a reward function, which combines the local operational stability of nodes with the overall efficiency of the system.

[0097] Distributed optimization algorithms are used to process data exchanged between nodes to generate distributed cognitive data.

[0098] Specifically, this distributed cognitive enhancement module embeds lightweight intelligent agents into water system nodes, enabling node-level state perception and action decision-making capabilities. It generates distributed cognitive data, providing feedback input to the cross-domain data fusion engine and the self-organizing topology mapping module. Deployed on an edge device network, the distributed cognitive enhancement module connects to water system nodes via a communication interface, receives fused feature sets as initial input, and outputs distributed cognitive data to other modules, supporting the optimization of system operation modes.

[0099] The water system nodes include pumping stations, pipelines, valves, and wastewater treatment equipment. Each node embeds a lightweight intelligent agent, which runs the agent program using an embedded processor. The agent records local state. in, The node state vector representing time t includes parameters such as water pressure, flow rate, and equipment power. This represents node actions, such as adjusting valve opening. State information exchange between nodes uses an asynchronous communication protocol, achieving low-latency data transmission through efficient binary serialization. A distributed cognitive network is defined as... in, Let q represent the set of nodes, and q represent the total number of nodes. The communication edge is determined based on physical network distance and data correlation. Communication weight. Calculated by the distance between nodes:

[0100] ;

[0101] in, Indicates the physical distance between nodes. The decay parameter is represented by an asynchronous communication protocol that transmits state data via a TCP connection. Node action optimization is based on reinforcement learning, employing an asynchronous Q-learning algorithm and evaluating action effectiveness through a reward function. The reward function, which considers both the local stability of the node and the overall system efficiency, is defined as:

[0102] ;

[0103] in, Representing local rewards, calculating the variance of state fluctuations. This represents the overall reward, calculated as the ratio of total water supply to total energy consumption. The balance parameters indicate that the motion optimization has passed. The formula for updating the Q value is:

[0104] ;

[0105] in =0.1 indicates a learning rate =0.9 represents the discount factor.

[0106] The distributed optimization algorithm uses asynchronous Q-learning to process interactive data and generate distributed cognitive data. It includes optimized states and actions.

[0107] The steps for dynamically constructing the topology network of a water system using self-organizing topology mapping based on complex network theory, combined with fused feature sets, dynamic causal graphs, and distributed cognitive data, include:

[0108] The water system is abstracted as a dynamic network, which includes nodes, edges and edge weights. Nodes represent equipment or subsystems, and edges represent water flow or energy flow.

[0109] The network structure is initialized based on the correlation between the physical pipeline layout and the water system operation status;

[0110] By combining fused feature sets, dynamic causal graphs, and distributed cognitive data, feature similarity, causal strength, and cognitive association are calculated, and edge weights are dynamically adjusted.

[0111] The topology network structure is updated in real time using an online incremental learning algorithm.

[0112] Specifically, the self-organizing topology mapping module, based on complex network theory, dynamically constructs a topological network of the water system by combining the fusion feature set provided by the cross-domain data fusion engine, the dynamic causal graph provided by the temporal causal reasoning engine, and the distributed cognitive data provided by the distributed cognitive enhancement module. This reflects the spatiotemporal evolution characteristics of the operating state and provides a network foundation for the dynamic game optimization module and the synchronous state graph visualization engine. Deployed on a distributed computing server, this module receives input data through a data interface and outputs the topological network to downstream modules. The water system is abstracted as a dynamic network, where nodes represent equipment or subsystems, such as pumping stations, pipeline valves, or sewage treatment plants, and edges represent water flow or energy flow, such as pipeline water flow or power transmission. Network initialization is based on geographic information system data to determine node locations and initial edge connections, and initial edge weights are calculated using water system operating state data. Weights reflect the interaction strength between nodes, such as water pressure correlation or energy flow efficiency, and are determined based on the statistical correlation of physical pipeline layout and operating state. Edge weight adjustment combines the fusion feature set, dynamic causal graph, and distributed cognitive data. The fusion feature set provides spatiotemporal correlation information between the water system's operational status and the external environment; the dynamic causal graph provides the strength of causal relationships between variables; and the distributed cognitive data provides node status and action optimization results. Feature similarity is calculated by comparing node feature vectors, causal strength is extracted from the dynamic causal graph, and cognitive association is determined based on node interactions in the distributed cognitive data. Weight adjustment is achieved through a weighted combination of these three factors, updated every minute to ensure the network reflects the latest system status.

[0113] The online incremental learning employs an online self-organizing map method to update the network structure in real time. Each time new input data is received, the module adjusts node positions and edge connections through competitive learning to maintain consistency between the topology and the dynamic changes of the water system. The update process is executed in parallel on the distributed computing framework Hadoop, handling large-scale network data. The self-organizing topology mapping module receives fused feature sets, dynamic causal graphs, and distributed cognitive data through a data interface, outputting the topology network to the dynamic game optimization module and the synchronous state graph visualization engine. The topology network is also fed back to the dynamic game optimization module, supporting the generation of multi-objective collaborative control strategies.

[0114] The steps for generating multi-objective collaborative regulation strategies through dynamic game optimization based on multi-agent game theory, combined with topological networks, dynamic causal graphs, and distributed cognitive data, include:

[0115] The water management subsystem is modeled as an intelligent agent, with each agent having a utility function, which integrates the operational stability of the subsystem and the overall efficiency of the system.

[0116] Based on Nash equilibrium optimization of the global objective, a multi-objective collaborative regulation strategy is generated.

[0117] By combining dynamic causal graphs and distributed cognitive data, the utility function weights of each agent are optimized through a dynamic weight adjustment mechanism.

[0118] The problem is decomposed and optimized using the alternating direction multiplier method to ensure real-time performance.

[0119] Specifically, this dynamic game optimization module, based on multi-agent game theory, utilizes the topological network provided by the self-organizing topology mapping module, the dynamic causal graph provided by the temporal causal inference engine, and the distributed cognitive data provided by the distributed cognitive enhancement module to generate multi-objective collaborative regulation strategies, providing optimized outputs for the regulation command generation and execution module. This module is deployed on a high-performance computing server, receiving input data through a data interface and outputting regulation strategies to the regulation command generation and execution module.

[0120] The water management subsystem, comprising water supply, drainage, and wastewater treatment subsystems, is modeled as intelligent agents. Each agent records the subsystem's state, such as water pressure and flow rate, as well as actions, such as adjusting pump power or valve opening. The utility function integrates subsystem operational stability and overall system efficiency, using weighted summation to ensure a balance between subsystem and global objectives. Global objective optimization is based on Nash equilibrium, generating a multi-objective collaborative control strategy by coordinating the actions of each agent. The optimization process ensures that no single agent can improve its own utility by unilaterally changing its actions, while maximizing global performance. The optimization algorithm runs every minute, updating the strategy based on the latest topology and causal graph. A dynamic weight adjustment mechanism utilizes the causal relationship strength of the dynamic causal graph and node interaction information from distributed cognitive data to update the weights of each agent's utility function in real time. Weight adjustments are made based on inter-subsystem dependencies and changes in operational state, ensuring the control strategy adapts to the dynamic environment.

[0121] The optimization problem decomposition employs the alternating direction multiplier method, breaking down the global optimization into local subproblems and improving efficiency through parallel computation. Consensus constraints are introduced during the decomposition process to ensure coordinated actions among agents, meeting real-time requirements. The dynamic game optimization module utilizes the TensorFlow framework for parallel computation, receiving topological networks, dynamic causal graphs, and distributed cognitive data through a data interface, and outputting control strategies to the control instruction generation and execution module. These strategies are also fed back to the synchronous state graph visualization engine, supporting the display of runtime status. The module is deployed on a high-availability server to ensure continuous operation and rapid response.

[0122] The steps for generating a state graph of a water system using a synchronous state graph visualization engine based on topology networks and dynamic causal graphs include:

[0123] Based on topological networks and dynamic causal graphs, a visual state graph is generated, which includes nodes, edges, and visual attributes.

[0124] WebGL real-time rendering technology is used to dynamically update the graph, highlighting key nodes and risky paths;

[0125] The fusion feature set of the distributed cognitive data and cross-domain data fusion engine, which integrates the distributed cognitive enhancement module, annotates the local state of nodes and the influence of the external environment;

[0126] It provides an interactive interface that allows users to view causal paths and regulatory suggestions, including the ability to zoom in on key areas and filter risk paths.

[0127] Specifically, this synchronous state graph visualization engine generates a state graph of the water system based on the topology network provided by the self-organizing topology mapping module and the dynamic causal graph provided by the temporal causal inference engine. It displays key nodes and risk paths, providing users with intuitive decision support. This module is deployed on a front-end server and receives topology network, dynamic causal graph, distributed cognitive data from the distributed cognitive enhancement module, and fused feature set from the cross-domain data fusion engine via API, outputting the visualized graph to the user interface. The visualized state graph includes nodes, edges, and visualization attributes. Nodes represent water system equipment or subsystems, such as pumping stations or pipelines; edges represent water flow or causal relationships; and visualization attributes include node color (red for abnormal states, green for normal states) and edge thickness (reflecting causal strength). The graph generation process integrates the structural information of the topology network and the causal relationships of the dynamic causal graph, mapping node states and edge weights to visualization elements. Dynamic updates utilize WebGL real-time rendering technology. The rendering process loads the topology network as a geometric model, applies the dynamic causal graph to update edge attributes, and refreshes the graph every second, highlighting key nodes and risk paths. Distributed cognitive data provides node state information, and the fused feature set provides information on external environmental influences. These data are used to label nodes, displaying state values ​​and environmental effects. Labeling is achieved through overlays, with color gradients indicating the degree of impact.

[0128] The interactive interface is built on HTML5 and JavaScript, allowing users to click on nodes to view causal paths and control suggestions. The interface provides zoom and filtering functions to ensure users can quickly locate problem areas. The synchronous state graph visualization engine is accessed via a browser and supports simultaneous operation by multiple users. Input data is received from upstream modules via API, and graph data is stored in a memory cache to ensure real-time performance and interactivity.

[0129] The steps involved in collecting multi-source data related to the operation of the water system through IoT devices, including sensor data, SCADA system data, and external data, and preprocessing it to generate a dataset in a unified format, include:

[0130] Deploy a distributed sensor network to collect water pressure, flow rate, and water quality data;

[0131] The operating status of the equipment, including pump station power and valve opening, is obtained through the SCADA system.

[0132] Integrate external data sources, including weather forecasts, geographic information, and user water usage behavior data;

[0133] Data cleaning and standardization are performed on edge computing devices, including median filtering to remove noise and missing value imputation, to generate datasets in a uniform format;

[0134] Specifically, the multi-source data acquisition and preprocessing module collects multi-source data related to the operation of the water system through IoT devices, cleans and standardizes it, and generates a unified-format dataset to provide input for the cross-domain data fusion engine. This module is deployed on an edge computing network, acquiring data through sensor interfaces, SCADA system interfaces, and external data APIs, and outputting the unified-format dataset to the cross-domain data fusion engine. The distributed sensor network includes pressure sensors, flow meters, and pH sensors, deployed at nodes such as water supply pipelines, drainage systems, and wastewater treatment plants within the water system. Sensors are used to collect data, generating time-series data including water pressure, flow rate, and water quality parameters. The data is transmitted to edge devices via the LoRaWAN protocol, ensuring low power consumption and long-distance communication. The SCADA system obtains device operating status via the OPCUA protocol, covering pump station power and valve opening, records real-time device operating parameters, stores them in a local buffer, and periodically transmits them to edge computing devices via the MQTT protocol.

[0135] External data sources are integrated with weather forecasts, geographic information, and user water usage behavior data via RESTful APIs. This external data is updated hourly, stored in a cloud database, and retrieved via edge devices. Data cleaning is performed on edge computing devices, using median filtering to remove noise and smooth out abnormal fluctuations. Missing value imputation uses linear interpolation, estimating missing values ​​based on preceding and following data points to ensure data continuity. A standardization process scales the data to the [0,1] range, unifying data formats across different dimensions and generating a unified dataset containing timestamps, measurements, and metadata. The multi-source data acquisition and preprocessing module uses edge devices to support high-concurrency data processing, and the output dataset is transmitted to the cross-domain data fusion engine via the MQTT protocol. The module supports dynamic expansion, allowing the addition of sensors and edge devices based on the scale of the water system to ensure coverage of all critical nodes.

[0136] The steps involved in generating control commands based on a multi-objective collaborative control strategy, executing them via IoT devices, monitoring feedback in real time, and updating the dataset include:

[0137] The multi-objective coordinated control strategy is translated into equipment instructions, including adjusting pump station power and valve opening.

[0138] Send device commands to IoT devices via the MQTT protocol;

[0139] The system status after execution is collected in real time and fed back to the cross-domain data fusion engine and the time-series causal reasoning engine.

[0140] The dataset is updated based on feedback to optimize subsequent cross-domain data fusion and temporal causal inference processes.

[0141] Specifically, the implementation method (second stage: control command generation and execution module)

[0142] The steps involved in generating control commands based on a multi-objective collaborative control strategy, executing them via IoT devices, monitoring feedback in real time, and updating the dataset include:

[0143] The multi-objective coordinated control strategy is translated into equipment instructions, including adjusting pump station power and valve opening.

[0144] Send device commands to IoT devices via the MQTT protocol;

[0145] The system status after execution is collected in real time and fed back to the cross-domain data fusion engine and the time-series causal reasoning engine.

[0146] The dataset is updated based on feedback to optimize subsequent cross-domain data fusion and temporal causal reasoning processes.

[0147] Specifically, the control command generation and execution module generates device commands based on the multi-objective collaborative control strategy provided by the dynamic game optimization module. These commands are then executed by IoT devices, and feedback data is collected to update the dataset, forming a closed-loop control. This module is deployed on the control center server, connects to IoT devices via an MQTT interface, receives the multi-objective collaborative control strategy, and outputs feedback data to the cross-domain data fusion engine and the time-series causal inference engine.

[0148] The multi-objective coordinated control strategy is provided by a dynamic game optimization module, encompassing optimized actions for each subsystem, such as adjusting pump station power or valve opening. The command conversion process uses a predefined equipment parameter mapping table to map actions to specific commands. For example, pump station power adjustment is converted into voltage or current control signals, and valve opening is converted into angle control values. The mapping table is stored in a server database, containing equipment type, parameter range, and control protocol. Commands are issued using the MQTT protocol, transmitted to IoT devices such as smart valves and pump station controllers via a topic subscription model. Command packets include timestamps, device IDs, and parameter values ​​to ensure real-time performance. The MQTT client runs on the IoT devices, ensuring data security through an encrypted channel. After execution, system status is collected in real-time via a distributed sensor network, including water pressure, flow rate, and equipment operating parameters. Feedback data is transmitted back to the control center server via the MQTT protocol in time-series format, including timestamps and measured values.

[0149] Feedback data is used to update the unified format dataset by appending new records and merging them into the existing dataset. The updated dataset is then transmitted to the cross-domain data fusion engine to optimize the feature extraction process; simultaneously, it is transmitted to the time-series causal inference engine to update the dynamic causal graph, ensuring that subsequent analysis reflects the latest system state. The control instruction generation and execution module uses industrial-grade servers to support high availability and low latency operation. Communication with IoT devices is achieved through an MQTT interface to ensure system real-time performance and stability.

[0150] The platform achieves closed-loop operation through the synergistic effect of a cross-domain data fusion engine, a temporal causal reasoning engine, a distributed cognitive enhancement module, a self-organizing topology mapping, dynamic game optimization, and a synchronous state graph visualization engine, including:

[0151] The cross-domain data fusion engine provides a fused feature set as input to the temporal causal inference engine and the self-organizing topology mapping;

[0152] The dynamic causal graph of the temporal causal reasoning engine and the distributed cognitive data of the distributed cognitive enhancement module are used to optimize the feature extraction of the cross-domain data fusion engine through a feedback optimization mechanism.

[0153] Self-organizing topology mapping combines fusion feature sets, dynamic causal graphs, and distributed cognitive data to generate topology networks for dynamic game optimization.

[0154] Dynamic game optimization utilizes topological networks, dynamic causal graphs, and distributed cognitive data to generate multi-objective collaborative regulation strategies.

[0155] The synchronous state graph visualization engine, based on topology networks and dynamic causal graphs, provides intuitive decision support, including real-time display of the water system's operating status and the impact of the external environment.

[0156] Specifically, the platform forms a closed-loop system from data acquisition to control execution through inter-module collaborative operation and data flow interaction. The collaborative mechanism is deployed on a central coordination server, using Kafka message queues to manage data flow, ensuring consistency and real-time communication between modules. The cross-domain data fusion engine integrates sensor data, SCADA system data, and external data to generate a fused feature set, which is transmitted to the time-series causal inference engine and the self-organizing topology mapping module as input for causal analysis and network modeling. The dynamic causal graph of the time-series causal inference engine and the distributed cognitive data of the distributed cognitive enhancement module are fed back to the cross-domain data fusion engine to improve data quality by optimizing feature extraction parameters. The self-organizing topology mapping module combines the fused feature set, dynamic causal graph, and distributed cognitive data to generate a topology network reflecting the operating status of the water system, which is transmitted to the dynamic game optimization module and the synchronous state graph visualization engine. The dynamic game optimization module uses the topology network, dynamic causal graph, and distributed cognitive data to generate a multi-objective collaborative control strategy, which is transmitted to the control command generation and execution module. The synchronous state graph visualization engine generates real-time state graphs based on topological networks and dynamic causal graphs, displaying the operating status of water systems and the impact of the external environment, supporting user decision-making.

[0157] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1.A water system digital twin driven intelligent early warning and adaptive regulation platform, comprising a processor and a memory, wherein the memory stores instructions, and the platform is characterized in that, The instructions are executed by the processor to implement the following steps: Collecting water system operation related multi-source data through Internet of Things devices, including sensor data, SCADA system data and external data, and preprocessing to generate a unified format data set; The multi-source data is integrated through a cross-domain data fusion engine, a spatio-temporal correlation pattern between a water system operation state and an external environment is mined, and a fusion feature set is generated. Parameters of a model used to generate the fusion feature set are based on joint feedback provided by time-series causal reasoning and distributed cognitive enhancement, are updated online through a target function containing a causal constraint term and a cognitive consistency term, and the target function is: wherein, is a traditional embedding loss based on a feature space distance; is a causal constraint term, which is constructed by calculating a KL divergence of a feature space similarity and a causal strength between corresponding variables in a dynamic causal graph; is a cognitive consistency term, which is constructed by calculating a cosine similarity of a fusion feature and a node state vector in distributed cognitive data; and is a hyperparameter used to balance weights of each term. Analyzing the causal relationship between variables in multi-source data through a time series causal reasoning engine, generating a dynamic causal diagram, and dynamically adjusting the probability distribution of the causal diagram through an online incremental update algorithm; Through a distributed cognitive enhancement module, the water system nodes are given node-level state perception and action decision-making capabilities, and through inter-node information interaction, the system operation mode is optimized, and distributed cognitive data is generated, the water system nodes including pump stations, pipeline valves and sewage treatment equipment; Through self-organizing topology mapping based on complex network theory, combined with the fusion feature set, dynamic causal diagram and distributed cognitive data, the topology network of the water system is dynamically constructed; the dynamic construction includes abstracting the water system as a dynamic network containing nodes, edges and edge weights, the nodes representing devices or subsystems, and the edges representing water flow or energy flow; based on the correlation between physical pipe network layout and water system operation state, the network structure is initialized; combined with the fusion feature set, dynamic causal diagram and distributed cognitive data, the feature similarity, causal strength and cognitive association are calculated, and the edge weight is dynamically adjusted; wherein the topology edge weight is determined by the weighted combination of feature similarity, causal strength and cognitive association, and is updated incrementally online within a predetermined period; Through dynamic game optimization based on multi-agent game theory, combined with the topology network, dynamic causal diagram and distributed cognitive data, a multi-objective collaborative control strategy is generated; the multi-objective collaborative control introduces a dynamic weight formed by the topology, causal strength and distributed cognitive data, and uses the alternating direction multiplier method to decompose and solve the global problem in parallel to meet the real-time constraint; Through a synchronous state atlas visualization engine based on the topology network and dynamic causal diagram, a state atlas of the water system is generated; According to the multi-objective collaborative control strategy, a control instruction is generated, which is executed through the Internet of Things device, and real-time monitoring feedback is performed, and the data set is updated; the system state after execution is used as feedback, which is used to update the model parameters for generating the fusion feature set, the causal strength metric of the dynamic causal diagram and the edge weight of the topology network, thereby forming a self-adaptive optimization closed loop based on joint feedback, integrating perception, analysis, decision-making and execution.

Images