Drainage basin environment data simulation deduction system based on digital twinborn technology

Abstract

The invention relates to the technical field of digital twinborn and drainage basin environment monitoring simulation, in particular to a drainage basin environment data simulation deduction system based on a digital twinborn technology, which comprises a digital twinborn management center used for calling multi-source heterogeneous monitoring data of a drainage basin environment to obtain an aligned environment data sequence; the causal coupling analysis unit is used for obtaining a steady-state logic signal or a dynamic decoupling signal; when a dynamic decoupling signal is generated, the topology adaptive reconstruction unit is used for performing structural mutation recognition feedback analysis on a calculation map of the digital twin model to obtain a reconstructed topology map; when the steady-state logic signal is generated, the simulation deduction execution unit is used for performing environment evolution deduction analysis on the aligned environment data sequence based on the current calculation map to obtain a simulation deduction result; according to the method, model calculation divergence or misleading output caused by sensor faults is effectively avoided, and the calculation problem caused by multi-source data heterogeneity is solved.

Description

Technical Field

[0001] This invention relates to the field of digital twin and watershed environment monitoring and simulation technology, specifically a watershed environment data simulation and deduction system based on digital twin technology. Background Technology

[0002] With the in-depth application of digital twin and Internet of Things technologies in the field of watershed environmental monitoring, the scale and dimensions of multi-source heterogeneous data have shown explosive growth; the complexity and real-time nature of this data pose a severe challenge to the computing architecture and logical processing capabilities of environmental simulation and simulation systems. Currently, existing watershed environment simulation technologies typically rely on static numerical calculation models with fixed structures. These models are constructed based on preset physical boundary conditions and parameters, which have significant limitations when facing complex watershed environments. Specifically, multi-source monitoring data often exhibit differences in sampling frequency and spatial distribution, leading to spatiotemporal semantic misalignment. Direct input can cause model calculations to diverge. More critically, the real physical environment often experiences unexpected abrupt changes, such as tributary diversion or the addition of new pollution sources, and the calculation patterns of static models cannot be automatically adjusted accordingly. This logical disconnect between the preset model and the dynamic real environment prevents the model from passing causal verification, resulting in the simulation results losing logical consistency in dynamic and complex environments, severely impacting the accuracy of water quality evolution and hydrodynamic diffusion predictions. Therefore, how to achieve accurate spatiotemporal alignment of multi-source heterogeneous data and enable simulation models to autonomously reconstruct topology and calibrate logic based on structural changes in the physical environment has become a pressing technical problem to be solved in this field. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a watershed environment data simulation and extrapolation system based on digital twin technology. Specifically, the technical solution of this invention includes: The system includes a digital twin management center, a spatiotemporal semantic alignment unit, a topology adaptive reconstruction unit, a causal coupling analysis unit, and a simulation and deduction execution unit. The digital twin management center is used to retrieve multi-source heterogeneous monitoring data of the watershed environment and send the multi-source heterogeneous monitoring data to the spatiotemporal semantic alignment unit for spatiotemporal asynchronous calibration analysis to obtain an aligned environmental data sequence. The causal coupling analysis unit is used to perform causal chain coupling confidence acquisition analysis on the alignment environment data sequence, process the obtained node association strength value and time delay consistency value to obtain the coupling confidence coefficient, and perform discrimination processing with the coupling confidence coefficient and preset confidence threshold to obtain steady-state logic signal or dynamic decoupling signal; When a dynamic decoupling signal is generated, the topology adaptive reconstruction unit is used to perform structural mutation identification and feedback analysis on the computational graph of the digital twin model to obtain the reconstructed topology graph. When a steady-state logic signal is generated, the simulation inference execution unit is used to perform environmental evolution inference analysis on the aligned environment data sequence based on the current computational graph, and obtain simulation inference results.

[0004] Preferably, the spatiotemporal asynchrony calibration analysis process is as follows: The acquisition timestamps of multi-source heterogeneous monitoring data are collected and set as time series reference points. The spatiotemporal distribution characteristics of the multi-source heterogeneous monitoring data within a preset time window are obtained. The spatiotemporal distribution characteristics include spatial coordinate vectors and time lag steps. Interpolation mapping is performed on the multi-source heterogeneous monitoring data. If an interpolated data sequence is generated, an aligned environment data sequence is generated. If no interpolated data sequence is generated, a default marker instruction is generated, and the data node corresponding to the default marker instruction is set as a node to be completed.

[0005] Preferably, the causal chain coupling confidence acquisition and analysis process is as follows: The interrelation information between each data node in the aligned environment data sequence is obtained. This interrelation information includes node correlation strength values ​​and time-delay consistency values. The node correlation strength values ​​and time-delay consistency values ​​are compared and analyzed with preset correlation strength thresholds and preset time-delay consistency thresholds. The number of node correlation strength values ​​and time-delay consistency values ​​that are greater than or equal to the corresponding preset correlation strength thresholds and preset time-delay consistency thresholds is obtained, and this number is set as the coupling confidence coefficient.

[0006] Preferably, the node association strength value represents the product obtained by multiplying the fluctuation amplitude of the upstream node data and the response amplitude of the downstream node data in the aligned environment data sequence after data normalization; the time delay consistency value represents the complement of the difference obtained by subtracting the actual observed peak propagation time from the theoretical propagation time calculated by the preset physical model in the aligned environment data sequence after data normalization.

[0007] Preferably, the structural mutation identification feedback analysis process is as follows: the target data node corresponding to the generated dynamic decoupling signal is obtained, the target data node is set as the mutation source node, the connection edge weight information of the mutation source node in the current computational graph is obtained, the connection edge weight information includes the in-degree weight value and the out-degree weight value, the gradient sensitivity test is performed on the connection edge weight information to obtain the gradient sensitivity feedback value, the gradient sensitivity feedback value is compared and analyzed with the preset sensitivity threshold to obtain the retain connection instruction or disconnect and reconnect instruction.

[0008] Preferably, when a disconnect and reconnection instruction is generated, the topology adaptive reconstruction unit performs the following operations: obtains the set of potential associated nodes of the mutation source node, calculates the causal contribution value of various potential connection relationships between the mutation source node and each node in the set of potential associated nodes, selects the connection relationship with the largest causal contribution value as the new connection edge, updates the adjacency matrix of the current computation graph based on the new connection edge, and sets the computation graph after updating the adjacency matrix as the reconstructed topology graph.

[0009] Preferably, the environmental evolution simulation analysis process is as follows: The aligned environmental data sequence is obtained as the input boundary condition; the input boundary condition is loaded onto the corresponding input node of the current computational graph; the node transfer function in the current computational graph is activated; the numerical sequence of the output node of the current computational graph is obtained; and the numerical sequence is set as the simulation result. When the actual monitoring data for the next time moment is obtained, the deviation between the simulation result and the actual monitoring data for the next time moment is calculated; and the deviation is set as the simulation error feedback signal.

[0010] Preferably, the simulation results include water quality evolution trend values ​​and hydrodynamic diffusion range values; the water quality evolution trend value represents the peak value of pollutant concentration in the target water area within a preset time period after data normalization; the hydrodynamic diffusion range value represents the ratio of the total number of grid cells with flow velocities greater than a preset flow velocity threshold in the velocity vector field of the target water area to the total number of grid cells.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention establishes a unified temporal reference point through a spatiotemporal semantic alignment unit and eliminates spatiotemporal semantic misalignment between multi-source monitoring data by using interpolation mapping technology; in particular, it generates default label instructions instead of simple zero-value filling when data is missing, which effectively avoids model calculation divergence or misleading output caused by sensor failure, provides a high-precision synchronous input basis for subsequent simulation, and solves the calculation problem caused by the heterogeneity of multi-source data. 2. This invention introduces a causal coupling analysis mechanism, which achieves dual verification of the model's physical logic by calculating the node correlation strength and time delay consistency. The system not only focuses on the fitting of data values, but also verifies the degree of consistency between energy transfer and the physical laws of time delay. This mechanism ensures that the model is only deemed valid when the physical wave propagation is highly consistent with the theoretical time series, avoiding simulation distortion caused by traditional models that rely solely on data fitting while ignoring physical causality. 3. This invention has the ability to adaptively reconstruct topology, and can automatically identify structural abrupt changes in the physical environment through gradient sensitivity testing; when the model is detected to be decoupled from the real logic, the system can automatically disconnect the failed connection and establish a new effective connection based on the causal contribution value to update the adjacency matrix; this self-repair mechanism, similar to neural plasticity, ensures that the computational graph can follow sudden changes such as tributary diversion in real time and maintain high-precision inference capability. 4. This invention outputs standardized water quality evolution trends and hydrodynamic diffusion range values ​​through a simulation and deduction execution unit combined with an error feedback mechanism. The system transforms complex three-dimensional flow field data into intuitive risk coverage indicators, directly reflecting the absolute risk level of pollution diffusion. This not only significantly improves the accuracy of peak prediction under sudden sewage discharge events, but also provides efficient and quantitative decision-making basis for environmental emergency command. Attached Figure Description

[0012] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0014] Example 1: Please see Figure 1 The watershed environment data simulation and deduction system based on digital twin technology includes: a digital twin management center, a spatiotemporal semantic alignment unit, a topology adaptive reconstruction unit, a causal coupling analysis unit, and a simulation and deduction execution unit; The digital twin management center is used to retrieve multi-source heterogeneous monitoring data of the watershed environment and send the multi-source heterogeneous monitoring data to the spatiotemporal semantic alignment unit for spatiotemporal asynchronous calibration analysis to obtain an aligned environmental data sequence. The causal coupling analysis unit is used to perform causal chain coupling confidence acquisition analysis on the alignment environment data sequence, process the obtained node association strength value and time delay consistency value to obtain the coupling confidence coefficient, and perform discrimination processing with the coupling confidence coefficient and preset confidence threshold to obtain steady-state logic signal or dynamic decoupling signal; When a dynamic decoupling signal is generated, the topology adaptive reconstruction unit is used to perform structural mutation identification and feedback analysis on the computational graph of the digital twin model to obtain the reconstructed topology graph. When a steady-state logic signal is generated, the simulation inference execution unit is used to perform environmental evolution inference analysis on the aligned environment data sequence based on the current computational graph, and obtain simulation inference results.

[0015] This embodiment details the overall architecture and core logic of the system, aiming to solve the technical challenge of the disconnect between the preset simulation model and the logic of the real environment caused by abrupt changes in physical boundary conditions in complex watershed environments. Instead of using a static numerical calculation model, the system constructs a virtual mapping entity with self-evolution capabilities. The digital twin management center, acting as the system's data throughput hub, performs data retrieval tasks. The multi-source heterogeneous monitoring data originates from water quality sensors, hydrological stations, and meteorological satellites deployed in the target watershed, covering physical quantities of different dimensions such as ammonia nitrogen, COD, flow velocity, water level, and rainfall. The spatiotemporal semantic alignment unit receives these raw data with different sampling frequencies and data formats, and maps them uniformly to the calculation step size of the digital twin model, eliminating semantic misalignment between data sources. Based on this, the causal coupling analysis unit, as the core logic judgment module, comprehensively judges the degree of consistency between the current model structure and physical laws by calculating the node association strength value and time delay consistency value. In response to the coupling confidence coefficient being less than the threshold, the system determines that a structural change has occurred in the physical environment, such as tributary diversion or the access of a new pollution source, and then generates a dynamic decoupling signal to activate the topology adaptive reconstruction unit. Conversely, in response to the coupling confidence coefficient being greater than or equal to the threshold, the system generates a steady-state logic signal, instructing the simulation inference execution unit to output simulation inference results such as a water quality diffusion map for the next 72 hours, based on the current calculation map or reconstructed topology map and with the environmental data sequence as input. This embodiment breaks through the limitations of traditional water conservancy simulation models, which have fixed structures and fine-tuned parameters, by introducing causal coupling analysis and topology adaptive reconstruction mechanism. When unexpected changes occur in the watershed environment, the system can automatically sense the logical decoupling between the model and reality and autonomously reconstruct the computational topology, ensuring the logical consistency and prediction accuracy of the simulation in dynamic and complex environments.

[0016] Example 2: The spatiotemporal asynchrony calibration analysis process is as follows: The acquisition timestamps of multi-source heterogeneous monitoring data are collected and set as time series reference points. The spatiotemporal distribution characteristics of the multi-source heterogeneous monitoring data within a preset time window are obtained. The spatiotemporal distribution characteristics include spatial coordinate vectors and time lag steps. Interpolation mapping is performed on the multi-source heterogeneous monitoring data. If an interpolated data sequence is generated, an aligned environment data sequence is generated. If no interpolated data sequence is generated, a default marker instruction is generated, and the data node corresponding to the default marker instruction is set as a node to be completed.

[0017] This embodiment deepens the spatiotemporal asynchrony calibration analysis process in Embodiment 1. Since the data acquisition from multi-source sensors is often asynchronous, directly inputting it into the model will lead to computational divergence. Therefore, it is necessary to establish a unified spatiotemporal reference. The system extracts the time information from the header of the multi-source heterogeneous monitoring data packet and establishes it as the time series reference point. The system analyzes the spatiotemporal distribution characteristics of data within a preset time window, such as the past hour; where: spatial coordinate vector The source is the sensor's GPS positioning module; its physical meaning is the longitude, latitude, and elevation information of the data source in the geographical space of the watershed; the time lag step size. The source is the difference between the system master clock and the data acquisition timestamp, and its physical meaning is the delay of the data source relative to the system synchronization time. The system performs interpolation mapping, using spatial interpolation algorithms such as Kriging interpolation and temporal interpolation algorithms such as spline interpolation to attempt to map discrete data onto standardized spatiotemporal grid points. In response to successful interpolation, the system outputs time-synchronized and spatially aligned tensor data, i.e., an aligned environmental data sequence. In response to sensor malfunctions or communication interruptions preventing the generation of valid interpolations, the system does not perform simple zero-value filling but instead generates default labeling instructions, such as labeling as... Or a specific mask, and lock the corresponding data node as the node to be completed; This embodiment effectively solves the asynchronous problem of multi-source data in time and space by establishing a unified time-series reference point and interpolation mapping mechanism; in particular, the introduction of default labeling instructions avoids model calculation crashes or misleading zero-value outputs caused by missing data, providing a high-quality input foundation for subsequent causal analysis.

[0018] Example 3: The process for obtaining and analyzing the confidence level of causal chain coupling is as follows: The system obtains the correlation information between each data node in the aligned environment data sequence. The correlation information includes the node correlation strength value and the time delay consistency value. The system compares and analyzes the node correlation strength value and the time delay consistency value with the preset correlation strength threshold and the preset time delay consistency threshold. The system obtains the number of node correlation strength values ​​and time delay consistency values ​​that are greater than or equal to the corresponding preset correlation strength threshold and the preset time delay consistency threshold. The system sets the number of node correlation strength values ​​and time delay consistency values ​​that are greater than or equal to the corresponding preset correlation strength threshold and the preset time delay consistency threshold as the coupling confidence coefficient. The node association strength value represents the product obtained by multiplying the fluctuation amplitude of the upstream node data and the response amplitude of the downstream node data in the aligned environment data sequence after data normalization. The time delay consistency value represents the complement of the difference between the actual observed peak propagation time in the aligned environment data sequence and the theoretical propagation time calculated by the preset physical model after data normalization.

[0019] This embodiment details the specific logic of obtaining and analyzing the confidence level of causal chain coupling, and defines key quantitative indicators. This process aims to verify whether the model understands the current physical environment through dual indicators; the system extracts the fluctuation amplitude of upstream node data from the aligned environment data sequence. and downstream node data response amplitude All calculations are performed using the root mean square (RMS) or standard deviation within a sliding time window to ensure that the dimensions are the physical quantity itself rather than its square. The calculation formula is modified as follows: in, The number of data points within the window. For the first in the window Each sampling data point The normalization function is the arithmetic mean of all data points within the window, used to quantify the dispersion of the data, i.e., the fluctuation energy; the normalization function involved in this embodiment... The specific calculation formula is as follows: in, and These are the minimum and maximum values ​​of the corresponding data sequence in the current statistical period, consistent with the aforementioned preset time window, such as the minimum and maximum values ​​within 1 hour. To prevent the preset minimum value where the denominator is zero, such as System calculation time delay consistency value To avoid the loss of absolute time error caused by separate normalization, this embodiment adopts a global error normalization method, the formula of which is modified as follows: in, To actually observe the wave crest propagation time, The theoretical transmission time is calculated. Use the preset maximum allowable time delay tolerance value Perform normalization processing; The value is determined based on the hydraulic residence time characteristics of the target watershed, specifically set to the longest observation time from upstream peak to downstream in historical monitoring data. For example, the value is 1.2 times the historical maximum lag time, which is set to 3600 seconds in this embodiment. This processing ensures that the normalized value can objectively reflect the relative proportion of time lag error within the physical limit range, i.e. This calculation method ensures that the calculation is performed only when the absolute error between the actual and theoretical values ​​is extremely small. Only then does it approach a high confidence value, meaning the theoretically calculated propagation time in physical terms; specifically, the Saint-Venant equations include: The continuity equation for a one-dimensional unsteady flow is given by the following formula: The momentum equation is as follows: in, Indicates the distance coordinates along the river's flow direction. Represents a time variable; The cross-sectional area of ​​the water passage. For traffic, For water level, The friction slope is determined by Manning's formula: in, The roughness coefficient of the river channel is not arbitrarily set, but is obtained through calibration using historical hydrological data of the watershed. Specifically, the measured water level-discharge relationship curves of the watershed during the wet and dry seasons are selected, and the comprehensive roughness value at different water level levels is derived using the least squares method. A water level-roughness lookup table is then established for dynamic retrieval. The hydraulic radius is determined by the cross-sectional area of ​​the water passage. With wet week The ratio is determined, that is / Among them, wetted perimeter Calculated based on measured river cross-sectional topographic data; It is a side inflow. Let gravitational acceleration be the value. ; System utilization The implicit finite difference method of the scheme is used to discretize and solve the above equations, calculating the number of time steps required for the wave crest to propagate from the upstream node to the downstream node under preset standard hydraulic conditions, thus obtaining a computable solution. Value; introduced during the solution process The source is a preset constant, whose physical meaning is to prevent tiny values ​​where the denominator is zero; based on this, the system calculates the node association strength value. and The results are compared with a preset association strength threshold, such as 0.7, and a preset time delay consistency threshold, such as 0.8. The number of all node pairs that simultaneously satisfy both indicators greater than or equal to the threshold is counted, and this statistical value is established as the coupling confidence coefficient. This embodiment achieves precise quantification of model credibility through dual verification of energy transfer efficiency and time lag consistency; the model is only deemed valid when the physical wave propagation is highly consistent with the theoretical time sequence logic, thereby avoiding simulation distortion caused by ignoring physical causality based solely on data fitting.

[0020] Example 4: The structural mutation identification feedback analysis process is as follows: The target data node corresponding to the generated dynamic decoupling signal is obtained, the target data node is set as the mutation source node, the connection edge weight information of the mutation source node in the current computation graph is obtained, the connection edge weight information includes the in-degree weight value and the out-degree weight value, the gradient sensitivity test is performed on the connection edge weight information to obtain the gradient sensitivity feedback value, the gradient sensitivity feedback value is compared and analyzed with the preset sensitivity threshold to obtain the retain connection instruction or disconnect and reconnect instruction; When a disconnect and reconnect command is generated, the topology adaptive reconstruction unit performs the following operations: obtains the set of potential associated nodes of the mutation source node, calculates the causal contribution value of various potential connection relationships between the mutation source node and each node in the set of potential associated nodes, selects the connection relationship with the largest causal contribution value as the new connection edge, updates the adjacency matrix of the current computation graph based on the new connection edge, and sets the computation graph after updating the adjacency matrix as the reconstructed topology graph.

[0021] This embodiment describes how, when the system detects model failure, it repairs the complete closed loop of the model through structural mutation identification feedback analysis and topology adaptive reconstruction. The system identifies the target data node causing the decrease in coupling confidence, i.e., the node with the largest prediction error, and marks it as the mutation source node. The system extracts the connection edge weight information of this node in the computational graph, performs gradient sensitivity testing on these weights, calculates the partial derivative of the prediction error loss function with respect to the weights, and obtains the gradient sensitivity feedback value. The calculation formula is as follows: in, The source is backpropagation calculation, and its physical meaning is the sensitivity of the connecting edges to the model's prediction error; The source is a preset loss function, such as mean squared error (MSE), which physically represents the total error between the model's predicted values ​​and the actual observed values. Its specific calculation formula is as follows: in, This represents the batch sample size. These are actual observations. These are the model's predicted values; The source is the learnable attention weight of each connection edge in the current computational graph, i.e., the weight matrix below. The corresponding element in the range of values ​​is a continuous interval. This is used to transmit gradient information during backpropagation; based on this, the logic for generating the disconnect and reconnection instruction is as follows: in response to Below the preset sensitivity threshold, and the corresponding weight parameters If the connection decays to below a preset threshold during training, such as 0.01, the system determines that the connection has failed physically and logically, and generates a disconnect and reconnection instruction, setting the corresponding position in the adjacency matrix to zero; the topology adaptive reconstruction unit intervenes to obtain a set of potential associated nodes of the mutation source node, such as geographically adjacent nodes that are not currently connected. Based on this, the system uses the transfer entropy algorithm to calculate the causal contribution value between the mutation source node and each node in the set. This value quantifies the directed information flow between the two time series; specifically, the mutation source node sequence potential associated node sequences Transitive entropy The calculation formula is as follows: Among them, the summation symbol Represents the variable Iterate through and sum all possible values ​​of the state; and Representing the mutation source node sequence and potential associated node sequences At any moment The historical embedding vector; Indicates the discrete time step index; The joint probability distribution is represented by the Gaussian kernel density estimation method, and its kernel function is... Using a standard Gaussian distribution: Furthermore, the kernel bandwidth parameter is adaptively determined according to Scott's rule, i.e., the kernel bandwidth calculation formula is: in, For the sample size, The dimension of the random variable. The standard deviation is the sample standard deviation. and Representing sequences respectively and At any moment Historical embedding vector, embedding dimension and All are preset to 3; this calculation process can accurately capture nonlinear causal driving relationships; the system selects the connection relationship with the largest causal contribution value to establish a new connection edge, and updates the adjacency matrix of the computation graph to complete the graph reconstruction; This embodiment implements a self-repair mechanism similar to neural plasticity; when the physical environment undergoes a sudden change such as the interruption of a water flow path or the creation of a new path, the system can automatically cut off invalid old connections and establish valid new connections, thereby ensuring that the simulation model can follow the structural changes of the physical world in real time and maintain high-precision inference capabilities.

[0022] Example 5: The environmental evolution simulation and analysis process is as follows: The alignment environment data sequence is obtained as the input boundary condition. The input boundary condition is loaded into the corresponding input node of the current computation graph, the node transfer function in the current computation graph is activated, the numerical sequence of the output node of the current computation graph is obtained, and the numerical sequence is set as the simulation result. When the actual monitoring data of the next moment is obtained, the deviation between the simulation result and the actual monitoring data of the next moment is calculated, and the deviation is set as the simulation error feedback signal. The simulation results include water quality evolution trend values ​​and hydrodynamic diffusion range values; The water quality evolution trend value represents the peak value of pollutant concentration in the target water area within a preset time period after data normalization; the hydrodynamic diffusion range value represents the ratio of the total number of grid cells with flow velocities greater than a preset flow velocity threshold in the velocity vector field of the target water area to the total number of grid cells.

[0023] This embodiment details the specific execution steps and physical definition of the output results of the environmental evolution simulation analysis. After obtaining the steady-state or reconstructed computational graph, the system loads the aligned environmental data sequence as input boundary conditions onto the input nodes of the graph. The system activates the node transfer functions in the graph, such as the aggregation and update functions in a graph neural network, to simulate the migration and diffusion process of pollutants in the topological network. The aggregation and update functions specifically employ the following message passing mechanism, and their calculation formula is as follows: in, Represents a node In the The state vector at any given moment corresponds, at the physical level, to the water quality concentration and flow velocity characteristic vectors of the monitoring section. For nodes The set of adjacent nodes corresponds to the flow nodes in the physical topology. All upstream tributary nodes; As a aggregation operator, its physical meaning is to simulate the mass conservation superposition in the process of material transport, that is, to aggregate the pollutant flux input from all upstream tributaries; and Let be a learnable weight matrix obtained by training with historical data, where is the matrix. The numerical values ​​of the elements in the model reflect the weighting of the influence of different upstream tributaries on downstream nodes, i.e., simulating the difference in the flow ratio between the main stream and tributaries, thus ensuring that the neural network's calculation process conforms to the convection and diffusion laws of hydrodynamics. For non-linear activation functions, such as Weight matrix and The training process is as follows: The system uses the Xavier initialization method to initialize... and Perform random initialization; Using historical alignment environment data sequences as training samples, a training loss function is constructed, and its calculation formula is as follows: in, This represents the total number of time steps for the training samples. Backpropagation is performed using the Adam optimization algorithm, and the gradient is calculated based on the chain rule. The specific gradient calculation formula is as follows: as well as Based on parameter update rules and Iteratively update the weights, where, A preset learning rate, such as 0.01, is used until the loss function converges. This function simulates the physical convection and diffusion processes of matter in the watershed topology. The system extracts the numerical sequence of the output nodes as the simulation result and calculates the deviation between this result and the actual monitoring data at the next moment in real time, generating a simulation error feedback signal for fine-tuning the weights in the next round. This simulation result includes two core indicators: water quality evolution trend value. The calculation formula is as follows: in, The source is a predicted sequence derived from simulation, and its physical meaning is the peak pollutant concentration in the target water area within a preset time period, such as 24 hours; here The same minimum-maximum normalization method was used for processing; hydrodynamic diffusion range value The calculation formula is revised as follows: in: The total number of grid cells with a flow rate greater than a preset flow rate threshold, such as 0.5 m / s; This represents the total number of grid cells in the target water area. Regarding the logic for the hydrodynamic diffusion range value, this system directly uses the above ratio as the output result without performing Min-Max normalization. This is because the quotient itself is... The dimensionless physical quantities within the interval directly reflect the absolute risk level of pollution diffusion. However, secondary normalization might amplify minute fluctuations into full-scale risk signals when the overall diffusion range is small, leading to misjudgments in command. Ensuring the dimensionless consistency of the indicators within the [0,1] interval is crucial to avoid numerical deviations caused by varying grid density. This embodiment transforms complex simulation data into intuitive decision-making data through standardized quantitative indicator output; in particular, the definition of the hydrodynamic diffusion range transforms flow field data into risk coverage, significantly improving the efficiency of pollution diffusion situation assessment in environmental emergency command. To verify this embodiment, a 30-day comparative test was conducted in a real watershed, covering two typical operating conditions: the dry season and the wet season. Experimental results show that, compared to traditional hydrodynamic models without causal coupling analysis, this system improves the accuracy of water quality peak prediction by 18.5% under sudden sewage discharge events, reduces the root mean square error (RMSE) from 0.45 mg / L to 0.12 mg / L, and shortens the model adaptation time after structural abrupt changes such as tributary diversion from an average of 48 hours to less than 2 hours. This effectively demonstrates the superiority of the dynamic topology reconstruction mechanism in the evolutionary deduction of complex environments.

[0024] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A watershed environment data simulation and extrapolation system based on digital twin technology, characterized in that, It includes a digital twin management center, a spatiotemporal semantic alignment unit, a topology adaptive reconstruction unit, a causal coupling analysis unit, and a simulation and deduction execution unit; The digital twin management center is used to retrieve multi-source heterogeneous monitoring data of the watershed environment and send the multi-source heterogeneous monitoring data to the spatiotemporal semantic alignment unit for spatiotemporal asynchronous calibration analysis to obtain an aligned environmental data sequence. The causal coupling analysis unit is used to perform causal chain coupling confidence acquisition analysis on the alignment environment data sequence, process the obtained node association strength value and time delay consistency value to obtain the coupling confidence coefficient, and perform discrimination processing with the coupling confidence coefficient and preset confidence threshold to obtain steady-state logic signal or dynamic decoupling signal; When a dynamic decoupling signal is generated, the topology adaptive reconstruction unit is used to perform structural mutation identification and feedback analysis on the computational graph of the digital twin model to obtain the reconstructed topology graph. When a steady-state logic signal is generated, the simulation inference execution unit is used to perform environmental evolution inference analysis on the aligned environment data sequence based on the current computational graph, and obtain simulation inference results.

2. The watershed environment data simulation and extrapolation system based on digital twin technology according to claim 1, characterized in that, The spatiotemporal asynchrony calibration analysis process is as follows: The acquisition timestamps of multi-source heterogeneous monitoring data are collected and set as time series reference points. The spatiotemporal distribution characteristics of the multi-source heterogeneous monitoring data within a preset time window are obtained. The spatiotemporal distribution characteristics include spatial coordinate vectors and time lag steps. Interpolation mapping is performed on the multi-source heterogeneous monitoring data. If an interpolated data sequence is generated, an aligned environment data sequence is generated. If no interpolated data sequence is generated, a default marker instruction is generated, and the data node corresponding to the default marker instruction is set as a node to be completed.

3. The watershed environment data simulation and extrapolation system based on digital twin technology according to claim 1, characterized in that, The process of obtaining and analyzing the causal chain coupling confidence is as follows: The interrelation information between data nodes in the aligned environment data sequence is obtained. This interrelation information includes node association strength values ​​and time-delay consistency values. The node association strength values ​​and time-delay consistency values ​​are compared and analyzed with preset association strength thresholds and preset time-delay consistency thresholds. The number of node association strength values ​​and time-delay consistency values ​​that are greater than or equal to the corresponding preset association strength thresholds and preset time-delay consistency thresholds is obtained. This number is then set as the coupling confidence coefficient.

4. The watershed environment data simulation and extrapolation system based on digital twin technology according to claim 3, characterized in that, The node association strength value represents the product obtained by multiplying the fluctuation amplitude of the upstream node data and the response amplitude of the downstream node data in the aligned environment data sequence after data normalization; the time delay consistency value represents the complement of the difference obtained by subtracting the actual observed peak propagation time from the theoretical propagation time calculated by the preset physical model in the aligned environment data sequence after data normalization.

5. The watershed environment data simulation and extrapolation system based on digital twin technology according to claim 1, characterized in that, The structural mutation identification feedback analysis process is as follows: the target data node corresponding to the generated dynamic decoupling signal is obtained, the target data node is set as the mutation source node, the connection edge weight information of the mutation source node in the current computation graph is obtained, the connection edge weight information includes the in-degree weight value and the out-degree weight value, the gradient sensitivity test is performed on the connection edge weight information to obtain the gradient sensitivity feedback value, the gradient sensitivity feedback value is compared and analyzed with the preset sensitivity threshold to obtain the retain connection instruction or disconnect and reconnect instruction.

6. The watershed environment data simulation and extrapolation system based on digital twin technology according to claim 5, characterized in that, When a disconnect and reconnect command is generated, the topology adaptive reconstruction unit performs the following operations: obtains the set of potential associated nodes of the mutation source node, calculates the causal contribution value of various potential connection relationships between the mutation source node and each node in the set of potential associated nodes, selects the connection relationship with the largest causal contribution value as the new connection edge, updates the adjacency matrix of the current computation graph based on the new connection edge, and sets the computation graph after updating the adjacency matrix as the reconstructed topology graph.

7. The watershed environment data simulation and extrapolation system based on digital twin technology according to claim 1, characterized in that, The environmental evolution simulation analysis process is as follows: the aligned environmental data sequence is obtained as the input boundary condition, the input boundary condition is loaded into the corresponding input node of the current computational graph, the node transfer function in the current computational graph is activated, the numerical sequence of the output node of the current computational graph is obtained, and the numerical sequence is set as the simulation result; when the actual monitoring data of the next moment is obtained, the deviation value between the simulation result and the actual monitoring data of the next moment is calculated, and the deviation value is set as the simulation error feedback signal.

8. The watershed environment data simulation and extrapolation system based on digital twin technology according to claim 7, characterized in that, The simulation results include water quality evolution trend values ​​and hydrodynamic diffusion range values; the water quality evolution trend value represents the peak pollutant concentration in the target water area within a preset future time period after data normalization. The hydrodynamic diffusion range value represents the ratio of the total number of grid cells with flow velocities greater than a preset flow velocity threshold in the velocity vector field of the target water area to the total number of grid cells.

Images