Digital twin water conservancy super-fusion method coupled with multi-source data and all-in-one machine

Abstract

The application provides a digital twin water conservancy super-fusion method and an all-in-one machine for coupling multi-source data, and relates to the technical field of intelligent water conservancy.The application realizes efficient fusion of multi-source heterogeneous data such as hydrology, meteorology, engineering operation, etc., solves the problem of water conservancy system data island by constructing a unified water conservancy space-time data chain protocol; a coupling distributed hydrological model and a hydrodynamic model are established to realize real-time and accurate simulation of the hydrological process of a river basin and the hydrodynamic process of a river channel; a multi-objective dynamic weight reinforcement learning algorithm is used to optimize the water conservancy scheduling scheme, and the model parameters are dynamically corrected in real time through a time sequence error regression analysis model to realize closed-loop adaptive optimization; the application realizes the closed-loop cooperation of water conservancy data fusion, real-time and accurate simulation, intelligent scheduling optimization and dynamic parameter self-correction, and significantly improves the real-time and accuracy of intelligent water conservancy decision-making.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart water conservancy, in particular to a digital twin water conservancy super-fusion method coupled with multi-source data and an all-in-one machine. BACKGROUND

[0002] The safe and efficient operation of water conservancy projects is of great significance for flood control and disaster reduction and optimal allocation of water resources. First, water conservancy data is diverse, including hydrological, meteorological, and engineering operation data scattered in different departments and systems, with non-uniform data standards, lack of effective data fusion means, and difficulty in achieving unified management and efficient utilization, resulting in a serious "data island" problem, which seriously affects the overall effectiveness of water conservancy decision-making and coordinated scheduling.

[0003] Secondly, traditional hydrological and hydrodynamic simulation models generally rely on a large amount of historical data for parameter calibration, and have the problems of model parameter staticization and lack of real-time performance. In particular, in emergency scenarios such as sudden floods, it is difficult to quickly and dynamically respond to the real-time needs of water conservancy scheduling decisions, and it is unable to meet the precise and real-time prediction capabilities required by the current complex and changing water conservancy environment.

[0004] Furthermore, the hardware resources (such as monitoring sensors and computing servers) and software systems (including data monitoring, simulation, and optimization scheduling) of existing water conservancy systems are often independent of each other, lack effective coordination mechanisms, and are difficult to form a unified and efficient overall scheduling system, which seriously reduces the response speed and decision-making efficiency of the water conservancy system.

[0005] In summary, how to break the isolated state of water conservancy data and build a unified data fusion system, how to improve the real-time performance and prediction accuracy of water conservancy simulation models to meet dynamic response needs, and how to achieve efficient coordination between hardware resources and software systems have become core technical problems that urgently need to be solved in the field of smart water conservancy. SUMMARY

[0006] The present application aims to at least solve one of the technical problems existing in the prior art; the present application provides a digital twin water conservancy super-fusion method coupled with multi-source data and an all-in-one machine.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0008] In a first aspect, the present application provides a digital twin water conservancy super-fusion method coupled with multi-source data, comprising:

[0009] A water conservancy space-time data chain protocol is established to realize water conservancy monitoring data formatting processing through hierarchical data modeling; an edge computing node is determined based on the formatted water conservancy monitoring data and water data, a predicted node future load value is obtained based on the edge computing node, and a cloud fusion data set is generated by selecting an adaptive data transmission format;

[0010] constructing a coupled distributed hydrological model and a hydrodynamic model based on the cloud fusion data set, generating an initial scheduling scheme for a prediction period according to rain information and performing forward simulation rehearsal to obtain an initial simulation result;

[0011] According to the difference data between the initial simulation result and the preset scheduling target, the control parameters of the initial scheduling scheme for the prediction period are optimized and adjusted in reverse by using a multi-objective dynamic weight reinforcement learning algorithm, and an optimized scheduling scheme for the prediction period is generated.

[0012] Based on the optimized scheduling scheme for the prediction period, forward simulation rehearsal is performed again to obtain an optimized simulation result, and according to the real-time deviation of the optimized simulation result and the preset scheduling target, a time series error regression analysis model is used to dynamically correct the model parameters of the coupled distributed hydrological model and the hydrodynamic model, so as to realize dynamic self-correction of the model parameters.

[0013] In a second aspect, the present application provides a digital twin water conservancy super-fusion all-in-one machine coupled with multi-source data, which is provided with a plurality of modules for realizing the above-mentioned digital twin water conservancy super-fusion method coupled with multi-source data, and the modules include:

[0014] A data fusion module is used to establish a water conservancy space-time data chain protocol, realize water conservancy monitoring data formatting processing through hierarchical data modeling, determine an edge computing node based on formatted water conservancy monitoring data and water data, obtain a predicted node future load value based on the edge computing node, and select an adaptive data transmission format to generate a cloud fusion data set.

[0015] A coupled hydrological simulation module is used to construct a coupled distributed hydrological model and a hydrodynamic model based on the cloud fusion data set, generate an initial scheduling scheme for a prediction period according to rain information, and perform forward simulation rehearsal to obtain an initial simulation result.

[0016] A scheduling optimization module is used to take the difference data as feedback, and use a multi-objective dynamic weight reinforcement learning algorithm to optimize and adjust the control parameters of the initial scheduling scheme for the prediction period in reverse, so as to generate an optimized scheduling scheme for the prediction period.

[0017] A parameter self-correction module is used to perform forward simulation rehearsal again based on the optimized scheduling scheme for the prediction period, to obtain an optimized simulation result, and according to the real-time deviation of the optimized simulation result and the preset scheduling target, a time series error regression analysis model is used to dynamically correct the model parameters of the coupled distributed hydrological model and the hydrodynamic model, so as to realize dynamic self-correction of the model parameters.

[0018] Compared with the prior art, the present application has the following advantages:

[0019] The application realizes efficient fusion and unified processing of multi-source heterogeneous water conservancy monitoring data by establishing a unified water conservancy space-time data chain protocol, breaks the traditional water conservancy system "data island" problem, and effectively improves the integration and collaborative utilization efficiency of water conservancy monitoring data.

[0020] The application significantly improves the real-time performance and dynamic response capability of the simulation model by constructing an edge computing driven coupled distributed hydrology and hydrodynamic simulation model combined with a multi-objective dynamic weight reinforcement learning algorithm, and ensures accurate prediction and rapid decision-making capability under extreme hydrological events such as sudden floods.

[0021] The application realizes efficient closed-loop collaboration between hardware resources and software systems by introducing a time series error regression analysis model for dynamic real-time self-correction of model parameters, ensuring sustained high accuracy and dynamic optimization capability of the water conservancy system in long-term operation, and effectively improving the overall intelligent management level of water conservancy projects.

[0022] The application realizes closed-loop collaboration of water conservancy data fusion, real-time accurate simulation, intelligent scheduling optimization and dynamic parameter self-correction, significantly improving the real-time performance and accuracy of water conservancy intelligent decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 Flowchart of the coupled multi-source data digital twin water conservancy super-fusion method of embodiment 1.

[0024] Figure 2 Architecture diagram of the coupled multi-source data digital twin water conservancy super-fusion all-in-one machine of embodiment 2. DETAILED DESCRIPTION

[0025] The application will be further described below in conjunction with the drawings and examples in the specification.

[0026] Embodiment 1

[0027] Please refer to Figure 1 The application provides a coupled multi-source data digital twin water conservancy super-fusion method, which includes:

[0028] A water conservancy space-time data chain protocol is established to realize water conservancy monitoring data formatting processing through hierarchical data modeling; based on the formatted water conservancy monitoring data and water data, an edge computing node is determined, the predicted node future load value is obtained based on the edge computing node, and an adaptive data transmission format is selected to generate a cloud fusion data set;

[0029] Based on the cloud fusion data set, a coupled distributed hydrology model and a hydrodynamic model are constructed, an initial scheduling scheme for the forecast period is generated according to the rainwater information, and a forward simulation is performed to obtain an initial simulation result;

[0030] According to the difference data between the initial simulation result and the preset scheduling target, a multi-objective dynamic weight reinforcement learning algorithm is adopted to reversely optimize and adjust the control parameters of the initial scheduling scheme of the prediction period, to generate an optimized scheduling scheme of the prediction period, with the difference data as feedback;

[0031] Based on the optimized scheduling scheme of the prediction period, forward simulation pre-performance verification is performed again, to obtain an optimized simulation result, and according to the real-time deviation of the optimized simulation result from the preset scheduling target, a time sequence error regression analysis model is adopted to dynamically correct the model parameters of the coupled distributed hydrological model and the hydrodynamic model, to realize dynamic self-correction of the model parameters.

[0032] In view of the diversity of data sources in the embodiment, different types of sensors (due to differences in monitoring principles and functional positioning, the generated data structures are significantly different. Therefore, the process of water conservancy monitoring data formatting in the embodiment is as follows:

[0033] Hierarchical modeling is performed on the water conservancy monitoring data to obtain a multi-tuple data structure, which is represented as: , wherein, represents the unique identifier of the monitoring subject , which is used for globally unique identification of the water conservancy monitoring equipment or node, and can be a sensor number, device address or network node number.

[0034] represents the monitoring data value of the monitoring subject . According to different specific monitoring objects, it can be a water level value, a flow value, a water quality parameter, a meteorological parameter, etc. If the monitoring subject is a water level sensor, the corresponding monitoring data value is a water level value; if the monitoring subject is a flow sensor, the corresponding monitoring data value is a flow rate, a cross-sectional area, a flow calculation method, etc. If the monitoring subject is a water quality sensor, the corresponding monitoring data value is a water level value pH , a dissolved oxygen concentration, a conductivity, etc. and a unit mapping relationship.

[0035] represents the geographic location information of the monitoring subject , such as the national height datum, the latitude and longitude coordinates or the grid coordinates.

[0036] represents a collection timestamp, expressed in Coordinated Universal Time (UTC), which is used to unify the time base of data of different monitoring subjects, to realize accurate fusion and association of multi-source data. UTC

[0037] In a further embodiment, the multi-tuple data structure further includes: a monitoring subject ​Monitoring principle of the monitoring system , that is, a five-element data structure is obtained, which is expressed as: .

[0038] To determine the specific location of data processing, the embodiment determines the edge computing node by the following method:

[0039] According to water data, the monitoring area is divided into a plurality of sub-areas; the water data described in the embodiment can be the geographical environment, water system structure, node spatial layout, network communication characteristics, and the like of the monitoring area.

[0040] The spatial positions of the sensor nodes in each sub-area are obtained in real time by using the Beidou satellite positioning system, a topological score formula is established, and the topological centrality scores of the sensor nodes are calculated The spatial priority order of each sensor node is determined by the topological centrality scores . The expression form of the topological score formula is as follows:

[0041] ,

[0042] wherein, represents the total number of sensor nodes in the sub-area, represents the number of grids passed through during data transmission between the sensor node and the sensor node , represents the spatial distance between the sensor node and the sensor node , and respectively represent the maximum number of grids and the maximum spatial distance between any two sensor nodes, and respectively represent the weight coefficients of the network hop factor and the spatial distance factor, and satisfy ;

[0043] In each sub-area, a main edge computing node and a plurality of backup nodes are determined according to the topological centrality scores ; and the main edge computing node and the plurality of backup nodes in the embodiment satisfy the following dynamic setting:

[0044] A health score mechanism is established to monitor the health score of the main edge computing node in real time; when the health score of the main node is less than a preset health threshold, the node with the highest health score is selected from the backup nodes as a new main edge computing node. The health score mechanism described in the embodiment can be established based on the occupancy rate, memory usage rate, network communication delay, and the like. CPU

[0045] ​The health score obtains a health score between 0 and 1, wherein 1 represents the best node state, and 0 represents complete failure of the node.

[0046] For example, in a certain sub-region, the health score of the current master edge computing node is initially 0.95, and as the running load increases or the network delay rises, the health score of the master node gradually decreases. When the health score of the master node falls below the preset health threshold (for example, 0.6) (assuming 0.55), the system triggers the standby node switching mechanism; at this time, the node with the highest score (0.82) is automatically selected from the preset standby nodes (for example, nodes A with a score of 0.75, node B with a score of 0.82, and node C with a score of 0.68) in the sub-region as the new master edge computing node, thereby realizing adaptive dynamic switching of the nodes and ensuring that the data processing and communication efficiency in the entire monitoring region remains stable and reliable. B

[0047] Based on this, the predicted future load value of the node is obtained as follows:

[0048] Obtain the historical running load sequence of the master edge computing node and the historical bandwidth change sequence , and the historical running load sequence and the historical bandwidth change sequence are processed according to the time stamp to obtain a fused feature sequence.

[0049] In the specific implementation process, the historical running load sequence is composed of the monitoring historical sequence of the calculation resources such as the occupancy rate and the memory usage rate of the master edge computing node in a continuous time window in the past, and can be expressed as: CPU

[0050]

[0051] In the formula, represents the historical load data sequence of the master edge computing node, represents the running load measurement value of the master edge computing node at the th moment, q represents the current moment, t represents the length of the historical observation window, T represents the running load measurement value at the start moment of the historical observation window, and the like represents the running load measurement value at the current moment.

[0052] ​​​​​​The historical bandwidth change rate data is a sequence composed of the bandwidth change rate of the main edge computing node at each moment within the same continuous time window relative to the previous moment, which can be represented as:

[0053] ,

[0054] In the formula, This represents a sequence of historical bandwidth change rate data for the main edge computing nodes. Indicates the node at the 1st i The relative rate of change of bandwidth at each moment. Indicates the start time of the historical observation window The relative rate of change of bandwidth, and so on. For the current moment The relative rate of change of bandwidth.

[0055] A load prediction data encoding model is deployed on a defined edge computing node. This model embeds a multi-head attention mechanism, which includes... Attention units set up in parallel.

[0056] The load prediction data encoding model described in this embodiment adopts a gated cyclic unit ( GRU The network serves as the basic structure to capture the temporal dependencies in historical sequence data, thereby accurately predicting future node operating load trends.

[0057] The multi-head attention mechanism outputs a feature vector which is then processed by a fully connected layer and a linear activation function to obtain the predicted future load value of the node.

[0058] exist GRU Based on the network, this embodiment further embeds a multi-head attention mechanism, which includes several attention units set up in parallel, each attention unit focusing on the... GRU The historical data feature vectors output by the network are independently weighted to enhance the model's differentiated expression of the importance of node load features at different times, thereby improving the accuracy and reliability of node load trend prediction.

[0059] Specifically, the comprehensive feature vector of the multi-head attention mechanism is obtained by concatenating the outputs of each attention unit and then performing a linear mapping. The specific calculation expression is as follows:

[0060] ,

[0061] in, This represents the combined feature vector output by the multi-head attention mechanism. The first attention unit outputs a sub-output feature vector. output a sub-output feature vector for the second attention unit, an output feature vector of the first attention unit, a full connection mapping matrix obtained through training, denotes a concatenation operation.

[0062] The load prediction data encoding model described in the embodiment is trained and optimized using a joint loss function. Further, a joint loss function composed of node running load prediction error and bandwidth prediction error is used as an optimization objective. The specific joint loss function can be represented as:

[0063] ,

[0064] In the formula, represents the value of the joint loss function, represents the number of total sample pairs in the training data set, represents the predicted value of the running load of the i-th sample node by the load prediction data encoding model, represents the predicted value of the bandwidth of the i-th sample node by the model, represents the actual measured value of the running load of the i-th sample node, represents the actual measured value of the bandwidth of the i-th sample node. and respectively represent the relative weight coefficients of the load prediction error and the bandwidth prediction error in the joint loss function. It should be noted that the weight coefficients are set by the technician according to the importance of the load prediction error and the bandwidth prediction error according to the actual scene, and satisfy .

[0065] The predicted future node load value is obtained by using the above technical solution, and the data transmission format is further determined, and the specific process is as follows: The future node load value output by the node load prediction model is compared with the preset load threshold value: if the predicted future node load is greater than or equal to the preset load threshold value, it indicates that the node will enter a high load state, and the node computing resource is insufficient to support additional data compression operation. At this time, the uncompressed

[0066] data format is directly used for data transmission to reduce the computing load and ensure the stable operation of the node.

[0067] JSON

[0068] ​​​​If the predicted future load of the node is less than the preset load threshold, a data coding and transmission format is determined according to a relationship between the model-predicted future bandwidth of the node and a preset bandwidth threshold: if the model-predicted future bandwidth of the node is sufficient and greater than or equal to the preset bandwidth threshold, an uncompressed data format is selected for transmission, so as to avoid unnecessary compression calculation overhead; if the model-predicted future bandwidth of the node is less than the preset bandwidth threshold, a compressed data format with a high compression ratio and relatively low calculation overhead is selected for data coding and transmission, so as to reduce the data transmission burden and improve the bandwidth utilization rate. JSON CBOR

[0069] After the above load prediction and data coding format selection process, the node encodes the monitoring data in the determined data coding format, forms a fused data set, and uploads the data to the cloud.

[0070] In one embodiment, the coupled distributed hydrological model and the hydrodynamic model are a multi-scale coupled model embedded with three lines of defense, and the construction process thereof includes:

[0071] Based on the cloud-fused data set, a distributed hydrological model is constructed; a spatial weight matrix is established based on the relief degree of the sub-basin and the spatial heterogeneity of rainfall, and the model structure of the distributed hydrological model is utilized by using the spatial weight matrix; the spatial weight matrix differentially adjusts the model parameters and weights at different spatial positions by comprehensively considering the terrain of the sub-basin and the distribution characteristics of rainfall, so that the model can more accurately reflect the spatial difference of hydrological response in the actual basin, thereby effectively improving the simulation accuracy of the model.

[0072] A hydrodynamic model is established, and a local water level gradient correction term is introduced in the numerical discrete format, and the hydrodynamic model is corrected by the local water level gradient correction term; it should be noted that the hydrodynamic model used in this embodiment is based on Saint - Venant non-constant flow equation set for simulating the water flow movement process in the river channel; in order to effectively reduce the numerical solution error, this embodiment introduces a local water level gradient correction term for error compensation and correction on the basis of the traditional Saint - Venant equation numerical discrete solution format.

[0073] The flow data output by the distributed hydrological model and the boundary conditions of the hydrodynamic model are associated and docked through an asynchronous data interaction mode, so as to ensure the real-time and coordination of data transmission between the two models, and finally form a multi-scale coupled model embedded with three lines of defense.

[0074] ​​It should be noted that the data interaction mode of coupling the distributed hydrological model and the hydrodynamic model is asynchronous interaction, that is, the two models are independently operated at different time scales, and the data required for model calculation is exchanged through the data interface; Specifically, the outlet flow data of the basin calculated by the distributed hydrological model is input as the boundary condition of the hydrodynamic model, and the time steps between the two are inconsistent;

[0075] It should be understood that the distributed hydrological model generally uses a larger time step for calculation, for example, the model time step is 1 hour, the purpose is to ensure the calculation efficiency of the hydrological process; While the hydrodynamic model needs to use a smaller time step, for example, 10 minutes or less, to more finely simulate the water flow dynamic change process in the river channel; Therefore, there is a significant difference in the data time scale between the two models, and the flow data cannot be directly and correspondingly used, and an asynchronous data exchange mode must be used for processing.

[0076] In order to facilitate the implementation of the above technical scheme, the method for determining the spatial weight matrix described in the embodiment comprises:

[0077] The basin is divided into a plurality of sub-units, and the terrain relief degree and the rainfall spatial variation coefficient of each sub-unit are obtained respectively.

[0078] The geomorphology weight of each sub-unit is determined according to the terrain relief degree and the rainfall spatial variation coefficient And the rainfall weight ; The specific calculation formula is as follows:

[0079] 、 ,

[0080] In the formula, and respectively represent the maximum value and the minimum value of the terrain relief degree of all sub-units, represents the rainfall weight, represents the rainfall spatial variation coefficient of the sub-unit, and represent the minimum value and the maximum value of the rainfall spatial variation coefficient of all sub-units.

[0081] The spatial weight of each sub-unit is obtained by comprehensively considering the geomorphology weight and the rainfall weight, and the spatial weight matrix is constructed by the spatial weight of each sub-unit.

[0082] It should be noted that the spatial weight matrix is composed of spatial weight values determined by the geomorphology weight and the rainfall weight of the sub-unit, and the specific calculation method is as follows:

[0083] ,

[0084] In the formula, The space weight value of the subunit, the weight coefficient and respectively represent the relative importance of the landform feature and the rainfall feature in the space weight, and satisfy The weight coefficient is obtained according to field experience and long-term data analysis.

[0085] The construction of the space weight matrix can be used for optimizing the parameter configuration of the distributed hydrological model, so as to realize the accurate optimization of the model structure in the spatial scale.

[0086] In a specific implementation, the local water level gradient correction term is used for correcting the local water level gradient in the calculation of the Saint-Venant equation, and the calculation manner is as follows:

[0087]

[0088] In the formula, denotes the corrected local water level gradient, denotes the local water level gradient calculated by the model preliminarily, denotes the local water level gradient correction coefficient to be self-corrected; and the correction coefficient is dynamically adjusted so as to effectively reduce the error between the model water level simulation and the actual observation;

[0089] In another embodiment, the preset scheduling target includes, but is not limited to, a cross-section flow control target in a river channel, a river water level over-warning duration control target, a submerged range control target outside the river channel, and a submerged duration control target, and the like.

[0090] The difference data is determined according to the difference between the initial simulation result index and the preset scheduling target, and the calculation formula of the difference data is as follows, including:

[0091] The initial simulation result index and the preset scheduling target are calculated, and the calculation formula is as follows:

[0092]

[0093] In the formula, denotes the difference data, denotes the initial simulation result index, denotes the preset scheduling target.

[0094] The difference data is judged, if is greater than zero, it indicates that the simulation index exceeds the preset target and needs to be optimized and reduced; if is less than zero, it indicates that the simulation index does not reach the preset target and needs to be optimized and increased; if is equal to zero, it indicates that the simulation index reaches the preset target and does not need to be optimized.

[0095] ​​The difference data is determined according to the difference between the initial simulation result and the preset scheduling target, and the difference data is taken as feedback input of the multi-target dynamic weight reinforcement learning algorithm, which is used to guide the reverse optimization adjustment of the control parameter, and then an optimized prospective scheduling scheme is generated.

[0096] Further, the reinforcement learning strategy network is used to optimize the control parameter based on the feedback difference data; in the embodiment, the reinforcement learning strategy network is trained by using a deep deterministic policy gradient algorithm to realize efficient prediction of the control parameter optimization.

[0097] In another implementation, according to the specific control parameter setting in the optimized scheduling scheme, the distributed hydrological model and the hydrodynamic model are re-called to perform joint forward simulation, to obtain the predicted values of various hydrological indexes reflecting the actual effect of the optimized scheduling scheme, and to form an optimized simulation result; however, the simulation model itself has certain parameter errors and simulation errors, so that there is still a certain real-time deviation between the optimized simulation result and the actual preset scheduling target; in the embodiment, a time series error regression analysis model is used to dynamically correct the model parameters, to realize dynamic self-correction of the parameters of the distributed hydrological model and the hydrodynamic model, thereby improving the simulation accuracy of the model and the accuracy of future prediction.

[0098] For example, the coefficient in the local water level gradient is dynamically adjusted to effectively reduce the error between the model water level simulation and the actual observation; the weight coefficients and the weight coefficients are dynamically adjusted to ensure that the spatial weight matrix reflects the current rainfall and topographic feature changes of the basin in real time, thereby improving the spatial prediction accuracy of the hydrological model. The parameter self-correction process described in the embodiment is realized by using a time series error regression analysis model.

[0099] Specifically, the time series error regression analysis model is a long and short term memory network that fuses error accumulation trends, takes the error time series data between the simulation result of the optimized scheduling scheme and the actual scheduling target as input, takes the model parameter correction amount as output to construct a training set, and takes the root mean square error between the error prediction value and the actual error as a loss function to train, and outputs a regression model for correcting the parameters of the distributed hydrological model and the hydrodynamic model.

[0100] In summary, in view of the data interaction and calculation accuracy between the hydrological model and the hydrodynamic model, a multi-scale coupled model with three defense lines is formed by setting the following multiple calculation links, and the three defense lines are embodied in combination with the above-mentioned technologies.

[0101] The first defense line is to use a spatial weight matrix structure based on sub-basin relief and rainfall spatial heterogeneity within the distributed hydrological model to accurately capture the spatial heterogeneity characteristics of the basin hydrological response, and to prevent systematic errors caused by insufficient description of spatial heterogeneity from the model structure.

[0102] The second defense line is to introduce a local water level gradient correction term in the numerical solution process of the hydrodynamic model, and to dynamically correct the local water level gradient through real-time monitoring data feedback, to prevent the propagation and accumulation of local errors in the model from the calculation link.

[0103] The third defense line is to use an asynchronous interaction combined with interpolation processing data fusion mechanism at the data interaction interface of the distributed hydrological model and the hydrodynamic model, to prevent data error amplification caused by time scale mismatch from the data transmission link, and to ensure the real-time and coordination of the data transmission of the two models.

[0104] Embodiment 2

[0105] As shown in Figure 2 The embodiment discloses a digital twin water conservancy super-fusion all-in-one machine coupled with multi-source data, which is used to implement the digital twin water conservancy super-fusion method of embodiment 1, and comprises:

[0106] A data fusion module is configured to establish a water conservancy space-time data chain protocol, realize water conservancy monitoring data formatting processing through hierarchical data modeling, determine an edge computing node based on the formatted water conservancy monitoring data and water data, obtain a predicted node future load value after edge computing processing based on the edge computing node, and select an adaptive data transmission format to generate a cloud fusion data set.

[0107] A coupled hydrological simulation module is configured to construct a coupled distributed hydrological model and a hydrodynamic model based on the cloud fusion data set, generate a forecast period initial scheduling scheme according to rainwater information, and perform forward simulation rehearsal to obtain an initial simulation result.

[0108] A scheduling optimization module is configured to perform reverse optimization adjustment on control parameters of the forecast period initial scheduling scheme based on difference data between the initial simulation result and a preset scheduling target, use the difference data as feedback, and use a multi-objective dynamic weight reinforcement learning algorithm to generate a forecast period optimization scheduling scheme.

[0109] A parameter self-correction module is configured to perform forward simulation rehearsal verification again based on the forecast period optimization scheduling scheme, obtain an optimization simulation result, and use a time series error regression analysis model to dynamically correct model parameters of the coupled distributed hydrological model and the hydrodynamic model according to real-time deviations between the optimization simulation result and the preset scheduling target, to realize dynamic self-correction of the model parameters.

[0110] Part of data in the above formula is calculated by removing dimension, and the formula is obtained by software simulation of a large amount of collected data to be closest to the real situation; the preset parameters and the preset threshold in the formula are set by the person skilled in the art according to the actual situation or obtained by a large amount of data simulation.

[0111] The above embodiments are only used to illustrate the technical method of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. A digital twin water conservancy super-fusion method for coupling multi-source data, characterized in that, The application relates to a water conservancy spatiotemporal data chain protocol, and comprises the following steps. A water conservancy monitoring data formatting process is established by hierarchical data modeling; Based on the formatted water conservancy monitoring data and water data, an edge computing node is determined, a predicted node future load value is obtained based on the edge computing node, and an adaptive data transmission format is selected to generate a cloud fusion data set; Based on the cloud fusion data set, a coupled distributed hydrological model and a hydrodynamic model are constructed, an initial scheduling scheme for a prediction period is generated according to rainwater information, and forward simulation pre-performance is carried out to obtain an initial simulation result; According to the difference between the initial simulation result and a preset scheduling target, the difference is used as feedback, a multi-objective dynamic weight reinforcement learning algorithm is used to reversely optimize and adjust the control parameters of the initial scheduling scheme for the prediction period, and an optimized scheduling scheme for the prediction period is generated; Based on the optimized scheduling scheme for the prediction period, forward simulation pre-performance is carried out again to obtain an optimized simulation result, and according to the real-time deviation between the optimized simulation result and the preset scheduling target, a time sequence error regression analysis model is used to dynamically correct model parameters of the coupled distributed hydrological model and the hydrodynamic model, so that dynamic self-correction of the model parameters is realized; The coupled distributed hydrological model and the hydrodynamic model are multi-scale coupled models embedded with three lines of defense, and the three lines of defense comprise a first line of defense, a second line of defense and a third line of defense. The coupled distributed hydrological model and the hydrodynamic model comprise a distributed hydrological model, the distributed hydrological model has a spatial weight matrix structure based on sub-basin relief and rainfall spatial heterogeneity, and the spatial weight matrix structure forms the first line of defense. The coupled distributed hydrological model and the hydrodynamic model comprise a hydrodynamic model coupled to the distributed hydrological model, a local water level gradient correction term is introduced into the hydrodynamic model, the hydrodynamic model is corrected through the local water level gradient correction term, and the second line of defense is formed. At a data interaction interface of the distributed hydrological model and the hydrodynamic model, an asynchronous interaction combined with interpolation processing data fusion mechanism is used to form the third line of defense.

2. The digital twin water super fusion method of coupling multi-source data according to claim 1, characterized in that, The water conservancy monitoring data formatting process is as follows: A hierarchical modeling of water conservancy monitoring data yields a tuple data structure, which is represented as follows: ,in, Indicates the monitoring entity Unique identifier, As the main monitoring subject The monitoring data values, As the main monitoring subject Location information For collection timestamps.

3. The digital twin water super fusion method of coupling multi-source data according to claim 1, wherein, The determination process of the edge computing node comprises: According to water data, the monitoring area is divided into several sub-areas; the spatial position information of the sensor nodes in each sub-area is obtained, a topological score formula is established, and the topological centrality scores of the sensor nodes are calculated , is the number of the sensor node; wherein, the expression form of the topological score formula is as follows: ， wherein, denotes the total number of sensor nodes within a sub-area, denotes the sensor node with which the data transmission is performed, the number of meshes passed between the sensor nodes denotes the sensor node with which the data transmission is performed, the spatial distance between the sensor nodes and denote the maximum number of meshes and the maximum spatial distance, respectively, between any two sensor nodes, and denote the weight coefficients of the network hop factor and the spatial distance factor, respectively. Within each sub-region, the nodes are ranked according to the topology centrality score determining a primary edge computing node and a plurality of backup nodes; the primary edge computing node and the plurality of backup nodes satisfy the following dynamic settings: A health score mechanism is established, the health score of the main edge computing node is monitored in real time, and when the health score of the main node is less than a preset health threshold, the node with the highest health score is selected from the standby nodes as a new main edge computing node.

4. The digital twin water super fusion method of coupling multi-source data according to claim 1, wherein, The predicted node future load value is obtained as follows: Acquire the historical running load sequence of the main edge computing node and the historical bandwidth change sequence and perform feature splicing processing according to the timestamps to obtain a fusion feature sequence A load prediction data encoding model is deployed at a determined edge computing node, the load prediction data encoding model embedding a multi-headed attention mechanism and a joint loss function, the multi-headed attention mechanism comprising attention units arranged in parallel; The fusion feature sequence is input into a first The attention unit outputs a sub-output feature vector The integrated feature vector is obtained by using a mapping relationship The integrated feature vector is output by the multi-head attention mechanism The predicted node future load value is obtained by processing through a full connection layer and a linear activation function. The expression form of the mapping relationship is as follows: ， In the formula, denotes a splicing operation, is a full connection mapping matrix obtained by training, is a first attention unit output sub-output feature vector, is a second attention unit output sub-output feature vector.

5. The digital twin water super fusion method of coupling multi-source data according to claim 1, characterized in that, The adaptive method of the data transmission format is as follows: The future node load value output by the node load prediction model is compared with a preset load threshold; If the predicted future load of the node is greater than or equal to the preset load threshold, the uncompressed data format is adopted JSON ; If the predicted node future load is less than the preset load threshold, the relationship between the predicted node future bandwidth of the load prediction data coding model and a preset bandwidth threshold is further determined to determine the data coding transmission format; If the predicted node future bandwidth is greater than or equal to a preset bandwidth threshold, an uncompressed data format is selected JSON ​ If the predicted future bandwidth of the node is less than a preset bandwidth threshold, then selecting CBOR compressed format.

6. The digital twin water super fusion method of coupling multi-source data according to claim 1, characterized in that, The determination method of the spatial weight matrix comprises: The sub-basin is divided into a plurality of sub-units, the terrain relief of each sub-unit and a rainfall spatial variation coefficient are obtained, and Determine the geomorphology weight and rainfall weight of each subunit according to the terrain relief degree and the rainfall spatial variation coefficient; Obtain the spatial weight value of each subunit by integrating the geomorphology weight and rainfall weight, and construct a spatial weight matrix with the spatial weight values of the subunits, and the calculation formula of the spatial weight value of the subunit is: ， wherein, represents the spatial weight value of the subunit, represents the topographic weight, represents the rainfall weight, weight coefficient and respectively represent the relative importance of the topographic feature and the rainfall feature in the spatial weight, and satisfy .

7. The digital twin water super fusion method of coupling multi-source data according to claim 1, characterized in that, The correction of the local water level gradient correction term is realized by a parameter self-correction module, the parameter self-correction module adopts a time series error regression analysis model, and dynamically adjusts the local water level gradient correction coefficient k in the local water level gradient correction term according to the real-time deviation between the optimization simulation result and the preset scheduling target.

8. The digital twin water conservancy super-converged appliance coupled with multi-source data, characterized in that, The all-in-one machine is provided with a plurality of modules for realizing the digital twin water conservancy super-fusion method of coupling multi-source data as claimed in any one of claims 1-7, and the modules include: A data fusion module is used to establish a water conservancy space-time data chain protocol, realize water conservancy monitoring data formatting processing through hierarchical data modeling, determine an edge computing node based on the formatted water conservancy monitoring data and water data, obtain a predicted node future load value based on the edge computing node, and select an adaptive data transmission format to generate a cloud fusion data set; A coupled hydrological simulation module is used to construct a coupled distributed hydrological model and a hydrodynamic model based on the cloud fusion data set, generate a forecast period initial scheduling scheme according to rain information and perform forward simulation rehearsal to obtain an initial simulation result; A scheduling optimization module is used to perform reverse optimization adjustment on the control parameters of the forecast period initial scheduling scheme according to the difference data between the initial simulation result and the preset scheduling target, generate a forecast period optimization scheduling scheme, and use the difference data as feedback; A parameter self-correction module is used to perform forward simulation rehearsal verification again based on the forecast period optimization scheduling scheme, obtain an optimization simulation result, and dynamically correct the model parameters of the coupled distributed hydrological model and the hydrodynamic model according to the real-time deviation between the optimization simulation result and the preset scheduling target, so as to realize dynamic self-correction of the model parameters.

Images