Drainage basin hydrological prediction method based on digital twinning

Abstract

The invention discloses a drainage basin hydrological prediction method based on digital twinning, and the method comprises the steps: obtaining multi-source heterogeneous data, carrying out the processing of the multi-source heterogeneous data, and obtaining a multi-source data set; constructing a multi-dimensional spatial feature set according to the multi-source data set; based on the multi-dimensional space feature set, a finite volume method is adopted to simulate a water circulation process, and a water circulation simulation result is obtained; obtaining an error value of the water circulation simulation result and actual observation data, and performing nonlinear relation mining on the multi-dimensional spatial feature set through a neural network algorithm based on the error value to obtain a data driving prediction result; fusing the water circulation simulation result and the data driving prediction result by adopting a weighted fusion algorithm to obtain a fused prediction result; and constructing a digital twinborn model according to the fusion prediction result, and predicting the hydrological state in real time based on the digital twinborn model. The method can provide more effective decision support for water resource management and flood control and disaster reduction.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, and in particular relates to a watershed hydrological prediction method based on digital twins. Background Art

[0002] River basin hydrological forecasting is crucial for water resources management, flood prevention and disaster reduction, and water conservancy project planning. Traditional river basin hydrological forecasting methods are primarily based on physical models or data-driven models, but these methods have limitations. Physical models require a large number of parameters and detailed topographic information, and the model construction and calibration process is complex, making it difficult to accurately reflect the dynamic changes in the river basin. While data-driven models can use historical data for predictions, they lack a deep understanding of physical processes, making it difficult to guarantee accurate predictions when faced with complex hydrological phenomena and extreme events. Summary of the Invention

[0003] To solve the above technical problems, the present invention proposes a watershed hydrological prediction method based on digital twins, which can solve the shortcomings of traditional hydrological prediction methods, improve the accuracy and reliability of watershed hydrological predictions, and provide more effective decision-making support for water resources management and flood prevention and disaster reduction.

[0004] The present invention provides a watershed hydrological prediction method based on digital twins, comprising:

[0005] Acquire multi-source heterogeneous data, process the multi-source heterogeneous data, and acquire a multi-source data set;

[0006] Constructing a multidimensional spatial feature set based on the multi-source data set;

[0007] Based on the multidimensional space feature set, a finite volume method is used to simulate the water cycle process to obtain a water cycle simulation result;

[0008] Obtaining an error value between the water cycle simulation result and actual observation data, and performing nonlinear relationship mining on a multidimensional spatial feature set using a neural network algorithm based on the error value to obtain a data-driven prediction result;

[0009] A weighted fusion algorithm is used to fuse the water cycle simulation results and the data-driven prediction results to obtain a fused prediction result;

[0010] According to the fusion prediction results, a digital twin model is constructed, and based on the digital twin model, the hydrological status is predicted in real time.

[0011] Optionally, processing the multi-source heterogeneous data to obtain a multi-source data set includes:

[0012] Use standardized protocols to convert the format of multi-source heterogeneous data, align the attributes of different data sources according to preset field mapping rules, and obtain multi-source heterogeneous data sets in a unified format;

[0013] The missing data in the multi-source heterogeneous datasets in a unified format are filled using a spatiotemporal interpolation algorithm. If there are gaps in the time or space dimensions of a data point, interpolation is calculated based on the adjacent data points to obtain the filled dataset.

[0014] A data consistency check is performed on the padded data set. If an outlier is detected, the outlier is removed or corrected using a preset threshold judgment rule to obtain the multi-source data set.

[0015] Optionally, constructing a multidimensional spatial feature set based on the multi-source data set includes:

[0016] extracting an initial feature space from the multi-source dataset;

[0017] The principal component analysis algorithm is used to reduce the dimension of the initial feature space to obtain the reduced dimension feature set;

[0018] If the variance contribution rate of the reduced dimension feature set is lower than the preset threshold, the parameters of the principal component analysis are adjusted, and the dimension reduction process is performed again to obtain the optimized feature set;

[0019] The multidimensional space feature set is constructed according to the optimized feature set.

[0020] Optionally, the initial feature space includes: rainfall features, evaporation features and runoff features.

[0021] Optionally, based on the multidimensional space feature set, a finite volume method is used to simulate the water cycle process, and obtaining the water cycle simulation results includes:

[0022] Extract key variables from the reduced-dimensional feature set and build a physical model based on hydrological principles;

[0023] For the physical model, the finite volume method is used to discretize the space and generate the discretized grid;

[0024] Initializing the hydrological variables in the discretized grid according to preset boundary conditions to obtain an initial state;

[0025] By using a time stepping method, the initial state is numerically simulated using explicit time integration to obtain the distribution of hydrological variables at each time step;

[0026] According to the distribution of the hydrological variables, key indicators of the water cycle are calculated to obtain water cycle simulation results.

[0027] Optionally, a weighted fusion algorithm is used to fuse the water cycle simulation results and the data-driven prediction results to obtain the fused prediction results, including:

[0028] Calculate the error distribution of historical data through the preset error analysis model to obtain the error distribution characteristics;

[0029] Using weighted fusion algorithm, the fusion weight is determined based on the error distribution characteristics to obtain the weight distribution scheme;

[0030] The water cycle simulation results and the data-driven prediction results are fused through a weight distribution scheme to obtain a preliminary fused prediction result;

[0031] If the deviation between the initial fusion prediction result and the historical data error distribution exceeds the preset threshold, the weight distribution scheme is adjusted through the iterative optimization algorithm to obtain the optimized fusion prediction result;

[0032] According to the optimized fusion prediction results, the deviation analysis method is used to calculate the residual distribution with historical data to obtain the residual distribution characteristics;

[0033] According to the residual distribution characteristics, the support vector regression algorithm is used to correct the optimized fusion prediction results to obtain the fusion prediction results.

[0034] Optionally, constructing a digital twin model based on the fusion prediction results, and predicting the hydrological status in real time based on the digital twin model includes:

[0035] Extracting features from the fusion prediction results, constructing a digital twin model, and simulating the hydrological state of the basin using the digital twin model;

[0036] Using a real-time stream processing architecture to process the watershed hydrological data obtained through real-time data collection and update the digital twin model;

[0037] If the deviation between the digital twin model output and the real-time hydrological data exceeds a preset threshold, the model parameters are adjusted through the Kalman filter algorithm to obtain an optimized digital twin model;

[0038] Based on the optimized digital twin model, hydrological conditions are predicted in real time.

[0039] Optionally, the basin hydrological prediction method further includes: generating a scheduling decision result based on the predicted hydrological status.

[0040] Optionally, the dispatch decision results generated based on the predicted hydrological conditions include:

[0041] If the hydrological status shows that the rainfall or water level exceeds the preset threshold, the flood risk assessment model is used to calculate and obtain the flood risk assessment results;

[0042] Based on the flood risk assessment results, a decision tree algorithm is used to analyze and generate a flood control scheduling plan;

[0043] Key parameters are extracted from the flood control scheduling plan, and scheduling decision results are generated through logical judgment.

[0044] Compared with the prior art, the present invention has the following advantages and technical effects:

[0045] The present invention first standardizes and interpolates multi-source data to extract key features, and then constructs a physical model based on hydrological principles to simulate the water cycle. When the simulation result has a large error, a neural network is used to mine nonlinear relationships, and the results of the two models are weighted and fused to improve the prediction accuracy. The present invention also constructs a digital twin model to update the hydrological status of the basin in real time and predict future hydrological trends based on time series analysis. When the flood risk is high, a flood control scheduling plan is automatically generated and executed. This method significantly improves the accuracy and real-time performance of hydrological predictions through the fusion of physical models and data-driven models, and the application of digital twin technology, providing powerful decision-making support for water resources management and flood prevention and disaster reduction, and has important practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0047] Figure 1 This is a flow chart of a watershed hydrological prediction method based on digital twins in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0049] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0050] This paper proposes a watershed hydrological prediction method based on digital twins, such as Figure 1 As shown, the specific steps include:

[0051] Acquire multi-source heterogeneous data, process the multi-source heterogeneous data, and obtain multi-source data sets;

[0052] Construct a multidimensional spatial feature set based on multi-source data sets;

[0053] Based on the multidimensional space feature set, the finite volume method is used to simulate the water cycle process and obtain the water cycle simulation results;

[0054] Obtain the error between the water cycle simulation results and the actual observation data, and based on the error, use a neural network algorithm to perform nonlinear relationship mining on the multidimensional spatial feature set to obtain data-driven prediction results;

[0055] A weighted fusion algorithm is used to fuse the water cycle simulation results and data-driven prediction results to obtain fused prediction results;

[0056] According to the fusion prediction results, a digital twin model is constructed, and based on the digital twin model, the hydrological status is predicted in real time.

[0057] Specifically, the present invention first standardizes and interpolates multi-source data to extract key features, and then constructs a physical model based on hydrological principles to simulate the water cycle. When the simulation result error is large, a neural network is used to mine nonlinear relationships, and the results of the two models are weighted and fused to improve the prediction accuracy. The present invention also constructs a digital twin model to update the hydrological status of the basin in real time and predict future hydrological trends based on time series analysis. When the flood risk is high, a flood control scheduling plan is automatically generated and executed. This method significantly improves the accuracy and real-time performance of hydrological predictions through the fusion of physical models and data-driven models, and the application of digital twin technology, providing powerful decision-making support for water resources management and flood prevention and disaster reduction, and has important practical value.

[0058] Furthermore, processing multi-source heterogeneous data to obtain multi-source datasets includes:

[0059] Use standardized protocols to convert the format of multi-source heterogeneous data, align the attributes of different data sources according to preset field mapping rules, and obtain multi-source heterogeneous data sets in a unified format;

[0060] The missing data in the multi-source heterogeneous datasets in a unified format are filled using a spatiotemporal interpolation algorithm. If there are gaps in the time or space dimensions of a data point, interpolation is calculated based on the adjacent data points to obtain the filled dataset.

[0061] A data consistency check is performed on the padded dataset. If outliers are detected, the outliers are removed or corrected using the preset threshold judgment rules to obtain a multi-source dataset.

[0062] Specifically, the process of acquiring multi-source, heterogeneous data within a watershed involves extracting raw data from various sensors and databases. For example, a temperature sensor at a weather station records temperature hourly, a river level monitoring station records water level every 30 minutes, and a soil moisture sensor records humidity data daily. This data is transmitted to a central database via a pre-defined collection interface, such as the MQTT protocol or a RESTful API, forming a multi-source, heterogeneous dataset. During the collection process, it is important to ensure the integrity of the data's timestamp and geolocation information for subsequent processing. In one possible implementation, format conversion is performed using standardized protocols for processing multi-source, heterogeneous datasets. For example, weather data might be stored in CSV format, water level data in JSON format, and soil moisture data in XML format. Using XML Schema or JSON Schema to define a unified data structure, fields such as temperature, water level, and humidity can be mapped to standard fields, such as "temp" to "temperature" and "level" to "water_level." This unified dataset facilitates subsequent analysis and reduces the processing complexity associated with data heterogeneity. Specifically, spatiotemporal interpolation algorithms are used to fill in missing data. For example, if a weather station within a watershed fails to record a temperature at a certain hour, an interpolated temperature can be calculated based on temperature data from neighboring stations using a distance-weighted average method. Assuming the temperatures at neighboring stations A and B are 20°C and 22°C, respectively, and their distances from the target point are 5 km and 10 km, respectively, the weighted interpolated temperature is approximately 20.67°C. Similarly, missing soil moisture data in the spatial dimension can be filled using kriging interpolation. This process improves data integrity and provides a reliable foundation for subsequent analysis. It is important to note that data consistency checks eliminate outliers using preset thresholds. For example, if water level data at a certain moment is recorded as 10 meters, while the historical average is 1 meter, and there are no similar fluctuations at neighboring stations, this is considered an anomaly and can be removed or corrected using neighboring values. Soil moisture values ​​exceeding 100% or negative also require correction. This step ensures the accuracy of the consistent dataset and prevents outliers from interfering with analytical results.

[0063] Furthermore, based on the multi-source data sets, constructing a multidimensional spatial feature set includes:

[0064] Extracting initial feature space from multi-source datasets;

[0065] The principal component analysis algorithm is used to reduce the dimension of the initial feature space to obtain the reduced dimension feature set;

[0066] If the variance contribution rate of the reduced dimension feature set is lower than the preset threshold, the parameters of the principal component analysis are adjusted, and the dimension reduction process is performed again to obtain the optimized feature set;

[0067] Based on the optimized feature set, a multidimensional space feature set is constructed.

[0068] Specifically, when extracting rainfall, evaporation, and runoff characteristics, an initial feature space can be constructed based on statistical indicators. For example, rainfall characteristics include average daily rainfall and the number of days of rainfall; evaporation characteristics include average daily evaporation and evaporation intensity; and runoff characteristics include peak discharge and discharge change rate. Assuming a watershed has an average daily rainfall of 10 mm, evaporation of 3 mm, and a peak runoff of 500 cubic meters per second over a seven-day period, these characteristics constitute the initial feature space, reflecting the basin's hydrological characteristics. It is understandable that principal component analysis (PCA) dimensionality reduction aims to retain key information. For example, if the initial feature space contains 10 features, PCA can extract the first three principal components, assuming their variance contributions are 40%, 30%, and 15%, respectively, for a cumulative contribution of 85%. If the preset threshold is 90%, the number of principal components needs to be adjusted, and a fourth principal component needs to be added to increase its contribution to 92%. This method reduces redundant features and optimizes the data structure. Specifically, the optimized feature set is obtained by adjusting the PCA parameters. For example, after reselecting the four principal components, the feature set includes core variables such as rainfall and peak runoff, and secondary features such as the number of days of rainfall are eliminated. The final feature space is based on these variables, with a more concentrated distribution, reflecting the hydrological response law of the basin. In one embodiment, the cluster analysis can use the K-means algorithm to group the final feature space. For example, based on the optimized feature set, the basin is divided into high rainfall and high runoff modes, low rainfall and low runoff modes, etc. Assuming that the characteristics of a certain period of time in a certain basin are rainfall of 20 mm and runoff of 300 cubic meters per second, the clustering result is classified as a high rainfall and high runoff mode. This classification helps to identify the environmental status of the basin and provide a basis for water resources management.

[0069] Furthermore, the initial feature space includes: rainfall features, evaporation features and runoff features.

[0070] Furthermore, based on the multidimensional space feature set, the finite volume method is used to simulate the water cycle process, and the water cycle simulation results obtained include:

[0071] Extract key variables from the reduced-dimensional feature set and build a physical model based on hydrological principles;

[0072] For the physical model, the finite volume method is used to discretize the space and generate the discretized grid;

[0073] Initialize the hydrological variables in the discretized grid according to the preset boundary conditions to obtain the initial state;

[0074] The initial state is numerically simulated using explicit time integration through the time stepping method to obtain the distribution of hydrological variables at each time step;

[0075] According to the distribution of hydrological variables, the key indicators of the water cycle are calculated to obtain the water cycle simulation results.

[0076] Specifically, starting with a reduced feature set, principal component analysis (PCA) projects high-dimensional variables such as rainfall, evaporation, and runoff into a low-dimensional space through linear transformation, preserving the key information. For example, assuming a watershed's original dataset contains daily rainfall, evaporation, and runoff depth, PCA can reduce these variables to two principal components, explaining 90% of the variance. For example, if a watershed has daily rainfall of 50 mm, evaporation of 5 mm, and runoff depth of 20 mm, the resulting principal components, after PCA, may be linear combinations of rainfall and runoff, reflecting the core trends in water volume fluctuations. This approach simplifies the complexity of subsequent modeling. In one possible implementation, when extracting key variables from the reduced feature set, variables with significant impact on the water cycle can be selected based on hydrological principles. For example, rainfall and soil moisture are often selected as key variables because they directly affect runoff generation. Preferably, variables can be screened through correlation analysis. Assuming the correlation coefficient between rainfall and runoff is 0.85, rainfall is retained first. Based on these variables, the physical model constructed may be a simple hydrological model based on water balance, with model parameters including soil permeability and evaporation coefficient, and a single-layer model structure that inputs rainfall and outputs runoff. This model is easy to parameterize and computationally efficient. It should be noted that when the finite volume method is used for spatial discretization, the watershed is divided into a regular grid. For example, a 100-square-kilometer watershed can be divided into 1-square-kilometer grid cells, each of which records rainfall and runoff values. The discretized grid must meet boundary conditions, such as zero-flow boundaries at the edges of the watershed. For example, assuming that there is no external water input at the watershed boundary, the runoff of the boundary grid is set to 0 during initialization. This initialization ensures that the simulation meets actual hydrological constraints. In one embodiment, time stepping uses explicit time integration to simulate the water cycle. For example, the time step is set to 1 hour, and the current runoff is calculated based on the rainfall and soil moisture content at the previous moment. Assuming that the initial soil moisture content of a grid cell is 30 mm and the rainfall is 10 mm, the model predicts a runoff of 2 mm. This method is simple to calculate and suitable for rapid simulation. The distribution of hydrological variables generated by the simulation can reflect the changes in runoff of each grid unit over time. It is understandable that when calculating key indicators of the water cycle, factors such as total runoff and peak runoff time can be extracted. For example, the simulation results show that the total runoff of a certain basin in 24 hours is 1,000 cubic meters, and the peak occurs in the 6th hour. These indicators intuitively reflect the characteristics of the water cycle. Furthermore, spatial and temporal variation characteristics are extracted from the simulation results to generate dynamic distribution data. For example, the runoff of the grid in the middle of a basin increases over time, while the runoff of the edge grid remains low, reflecting the impact of terrain on water convergence. This dynamic distribution data provides a spatial decision-making basis for water resources management.

[0077] Furthermore, a weighted fusion algorithm is used to fuse the water cycle simulation results and the data-driven prediction results to obtain the following fusion prediction results:

[0078] Calculate the error distribution of historical data through the preset error analysis model to obtain the error distribution characteristics;

[0079] Using weighted fusion algorithm, the fusion weight is determined based on the error distribution characteristics to obtain the weight distribution scheme;

[0080] The water cycle simulation results and data-driven prediction results are fused through a weight distribution scheme to obtain preliminary fusion prediction results.

[0081] If the deviation between the initial fusion prediction result and the historical data error distribution exceeds the preset threshold, the weight distribution scheme is adjusted through the iterative optimization algorithm to obtain the optimized fusion prediction result;

[0082] According to the optimized fusion prediction results, the deviation analysis method is used to calculate the residual distribution with historical data to obtain the residual distribution characteristics;

[0083] According to the residual distribution characteristics, the support vector regression algorithm is used to correct the optimized fusion prediction results to obtain the fusion prediction results.

[0084] Specifically, in water cycle simulations, simulation results and data-driven predictions are typically based on observed data such as rainfall, evaporation, and runoff in a given basin. For example, simulation results for a given basin show an average annual runoff of 500 mm, while the data-driven prediction yields 480 mm. The discrepancy between the two results requires further analysis to determine a fusion strategy. This approach, by comparing data from different sources, identifies the source of error and provides a basis for subsequent fusion. In one possible implementation, the error analysis model can use statistical distribution methods to calculate the error distribution of historical data. For example, based on the past 10 years of runoff observations, the error distribution might exhibit a normal distribution with a mean of 5 mm and a standard deviation of 2 mm. The error distribution characteristics can be visualized using a histogram, reflecting the deviation patterns of the simulation results. This approach helps quantify the error range and provides data support for fusion weight allocation. Specifically, the weighted fusion algorithm determines the weighting scheme based on the error distribution characteristics. For example, if the simulation result has a smaller error standard deviation, it is assigned a higher weight, such as 0.7, while the data-driven prediction result is given a weight of 0.3. The weight distribution scheme can calculate the fusion result through linear weighting, such as the fused runoff value = 0.7×500+0.3×480=494 mm. This method improves the prediction accuracy by balancing the reliability of the two results. Preferably, if the deviation between the preliminary fusion prediction result and the historical data exceeds a threshold (such as 10 mm), an iterative optimization algorithm can be used to adjust the weight. For example, the weight is gradually adjusted by the gradient descent method to make the fusion result closer to the observed value of 490 mm, and finally the optimized fusion prediction result of 492 mm is obtained. This iterative process can dynamically optimize the weight and reduce the deviation. In one embodiment, the deviation analysis method is used to calculate the residual distribution characteristics. For example, the residual between the optimized fusion result 492 mm and the observed value 490 mm is 2 mm, and the residual distribution may show that most of the residuals are concentrated in the range of -5 to 5 mm. This distribution feature provides a basis for subsequent corrections to ensure that the prediction results are closer to reality. For example, the support vector regression algorithm can correct the fusion prediction results based on the residual distribution characteristics. Assuming the residuals are correlated with rainfall and temperature, the algorithm can learn from historical data relationships and predict a corrected residual value, such as adjusting 492 mm to 491 mm. This approach further improves forecast accuracy by capturing nonlinear relationships.

[0085] Furthermore, a digital twin model is constructed based on the fusion prediction results. Based on the digital twin model, the real-time prediction of hydrological status includes:

[0086] Extract features from the fusion prediction results, build a digital twin model, and use the digital twin model to simulate the hydrological status of the basin;

[0087] Adopting a real-time stream processing architecture to process the basin hydrological data obtained from real-time data collection and update the digital twin model;

[0088] If the deviation between the digital twin model output and the real-time hydrological data exceeds a preset threshold, the model parameters are adjusted through the Kalman filter algorithm to obtain an optimized digital twin model;

[0089] Based on the optimized digital twin model, hydrological conditions are predicted in real time.

[0090] Specifically, in one possible implementation, features are extracted from the fused prediction results to construct a digital twin model. The extracted time series features include daily trends in water level and flow, and spatial distribution features reflect hydrological differences at different monitoring points within the basin. A neural network algorithm is used to train the model based on historical hydrological data to simulate the basin's hydrological state. For example, the model outputs a water level of 2.6 meters and a flow of 310 cubic meters per second on a particular day. This approach accurately reflects basin dynamics through feature extraction. Specifically, a real-time stream processing architecture is used to process collected hydrological data. Assuming that sensors within the basin update data hourly, the stream processing architecture uses distributed computing to integrate the data in real time and input it into the digital twin model. For example, if the water level recorded at a certain hour is 2.7 meters, the model updates its state accordingly, ensuring that the model is synchronized with the actual hydrological state. Preferably, if the digital twin model output deviates significantly from the real-time data—for example, if the model predicts a water level of 2.6 meters while the real-time water level is 2.8 meters, and the deviation exceeds a threshold of 0.1 meters—a Kalman filter algorithm is used to adjust the model parameters. Through iterative optimization, the model output water level was adjusted to 2.75 meters. This method enhances the adaptability of the model through dynamic correction.

[0091] Furthermore, the basin hydrological prediction method also includes: generating a scheduling decision result based on the predicted hydrological state.

[0092] Furthermore, the dispatching decision results generated based on the predicted hydrological conditions include:

[0093] If the hydrological status shows that the rainfall or water level exceeds the preset threshold, the flood risk assessment model is used to calculate and obtain the flood risk assessment results;

[0094] Based on the flood risk assessment results, a decision tree algorithm is used to analyze and generate a flood control scheduling plan;

[0095] Extract key parameters from the flood control scheduling plan and generate scheduling decision results through logical judgment.

[0096] Specifically, in one possible implementation, hydrological forecast trends are generated using a time series model. For example, water level changes over the next 24 hours are predicted based on seven days of historical water level and rainfall data. If the forecast indicates that the water level in a particular river section will exceed the 5-meter warning line, or if cumulative rainfall reaches 100 mm, a flood risk assessment is triggered. The flood risk assessment model can combine topographic data and historical flood records to calculate the risk level. For example, a low-lying area with poor drainage may receive a high risk rating from the assessment model, indicating a need for special attention. Specifically, flood control scheduling plans are generated using a decision tree algorithm. For example, given the risk level, reservoir storage, and downstream demand, the decision tree determines whether to release water from the reservoir or adjust the gate opening. If the risk is high and the reservoir water level is nearing the upper limit, a plan of "releasing 500 cubic meters per second" is generated. Scheduling decisions require further refinement, such as using logical reasoning to determine the timing of flood releases and the priority of river dredging. For example, if a river channel is severely silted, the decision tree may prioritize dredging operations while limiting the amount of flood release to prevent excessive pressure downstream.

[0097] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A watershed hydrological prediction method based on digital twins, characterized by: include: Acquire multi-source heterogeneous data, process the multi-source heterogeneous data, and acquire a multi-source data set; Constructing a multidimensional spatial feature set based on the multi-source data set; Based on the multidimensional space feature set, a finite volume method is used to simulate the water cycle process to obtain a water cycle simulation result; Obtaining an error value between the water cycle simulation result and actual observation data, and performing nonlinear relationship mining on a multidimensional spatial feature set using a neural network algorithm based on the error value to obtain a data-driven prediction result; A weighted fusion algorithm is used to fuse the water cycle simulation results and the data-driven prediction results to obtain a fused prediction result; According to the fusion prediction results, a digital twin model is constructed, and based on the digital twin model, the hydrological status is predicted in real time.

2. A watershed hydrological prediction method based on digital twin according to claim 1, characterized in that: Processing the multi-source heterogeneous data to obtain a multi-source data set includes: Use standardized protocols to convert the format of multi-source heterogeneous data, align the attributes of different data sources according to preset field mapping rules, and obtain multi-source heterogeneous data sets in a unified format; The missing data in the multi-source heterogeneous datasets in a unified format are filled using a spatiotemporal interpolation algorithm. If there are gaps in the time or space dimensions of a data point, interpolation is calculated based on the adjacent data points to obtain the filled dataset. A data consistency check is performed on the padded data set. If an outlier is detected, the outlier is removed or corrected using a preset threshold judgment rule to obtain the multi-source data set.

3. A watershed hydrological prediction method based on digital twin according to claim 1, characterized in that: Constructing a multidimensional spatial feature set based on the multi-source data set includes: extracting an initial feature space from the multi-source dataset; The principal component analysis algorithm is used to reduce the dimension of the initial feature space to obtain the reduced dimension feature set; If the variance contribution rate of the reduced dimension feature set is lower than the preset threshold, the parameters of the principal component analysis are adjusted, and the dimension reduction process is performed again to obtain the optimized feature set; The multidimensional space feature set is constructed according to the optimized feature set.

4. A watershed hydrological prediction method based on digital twin according to claim 3, characterized in that: The initial feature space includes: rainfall features, evaporation features and runoff features.

5. The watershed hydrological prediction method based on digital twin according to claim 1, characterized in that: Based on the multidimensional space feature set, the finite volume method is used to simulate the water cycle process, and the water cycle simulation results obtained include: Extract key variables from the reduced-dimensional feature set and build a physical model based on hydrological principles; For the physical model, the finite volume method is used to discretize the space and generate the discretized grid; Initializing the hydrological variables in the discretized grid according to preset boundary conditions to obtain an initial state; By using a time stepping method, the initial state is numerically simulated using explicit time integration to obtain the distribution of hydrological variables at each time step; According to the distribution of the hydrological variables, key indicators of the water cycle are calculated to obtain water cycle simulation results.

6. A watershed hydrological prediction method based on digital twins according to claim 1, characterized in that: The water cycle simulation results and the data-driven prediction results are fused using a weighted fusion algorithm to obtain fused prediction results including: Calculate the error distribution of historical data through the preset error analysis model to obtain the error distribution characteristics; Using weighted fusion algorithm, the fusion weight is determined based on the error distribution characteristics to obtain the weight distribution scheme; The water cycle simulation results and the data-driven prediction results are fused through a weight distribution scheme to obtain a preliminary fused prediction result; If the deviation between the initial fusion prediction result and the historical data error distribution exceeds the preset threshold, the weight distribution scheme is adjusted through the iterative optimization algorithm to obtain the optimized fusion prediction result; According to the optimized fusion prediction results, the deviation analysis method is used to calculate the residual distribution with historical data to obtain the residual distribution characteristics; According to the residual distribution characteristics, the support vector regression algorithm is used to correct the optimized fusion prediction results to obtain the fusion prediction results.

7. The watershed hydrological prediction method based on digital twin according to claim 1, characterized in that: According to the fusion prediction results, a digital twin model is constructed. Based on the digital twin model, real-time prediction of hydrological status includes: Extracting features from the fusion prediction results, constructing a digital twin model, and simulating the hydrological state of the basin using the digital twin model; Using a real-time stream processing architecture to process the watershed hydrological data obtained through real-time data collection and update the digital twin model; If the deviation between the digital twin model output and the real-time hydrological data exceeds a preset threshold, the model parameters are adjusted through the Kalman filter algorithm to obtain an optimized digital twin model; Based on the optimized digital twin model, hydrological conditions are predicted in real time.

8. The method for watershed hydrological prediction based on digital twin according to claim 1, characterized in that: The watershed hydrological prediction method further includes: generating a scheduling decision result according to the predicted hydrological state.

9. A watershed hydrological prediction method based on digital twins according to claim 8, characterized in that: The dispatching decision results generated based on the predicted hydrological state include: If the hydrological status shows that the rainfall or water level exceeds the preset threshold, the flood risk assessment model is used to calculate and obtain the flood risk assessment results; According to the flood risk assessment results, a decision tree algorithm is used for analysis to generate a flood control scheduling plan; key parameters are extracted from the flood control scheduling plan, and a scheduling decision result is generated through logical judgment.

Images