Dynamic Estimation Method of Basin Hydrological Model Parameters Based on Digital Twin Technology

Abstract

The present invention discloses a method for dynamic estimation of parameters of a watershed hydrological model based on digital twin technology, which relates to the technical field of watershed management, and includes the following steps: comprehensively modeling the hydrological, meteorological, and geographical characteristics of the watershed through multimodal data, terrain, and remote sensing data within the watershed; partitioning the watershed based on terrain information and meteorological data using a spatial clustering algorithm to find areas prone to flash floods; for each area prone to flash floods, using the collected multimodal data to extract key features that reflect the potential risk of flash floods in the region. The present invention constructs an accurate watershed model by fusing multimodal data and geographical features, uses spatial clustering to identify areas prone to flash floods, and predicts risks through feature engineering and machine learning. Combined with intelligent adjustment of spatiotemporal sensitivity, the system can respond to risk changes in real time, improve prediction accuracy and response speed, provide accurate decision-making support for watershed managers, and effectively prevent and mitigate the impact of flash floods.

Description

Technical Field

[0001] The present invention relates to the technical field of watershed management, and particularly to a method for dynamically estimating parameters of a watershed hydrological model based on digital twin technology. Background Art

[0002] Dynamically estimating parameters of a watershed hydrological model based on digital twin technology means that by collecting multi-source heterogeneous data (such as water level, flow rate, rainfall, temperature, etc.) in the watershed in real time, combined with the real-time feedback and adaptive correction mechanism of the digital twin model, the key parameters of the hydrological model are dynamically estimated and optimized. In this process, first, the multi-modal data fusion technology is used to perform spatio-temporal alignment and dynamic fusion on the heterogeneous data to generate a unified hydrological dynamic data representation; then, a method based on an adaptive correction algorithm is adopted to calculate the difference between the real-time data and the historical data to adjust the model parameters, thereby accurately reflecting the changes in the hydrological state in the watershed. In addition, combined with the Bayesian dynamic network or adversarial learning algorithm, the core parameters of the model are further optimized, and its impact on the future hydrological state is predicted, so as to improve the accuracy and adaptability of the hydrological model in the optimization of watershed water resources scheduling. Finally, the model can achieve automatic dynamic adjustment to ensure efficient response and accurate prediction of complex hydrological conditions.

[0003] The prior art has the following deficiencies:

[0004] In the adaptive partition, the dynamic weight allocation of regions usually depends on hydrological characteristics (such as rainfall intensity, flow rate fluctuation amplitude, etc.). However, some regions (such as flood-prone areas) may, due to terrain complexity or sudden local extreme climate conditions, cause the model to fail to adjust the weight in time. For example, a sudden strong flash flood occurred in a small watershed during a rainfall process, and the model failed to allocate sufficient weight to this small watershed. This will lead to insufficient estimation of the water level and flow rate in this small watershed, resulting in missed warnings and ultimately small-scale flash flood disasters, and even affecting the safety of downstream areas.

[0005] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, and thus it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] The objective of the present invention is to provide a method for dynamically estimating parameters of a watershed hydrological model based on digital twin technology. By integrating multi-modal data and geographical features, an accurate watershed model is established, and spatial clustering technology is used to identify areas prone to mountain floods. Key risk features are extracted through feature engineering technology and prediction is carried out using machine learning. Combining with the intelligent adjustment of spatio-temporal sensitivity, the system can dynamically respond to changes in regional risks according to real-time data and mountain flood outbreak prediction results, greatly improving the accuracy and response speed of mountain flood risk prediction, ensuring that watershed managers can obtain more accurate and dynamic decision-making support in a timely manner, thereby effectively preventing and mitigating the impact of mountain floods and solving the problems in the above background technology.

[0007] To achieve the above objective, the present invention provides the following technical solutions: A method for dynamically estimating parameters of a watershed hydrological model based on digital twin technology, comprising the following steps:

[0008] Comprehensively model the hydrological, meteorological, and geographical features of the watershed through multi-modal data, terrain, and remote sensing data within the watershed, wherein the multi-modal data includes real-time hydrological data and meteorological data;

[0009] Based on terrain information and meteorological data, use a spatial clustering algorithm to partition the watershed and identify areas prone to mountain floods;

[0010] For each area prone to mountain floods, extract key features reflecting the potential risk of mountain flood outbreaks in the area using the collected multi-modal data, deeply analyze and process the extracted features through feature engineering technology, and input the processed features as feature vectors into a machine learning model that has been trained and put into use to predict potential mountain flood outbreaks;

[0011] When there is a potential risk of mountain flood outbreaks in an area prone to mountain floods, combine the real-time outbreak prediction results of the area prone to mountain floods and its characteristics changing over time and space, and intelligently adjust the spatio-temporal sensitivity threshold of the area prone to mountain floods. In the area with potential mountain flood outbreaks, dynamically enhance the spatio-temporal sensitivity to enhance the response ability to risk changes, ensuring that watershed managers can obtain more accurate and timely decision-making support to cope with the potential threats brought by mountain flood outbreaks.

[0012] Preferably, the step of comprehensively modeling the hydrological, meteorological, and geographical features of the watershed through multi-modal data, terrain, and remote sensing data within the watershed includes:

[0013] Collect and integrate data from different sources, including real-time hydrological data, meteorological data, terrain information, and remote sensing image data;

[0014] Preprocess the collected data, including denoising, standardization, and spatio-temporal alignment, to ensure the quality and consistency of the data;

[0015] The river basin is geographically partitioned through a spatial analysis method into several sub-regions with similar hydrometeorological characteristics;

[0016] Statistical methods are used to analyze and extract features from various types of data, constructing a multi-dimensional model that can reflect the hydrological, meteorological, and geographical characteristics within the river basin. Through multi-modal data fusion technology, comprehensive modeling of hydrological, meteorological, and geographical characteristics is achieved.

[0017] Preferably, based on topographic information and meteorological data, the specific steps for partitioning the river basin using a spatial clustering algorithm include:

[0018] Collect and integrate the topographic and meteorological data within the river basin, and convert them into feature vectors suitable for processing by the clustering algorithm;

[0019] Select a clustering algorithm for clustering analysis. The clustering results will reveal different regions within the river basin and identify those regions highly correlated with flash floods, i.e., flash flood-prone areas;

[0020] Display the clustering results through a visualization tool, and combine multi-modal data fusion technology to comprehensively integrate hydrological, meteorological, and geographical characteristics to construct an all-round river basin model.

[0021] Preferably, for each flash flood-prone area, key features reflecting the potential risk of flash floods in the area are extracted using the collected multi-modal data. Among them, the extracted features include the rising and falling amplitudes of the water level and the frequency and amplitude of the flow rate change within the river basin. During the monitoring period, through feature engineering technology, the rising and falling amplitudes of the water level and the frequency and amplitude of the flow rate change within the river basin are deeply analyzed and processed to generate a water level fluctuation reference value and a flow rate change reference value respectively. The water level fluctuation reference value and the flow rate change reference value are used as feature vectors and input into a machine learning model that has been trained and put into use. The flash flood risk index generated by the machine learning model is used to predict potential flash floods.

[0022] Preferably, when predicting potential flash floods using a machine learning model that has been trained and put into use, the flash flood risk index generated is compared and analyzed with a pre-set flash flood risk index reference threshold to predict potential flash floods. The specific steps are as follows:

[0023] If the flash flood risk index is greater than the flash flood risk index reference threshold, a risk signal is generated, indicating that there is a risk of flash floods in this flash flood-prone area; if the flash flood risk index is less than or equal to the flash flood risk index reference threshold, a normal state signal is generated, indicating that the state of this flash flood-prone area is stable and there is no risk of flash floods.

[0024] Preferably, when there is a potential risk of mountain flood in a mountain flood-prone area, the spatio-temporal sensitivity threshold of the mountain flood-prone area is intelligently adjusted in combination with the real-time outbreak prediction results of the mountain flood-prone area and its characteristics changing with time and space. The specific steps are as follows:

[0025] First, calculate the spatio-temporal sensitivity threshold of the mountain flood-prone area. The spatio-temporal sensitivity threshold is dynamically adjusted according to the mountain flood risk index and the reference threshold of the mountain flood risk index. The calculation formula of the spatio-temporal sensitivity threshold is as follows: , where is the spatio-temporal sensitivity threshold, is the mountain flood risk index, indicating the mountain flood outbreak risk of the current basin, is the reference threshold of the mountain flood risk index, represents the time change rate of the mountain flood risk index, that is, the change speed of the mountain flood risk index with time, reflecting the change trend of the mountain flood outbreak risk, and are preset weighting coefficients, respectively representing the contribution weights of the current mountain flood risk index and the reference threshold of the mountain flood risk index to the spatio-temporal sensitivity threshold, represents the influence factor of time change, weighing the influence degree of time change in the basin on risk response;

[0026] When the spatio-temporal sensitivity threshold is adjusted, re-evaluate the mountain flood outbreak risk of the mountain flood-prone area, and determine the risk level according to the new spatio-temporal sensitivity threshold. When there is a potential risk of mountain flood in the mountain flood-prone area, intelligently adjust the spatio-temporal sensitivity threshold of the mountain flood-prone area. The dynamic adjustment expression of the new spatio-temporal sensitivity threshold is: , where: is the spatio-temporal response adjustment factor, used to control the dynamic amplification multiple of the spatio-temporal sensitivity threshold when the mountain flood outbreak risk changes, is the adjusted spatio-temporal sensitivity threshold, which is dynamically increased based on the influence of the deviation between the mountain flood risk index and the reference threshold of the mountain flood risk index, enhancing the response ability to the mountain flood outbreak risk.

[0027] Preferably, for each mountain flood-prone area, during the monitoring period, the specific steps of generating the water level fluctuation reference value by deeply analyzing and processing the rising and falling amplitudes of the water level in the basin through feature engineering technology are as follows:

[0028] For each time window , introduce the water level change amplification factor to identify the severity of the water level change in each time window. The calculation expression of the water level change amplification factor is: , where: represents the water level change amplification factor, used to quantify the fluctuation intensity of the water level within the time window, is the jThe water level value corresponding to a timestamp, is a parameter that controls the sensitivity of the water level fluctuation amplitude and controls the weighting degree of the change, is the time window The number of data points within, is the water level value at the previous time point;

[0029] After calculating the increase amplitude of the water level fluctuation within each time window, a water level fluctuation reference value is further generated through local outlier measure. The local outlier measure is used to measure the degree of deviation of the water level change within each time window from the overall trend. The local outlier measure is defined as: where: is the predicted water level value obtained using the locally weighted regression model, is a parameter that controls the sensitivity of the outlier measure;

[0030] By combining and weighting the outlier measure of each time window with the water level change increase factor, the final water level fluctuation reference value is obtained. The calculation expression of the water level fluctuation reference value is: where: is the water level fluctuation reference value, is the weight factor, which is used to weight the fluctuation conditions of each time window, i represents the i th time window during the monitoring period, that is, the basin water level data is divided into multiple time windows during the monitoring period, m represents the total number of time windows divided during the entire monitoring period, that is, the entire monitoring time period is divided into m time windows.

[0031] Preferably, for each mountain flood prone area, during the monitoring period, the specific steps for generating the flow change reference value by deeply analyzing and processing the frequency and amplitude of the flow change in the basin through feature engineering technology are as follows:

[0032] Perform high-order moment analysis on the flow data to reveal the complex patterns and abnormal fluctuations of the flow change. Use the skewness and kurtosis in the high-order moments to capture the asymmetry and extreme fluctuation characteristics of the flow distribution. The specific calculation formulas for skewness and kurtosis are as follows: where: is the flow at the k moment, is the mean value of the flow, is the standard deviation of the flow, n is the total number of data samples;

[0033] The complexity of flow fluctuations is evaluated by the fractal dimension, which is used to describe the self-similarity and complexity of the flow curve and is a method for measuring the details of flow fluctuations. The fractal dimension of the flow curve is calculated by the following formula, and the calculation expression is: where: represents the fractal dimension, represents the covering number of the flow curve at scale , that is, the level of detail of the flow change, is the scale parameter;

[0034] Combined with the analysis results of skewness , kurtosis and fractal dimension , the flow change reference value is calculated, and the calculation expression is: where: represents the flow change reference value, which comprehensively measures the change characteristics of the flow by combining skewness, kurtosis and fractal dimension.

[0035] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0036] The present invention establishes an accurate watershed model by integrating multi-modal data and geographical features, and uses spatial clustering technology to identify areas prone to mountain floods. Key risk features are extracted through feature engineering technology and prediction is carried out using machine learning. Combined with the intelligent adjustment of spatio-temporal sensitivity, the system can dynamically respond to changes in regional risks according to real-time data and mountain flood outbreak prediction results, greatly improving the accuracy and response speed of mountain flood risk prediction, ensuring that watershed managers can obtain more accurate and dynamic decision-making support in a timely manner, and thus effectively preventing and reducing the impact of mountain flood disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0038] Figure 1 is the method flow chart of the method for dynamically estimating the parameters of the watershed hydrological model based on the digital twin technology of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art.

[0040] The present invention provides a method for dynamically estimating parameters of a watershed hydrological model based on digital twin technology as Figure 1 shown, comprising the following steps:

[0041] Comprehensively model the hydrological, meteorological, and geographical characteristics of the watershed through multimodal data (such as rainfall, water level, flow rate, air temperature, soil humidity, etc.) within the watershed and topographic and remote sensing data, where the multimodal data includes real-time hydrological data and meteorological data;

[0042] The step of comprehensively modeling the hydrological, meteorological, and geographical characteristics of the watershed through multimodal data (such as rainfall, water level, flow rate, air temperature, soil humidity, etc.) within the watershed and topographic and remote sensing data includes:

[0043] First, collect and integrate data from different sources, including real-time hydrological data (such as rainfall, flow rate, water level, etc.), meteorological data (such as temperature, humidity, wind speed, etc.), and topographic information (such as slope, elevation, watershed morphology, etc.) and remote sensing image data (such as land use type, vegetation coverage, etc.).

[0044] Next, preprocess this data, including denoising, standardization, and spatio-temporal alignment, to ensure the quality and consistency of the data.

[0045] Then, geographically partition the watershed through spatial analysis methods (such as GIS technology), and divide the watershed into several sub-regions with similar hydrometeorological characteristics.

[0046] On this basis, use statistical methods to analyze and extract features from various types of data, construct a multi-dimensional model that can reflect the hydrological, meteorological, and geographical characteristics within the watershed, and achieve comprehensive modeling of hydrological, meteorological, and geographical characteristics through multimodal data fusion technology.

[0047] This modeling process provides a scientific basis for subsequent risk assessment, disaster prediction, and emergency response.

[0048] Based on topographic information (such as slope, watershed morphology, etc.) and meteorological data (such as precipitation, temperature change, etc.), use spatial clustering algorithms (such as K-means, DBSCAN) to partition the watershed and identify areas prone to mountain floods;

[0049] Based on topographic information (such as slope, basin morphology, etc.) and meteorological data (such as precipitation, temperature changes, etc.), the specific steps for partitioning a basin using spatial clustering algorithms (such as K-means, DBSCAN) include:

[0050] First, collect and integrate the topographic and meteorological data within the basin and convert it into feature vectors suitable for processing by the clustering algorithm. For example, use data such as slope, elevation, basin morphology, precipitation, temperature, etc. as feature inputs.

[0051] Then, select an appropriate clustering algorithm (such as K-means, DBSCAN, etc.) for clustering analysis. The K-means algorithm divides the basin into K sub-regions through iteration, with similar features in each sub-region; while DBSCAN identifies areas prone to flash floods through density and can automatically detect high-risk areas with intensive precipitation and special topographic features. The clustering results will reveal different regions within the basin and specifically identify those regions highly related to flash floods, that is, areas prone to flash floods.

[0052] Finally, display the clustering results through visualization tools (such as GIS software), combine multi-modal data fusion technology, comprehensively integrate hydrological, meteorological, and geographical features, and construct an all-round basin model to provide accurate data support for subsequent prediction and decision-making.

[0053] This process classifies regions with similar geographical and meteorological features within the basin through spatial clustering and accurately identifies areas prone to flash floods, providing support for disaster warning and resource scheduling.

[0054] For each area prone to flash floods, use the collected multi-modal data to extract key features reflecting the potential risk of flash floods in the area. Through feature engineering techniques, deeply analyze and process the extracted features, and use the processed features as feature vectors to input into a machine learning model that has been trained and put into use to predict potential flash floods;

[0055] For each area prone to flash floods, use the collected multi-modal data to extract key features reflecting the potential risk of flash floods in the area. Among them, the extracted features include the rising and falling amplitudes of the water level and the frequency and amplitude of changes in the flow within the basin. During the monitoring period, through feature engineering techniques, deeply analyze and process the rising and falling amplitudes of the water level and the frequency and amplitude of changes in the flow within the basin, and generate a water level fluctuation reference value and a flow change reference value respectively. Use the water level fluctuation reference value and the flow change reference value as feature vectors to input into a machine learning model that has been trained and put into use, and predict potential flash floods through the flash flood risk index generated by the machine learning model.

[0056] A trained machine learning model is the result of a process where machine learning algorithms learn from and optimize historical data. During this process, the model automatically identifies the non-linear or complex relationships between input features (such as water level fluctuations, flow rate changes, etc.) and flash floods from the data. The core objective of the training process is to find the underlying patterns in the data and gradually optimize the model's parameters through techniques such as backpropagation (in neural networks) and gradient descent (in regression models), making its predictions for input features more accurate.

[0057] Taking a neural network as an example, during the training process, the original hydrological data (including historical water level data, flow rate data, rainfall, etc.) is first input into the network. Through repeated forward and backward propagations, the network continuously adjusts its internal weights and biases until the model can accurately capture the complex relationships between features such as water level fluctuations and flow rate changes and flash floods. After training is completed, the model is tested using a validation set (i.e., data not used for training) to evaluate the accuracy and robustness of the model in practical applications, ensuring that it can effectively handle different hydrological and meteorological conditions.

[0058] The trained model can not only generate prediction results based on historical data but also respond quickly to the input real-time data, predicting the likelihood of flash floods. This enables the model to be used as a real-time warning system in practical applications. At this stage, the model is no longer just a theoretical tool but has been verified through practice and has the ability to continuously make effective predictions in a complex and dynamic environment. The trained model continuously optimizes and adjusts through the use of the backpropagation algorithm and different evaluation metrics (such as accuracy, recall, F1-score, etc.), ultimately achieving accurate prediction of flash flood risks.

[0059] A machine learning model put into use means that the model has completed training and has been applied in a real environment, capable of receiving new data in real-time and generating corresponding prediction results. In the prediction of flash floods, this stage is crucial because it ensures that the machine learning model can process real-time data in actual scenarios and provide practical and effective support for flash flood monitoring and warning.

[0060] For a machine learning model for flash flood risk prediction, being put into use means seamless integration in data flow, model calculation, and decision support systems. The machine learning model monitors multi-modal data such as water level, flow rate, and rainfall in real-time, inputs this data into the trained model, and generates prediction results regarding flash flood risks. For example, based on the input reference values of water level fluctuations and flow rate changes, the model calculates the "flash flood risk index" for the area using the patterns learned during previous training and provides warning information to basin managers based on this.

[0061] This process involves a "prediction - feedback - optimization" loop. The model predicts the mountain flood risk index of potential mountain floods, alerting relevant management departments to take preventive measures. Meanwhile, in practical applications, basin managers will provide feedback on the model's prediction results, and this feedback data will continuously supplement the training dataset, thereby further optimizing the model. With the continuous input of real - time data, the model will gradually improve the accuracy of its predictions. This "feedback - based learning" not only keeps the model efficient but also can cope with the uncertainties brought about by changes in climate, terrain, etc., ensuring the long - term effectiveness of the model.

[0062] In practical use, the implementation of the machine - learning model also means that it can be integrated into an automated early - warning system and work in coordination with other monitoring devices (such as meteorological radars, remote - sensing satellites, etc.). For example, when the water level in a mountain - flood - prone area exceeds a certain threshold, the model will automatically trigger an early - warning and send risk warning messages to relevant personnel via text messages, emails, APP push, etc. This makes the early warning and emergency response to mountain floods more rapid and efficient.

[0063] In addition, the implementation of the machine - learning model also emphasizes its adaptability to real - time data. The model can adapt to various changes under different climate and terrain conditions and generate accurate predictions. This ability makes the machine - learning model an indispensable tool in the prevention and mitigation of mountain floods. It is not just a static analysis tool but a dynamic and continuously optimized decision - making support system.

[0064] The rising and falling amplitudes of the water level in the basin are important indicators for measuring the potential risk of mountain floods. Generally, when the water - level change amplitude is large, it means that the precipitation suddenly increases or the water in the basin rapidly converges, resulting in a rapid rise in the water level. Such drastic changes often occur under extreme meteorological conditions, especially short - term heavy rain or concentrated precipitation events. In this case, the drainage capacity of the basin may be insufficient, and mountain floods are extremely likely to occur. In addition, a sharp drop in the water level may also reflect the rapid loss of water flow after a mountain flood, and the drastic change in the hydrological conditions in the basin indicates that the water resources in the basin have fluctuated greatly in a short period, which may trigger mountain floods. Therefore, areas with a large water - level fluctuation amplitude generally mean a higher risk of mountain floods, while areas with a smaller water - level fluctuation amplitude may mean a lower risk of mountain floods. By monitoring the trend of water - level changes, the potential risk of mountain floods can be effectively predicted and evaluated.

[0065] For each mountain - flood - prone area, during the monitoring period, the specific steps for generating the water - level fluctuation reference value through in - depth analysis and processing of the rising and falling amplitudes of the water level in the basin by feature - engineering techniques are as follows:

[0066] First, the water level fluctuation sequence needs to be segmented according to the time stamp, and the trend of water level change within each time period is calculated. Specifically, by calculating the local slope, the change characteristics of the water level within each short time window are analyzed. For each time window (i.e., the set of water level data of the selected basin within the time range), a water level change amplification factor is introduced to identify the severity of water level change within each small time window. The formula is expressed as:

[0067] , where: represents the water level change amplification factor, is the water level value corresponding to the j th time stamp, is a parameter that controls the sensitivity of the water level fluctuation amplitude, controlling the weighting degree of the change, is the time window the number of data points within, is the water level value of the previous time point. The water level change amplification factor is used to quantify the fluctuation intensity of the water level within a short time window;

[0068] The role of this step is to highlight local severe fluctuations through an adaptive amplification factor and retain the non-linear fluctuation characteristics of the data.

[0069] After calculating the water level fluctuation amplification within each time window, a water level fluctuation reference value is further generated through local anomaly measurement. The local anomaly measurement is used to measure the degree of deviation of the water level change within each time window from the overall trend, especially sudden events of rapid rise or fall. The local anomaly measurement is defined as:

[0070] , where: is the predicted water level value obtained using a local weighted regression model (such as Gaussian weighted regression), is a parameter that controls the sensitivity of the anomaly measurement;

[0071] Through this formula, the anomaly degree of water level fluctuation can be quantified, considering whether the change of water level in a certain time window significantly deviates from the overall trend. Finally, the water level fluctuation reference value is obtained by combining and weighting the anomaly measurement of each time window with the water level change amplification factor. The expression of the water level fluctuation reference value is: , where: is the water level fluctuation reference value, is the weight factor, used to weight the fluctuation conditions of each time window, i represents the iA time window means that the basin water level data is divided into multiple time windows during the monitoring period. m Indicates the total number of time windows divided during the entire monitoring period, that is, the entire monitoring time period is divided into m time windows;

[0072] The function of this step is to comprehensively evaluate the intensity of water level fluctuations and the degree of anomalies within different time windows, so as to obtain an overall water level fluctuation reference value with early warning capabilities. The water level fluctuation reference value can accurately reflect the violent fluctuations of the water level in the basin and is further used to evaluate the potential risk of flash floods.

[0073] From the water level fluctuation reference value, it can be seen that for each flash flood prone area, during the monitoring period, the larger the performance value of the water level fluctuation reference value generated by deeply analyzing and processing the rising and falling amplitudes of the water level in the basin through feature engineering techniques, the greater the potential risk of flash floods in the flash flood prone area. The water level fluctuation reference value is obtained by deeply analyzing the rising and falling amplitudes of the water level in the basin, and it comprehensively reflects the degree of rapid changes in the water level in a short period of time. When the water level rises or falls rapidly, it usually means a sharp increase in precipitation or a rapid accumulation of water bodies in the basin. Such violent water level fluctuations are omens of flash floods. In flash flood prone areas, an increase in the water level fluctuation reference value usually indicates that the hydrological conditions in the basin have changed rapidly, possibly exceeding the drainage capacity of the basin, resulting in an increased risk of flash floods. Therefore, the larger the performance value of the water level fluctuation reference value, the more violent the water level fluctuations, and the higher the likelihood of flash floods; conversely, if the water level fluctuation reference value is small, it indicates that the water level changes relatively smoothly, and the potential risk of flash floods is low.

[0074] For each flash flood prone area, a greater frequency and amplitude of flow changes in the basin usually indicate a higher risk of flash floods in that area. The reasons for frequent and large-amplitude flow changes are often related to sudden heavy precipitation, mountain terrain features, or other factors (such as soil saturation, river channel blockage, etc.). In flash flood prone areas, when the precipitation is large, the water flow is likely to quickly flow into streams or river channels, resulting in a sharp increase in flow. This rapid flow change reflects the response speed and intensity of the basin to precipitation. Frequent and violent flow fluctuations usually indicate that the drainage capacity of the area is poor, and the accumulation and rapid flow of water bodies may exceed the capacity of natural or artificial drainage facilities, thus creating conditions for flash floods. Therefore, the frequency and amplitude of flow changes are an important indicator of flash floods in flash flood prone areas.

[0075] For each flash flood prone area, during the monitoring period, the specific steps for generating a flow variation reference value by deeply analyzing and processing the frequency and amplitude of flow changes in the basin through feature engineering techniques are as follows:

[0076] Perform high - order moment analysis on the flow data to reveal the complex patterns and abnormal fluctuations of the flow changes. High - order moment analysis focuses on the asymmetry and kurtosis (i.e., the sharpness of the fluctuations) of the flow fluctuations. In this step, skewness and kurtosis are used to capture the asymmetry and extreme fluctuation characteristics of the flow distribution. The specific formulas are as follows: , where: is the flow at the k th moment, is the mean of the flow, is the standard deviation of the flow, n is the total number of data samples;

[0077] Skewness is used to quantify the degree of deviation of the flow changes (i.e., whether the changes tend to increase or decrease), and kurtosis is used to quantify the sharpness of the flow fluctuations (i.e., whether there are violent fluctuations or extreme changes). High skewness and high kurtosis values often indicate violent fluctuations in the flow, predicting the potential risk of flash floods;

[0078] This step is mainly used to capture the asymmetry and abnormal fluctuation information of the flow fluctuations, providing the basic non - linear characteristics for the calculation of the flow change index in the next step. This helps to reveal the complexity of the flow changes, especially in the context of sharp changes or abnormal fluctuations.

[0079] Evaluate the complexity of the flow fluctuations through the fractal dimension. The fractal dimension is used to describe the self - similarity and complexity of the flow curve and is a method to measure the details of the flow fluctuations. The fractal dimension of the flow curve can be calculated through the following formula, and the calculation expression is: , where: represents the fractal dimension, represents the covering number of the flow curve at the scale , that is, the level of details of the flow changes, is the scale parameter. Specifically, is the number of segments of the flow time series at different scales. As gradually decreases, the change of reflects the complexity of the flow fluctuations. If the fractal dimension is high, it indicates that the flow fluctuations have high complexity, which may indicate a strong hydrological response and potential flash flood risk;

[0080] This step can reveal the non - linear complexity of the flow fluctuations through the calculation of the fractal dimension, further enhancing the accurate description of the flow changes. A high fractal dimension means that the flow changes are more complex and irregular, usually closely related to the risk of sudden flash floods.

[0081] Combining skewness , kurtosis and the fractal dimension Based on the analysis results, calculate the reference value of flow rate variation. The calculation formula is as follows: where: represents the reference value of flow rate variation, which comprehensively measures the variation characteristics of flow rate by combining three important features: skewness, kurtosis, and fractal dimension. Skewness and kurtosis together reflect the asymmetry and extreme fluctuations of flow rate fluctuations, and the fractal dimension quantifies the complexity of flow rate fluctuations. Finally, the reference value of flow rate variation can quantify the degree and complexity of flow rate fluctuations, thereby providing an effective prediction basis for the potential risk of flash floods;

[0082] The reference value of flow rate variation can effectively aggregate the multi-dimensional flow rate fluctuation characteristics into a digital risk assessment index. By quantifying the asymmetry, extreme fluctuations, and complexity of the flow rate, it provides accurate quantitative data support for flash flood risk prediction.

[0083] It can be seen from the reference value of flow rate variation that for each flash flood prone area, during the monitoring period, the larger the value of the reference value of flow rate variation generated by deeply analyzing and processing the frequency and amplitude of flow rate changes in the basin through feature engineering techniques, the more intense and complex the flow rate fluctuations in the basin are, usually accompanied by a larger hydrological response, which is particularly important in flash flood prone areas. Frequent changes and large amplitude fluctuations of the flow rate usually indicate a higher probability of extreme precipitation or other extreme climate conditions occurring in the area in a short period of time, thereby increasing the risk of flash floods. The reference value of flow rate variation extracted through feature engineering techniques can quantify the intensity and complexity of these fluctuations. If the reference value of flow rate variation is high, it indicates that the hydrological characteristics of this flash flood prone area show relatively irregular and intense fluctuations, which are usually caused by the superposition effect of factors such as sudden precipitation, upstream water source injection, and terrain characteristics. These factors greatly increase the potential risk of flash floods. On the contrary, if the reference value of flow rate variation is low, it indicates that the hydrological fluctuations in the basin are relatively stable, the impact of extreme weather events on this area is small, and the potential flash flood risk is low. Therefore, the reference value of flow rate variation can not only be used as an effective quantitative standard, but also provide a scientific basis for predicting the risk of flash floods.

[0084] The machine learning model is not specifically limited here. Any machine learning model that can perform comprehensive analysis on the reference value of water level fluctuation and the reference value of flow rate variation to generate the flash flood risk index is acceptable. To implement the technical solution of the present invention, the present invention provides a specific implementation method; the calculation formula for generating the flash flood risk index is as follows: , where and are respectively the preset proportionality coefficients of the water level fluctuation reference value and the flow rate change reference value , and and are both greater than 0. The preset proportionality coefficients ( and ) refer to the coefficients for weighting the flash flood risk by the water level fluctuation reference value ( ) and the flow rate change reference value ( ) when generating the flash flood risk index. The role of these preset proportionality coefficients is to set the contribution ratio of different indicators to the final risk assessment according to the actual hydrological model or historical data. Specifically, is the influence weight of the water level fluctuation reference value ( ) on the flash flood risk, is the influence weight of the flow rate change reference value ( ) on the flash flood risk. In practical applications, these preset proportionality coefficients help the model dynamically adjust the influence degree on risk prediction according to the changes in water level and flow rate, ensuring that the output of the model (i.e., the flash flood risk index) can accurately reflect the relative contributions of different hydrological characteristics to the flash flood risk. Therefore, and are parameters obtained by fitting through historical data or domain experience by the model, and usually need to be tuned to ensure the prediction accuracy of the model.

[0085] It can be seen from the flash flood risk index that for each flash flood prone area, during the monitoring period, the larger the performance value of the water level fluctuation reference value generated by deeply analyzing and processing the rising and falling amplitudes of the water level in the basin through feature engineering techniques, and the larger the performance value of the flow rate change reference value generated by deeply analyzing and processing the frequency and amplitude of the flow rate change in the basin through feature engineering techniques, that is, the larger the performance value of the flash flood risk index generated when the machine learning model that has been trained and put into use predicts potential flash floods, the greater the risk of flash floods in this flash flood prone area, and vice versa, the smaller the risk of flash floods in this flash flood prone area.

[0086] Compare and analyze the flash flood risk index generated when the machine learning model that has been trained and put into use predicts potential flash floods with the preset flash flood risk index reference threshold to predict potential flash floods. The specific steps are as follows:

[0087] If the flash flood risk index is greater than the flash flood risk index reference threshold, a risk signal is generated, indicating that there is a risk of flash floods in the flash flood-prone area; if the flash flood risk index is less than or equal to the flash flood risk index reference threshold, a normal signal is generated, indicating that the state of the flash flood-prone area is relatively stable and there is no risk of flash floods.

[0088] When there is a potential flash flood risk in a flash flood prone area, the temporal and spatial sensitivity threshold of the flash flood prone area is intelligently adjusted based on the real-time flash flood forecast results and its characteristics of temporal and spatial changes. In the potential flash flood area, the temporal and spatial sensitivity is dynamically improved to enhance the ability to respond to risk changes, ensuring that watershed managers can obtain more accurate and timely decision support to deal with the potential threats brought by flash floods.

[0089] When there is a potential flash flood risk in a flash flood prone area, the temporal and spatial sensitivity thresholds of the flash flood prone area are intelligently adjusted based on the real-time flash flood forecast results and its characteristics of temporal and spatial changes. The specific steps are as follows:

[0090] First, the temporal and spatial sensitivity thresholds of flash flood prone areas are calculated. The temporal and spatial sensitivity thresholds are dynamically adjusted according to the flash flood risk index and the flash flood risk index reference threshold. The calculation formula of the temporal and spatial sensitivity thresholds is as follows: ,in, is the spatiotemporal sensitivity threshold, is the flash flood risk index, which indicates the flash flood risk in the current basin. It is the reference threshold of flash flood risk index, which is usually set through historical data, expert judgment or simulation analysis. It indicates the time change rate of the flash flood risk index, that is, the speed at which the flash flood risk index changes over time, reflecting the changing trend of flash flood risk, especially in the event of sudden weather or hydrological events. and are preset weighting coefficients, which respectively represent the contribution weights of the current flash flood risk index and the flash flood risk index reference threshold to the temporal and spatial sensitivity threshold. The impact factor representing temporal variation weighs the degree of influence of temporal variation on risk response within the basin;

[0091] The purpose of this step is to dynamically adjust the spatiotemporal sensitivity threshold according to the gap between the real-time flash flood risk index and the flash flood risk index reference threshold, as well as the speed at which the flash flood risk index changes over time. Dynamically increasing spatiotemporal sensitivity means that when the flash flood risk index approaches or exceeds the flash flood risk index reference threshold, the spatiotemporal sensitivity threshold will be increased to enhance the ability to respond to changes. This adjustment ensures that the system responds quickly when the risk changes dramatically.

[0092] After the spatio-temporal sensitivity threshold is adjusted, the flash flood risk in the flash flood prone area is re-evaluated, and the risk level is determined according to the new spatio-temporal sensitivity threshold. When there is a potential flash flood risk in the flash flood prone area (i.e., when the flash flood risk index is greater than the flash flood risk index reference threshold), the spatio-temporal sensitivity threshold of the flash flood prone area is intelligently adjusted. The dynamic adjustment expression of the new spatio-temporal sensitivity threshold is: , where: is the spatio-temporal response adjustment factor, which is used to control the dynamic amplification factor of the spatio-temporal sensitivity threshold when the flash flood risk changes, is the adjusted spatio-temporal sensitivity threshold, which is dynamically increased based on the influence of the deviation between the flash flood risk index and the flash flood risk index reference threshold, enhancing the response ability to flash flood risks.

[0093] By real-time monitoring and predicting the risk changes in the flash flood prone area, the spatio-temporal sensitivity threshold of this area is intelligently adjusted, thereby enhancing the system's response ability to the potential threat of flash floods. When the risk changes drastically, the system will dynamically increase the spatio-temporal sensitivity, making the early warning of flash floods more accurate and timely. Through this dynamic adjustment, it can be ensured that when facing sudden meteorological changes or hydrological anomalies, the basin managers can quickly obtain reliable risk assessments and take disaster prevention and mitigation measures in a timely manner, thereby minimizing the threat of flash floods to life and property and improving the efficiency and effectiveness of emergency responses. This process improves the flexibility of the early warning system, enabling it to effectively respond to flash floods of different risk levels and ensuring that prevention and control measures are effectively implemented at the most critical moments.

[0094] Through the dynamic estimation method of basin hydrological model parameters based on digital twin technology, the problems in the prior art that the model fails to adjust weights in time and misses the early warning of flash floods can be effectively solved. This method establishes an accurate basin model by integrating multi-modal data (such as rainfall, water level, flow rate, etc.) and geographical features, and uses spatial clustering technology to identify flash flood prone areas. By extracting key risk features through feature engineering technology and using machine learning for prediction, combined with the intelligent adjustment of spatio-temporal sensitivity, the system can dynamically respond to the changes in regional risks according to real-time data and flash flood prediction results. This solution greatly improves the accuracy and response speed of flash flood risk prediction, ensuring that basin managers can obtain more accurate and dynamic decision-making support in a timely manner, thereby effectively preventing and mitigating the impact of flash floods.

[0095] The above formulas are all dimensionless and take their numerical calculations. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0096] As described above, it is only the specific implementation manner of the present application. However, the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims described above.

[0097] Only some exemplary embodiments of the present invention have been described above by way of illustration. Undoubtedly, for those of ordinary skill in the art, the described embodiments can be modified in various different ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the protection scope of the claims of the present invention.

Claims

1. A method for dynamic estimation of basin hydrological model parameters based on digital twin technology, characterized in that: The following steps are involved: Comprehensively model the hydrological, meteorological and geographical characteristics of the basin through multimodal data, topographic and remote sensing data in the basin, where multimodal data includes real-time hydrological and meteorological data; Based on terrain information and meteorological data, spatial clustering algorithms are used to partition the watershed and identify areas prone to flash floods. For each flash flood prone area, the key features reflecting the potential risk of flash floods in the area are extracted using the collected multimodal data. The extracted features are deeply analyzed and processed through feature engineering technology, and the processed features are input as feature vectors into the trained and put into use machine learning model to predict potential flash floods. When there is a potential flash flood risk in a flash flood prone area, the temporal and spatial sensitivity threshold of the flash flood prone area is intelligently adjusted based on the real-time flash flood forecast results and its characteristics of temporal and spatial changes. In the potential flash flood area, the temporal and spatial sensitivity is dynamically improved to enhance the ability to respond to risk changes, ensuring that watershed managers can obtain more accurate and timely decision support to deal with the potential threats brought by flash floods. When there is a potential flash flood risk in a flash flood prone area, the temporal and spatial sensitivity thresholds of the flash flood prone area are intelligently adjusted based on the real-time flash flood forecast results and its characteristics of temporal and spatial changes. The specific steps are as follows: First, the temporal and spatial sensitivity thresholds of flash flood prone areas are calculated. The temporal and spatial sensitivity thresholds are dynamically adjusted according to the flash flood risk index and the flash flood risk index reference threshold. The calculation formula of the temporal and spatial sensitivity thresholds is as follows: ,in, is the spatiotemporal sensitivity threshold, is the flash flood risk index, which indicates the flash flood risk in the current basin. is the reference threshold of flash flood risk index, It represents the time change rate of the flash flood risk index, that is, the speed at which the flash flood risk index changes over time, reflecting the changing trend of the flash flood risk. and are preset weighting coefficients, which respectively represent the contribution weights of the current flash flood risk index and the flash flood risk index reference threshold to the temporal and spatial sensitivity threshold. The impact factor of temporal variation is used to weigh the impact of temporal variation on risk response within the basin; When the spatiotemporal sensitivity threshold is adjusted, the flash flood risk of the flash flood prone area is re-evaluated, and the risk level is determined according to the new spatiotemporal sensitivity threshold. When there is a potential flash flood risk in the flash flood prone area, the spatiotemporal sensitivity threshold of the flash flood prone area is intelligently adjusted. The dynamic adjustment expression of the new spatiotemporal sensitivity threshold is: ,in: is the spatiotemporal response adjustment factor, which is used to control the dynamic amplification of the spatiotemporal sensitivity threshold when the flash flood risk changes. The adjusted spatiotemporal sensitivity threshold is dynamically improved based on the impact of the deviation between the flash flood risk index and the flash flood risk index reference threshold, thereby enhancing the response capacity to flash flood risks.

2. The method for dynamic estimation of watershed hydrological model parameters based on digital twin technology according to claim 1 is characterized in that: The steps to comprehensively model the hydrological, meteorological, and geographical characteristics of a watershed using multimodal data, topographic, and remote sensing data include: Collect and integrate data from different sources, including real-time hydrological data, meteorological data, terrain information, and remote sensing image data; Preprocess the collected data, including denoising, standardization, and spatiotemporal alignment, to ensure data quality and consistency; The basin is geographically partitioned by spatial analysis methods, and the basin is divided into several sub-areas with similar hydrological and meteorological characteristics; Statistical methods are used to analyze and extract features from various types of data, and a multidimensional model that can reflect the hydrological, meteorological and geographical characteristics of the basin is constructed. Through multimodal data fusion technology, comprehensive modeling of hydrological, meteorological and geographical characteristics is achieved.

3. The method for dynamic estimation of watershed hydrological model parameters based on digital twin technology according to claim 1 is characterized in that: Based on terrain information and meteorological data, the specific steps of using spatial clustering algorithm to partition the watershed include: Collect and integrate topographic and meteorological data in the basin and convert them into feature vectors suitable for clustering algorithm processing; Select clustering algorithm for cluster analysis. The clustering results will reveal different areas in the basin and identify those areas that are highly correlated with flash floods, i.e. flash flood prone areas. The clustering results are displayed through visualization tools, and combined with multimodal data fusion technology, hydrological, meteorological and geographical characteristics are comprehensively integrated to build a comprehensive watershed model.

4. The method for dynamic estimation of watershed hydrological model parameters based on digital twin technology according to claim 1 is characterized in that: For each area prone to flash floods, the collected multimodal data are used to extract key features that reflect the potential risk of regional flash floods, among which the extracted features include the rise and fall of water levels and the frequency and magnitude of flow changes in the basin. During the monitoring period, the rise and fall of water levels and the frequency and magnitude of flow changes in the basin are deeply analyzed and processed through feature engineering technology to generate water level fluctuation reference values ​​and flow change reference values, respectively. The water level fluctuation reference values ​​and flow change reference values ​​are input as feature vectors into the trained and put into use machine learning model, and the flash flood risk index generated by the machine learning model is used to predict potential flash floods.

5. The method for dynamic estimation of watershed hydrological model parameters based on digital twin technology according to claim 4 is characterized in that: The flash flood risk index generated by the trained and put-in-use machine learning model when predicting potential flash floods is compared and analyzed with the pre-set flash flood risk index reference threshold to predict potential flash floods. The specific steps are as follows: If the flash flood risk index is greater than the flash flood risk index reference threshold, a risk signal is generated, indicating that there is a risk of flash floods in the flash flood-prone area; if the flash flood risk index is less than or equal to the flash flood risk index reference threshold, a normal signal is generated, indicating that the state of the flash flood-prone area is stable and there is no risk of flash floods.

6. The method for dynamic estimation of watershed hydrological model parameters based on digital twin technology according to claim 4 is characterized in that: For each flash flood prone area, during the monitoring period, the specific steps of deeply analyzing and processing the rise and fall of water levels in the basin through feature engineering technology to generate water level fluctuation reference values ​​are as follows: For each time window , the water level change amplification factor is introduced to identify the severity of water level changes in each time window. The calculation expression of the water level change amplification factor is: ,in: Represents the water level change amplification factor, which is used to quantify the intensity of water level fluctuations within the time window. It is j The water level value corresponding to the timestamp, is a parameter that controls the sensitivity of water level fluctuations and the degree of weighting of the control changes. It is the time window The number of data points in is the water level value at the previous time point; After completing the calculation of the water level fluctuation increase in each time window, the water level fluctuation reference value is further generated through local anomaly measurement. It is used to measure the deviation of water level change in each time window from the overall trend and local anomaly measurement. Defined as: ,in: is the predicted water level value obtained using the local weighted regression model, is a parameter that controls the sensitivity of the anomaly measurement; The final water level fluctuation reference value is obtained by combining the abnormal measurement of each time window with the water level change increase factor for weighted calculation. The calculation expression of the water level fluctuation reference value is: ,in: is the reference value of water level fluctuation, is the weight factor, which is used to weight the fluctuations in each time window. i Indicates the monitoring period i time windows, that is, the basin water level data is divided into multiple time windows during the monitoring period. m Indicates the total number of time windows divided during the entire monitoring period, that is, the entire monitoring period is divided into m time window.

7. The method for dynamic estimation of watershed hydrological model parameters based on digital twin technology according to claim 4 is characterized in that: For each flash flood prone area, during the monitoring period, the frequency and magnitude of flow changes in the basin are deeply analyzed and processed by feature engineering technology to generate flow change reference values. The specific steps are as follows: Perform high-order moment analysis on flow data to reveal complex patterns and abnormal fluctuations in flow changes, using skewness in high-order moments and Kurtosis To capture the asymmetry and extreme volatility of traffic distribution, skewness and Kurtosis The specific calculation formula is as follows: ,in: For the k The flow rate at the moment, is the mean flow rate, is the standard deviation of the flow rate, n is the total number of samples of the data; The complexity of traffic fluctuations is evaluated by fractal dimension. Fractal dimension is used to describe the self-similarity and complexity of traffic curves. It is a method to measure the details of traffic fluctuations. The fractal dimension of the traffic curve is calculated by the following formula: ,in: represents the fractal dimension, Indicated in scale The coverage number of the lower flow curve, that is, the level of detail of the flow change, is the scale parameter; Combined skewness , Kurtosis And fractal dimension The analysis results are used to calculate the flow change reference value. The calculation expression is: ,in: It represents the reference value of flow change, and comprehensively measures the change characteristics of flow by combining skewness, kurtosis and fractal dimension.

Images