A method and system for evaluating river regime change of high-intensity development river basin

Abstract

The application discloses a kind of high-strength development river basin's river hydrological regime change evaluation method and system, it is related to hydrological regime change evaluation technical field, the method includes: to high-strength development river basin is carried out day flow basic data backtracking, obtains long-sequence day runoff data;Carry out space-time correlation auxiliary data call, obtain auxiliary data set;Obtain transient interference event sequence;Reconstruct natural day runoff sequence;Generate interference natural contrast matrix;Carry out river hydrological regime change quantification, output interference degree quantification evaluation result, wherein the interference degree quantification evaluation result includes FSR comprehensive score and multiple single index degradation degree.The present application solves the technical problems that the prior art has difficulty in capturing daily scale transient interference, the data integrity is insufficient, resulting in low hydrological regime evaluation accuracy and weak ecological characterization ability, achieves accurate capture of basin transient interference, improves the accuracy of hydrological regime evaluation and ecological characterization ability.

Description

Technical Field

[0001] This invention relates to the field of hydrological situation change assessment technology, specifically to a method and system for assessing river hydrological situation changes in a highly developed watershed. Background Technology

[0002] Changes in hydrological conditions have become a key factor affecting the ecological structure and function of rivers. Conducting hydrological change assessments is crucial for river health management. The core of hydrological condition assessment lies in achieving effective characterization of river ecosystem responses through quantitative analysis of hydrological processes, thereby providing a scientific basis for ecological protection decisions. Currently, various hydrological assessment methods, such as IHA and RVA, are commonly used both domestically and internationally. These methods primarily focus on the statistical characteristics of hydrological elements, failing to incorporate the core needs of ecological processes like river reproduction and growth into their evaluation index design. Their ecological significance is unclear, making it difficult to accurately reflect the actual impact of hydrological process changes on key ecosystem processes such as fish spawning migration, benthic survival, and vegetation succession. Their ecological representation capabilities are inherently insufficient. While the traditional River Pressure Index (FSR) method can comprehensively assess the overall degree of change in hydrological conditions from multiple dimensions, its use of monthly runoff data results in low temporal resolution. This easily overlooks key characteristics such as short-duration hydrological abrupt changes and runoff rhythm variations during ecologically sensitive periods like fish breeding and fry fattening. These daily-scale hydrological characteristics are crucial links between hydrological processes and ecological responses. Ultimately, this method tends to be overly optimistic in assessing hydrological disturbances in watersheds with high levels of human activity, failing to objectively quantify the degree of disturbance to the watershed ecosystem caused by hydrological changes, further weakening the ecological representation value of the assessment results.

[0003] Existing technologies struggle to capture daily-scale transient disturbances and suffer from insufficient data integrity, resulting in low accuracy in hydrological situation assessment and weak ecological characterization capabilities. Summary of the Invention

[0004] This application provides a method and system for assessing changes in river hydrological conditions in high-intensity development basins, which addresses the technical problems in existing technologies, such as difficulty in capturing diurnal transient disturbances, insufficient data integrity, resulting in low accuracy of hydrological condition assessment and weak ecological characterization capabilities.

[0005] In view of the above problems, this application provides a method and system for assessing changes in river hydrological conditions in watersheds with high intensity of development.

[0006] The first aspect of this application provides a method for assessing changes in river hydrological conditions in watersheds with high-intensity development, the method comprising: Daily flow baseline data were backtracked for high-intensity development watersheds to obtain long-sequence daily runoff data. Auxiliary datasets were obtained by calling spatiotemporal correlation auxiliary data based on the start and end timestamps of the long-sequence daily runoff data. These auxiliary datasets included time-series changes in water area, hydropower dispatch logs, and watershed meteorological data. After imputation of missing data in the long-sequence daily runoff data using an LSTM neural network, transient disturbance events were captured by aligning the daily runoff data with the auxiliary dataset based on timestamps, resulting in a transient disturbance event sequence. A SWAT hydrological model was used to invert a scenario without human interference, reconstructing the natural daily runoff sequence. The transient disturbance event sequence was projected onto the natural daily runoff sequence based on timestamps for daily-scale ecological process comparison, generating a natural disturbance comparison matrix. Based on the natural disturbance comparison matrix, the changes in river hydrological conditions were quantified, outputting a quantitative assessment result of the disturbance degree, which included a comprehensive FSR score and the degradation degree of multiple individual indicators.

[0007] A second aspect of this application provides a system for assessing changes in river hydrological conditions in watersheds with high-intensity development, the system comprising: The system includes a long-sequence daily runoff data acquisition module, used to backtrack basic daily flow data in high-intensity development watersheds to obtain long-sequence daily runoff data; an auxiliary dataset acquisition module, used to retrieve spatiotemporal correlation auxiliary data based on the start and end timestamps of the long-sequence daily runoff data to obtain an auxiliary dataset, wherein the auxiliary dataset includes time-series changes in water area, hydropower dispatch logs, and watershed meteorological data; and a transient disturbance event sequence acquisition module, used to complete the missing data of the long-sequence daily runoff data based on LSTM neural network, and then complete the daily runoff data and the auxiliary dataset based on timestamp alignment to capture transient disturbance events. The system employs a three-stage approach: a transient disturbance event sequence acquisition module; a natural daily runoff sequence reconstruction module, used to invert the scenario without human disturbance using the SWAT hydrological model and reconstruct the natural daily runoff sequence; a disturbance natural comparison matrix generation module, used to project the transient disturbance event sequence onto the natural daily runoff sequence based on timestamps to compare daily-scale ecological processes and generate a disturbance natural comparison matrix; and an evaluation result output module, used to quantify the changes in river hydrological conditions based on the disturbance natural comparison matrix and output the disturbance degree quantitative evaluation result, wherein the disturbance degree quantitative evaluation result includes an FSR comprehensive score and the degradation degree of multiple individual indicators.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: Daily flow baseline data was backtracked in a high-intensity development watershed to obtain long-sequence daily runoff data. Auxiliary datasets were obtained by calling spatiotemporally correlated auxiliary data based on the start and end timestamps of these long-sequence daily runoff data. After imputation of missing data in the long-sequence daily runoff data using an LSTM neural network, transient disturbance events were captured by aligning the completed daily runoff data with the auxiliary dataset based on timestamps, resulting in a transient disturbance event sequence. A SWAT hydrological model was used to invert a scenario without human interference, reconstructing the natural daily runoff sequence. The transient disturbance event sequence was projected onto the natural daily runoff sequence based on timestamps for daily-scale ecological process comparison, generating a natural disturbance comparison matrix. Based on this natural disturbance comparison matrix, the changes in river hydrological conditions were quantified, outputting a quantitative assessment result of the disturbance degree. This approach achieved the technical effect of accurately capturing transient disturbances in the watershed, improving the accuracy of hydrological condition assessment and ecological characterization capabilities. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of the methodology for assessing changes in river hydrological conditions in high-intensity development watersheds, as described in this application. Figure 2 This is a schematic diagram of the structure of the river hydrological situation change assessment system for high-intensity development watersheds in this application.

[0011] Figure labeling: Long-series daily runoff data acquisition module 10, auxiliary dataset acquisition module 20, transient disturbance event sequence acquisition module 30, natural daily runoff sequence reconstruction module 40, disturbance natural comparison matrix generation module 50, evaluation result output module 60. Detailed Implementation

[0012] This application provides a method and system for assessing changes in river hydrological conditions in high-intensity development basins, which addresses the technical problems in existing technologies, such as difficulty in capturing diurnal transient disturbances, insufficient data integrity, resulting in low accuracy of hydrological condition assessments and weak ecological characterization capabilities.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] Example 1 like Figure 1 As shown, this application provides a method for assessing changes in river hydrological conditions in watersheds with high-intensity development, the method comprising: Step S100: Backtrack the daily flow baseline data of the high-intensity development watershed to obtain long-sequence daily runoff data.

[0015] Specifically, for watersheds with high-intensity development, a systematic review and source collection of full-cycle hydrological monitoring data is carried out. Historical daily flow observation data of each hydrological monitoring station in the watershed are comprehensively retrieved. The data is then processed to unify the format, conduct preliminary screening of outliers, and integrate them continuously over time, ultimately forming a long-sequence daily runoff data that covers a long time span and is continuous in time.

[0016] Step S200: Based on the start and end timestamps of the long-sequence daily runoff data, perform spatiotemporal correlation auxiliary data retrieval to obtain an auxiliary dataset, wherein the auxiliary dataset includes time-series changes in water body area, hydropower dispatch logs, and watershed meteorological data.

[0017] Specifically, the start and end timestamps of long-sequence daily runoff data are extracted in the time dimension. Using these timestamps as the spatiotemporal correlation benchmark, multi-source related data of high-intensity development basins are targeted for retrieval and synchronous access. Data on the time-series changes of water body area that dynamically changes over time within the basin, hydropower scheduling logs related to the operation of dams and gates of hydropower projects within the basin, and basin meteorological data of meteorological elements such as temperature and precipitation within the basin are integrated. After matching and structuring the three types of data in the time dimension, a complete auxiliary dataset is formed.

[0018] Step S300: After completing the missing data of long-sequence daily runoff data based on LSTM neural network, the daily runoff data and the auxiliary dataset are completed based on timestamp alignment to capture transient interference events and obtain a transient interference event sequence.

[0019] Specifically, the missing locations in long-sequence daily runoff data are first located, marked, and normalized. Artificial missing samples are obtained through sliding window truncation and random masking of data points. These samples are then used to train an LSTM neural network with K stacked LSTM hidden layers to obtain a runoff completion model. Next, the data before and after the missing locations are extracted and input into the model to complete the prediction of missing values. After confidence verification, the fidelity-preserving completed values ​​are embedded into the original data to obtain complete completed daily runoff data. Subsequently, the completed daily runoff data is precisely aligned with the auxiliary dataset using timestamps. The start and end timestamps of dam operation events are extracted from the hydropower dispatch logs. Runoff mutation detection is performed on the completed daily runoff data to obtain a runoff mutation candidate event sequence. Natural interference correction is performed by combining watershed meteorological data to obtain an artificial interference event sequence. Then, the spatial authenticity of the artificial interference event sequence is verified by the temporal change of water body area. Finally, a transient interference event sequence containing timestamps, event types, interference intensity, and ecological sensitivity period markers is output.

[0020] Step S400: Use the SWAT hydrological model to invert the scenario without human interference and reconstruct the natural daily runoff sequence.

[0021] Specifically, historical meteorological datasets, soil attribute datasets, and natural land use datasets from the pre-development period are first retrieved as basic input data for the high-intensity development watershed. After inputting the above multi-source basic data into the SWAT hydrological model, the artificial reservoir module and hydropower station module in the model are disabled to simulate the natural hydrological cycle process in the watershed without human development interference and output the initial natural daily runoff sequence. Then, based on the measured daily runoff data of the watershed before development, the hydrological consistency test is carried out on the initial natural daily runoff sequence. According to the test results, the relevant parameters of the SWAT hydrological model are iteratively optimized and adjusted. The model is repeatedly checked and optimized until the runoff sequence output by the model meets the hydrological consistency requirements. Finally, a natural daily runoff sequence that fits the natural hydrological characteristics of the watershed is reconstructed.

[0022] Step S500: Based on the timestamp, project the transient disturbance event sequence to the natural daily runoff sequence to compare the daily-scale ecological processes and generate a natural disturbance comparison matrix.

[0023] Specifically, the process begins by identifying ecologically sensitive periods in the reconstructed natural daily runoff sequence, and then introducing ecological response weighting coefficients to perform sensitive period readjustment, resulting in a weighted natural runoff sequence. Next, transient disturbance event sequences are precisely projected onto this weighted natural runoff sequence based on timestamps, and daily-scale ecological impact mapping is extracted to obtain an event-day ecological impact vector sequence. Subsequently, based on the event-day ecological impact vector sequence and the transient disturbance event sequence, deviation calculations are performed for each transient disturbance event to obtain an event ecological deviation feature set. Finally, the event ecological deviation feature set is matrix-structured and integrated, and dynamically correlated with the FSR index to construct a disturbance natural comparison matrix.

[0024] Step S600: Based on the natural contrast matrix of the disturbance, quantify the changes in river hydrological situation and output the quantitative assessment result of the disturbance degree, wherein the quantitative assessment result of the disturbance degree includes the FSR comprehensive score and the degradation degree of multiple individual indicators.

[0025] Specifically, the process first extracts multiple event subsets with aggregated individual FSR indicators from the interference natural comparison matrix according to the bound FSR indicator fields. Based on each event subset, the degradation degree of multiple individual indicators that can characterize the changes in various dimensions of river hydrology in the basin is calculated. Then, the degradation degree of all individual indicators is standardized to eliminate the dimensional differences between different indicators. Subsequently, the degradation degree of each standardized individual indicator is weighted and calculated according to the preset ecological impact weight. Finally, the comprehensive FSR score is output. The comprehensive FSR score and the degradation degree of multiple individual indicators together constitute the quantitative assessment result of the disturbance degree of river hydrological situation in the high-intensity development basin.

[0026] In one possible implementation, step S300 further includes: Step S310: After locating and marking multiple missing positions to be filled in the long-sequence daily runoff data, perform complete data segment normalization processing on the long-sequence daily runoff data to obtain a normalized runoff sequence.

[0027] Step S320: After extracting multiple consecutive daily runoff subsequences from the normalized runoff sequence using a sliding window, data points are randomly masked to obtain multiple artificially missing samples.

[0028] Step S330: Divide the multiple artificially missing samples into a training set, a test set, and a validation set in chronological order.

[0029] Step S340: Pre-construct an LSTM neural network, wherein the LSTM neural network includes an input layer, K stacked LSTM hidden layers, and an output layer using a linear activation function.

[0030] Step S350: Using mean squared error as the loss function, and the training set, test set, and validation set as training data, the Adam optimizer is used to train the LSTM neural network under the constraint of early stopping mechanism to obtain the runoff completion model.

[0031] Specifically, a point-by-point screening of long-series daily runoff data in high-intensity development basins is performed to accurately identify missing points in the data sequence and mark them one by one. Multiple missing locations to be filled are determined. Then, complete data segments without missing points are extracted from the long-series daily runoff data and subjected to standardized normalization processing to eliminate the influence of differences in data dimensions and numerical magnitudes, resulting in a time-continuous and numerically normalized runoff sequence.

[0032] For the obtained normalized runoff sequence, the sliding window method is used to continuously slide and truncate the sequence according to the preset window length and step size to obtain multiple time-continuous daily runoff subsequences. Then, the data points in each subsequence are randomly masked to simulate runoff data missing scenarios with different missing locations and different missing scales. Each subsequence after random masking is used as artificial missing samples, and finally, multiple sets of artificial missing samples for training the adaptation model are obtained.

[0033] For multiple artificially missing samples obtained through random masking, their temporal order features in the original normalized runoff sequence are preserved. According to a preset division ratio, the samples with earlier temporal order are divided into the training set for model training, the samples with middle temporal order are divided into the test set for model performance testing, and the samples with later temporal order are divided into the validation set for model effect verification. This ensures the temporal correlation and reasonable sample distribution of each dataset, and provides standardized dataset support for the subsequent phased training of the LSTM neural network.

[0034] To address the temporal characteristics and missing data completion requirements of long-series daily runoff data, a pre-built adaptive LSTM neural network model is constructed. This network architecture consists of a sequentially connected input layer, K stacked LSTM hidden layers, and an output layer. The input layer receives normalized runoff time-series data, the K stacked LSTM hidden layers capture long-term temporal dependency features of runoff data through multi-level gating structures, improving the model's fitting and prediction capabilities for time-series data, and the output layer uses a linear activation function to adapt to the prediction requirements of continuous runoff data. This establishes a neural network structure that fits the runoff missing data completion scenario.

[0035] The mean squared error is set as the loss function for model training to quantify the deviation between the model's predicted values ​​and the true values. The predefined training, test, and validation sets are then used as the training data for the entire process and input into a pre-built LSTM neural network for model training. During the training process, the Adam optimizer is used for adaptive optimization and iterative updating of network parameters. At the same time, an early stopping mechanism is introduced to constrain the training process. When the loss value of the validation set no longer decreases for several consecutive rounds, the training is terminated immediately to effectively avoid the model overfitting problem. Finally, the training is completed to obtain a runoff completion model adapted to fill in the missing data of long-sequence daily runoff.

[0036] In one possible implementation, step S300 further includes: Step S360: Extract multiple sets of front and back window data from the long-sequence daily runoff data based on the multiple missing locations to be filled.

[0037] Step S370: Input the multiple sets of front and rear window data into the runoff completion model to predict and complete missing values, and obtain multiple sets of candidate predicted value sequences.

[0038] Step S380: Perform confidence verification on the multiple sets of candidate predicted value sequences to obtain multiple fidelity complete values.

[0039] Step S390: Embed the multiple fidelity completion values ​​into the long-sequence daily runoff data according to the multiple missing positions to be filled, and output the completed daily runoff data.

[0040] Specifically, based on the multiple missing locations that have been located and marked in the long-series daily runoff data, and according to the preset window length and value rules, the two consecutive effective runoff data segments before and after each missing location are extracted in the time dimension. This forms multiple sets of front and back window data that correspond one-to-one with each missing location, fully preserving the runoff time series characteristics around the missing location, and providing targeted local time series data support for the accurate prediction and completion of subsequent missing values.

[0041] The multiple sets of front and back window data extracted for each missing position are normalized and standardized to match the input data specifications of the runoff completion model. Then, the processed sets of front and back window data are input into the runoff completion model that has been trained and has the ability to predict long-term runoff data. The model captures the temporal dependency features of the window data through K-layer stacked LSTM hidden layers, accurately predicts and calculates the runoff value for each missing position, and outputs multiple sets of candidate predicted value sequences corresponding to each missing position.

[0042] For the multiple candidate predicted value sequences output by the runoff completion model, multi-dimensional confidence verification is carried out. By calculating the confidence interval of each predicted value, verifying the temporal correlation and numerical rationality between the predicted values ​​and the surrounding runoff data, and removing abnormal prediction results in the sequence, the reliability of each candidate predicted value is quantitatively judged, and the predicted values ​​with confidence levels meeting the preset threshold are selected. Finally, multiple fidelity completion values ​​are obtained that correspond one-to-one with each missing position to be filled, ensuring the authenticity and accuracy of the completed values.

[0043] Based on multiple pre-located and marked missing locations in the long-series daily runoff data, and according to the one-to-one correspondence between location and value, multiple fidelity completion values ​​obtained after confidence verification are accurately backfilled to the corresponding missing points in the original long-series daily runoff data. This completes the numerical completion of the missing locations in the original data and restores the continuity of the sequence. Finally, it outputs time-continuous and missing completed daily runoff data, providing complete and accurate hydrological basic data support for subsequent transient disturbance event capture.

[0044] In one possible implementation, step S300 further includes: Multiple start and end timestamps of various dam operation events were extracted from the hydropower dispatch logs.

[0045] Based on the multiple start and end timestamps, runoff mutation detection is performed on the completed daily runoff data to obtain a sequence of candidate runoff mutation events.

[0046] The natural interference correction of the candidate runoff mutation event sequence is performed based on meteorological data to obtain the artificial interference event sequence.

[0047] The spatial authenticity of the artificial disturbance event sequence is verified by using the temporal change of the water body area, and the transient disturbance event sequence is output.

[0048] Specifically, the hydropower dispatch logs in the auxiliary dataset are subjected to structured analysis and refined information extraction. All kinds of operation events such as opening and closing, discharge regulation, and water storage scheduling of all dams and gates in the high-intensity development basin are comprehensively sorted out. The specific occurrence time and end time of each dam and gate operation event are determined one by one. The relevant time information is converted and recorded with standardized timestamps, and finally multiple sets of start and end timestamps corresponding to multiple dam and gate operation events are obtained.

[0049] Using the start and end timestamps of multiple dam operation events extracted from hydropower dispatch logs as the core time reference, targeted runoff mutation detection is carried out on the continuous daily runoff data. Focusing on the runoff flow change characteristics within and around each dam operation time interval, hydrological points with significant sudden increases or decreases in runoff flow within this time range are identified. Each mutation point is integrated, marked, and preliminarily recorded according to time sequence, and finally a sequence of candidate runoff mutation events associated with dam operation time is formed.

[0050] By combining natural driving factors such as precipitation and temperature in the basin meteorological data, the natural disturbance attribution and correction processing of the candidate event sequence of runoff mutations is carried out on an event-by-event basis. Runoff mutations directly caused by natural factors such as rainfall and snowmelt are eliminated, while hydrological disturbance events caused only by human activities such as dam and gate operation are retained, thus obtaining the sequence of artificial disturbance events after natural disturbances are eliminated.

[0051] Using the time-series change data of water body area in the auxiliary dataset, spatial authenticity verification is carried out for each event in the artificial interference event sequence. The runoff mutations corresponding to the dam scheduling are checked one by one to see if they are consistent with the spatiotemporal change trend of the water body area in the basin. Interference events with no spatial response or false matching are excluded. After the verification is completed, the data are integrated to form a transient interference event sequence with accurate time and reliable spatiality and output it.

[0052] In one possible implementation, step S300 further includes: Each transient interference event in the transient interference event sequence includes a timestamp, event type, interference intensity, and ecologically sensitive period marker.

[0053] Specifically, the transient disturbance event sequence after spatial authenticity verification is structured and labeled, and each transient disturbance event in the sequence is assigned and recorded with corresponding information: including the precise timestamp of the time when the event occurs, the event type that characterizes the type of dam scheduling behavior, the disturbance intensity that quantifies the magnitude and degree of runoff change, and the ecological sensitivity period marker that indicates whether the event is in an ecologically sensitive period of the watershed, thus forming a complete and standardized transient disturbance event sequence.

[0054] In one possible implementation, step S500 further includes: Step S510: After identifying the ecologically sensitive period of the natural daily runoff sequence, an ecological response weight coefficient is introduced to perform sensitive period readjustment to obtain a weighted natural runoff sequence.

[0055] Step S520: Project the transient disturbance event sequence to the weighted natural runoff sequence based on the timestamp, and perform daily-scale ecological impact mapping extraction to obtain the daily ecological impact vector sequence of the event.

[0056] Step S530: Based on the daily ecological impact vector sequence and transient disturbance event sequence, perform event-by-event deviation calculation to obtain the event ecological deviation feature set.

[0057] Step S540: Use the event ecological deviation feature set to perform matrix structured integration and dynamic correlation of FSR indicators to construct the interference natural comparison matrix.

[0058] Specifically, ecologically sensitive periods are identified in the natural daily runoff sequence based on watershed eco-hydrological goals and biological life cycle characteristics. Key periods sensitive to flow processes, such as fish spawning, reproduction, and migration, are determined. Then, corresponding ecological response weight coefficients are introduced for different sensitivity levels. The runoff data in the natural daily runoff sequence during sensitive periods are weighted and corrected to strengthen the expression of hydrological characteristics of ecologically critical periods. Finally, a weighted natural runoff sequence that highlights the impact of ecologically sensitive periods is obtained.

[0059] Using the precise timestamps corresponding to each event in the transient disturbance event sequence as the matching benchmark, each disturbance event is projected one by one into the weighted natural runoff sequence. At the daily scale, the spatiotemporal matching and ecological impact mapping of the actual disturbed runoff and the natural weighted runoff are completed. Multi-dimensional hydrological and ecological characteristics such as flow, amplitude, and duration on the day the event occurs are extracted, and finally, a daily ecological impact vector sequence containing complete ecological impact information is formed.

[0060] Based on the daily ecological impact vector sequence and transient disturbance event sequence, for each disturbance event, multi-dimensional deviation calculations are generated between the actual runoff and the weighted natural runoff, including flow rate, amplitude, duration, and fluctuation rate. This quantifies the magnitude of changes and ecological impact of dam scheduling on natural hydrological processes, ultimately forming an event ecological deviation feature set containing complete deviation characteristics.

[0061] The deviation values ​​of each event in the ecological deviation feature set are standardized and arranged according to the row and column dimensions to achieve matrix structure integration. Each row represents a transient disturbance event, and each column represents ecological deviation features such as flow, amplitude, duration, and fluctuation rate. Then, the deviation features in each column are dynamically correlated with the FSR hydrological situation evaluation index, and the feature values ​​are converted into index values ​​according to the preset mapping relationship to form a disturbance natural comparison matrix with events as samples, FSR indexes as dimensions, and quantitative values ​​as elements.

[0062] In one possible implementation, step S400 further includes: Step S410: Retrieve multi-source basic data for high-intensity development watersheds before development to obtain historical meteorological datasets, soil attribute datasets, and natural land use datasets.

[0063] Step S420: After inputting the historical meteorological dataset, soil attribute dataset, and natural land use dataset into the SWAT hydrological model, disable the artificial reservoir module and hydropower station module to simulate the hydrological cycle process without human interference and output the initial natural daily runoff sequence.

[0064] Step S430: Based on the measured daily runoff data before development, perform hydrological consistency verification on the initial natural daily runoff sequence, and perform iterative optimization of the parameters of the SWAT hydrological model based on the verification results until the natural daily runoff sequence that meets the hydrological consistency requirements is output.

[0065] Specifically, for watersheds targeted for high-intensity development, multi-source basic data prior to large-scale development of the watershed are retrieved and integrated, including long-sequence historical meteorological data, watershed soil attribute data, and natural land use data that have not been disturbed by large-scale engineering projects. After standardization processing, standardized historical meteorological datasets, soil attribute datasets, and natural land use datasets are formed.

[0066] Historical meteorological datasets, soil attribute datasets, and natural land use datasets were spatially registered and standardized according to the SWAT hydrological model input format before being imported into the model. In the model configuration file, modules related to reservoir storage, water release, and water allocation, as well as modules related to hydropower station power generation and daily scheduling control, were disabled through parameter settings. At the same time, all interference modules related to artificial water conservancy projects and human development activities, such as the opening and closing of river sluice gates and dams within the basin, artificial water lifting for irrigation, industrial and domestic water intake, inter-basin water transfer, water obstruction by dike construction and damming, and the hydrological impact of tidal flat reclamation, were closed. All human interventions on the hydrological cycle were completely eliminated, and only physical processes such as natural rainfall, infiltration, evapotranspiration, slope runoff, and river runoff were retained. The natural evolution and transport process of hydrological elements within the basin were simulated strictly according to the natural underlying surface conditions under undeveloped conditions. The formation and change process of runoff were accurately calculated for each time period, and the initial natural daily runoff sequence was finally output.

[0067] Using pre-development, un-human-interferenced measured daily runoff data as a benchmark, this study first employs statistical methods to conduct a hydrological consistency test on the initial natural daily runoff sequence. Core evaluation indicators such as the Nash efficiency coefficient (NSE), coefficient of determination (R²), and root mean square error (RMSE) are calculated to quantify the fitting deviation between the initial simulated and measured sequences. Subsequently, a closed-loop optimization process of parameter sensitivity analysis, iterative optimization, and accuracy verification is constructed. First, the Morris screening method is used to identify parameters in the SWAT model that significantly affect runoff simulation results, such as soil saturated hydraulic conductivity, crop evapotranspiration coefficient, and channel Manning coefficient. Then, the particle swarm optimization algorithm (PSO) is used for iterative parameter optimization. Using NSE maximization as the objective function, a particle population containing sensitive parameters is initialized. The position and velocity of the particles are updated iteratively. The optimization effect of each particle is evaluated using the fitness function, i.e., the NSE value, and the parameter combination corresponding to the globally optimal particle is retained. The optimized parameters are substituted into the SWAT model and run again to obtain a new natural daily runoff sequence. Hydrological consistency is checked again. If the check index does not reach the preset threshold, such as NSE ≥ 0.75 and R² ≥ 0.8, the above parameter optimization and model operation process is repeated until the consistency index between the simulated output natural daily runoff sequence and the measured data meets the preset accuracy requirements. Finally, a natural daily runoff sequence that meets hydrological consistency is determined.

[0068] In one possible implementation, step S600 further includes: Step S610: Extract multiple event subsets that aggregate multiple individual FSR indicators from the interference natural contrast matrix according to the bound FSR indicator field.

[0069] Step S620: Calculate the degradation degree of the multiple individual indicators based on the multiple sets of event subsets.

[0070] Step S630: After standardizing and transforming the degradation degree of the multiple individual indicators, perform weighted summation based on preset ecological impact weights to output the FSR comprehensive score, wherein the FSR comprehensive score and the degradation degree of the multiple individual indicators constitute the quantitative assessment result of the interference degree.

[0071] Specifically, based on the field names in the natural interference contrast matrix that are pre-established with dynamic associations with FSR indicators, the column vector data of individual FSR indicators such as flow rate, amplitude, duration, and fluctuation rate are located through field indexing and condition filtering. The feature values ​​of the same transient interference event are row-aligned and grouped according to the indicator dimension. After removing null values ​​and outliers, multiple sets of event subsets with well-organized structures are formed, classified according to individual FSR indicators.

[0072] For each subset of events grouped by a single FSR indicator, the relative deviation between the actual runoff characteristic value and the natural runoff characteristic value is calculated. The attenuation, change and deviation of hydrological characteristics are quantified for each event and each indicator in the form of (natural value - actual value) / natural value. Then, the calculation results of all events under the same indicator are statistically aggregated to finally obtain the index degradation degree of multiple single FSR indicators such as flow, amplitude, duration and fluctuation rate.

[0073] The extreme value standardization method is used to perform dimensionless transformation on the degradation degree of multiple individual FSR indicators such as flow, amplitude, duration, and fluctuation rate, and the degradation degree of each indicator is uniformly mapped to the interval [0, 1] to eliminate the dimensional differences between different indicators. Then, combined with the watershed ecological protection goals and hydrological situation response characteristics, the preset ecological impact weights of each individual indicator are retrieved. For example, the flow indicator is given a higher weight during the fish spawning period. Based on the standardized degradation degree of each individual indicator, the weighted summation formula is used to complete the weighted comprehensive calculation of multiple indicators: comprehensive score = ∑(standardized degradation degree × ecological impact weight). Finally, the comprehensive FSR score that can reflect the overall disturbance degree is output. This comprehensive FSR score and the degradation degree of each individual indicator are integrated into a complete quantitative assessment result of the disturbance degree, which comprehensively presents the impact characteristics and overall intensity of human disturbance on the watershed hydrological situation.

[0074] In one possible implementation, step S600 further includes: Multiple river disturbance classification and early warning slices are matched based on the degradation degree of the aforementioned individual indicators.

[0075] Global risk warning based on the FSR comprehensive score.

[0076] The global risk warning and multiple river disturbance graded warning slices are fused to output a multi-level river hydrological situation warning map for river hydrological situation status warning.

[0077] Specifically, based on the degradation degree of multiple individual indicators such as flow rate, amplitude, duration, and fluctuation rate, each indicator is compared and judged against the pre-set thresholds for mild, moderate, severe, and extreme interference. A graded early warning result corresponding to the interference level is generated for each FSR indicator, and the result is stored and labeled in the form of independent data slices, forming multiple graded early warning slices of river interference that reflect the degree of interference in different hydrological dimensions.

[0078] Multiple global risk warning threshold ranges are pre-defined, such as [0, 0.2) for mild risk, [0.2, 0.5) for moderate risk, [0.5, 0.8) for severe risk, and [0.8, 1.0] for extreme risk. The FSR comprehensive score is accurately compared with the preset threshold system through a numerical range matching algorithm to determine the risk level range to which the score belongs. Then, the warning level mapping table is called to complete the one-click matching of the score to the global risk warning level, including the level name, warning symbol, and risk description. Finally, a standardized global risk warning result is output.

[0079] Using a unified spatiotemporal benchmark as a framework, the global risk early warning is taken as the top-level global early warning layer, and multiple river disturbance graded early warning slices such as flow, amplitude, duration, and fluctuation rate are taken as indicator sub-item layers. They are superimposed, merged and graded according to layer weights. Visualization methods such as color mapping, graded color filling and indicator labeling are used to encode and display different indicators and different risk levels, and construct a multi-level early warning map of river hydrological situation that combines overall situation and sub-item characteristics. This enables a layered, graded and whole-area comprehensive early warning of the state of watershed hydrological situation affected by human activities.

[0080] Example 2 Based on the same inventive concept as the method for assessing changes in river hydrological conditions in a high-intensity development watershed described in the foregoing embodiments, such as Figure 2 As shown, this application provides a system for assessing changes in river hydrological conditions in watersheds with high-intensity development. The system and method embodiments in this application are based on the same inventive concept. The system includes: The long-series daily runoff data acquisition module 10 is used to backtrack the basic daily flow data of high-intensity development watersheds to obtain long-series daily runoff data.

[0081] The auxiliary dataset acquisition module 20 is used to retrieve spatiotemporal correlation auxiliary data based on the start and end timestamps of the long-sequence daily runoff data to obtain an auxiliary dataset, wherein the auxiliary dataset includes time-series changes in water body area, hydropower scheduling logs, and watershed meteorological data.

[0082] The transient disturbance event sequence acquisition module 30 is used to capture transient disturbance events by completing the missing data of long-sequence daily runoff data based on LSTM neural network, and then completing the daily runoff data and the auxiliary dataset based on timestamp alignment, thereby obtaining a transient disturbance event sequence.

[0083] The Natural Daily Runoff Sequence Reconstruction Module 40 is used to reconstruct the natural daily runoff sequence by inverting the scenario without human interference using the SWAT hydrological model.

[0084] The interference natural contrast matrix generation module 50 is used to project the transient interference event sequence to the natural daily runoff sequence based on the timestamp to compare the daily-scale ecological processes and generate the interference natural contrast matrix.

[0085] The evaluation result output module 60 is used to quantify the changes in river hydrological conditions based on the natural contrast matrix of the disturbance, and output the quantitative evaluation result of the disturbance degree, wherein the quantitative evaluation result of the disturbance degree includes the FSR comprehensive score and the degradation degree of multiple individual indicators.

[0086] Furthermore, the system is also used to implement the following functions: After locating and marking multiple missing positions in the long-sequence daily runoff data, the long-sequence daily runoff data is normalized to obtain a normalized runoff sequence. A sliding window is used to extract multiple consecutive daily runoff subsequences from the normalized runoff sequence, and data points are randomly masked to obtain multiple artificially missing samples. These artificially missing samples are then divided into training, testing, and validation sets according to time sequence. An LSTM neural network is pre-constructed, comprising an input layer, K stacked LSTM hidden layers, and an output layer using a linear activation function. Using mean squared error as the loss function, and the training, testing, and validation sets as training data, the Adam optimizer is used to train the LSTM neural network with early stopping mechanism constraints, resulting in a runoff completion model.

[0087] Furthermore, the system is also used to implement the following functions: Multiple sets of preceding and following window data are extracted from the long-series daily runoff data based on the multiple missing locations to be filled; the multiple sets of preceding and following window data are input into the runoff completion model to predict and complete missing values, resulting in multiple sets of candidate predicted value sequences; the confidence of the multiple sets of candidate predicted value sequences is verified to obtain multiple fidelity completed values; the multiple fidelity completed values ​​are embedded into the long-series daily runoff data based on the multiple missing locations to be filled, and the completed daily runoff data is output.

[0088] Furthermore, the system is also used to implement the following functions: Multiple start and end timestamps of multiple dam operation events are extracted from the hydropower dispatch log; runoff mutation detection is performed on the completed daily runoff data based on the multiple start and end timestamps to obtain a runoff mutation candidate event sequence; natural interference correction is performed on the runoff mutation candidate event sequence based on meteorological data to obtain an artificial interference event sequence; the spatial authenticity of the artificial interference event sequence is verified by the temporal change of water body area, and the transient interference event sequence is output.

[0089] Furthermore, the system is also used to implement the following functions: Each transient interference event in the transient interference event sequence includes a timestamp, event type, interference intensity, and ecologically sensitive period marker.

[0090] Furthermore, the system is also used to implement the following functions: After identifying the ecologically sensitive periods of the natural daily runoff sequence, an ecological response weight coefficient is introduced for sensitive period readjustment to obtain a weighted natural runoff sequence. Based on the timestamp, the transient disturbance event sequence is projected onto the weighted natural runoff sequence, and a daily-scale ecological impact mapping is performed to extract the event daily ecological impact vector sequence. Based on the event daily ecological impact vector sequence and the transient disturbance event sequence, event-by-event deviation calculation is performed to obtain the event ecological deviation feature set. The event ecological deviation feature set is used for matrix structure integration and dynamic association of FSR indicators to construct the disturbance natural comparison matrix.

[0091] Furthermore, the system is also used to implement the following functions: Before development, multi-source basic data are retrieved for high-intensity development watersheds to obtain historical meteorological datasets, soil attribute datasets, and natural land use datasets. These datasets are then input into the SWAT hydrological model, and the artificial reservoir and hydropower station modules are disabled to simulate a hydrological cycle process without human interference, outputting an initial natural daily runoff sequence. Based on the measured daily runoff data before development, the initial natural daily runoff sequence is subjected to a hydrological consistency check, and the parameters of the SWAT hydrological model are iteratively optimized based on the check results until a natural daily runoff sequence that meets the hydrological consistency requirements is output.

[0092] Furthermore, the system is also used to implement the following functions: Multiple event subsets with multiple individual FSR indicators are extracted and aggregated from the interference natural comparison matrix according to the bound FSR indicator fields; the degradation degree of the multiple individual indicators is calculated based on the multiple event subsets; after standardizing the degradation degree of the multiple individual indicators, a weighted comprehensive score is generated based on the preset ecological impact weight, and the FSR comprehensive score is output. The FSR comprehensive score and the degradation degree of the multiple individual indicators constitute the quantitative assessment result of the interference degree.

[0093] Furthermore, the system is also used to implement the following functions: Based on the degradation degree of the multiple individual indicators, multiple river disturbance graded early warning slices are matched; a global risk early warning is matched with the FSR comprehensive score; the global risk early warning and multiple river disturbance graded early warning slices are risk-fused to output a multi-level early warning map of river hydrological situation for early warning of river hydrological situation status.

[0094] It should be noted that the order of the embodiments described above is for descriptive purposes only and does not represent the superiority or inferiority of the embodiments. Specific embodiments of this specification have been described above. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0095] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0096] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for assessing changes in river hydrological conditions in a high-intensity development watershed, characterized in that, The method includes: By retrospectively analyzing the basic daily flow data of watersheds with high-intensity development, long-sequence daily runoff data can be obtained. Based on the start and end timestamps of the long-sequence daily runoff data, spatiotemporal correlation auxiliary data is retrieved to obtain an auxiliary dataset, which includes time-series changes in water body area, hydropower dispatch logs, and watershed meteorological data. After imputing missing data in long-sequence daily runoff data using LSTM neural network, the daily runoff data and the auxiliary dataset are then aligned based on timestamps to capture transient interference events, thus obtaining a transient interference event sequence. The SWAT hydrological model was used to invert the natural daily runoff sequence under the scenario of no human interference. Based on the timestamp projection of the transient disturbance event sequence to the natural daily runoff sequence, a daily-scale ecological process comparison is performed to generate a natural disturbance comparison matrix. Based on the aforementioned natural contrast matrix of disturbances, the changes in river hydrological conditions are quantified, and the quantitative assessment results of the disturbance degree are output. The quantitative assessment results of the disturbance degree include the FSR comprehensive score and the degradation degree of multiple individual indicators.

2. The method for assessing changes in river hydrological conditions in high-intensity development basins according to claim 1, characterized in that, The method further includes: After locating and marking multiple missing positions to be filled in the long-series daily runoff data, the long-series daily runoff data is normalized to a complete data segment to obtain a normalized runoff sequence. After extracting multiple consecutive daily runoff subsequences from the normalized runoff sequence using a sliding window, data points are randomly masked to obtain multiple artificially missing samples. The multiple artificially missing samples are divided into a training set, a test set, and a validation set according to time sequence; A pre-constructed LSTM neural network is provided, wherein the LSTM neural network includes an input layer, K stacked LSTM hidden layers, and an output layer employing a linear activation function; Using mean squared error as the loss function and the training set, test set, and validation set as training data, the Adam optimizer is used to train the LSTM neural network under the constraint of early stopping mechanism, thus obtaining the runoff completion model.

3. The method for assessing changes in river hydrological conditions in high-intensity development basins according to claim 2, characterized in that, The method for imputing missing data in long-sequence daily runoff data based on LSTM neural networks includes: Multiple sets of front and back window data are extracted from the long-sequence daily runoff data based on the multiple missing locations to be filled. The multiple sets of front and rear window data are input into the runoff completion model to predict and complete missing values, resulting in multiple sets of candidate predicted value sequences. The confidence level of the multiple sets of candidate predicted value sequences is verified to obtain multiple fidelity complete values; The multiple fidelity completion values ​​are embedded into the long-series daily runoff data based on the multiple missing positions to be filled, and the completed daily runoff data is output.

4. The method for assessing changes in river hydrological conditions in high-intensity development basins according to claim 1, characterized in that, Transient disturbance events are captured based on timestamp-aligned and completed daily runoff data and the auxiliary dataset to obtain a sequence of transient disturbance events. The method includes: Extract multiple start and end timestamps from the hydropower dispatch logs for multiple dam operation events; Based on the multiple start and end timestamps, runoff mutation detection is performed on the completed daily runoff data to obtain a sequence of candidate runoff mutation events. The natural interference correction of the candidate runoff mutation event sequence is performed based on meteorological data to obtain the artificial interference event sequence; The spatial authenticity of the artificial disturbance event sequence is verified by using the temporal change of the water body area, and the transient disturbance event sequence is output.

5. The method for assessing changes in river hydrological conditions in high-intensity development basins according to claim 4, characterized in that, Each transient interference event in the transient interference event sequence includes a timestamp, event type, interference intensity, and ecologically sensitive period marker.

6. The method for assessing changes in river hydrological conditions in high-intensity development basins according to claim 5, characterized in that, Based on the projection of the transient disturbance event sequence to the natural diurnal runoff sequence using timestamps, a diurnal-scale ecological process comparison is performed to generate a natural disturbance comparison matrix. The method includes: After identifying the ecologically sensitive periods of the natural daily runoff sequence, an ecological response weight coefficient is introduced for sensitive period readjustment to obtain a weighted natural runoff sequence. Based on the timestamp projection of the transient disturbance event sequence to the weighted natural runoff sequence, daily-scale ecological impact mapping is extracted to obtain the daily ecological impact vector sequence of the event; Based on the daily ecological impact vector sequence and transient disturbance event sequence, an event-by-event deviation calculation is performed to obtain the event ecological deviation feature set; The interference natural comparison matrix is ​​constructed by using the event ecological deviation feature set for matrix structure integration and dynamic correlation of FSR indicators.

7. The method for assessing changes in river hydrological conditions in high-intensity development basins according to claim 1, characterized in that, The natural daily runoff sequence is reconstructed by inverting the SWAT hydrological model under a scenario without human disturbance. The method includes: Before development, multi-source basic data were retrieved from watersheds with high-intensity development to obtain historical meteorological datasets, soil property datasets, and natural land use datasets. After inputting the historical meteorological dataset, soil property dataset, and natural land use dataset into the SWAT hydrological model, the artificial reservoir module and hydropower station module are disabled to simulate the hydrological cycle process without human interference and output the initial natural daily runoff sequence. Based on the measured daily runoff data before development, the initial natural daily runoff sequence is subjected to hydrological consistency verification, and the parameters of the SWAT hydrological model are iteratively optimized based on the verification results until the natural daily runoff sequence that meets the hydrological consistency requirements is output.

8. The method for assessing changes in river hydrological conditions in high-intensity development basins according to claim 1, characterized in that, The method involves quantifying changes in river hydrological conditions based on the aforementioned natural contrast matrix of disturbances, and outputting a quantitative assessment result of the degree of disturbance. Extract multiple event subsets that aggregate multiple individual FSR indicators from the interference natural contrast matrix based on the bound FSR indicator fields; The degradation degree of the multiple individual indicators is calculated based on the multiple sets of event subsets; After standardizing and transforming the degradation degree of the multiple individual indicators, a weighted comprehensive score is generated based on the preset ecological impact weights to output the FSR comprehensive score. The FSR comprehensive score and the degradation degree of the multiple individual indicators constitute the quantitative assessment result of the interference degree.

9. The method for assessing changes in river hydrological conditions in high-intensity development basins according to claim 8, characterized in that, Also includes: Multiple river disturbance classification and early warning slices are matched based on the degradation degree of the aforementioned individual indicators; Global risk warning matching the FSR comprehensive score; The global risk warning and multiple river disturbance graded warning slices are fused to output a multi-level river hydrological situation warning map for river hydrological situation status warning.

10. A system for assessing changes in river hydrological conditions in a high-intensity development basin, characterized in that, The system is used to implement the method for assessing changes in river hydrological conditions in high-intensity development watersheds as described in any one of claims 1-9, and the system comprises: The long-series daily runoff data acquisition module is used to backtrack the basic daily flow data of high-intensity development watersheds to obtain long-series daily runoff data; The auxiliary dataset acquisition module is used to retrieve spatiotemporally correlated auxiliary data based on the start and end timestamps of the long-sequence daily runoff data to obtain an auxiliary dataset, wherein the auxiliary dataset includes time-series changes in water body area, hydropower dispatch logs, and watershed meteorological data. The transient disturbance event sequence acquisition module is used to capture transient disturbance events by completing the missing data of long-sequence daily runoff data based on LSTM neural network, and then completing the daily runoff data based on timestamp alignment and the auxiliary dataset. The Natural Daily Runoff Sequence Reconstruction Module is used to reconstruct the natural daily runoff sequence by inverting the SWAT hydrological model under the scenario of no human disturbance. The interference natural comparison matrix generation module is used to project the transient interference event sequence to the natural daily runoff sequence based on the timestamp to compare daily-scale ecological processes and generate an interference natural comparison matrix. The evaluation result output module is used to quantify the changes in river hydrological conditions based on the natural contrast matrix of the disturbance, and output the quantitative evaluation result of the disturbance degree, wherein the quantitative evaluation result of the disturbance degree includes the FSR comprehensive score and the degradation degree of multiple individual indicators.

Images