Artificial intelligence-based flood forecasting and early warning method

Abstract

The invention relates to the technical field of flood early warning, and discloses a flood forecasting and early warning method based on artificial intelligence, and the method comprises the steps: correcting deviation data segments in multi-source hydrological original data of a target region, and obtaining homologous consistent data; performing feature fusion according to the data internal mapping relation to obtain fused feature data; integrating trend information evolved along with time, periodic features and association rules in the fused feature data into time sequence feature data; evaluating the similarity and difference between the real-time and historical critical hydrological features, and generating a dynamic early warning threshold value; according to the core influence factor, the time sequence evolution characteristic and the real-time change rule of the time sequence characteristic data, cooperatively deducing the hydrological evolution trend to obtain a preliminary flood forecasting result; based on a dynamic early warning threshold value, integrating the preliminary flood forecasting result meeting the early warning triggering condition and the regional features in the homologous consistent data into flood early warning information; the flood forecasting and early warning efficiency can be improved.

Description

Technical Field

[0001] This invention relates to the field of flood early warning technology, and in particular to a flood forecasting and early warning method based on artificial intelligence. Background Technology

[0002] In the field of flood forecasting and early warning, existing technologies have significant shortcomings in processing multi-source hydrological raw data. They are unable to effectively filter out noise components in the data and cannot unify the storage format and description standards of data from different sources, resulting in prominent data deviation problems. It is difficult to form a consistent data foundation, and subsequent feature analysis and trend inference based on these data lack reliable support, which directly affects the accuracy of flood forecast results and fails to accurately reflect the actual hydrological conditions of the target area.

[0003] Existing technologies have shortcomings in setting early warning thresholds and predicting hydrological evolution trends. Early warning thresholds are mostly fixed values ​​and are not dynamically adjusted based on the differences between real-time hydrological characteristics and historical critical characteristics, making it difficult to adapt to real-time changes in hydrological conditions in the target area. Furthermore, when predicting hydrological evolution trends, core influencing factors, temporal evolution characteristics, and real-time change patterns are not fully integrated, resulting in significant deviations between the predicted results and actual hydrological evolution. The generation of early warning information lacks scientific rigor and cannot provide timely and accurate effective guidance for flood control. Therefore, improving the efficiency of flood forecasting and early warning has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides an artificial intelligence-based flood forecasting and early warning method to solve the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides a flood forecasting and early warning method based on artificial intelligence, comprising: S1: Correct the biased data segments of the target area in the multi-source hydrological raw data to obtain consistent data of the target area from the same source; S2: Based on the inherent mapping relationship between different data in the same source consistent data, feature fusion is performed on the same source consistent data to obtain the fused feature data of the same source consistent data; S3: Integrate the trend information, periodic features and correlation patterns that evolve over time in the fused feature data into the time-series feature data of the target region; S4: Based on the real-time change trend and evolution rate in the time series feature data, evaluate the similarity and difference between the real-time hydrological features in the time series feature data and the historical critical hydrological features, so as to generate a dynamic early warning threshold adapted to the target area. S5: Based on the core influencing factors, time-series evolution characteristics, and real-time change patterns of the time-series characteristic data, the hydrological evolution trend of the target area is synergistically deduced to obtain the preliminary flood forecast results for the target area; S6: Based on the dynamic early warning threshold, the preliminary flood forecast results that meet the early warning triggering conditions are integrated with the regional features in the consistent data from the same source to form the flood early warning information for the target area.

[0006] In a preferred embodiment, the deviation data segments of the target area in the multi-source hydrological raw data are corrected to obtain consistent data of the target area, including: Collect multi-source raw hydrological data related to flood forecasting in the target area; Noise filtering is performed on the multi-source hydrological raw data to obtain clean multi-source hydrological data for the target area. By unifying the storage format and description specifications of the multi-source hydrological clean data, standardized hydrological data for the target area can be obtained. By correcting the biased data segments in the standardized hydrological data, consistent data from the same source for the target area are obtained.

[0007] In a preferred embodiment, the step of performing feature fusion on the consistent data based on the inherent mapping relationship between different data in the consistent data to obtain fused feature data of the consistent data includes: Define the attribute descriptions and application scenarios corresponding to different data in the same source data to obtain a list of data attributes of the same source data; The data attribute list is mapped to preset element association rules to obtain a preliminary mapping relationship set of the consistent data from the same source. By removing false associations with no actual hydrological significance from the preliminary mapping relationship set, the effective intrinsic mapping relationship set of the consistent data from the same source is obtained; Based on the effective intrinsic mapping relationship set, the data that are related in the same source and consistent data are classified by features to obtain the related feature group of the same source and consistent data. The core attribute information in the associated feature group is integrated with the basic attribute data of the target region to form the fused feature data of the same source and consistent data.

[0008] In a preferred embodiment, the step of integrating the core attribute information in the associated feature group with the basic attribute data of the target region to form the fused feature data of the consistent data includes: The hydrological representation dimension of the core attribute information and the scenario adaptation dimension of the basic attribute data are decomposed, and the functional positioning of the core attribute information and the basic attribute data in the description of hydrological evolution is determined, generating a dimension function correspondence table of the homogeneous and consistent data. Based on the dimension function correspondence table, a bidirectional association mapping between the core attribute information and the basic attribute data is established to obtain the attribute association mapping set of the consistent data from the same source. Eliminate the differences in description perspective and expression standard of different attributes of the associated pairs in the attribute association mapping set to obtain the logically unified attribute association set of the consistent data from the same source; By mining the hidden hydrological synergistic patterns in the logically unified attribute association set, the association information that can jointly reflect the hydrological evolution characteristics of the target area is strengthened, and the synergistic strengthening attribute set of the consistent data from the same source is obtained. The core collaborative features of the collaborative enhancement attribute set are extracted and integrated into the fusion feature data of the homogeneous and consistent data.

[0009] In a preferred embodiment, integrating the trend information, periodic features, and correlation patterns that evolve over time from the fused feature data into time-series feature data for the target region includes: By sorting out the temporal correlation clues of the fused feature data, clarifying the time record benchmarks corresponding to different types of features, and generating the temporal dimension identifier set of the fused feature data; Based on the time dimension identifier set, the feature content that shows a continuous change over time in the fused feature data is separated to obtain the trend feature subset of the fused feature data; Identify recurring feature patterns in the fused feature data to obtain a periodic feature subset of the fused feature data; The temporal relationship between the trend feature subset and the periodic feature subset is extracted to obtain the interaction relationship between the trend feature subset and the periodic feature subset; The interaction relationships are subjected to feature association logic condensation to obtain a subset of association patterns in the fused feature data; Using the time dimension identifier set as a unified benchmark, the time description specifications of the trend feature subset, the periodic feature subset, and the correlation pattern subset are calibrated to obtain the time-series feature data of the fused feature data.

[0010] In a preferred embodiment, the step of evaluating the similarity and difference between real-time hydrological features and historical critical hydrological features in the time-series feature data based on the real-time change trend and evolution rate in the time-series feature data, in order to generate a dynamic early warning threshold adapted to the target area, includes: Extract the real-time change trend sequence and evolution rate sequence from the time-series feature data; Filter out historical critical hydrological feature sequences from the historical hydrological database of the target area that correspond to the geographical features and hydrological characteristics of the target area; The real-time trend sequence and the historical critical hydrological feature sequence are aligned on a time scale, and the degree of morphological matching between the real-time trend sequence and the historical critical hydrological feature sequence is identified. Assess the degree of deviation between the evolution rate sequence and the historical critical hydrological characteristic sequence; By integrating the matching degree and the deviation degree, and combining them with the real-time water level, rainfall and soil moisture data of the target area, a dynamic early warning threshold for the target area is obtained.

[0011] In a preferred embodiment, assessing the degree of difference between the evolution rate sequence and the historical critical hydrological characteristic sequence includes: Noise filtering is applied to the evolution rate sequence and the historical critical hydrological feature sequence to obtain a smoothed evolution rate sequence and a smoothed historical sequence for the target region. According to the corresponding time points, the smoothed evolution rate sequence and the smoothed history sequence are differentially processed to construct the residual sequence of the target region; The frequency of residual data in the residual sequence is statistically analyzed within different numerical intervals to obtain the frequency of the residual data. The frequency of the residual data is then structured and regularized to obtain the residual frequency distribution of the residual sequence. Based on the skewness and kurtosis characteristics in the residual frequency distribution, the degree of deviation between the evolution rate sequence and the historical critical hydrological characteristic sequence is comprehensively quantified.

[0012] In a preferred embodiment, the step of statistically analyzing the frequency of residual data occurrences in the residual sequence within different numerical intervals to obtain the frequency of the residual data, and then performing structured regularization on the frequency of the residual data to obtain the residual frequency distribution of the residual sequence, includes: Based on the hydrological critical state reference standard corresponding to the target area, and combined with the hydrological evolution deviation characteristics reflected by the residual sequence, the residual classification intervals with clear hydrological significance in the residual sequence are divided to obtain the hydrological adaptation classification interval set of the residual sequence. Traverse the residual data in the residual sequence and, according to the interval definition rules of the hydrological adaptation classification interval set, assign the residual data to the corresponding classification interval to obtain the interval classification result of the residual data; By analyzing the residual data correlation information corresponding to different classification intervals in the interval classification results, the correlation characterization data of the interval classification results are obtained. The associated representation data is structured and organized to clarify the priority and interaction relationships of different classification intervals, thereby obtaining the residual frequency distribution of the residual sequence.

[0013] In a preferred embodiment, the step of collaboratively deducing the hydrological evolution trend of the target area based on the core influencing factors, temporal evolution characteristics, and real-time change patterns of the time-series characteristic data to obtain preliminary flood forecast results for the target area includes: Principal component analysis was performed on the time-series feature data to select the features with the largest variance contribution rate from the time-series feature data, forming the core influencing factor set of the time-series feature data; Based on the real-time status of the core impact factor set, a status difference comparison is performed on the core impact factor set, and the recurring change paths in the status difference comparison results are extracted. Based on the aforementioned change path, the state transition pattern of the core influencing factor set over time is deduced to obtain the temporal evolution pattern of the target region; By identifying frequently co-occurring feature combinations in the time-series feature data, a set of real-time change patterns of the target region is obtained; Based on the core influencing factor set, time-series evolution pattern, and real-time change pattern set, the hydrological evolution trend of the target area is synergistically deduced; The hydrological evolution trend is de-standardized to obtain preliminary flood forecast results for the target area.

[0014] In a preferred embodiment, the step of integrating the preliminary flood forecast results that meet the early warning triggering conditions with the regional characteristics in the consistent data based on the dynamic early warning threshold to form flood early warning information for the target area includes: The preliminary flood forecast results are compared point by point with the dynamic early warning threshold, and the time points in the preliminary flood forecast results that exceed the threshold are marked. The elevation distribution, river network, and land use type features in the consistent data are integrated into a regional topographic and water system feature set for the target area. Using the time point as a time reference, the inundation boundary and water depth changes of the topographic and water system feature set of the region are tracked to obtain the flood impact range and depth of the target area; Based on a preset warning level mapping library, and combined with the flood impact range and the depth, the warning level of the target area is divided; The warning level, the flood impact range, and the time point are integrated into flood warning information for the target area.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention corrects deviation segments in multi-source hydrological raw data of the target area by first filtering data noise, unifying storage formats and description standards to obtain consistent data from the same source, ensuring the accuracy and consistency of the data foundation; then, based on the inherent mapping relationship of the data, it eliminates false correlations, classifies the characteristics of the correlated data, and integrates core attributes and regional basic attributes to form fused feature data, strengthening the effective correlation between data; subsequently, it integrates the temporal evolution trend, periodic characteristics, and correlation patterns of the fused feature data into time-series feature data, fully preserving the time dimension information of the hydrological data, and providing high-quality data support for subsequent forecasting and early warning.

[0016] 2. This invention evaluates the similarity and difference between real-time and historical critical hydrological features in time-series characteristic data, and generates dynamic early warning thresholds by combining real-time water level, rainfall, and other data, thereby improving the adaptability of the thresholds to regional real-time hydrological conditions. Based on the core influencing factors, time-series evolution characteristics, and real-time change patterns of the time-series characteristic data, hydrological trends are synergistically inferred to obtain accurate preliminary flood forecast results. Finally, the forecast results that meet the early warning conditions are integrated with regional characteristics to generate flood early warning information, effectively improving the accuracy of flood forecasts and the timeliness of early warnings, and providing reliable guidance for flood prevention and control. Attached Figure Description

[0017] Figure 1 A flowchart illustrating an artificial intelligence-based flood forecasting and early warning method according to an embodiment of the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides an artificial intelligence-based flood forecasting and early warning method. The executing entity of this AI-based flood forecasting and early warning method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application embodiment: a server, a terminal, etc. In other words, the AI-based flood forecasting and early warning method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0020] Reference Figure 1 The diagram shown is a flowchart illustrating an artificial intelligence-based flood forecasting and early warning method according to an embodiment of the present invention. In this embodiment, the artificial intelligence-based flood forecasting and early warning method includes: S1: Correct the biased data segments of the target area in the multi-source hydrological raw data to obtain consistent data of the target area from the same source; In this embodiment of the invention, the correction of the deviation data segments of the target area in the multi-source hydrological raw data to obtain consistent data of the target area includes: Collect multi-source raw hydrological data related to flood forecasting in the target area; Noise filtering is performed on the multi-source hydrological raw data to obtain clean multi-source hydrological data for the target area. By unifying the storage format and description specifications of the multi-source hydrological clean data, standardized hydrological data for the target area can be obtained. By correcting the biased data segments in the standardized hydrological data, consistent data from the same source for the target area are obtained.

[0021] Hydrological data recorded by various monitoring devices such as water level stations, rain gauge stations, flow stations, and soil moisture sensors within the target area are collected. The data covers indicators directly related to flood forecasting, such as water level, rainfall, flow, soil moisture, river runoff, and topographic elevation. Historical data from the past 5 years and real-time updated current data are collected at fixed time intervals to ensure that the data covers key monitoring points and flood-prone areas throughout the target area. After complete collection, the data is summarized to form multi-source hydrological raw data for the target area.

[0022] The sliding window method was used to filter noise from multi-source hydrological raw data. A fixed-length sliding window containing 10 consecutive data points was set. Starting from the beginning of the data, the window was used to cover the data segment in turn. The average value of all data points in each window was calculated. Data points whose difference from the average value exceeded one-tenth of the average value were identified as noise data and replaced with the average value of the window. The entire set of data was traversed window by window. After all the noise data was replaced, the clean multi-source hydrological data of the target area was obtained.

[0023] The storage format of multi-source hydrological clean data is standardized to CSV format. The file naming rule is defined as target area name-data type-collection date. Data fields are arranged in a fixed order according to the timestamp, monitoring indicator name, and data value. The timestamp adopts the standard format of year-month-day hour:minute:second. The monitoring indicator name strictly follows the unified standard of the hydrological industry. The data value is uniformly retained to two decimal places. The format and description of all data files are adjusted one by one to obtain standardized hydrological data of the target area.

[0024] Standardized hydrological historical data for the same period over the past 10 years were retrieved from the target area and archived according to data type. Simultaneously, standardized hydrological data from adjacent monitoring points were collected, and the geographical distance and hydrological correlation of each monitoring point were marked. Differential weights were assigned based on the strength of the correlation between different data types and the hydrological status, with the highest weight given to types directly reflecting the core hydrological state of the basin and exhibiting the strongest correlation, and lower weight given to types significantly affected by local factors and exhibiting weaker short-term fluctuations. Weather data for the same period from the target area and adjacent monitoring points were also collected. Corresponding historical fluctuation range adjustment rules and reasonable difference range adjustment rules for adjacent monitoring points were formulated for special weather scenarios such as heavy rain, drought, and typhoons, as well as normal weather scenarios. When comparing each data segment in the current standardized hydrological data, the percentage difference between the original fluctuation range of that data segment and the historical data for the same period was first calculated, and then the percentage difference was used to determine the correct data segment. The comprehensive deviation is obtained by multiplying the comprehensive deviation by the corresponding data type weight. When the original difference corresponding to the comprehensive deviation exceeds 30%, and the difference between the data segment and the data of the adjacent monitoring points in the same period is judged to be beyond the reasonable range after weather adaptation, the data segment is determined to be a deviation data segment. In the correction process, the historical change pattern of the high-weight data type is the core, combined with the hydrological change logic under the weather scenario, and the change trend of the effective data before and after the deviation data segment is referenced. The value of the deviation data segment is manually adjusted to a reasonable range with priority to weight and weather adaptation, so as to ensure that there are no abrupt breaks between the adjusted data and the effective data before and after. After the correction of all deviation data segments is completed, the correlation between the target area data and the data of the adjacent monitoring points is checked again to ensure that the overall data conforms to the hydrological pattern of the same weather and the same watershed, and finally the same source consistent data of the target area is obtained.

[0025] The beneficial effects are that by comprehensively collecting multi-source hydrological raw data, using clear methods to filter noise, standardize formats, and correct deviations, the accuracy, integrity, and consistency of data from the same source are ensured. Invalid interference components in the data are eliminated, and problems such as inconsistent formats and data deviations in multi-source data are solved. This provides a high-quality and reliable data foundation for subsequent work such as feature fusion and time-series feature extraction based on the inherent mapping relationship of the data, and ensures the smooth implementation of subsequent steps in flood forecasting and early warning methods and the accuracy of results.

[0026] S2: Based on the inherent mapping relationship between different data in the same source consistent data, feature fusion is performed on the same source consistent data to obtain the fused feature data of the same source consistent data; In this embodiment of the invention, the step of performing feature fusion on the consistent data based on the inherent mapping relationship between different data in the consistent data to obtain fused feature data of the consistent data includes: Define the attribute descriptions and application scenarios corresponding to different data in the same source data to obtain a list of data attributes of the same source data; The data attribute list is mapped to preset element association rules to obtain a preliminary mapping relationship set of the consistent data from the same source. By removing false associations with no actual hydrological significance from the preliminary mapping relationship set, the effective intrinsic mapping relationship set of the consistent data from the same source is obtained; Based on the effective intrinsic mapping relationship set, the data that are related in the same source and consistent data are classified by features to obtain the related feature group of the same source and consistent data. The core attribute information in the associated feature group is integrated with the basic attribute data of the target region to form the fused feature data of the same source and consistent data.

[0027] The process of integrating the core attribute information in the associated feature group with the basic attribute data of the target region to form the fused feature data of the same source and consistent data includes: The hydrological representation dimension of the core attribute information and the scenario adaptation dimension of the basic attribute data are decomposed, and the functional positioning of the core attribute information and the basic attribute data in the description of hydrological evolution is determined, generating a dimension function correspondence table of the homogeneous and consistent data. Based on the dimension function correspondence table, a bidirectional association mapping between the core attribute information and the basic attribute data is established to obtain the attribute association mapping set of the consistent data from the same source. Eliminate the differences in description perspective and expression standard of different attributes of the associated pairs in the attribute association mapping set to obtain the logically unified attribute association set of the consistent data from the same source; By mining the hidden hydrological synergistic patterns in the logically unified attribute association set, the association information that can jointly reflect the hydrological evolution characteristics of the target area is strengthened, and the synergistic strengthening attribute set of the consistent data from the same source is obtained. The core collaborative features of the collaborative enhancement attribute set are extracted and integrated into the fusion feature data of the homogeneous and consistent data.

[0028] Each type of data in the consistent data source is analyzed one by one, and the corresponding monitoring indicator name, data type, unit of measurement, statistical caliber, and other attribute descriptions are clarified. For example, the attribute description of water level data is: monitoring indicator water level data type continuous data unit of measurement meter statistical caliber instantaneous value. At the same time, combined with the actual process of flood forecasting, the application scenarios of each type of data are determined. For example, rainfall data is used to determine the source of flood replenishment, water level data is used to monitor the water storage status of river channels, and flow data is used to analyze the confluence capacity. The attribute descriptions and application scenarios of all data are recorded one by one to form a data attribute list of consistent data source in the target area.

[0029] The preset element association rules are a fixed set of rules based on general laws in the hydrological field and historical hydrological data of the target area. They include common element association relationships such as rainfall and runoff, topography and runoff velocity, and soil moisture and infiltration. Each data attribute in the data attribute list is compared with the elements in the preset element association rules one by one. If the element corresponding to the data attribute matches the element in the rule, the association relationship between the data attribute and the corresponding element is recorded. After summarizing all the successfully matched association relationships, a preliminary mapping relationship set of consistent data from the same source is obtained.

[0030] By comparing the hydrological cycle principle and actual hydrological process of the target area, each correlation in the preliminary mapping relationship set is checked one by one to determine whether there is an actual hydrological interaction between the two parties in the correlation. If the two elements in a certain correlation do not have a direct or indirect mutual influence in the hydrological process, such as the correlation between soil moisture and topographic elevation, there is no actual hydrological significance. Therefore, it is determined to be a false correlation. All false correlations are removed from the preliminary mapping relationship set, and the remaining correlations form the effective intrinsic mapping relationship set of consistent data from the same source.

[0031] Based on the association relationships in the effective intrinsic mapping relationship set, all data with direct correlations in the same source data are classified into the same group. Within the same group, the core association attributes and secondary association attributes are clearly defined. The core association attribute is the key element driving the association of the data in this group. For example, when rainfall is the core association attribute, data related to it such as runoff, water level, and soil moisture are grouped into one group. Each group of data is labeled with an association logic description. After all groupings are completed, the association feature group of the same source data is obtained.

[0032] Core attribute information that directly reflects the core laws of hydrological changes is extracted from the associated feature groups. Its hydrological representation dimensions are decomposed, such as the representation dimension of rainfall including precipitation intensity, duration, and spatial distribution. At the same time, basic attribute data such as topography, geomorphology, watershed area, and river morphology of the target area are collected and their scene adaptation dimensions are decomposed, such as the adaptation dimension of topography including slope, aspect, and altitude. Combined with the needs of hydrological evolution description in flood forecasting, the core attribute information is determined to quantify the dynamic changes of hydrological elements, and the basic attribute data is used to define the boundary conditions of hydrological evolution. The dimension decomposition results and functional positioning are matched one by one to generate a dimension function correspondence table of consistent data from the same source.

[0033] Based on the dimensional association and functional complementarity between the core attribute information and the basic attribute data in the dimensional function correspondence table, a bidirectional association mapping is established. That is, a certain representation dimension of the core attribute information corresponds to the matching dimension of the basic attribute data, and at the same time, the matching dimension of the basic attribute data is inversely associated with the representation dimension of the core attribute information. For example, the precipitation intensity dimension of rainfall corresponds to the slope dimension of topography, and the slope dimension is inversely associated with the precipitation intensity dimension. The specific content of each bidirectional association is recorded to form an attribute association mapping set of consistent data from the same source.

[0034] Each association pair in the attribute association mapping set is checked one by one. For the differences in the descriptive perspectives of different attributes, such as core attribute information focusing on describing dynamic changes and basic attribute data focusing on describing static characteristics, the logical perspective of the association expression is unified, and the association logic between dynamic data and static data is clarified. For the differences in expression standards, such as inconsistent index names and different units of measurement, the expression method is adjusted according to the unified standards of the hydrological industry to ensure that the description of the association pairs is unified, standardized and unambiguous, and to obtain a logically unified attribute association set of data with the same source.

[0035] Continuous time series trend analysis is performed on all associated data in the logically unified attribute association set to observe the synchronicity and interaction of different attribute data during the formation and evolution of floods, and to uncover hidden hydrological synergistic patterns, such as the faster growth of runoff in areas with steeper slopes under the same precipitation intensity. These associations that can jointly reflect the hydrological evolution characteristics of the target area are strengthened, highlighting the impact of synergistic effects on hydrological changes. The strengthened associations are then integrated to obtain a synergistic strengthened attribute set of consistent data from the same source.

[0036] The importance of all attribute information in the synergistic reinforcement attribute set is ranked, and the core synergistic features with the greatest impact and strongest correlation on the hydrological evolution trend are selected, such as the synergistic features of rainfall-slope-runoff and the synergistic features of rainfall duration-soil moisture-water level. The core synergistic features are systematically organized according to the time dimension and correlation strength to ensure that the feature information is complete and logically coherent. The organized core synergistic features are integrated into a structured dataset to obtain fused feature data of consistent data from the same source.

[0037] The beneficial effects are achieved through a series of meticulous operations, including gradually clarifying data attributes, establishing correlations, eliminating false correlations, classifying correlation features, decomposing dimensions, establishing bidirectional mapping, unifying logic, mining synergistic patterns, and refining core features. This ensures that the fused feature data can accurately reflect the inherent correlations and synergistic evolution patterns of hydrological elements in the target area. The correlation, integrity, and effectiveness of the data are significantly improved, providing high-quality core data support for subsequent extraction of time-series feature data, generation of dynamic early warning thresholds, and prediction of hydrological evolution trends. This ensures the scientific nature and accuracy of flood forecasting and early warning work.

[0038] S3: Integrate the trend information, periodic features and correlation patterns that evolve over time in the fused feature data into the time-series feature data of the target region; In this embodiment of the invention, integrating the trend information, periodic features, and correlation patterns that evolve over time from the fused feature data into time-series feature data for the target region includes: By sorting out the temporal correlation clues of the fused feature data, clarifying the time record benchmarks corresponding to different types of features, and generating the temporal dimension identifier set of the fused feature data; Based on the time dimension identifier set, the feature content that shows a continuous change over time in the fused feature data is separated to obtain the trend feature subset of the fused feature data; Identify recurring feature patterns in the fused feature data to obtain a periodic feature subset of the fused feature data; The temporal relationship between the trend feature subset and the periodic feature subset is extracted to obtain the interaction relationship between the trend feature subset and the periodic feature subset; The interaction relationships are subjected to feature association logic condensation to obtain a subset of association patterns in the fused feature data; Using the time dimension identifier set as a unified benchmark, the time description specifications of the trend feature subset, the periodic feature subset, and the correlation pattern subset are calibrated to obtain the time-series feature data of the fused feature data.

[0039] The analysis of various feature contents in the fused feature data is carried out one by one. The collection time nodes, collection intervals and time recording basis of each type of feature data are traced. The temporal connection and synchronous correspondence between different features are sorted out. The time recording benchmark corresponding to each type of feature is clarified. For example, some features use the hour as the time recording benchmark and some features use the natural day as the statistical period. The time association clues, time recording benchmarks and time attribute information of all features are sorted out one by one to form the time dimension identifier set of the fused feature data of the target area.

[0040] Based on the clearly defined time record benchmarks of the time dimension, the fused feature data is sorted in chronological order. The changes in the sorted data are observed segment by segment. Feature content that shows a continuous rise, fall or stable fluctuation over time without abrupt interruptions or jumps is selected. These feature contents maintain a consistent trend of change within a continuous time interval. All feature contents that meet this condition are integrated to obtain the trend feature subset of the fused feature data.

[0041] According to the statistical period division standard of the time dimension identifier set, the fused feature data is divided into multiple time periods of the same length. The feature change patterns in different time periods are compared to find the feature combinations and change trajectories that recur in multiple non-continuous time periods, such as the fixed combination of rainfall and runoff in a specific season. These recurring feature patterns are completely extracted and summarized to obtain the periodic feature subset of the fused feature data.

[0042] Using the time dimension identifier set as a unified time reference, the trend feature subset and the cyclic feature subset are aligned on the time axis. The superposition effect and mutual influence of the two are analyzed in time period by time. The role of the cyclic feature in the trend change process is observed, such as whether the cyclic feature will enhance or weaken the trend change amplitude. At the same time, the constraint relationship between the trend feature and the cyclic feature is recorded, such as whether the overall trend direction limits the fluctuation range of the cyclic feature. These observed interactions are systematically sorted out to obtain the interaction relationship between the trend feature subset and the cyclic feature subset.

[0043] The identified interactions are analyzed in depth to extract stable correlations, eliminate occasional temporary correlations, and clarify the triggering conditions and impacts of different interactions. For example, when a trend feature shows an upward trend, a specific cyclical feature will amplify the trend change by a fixed amount. These extracted stable correlations are compiled into a book to obtain a subset of correlation patterns in the integrated feature data.

[0044] Using the time record benchmark and time specification of the time dimension identifier set as a unified standard, the time description methods of the trend feature subset, periodic feature subset and correlation pattern subset are adjusted, and the timestamp format, time statistical interval and time expression terminology of each subset are unified to ensure that the three subsets are completely synchronized and consistent in the time dimension, without time deviation and expression ambiguity. The three calibrated subsets are systematically integrated according to the time axis order and correlation logic to obtain the time series feature data of the fused feature data of the target area.

[0045] The beneficial effects are that by systematically sorting out the time-dimensional correlation clues, separating trends and periodic features, extracting the interaction relationship, condensing the correlation rules, and calibrating the time standard, the time series feature data fully includes the time evolution trend, periodic change pattern and inherent correlation rules of the fused feature data, and the time description is unified and standardized. This provides a high-quality data foundation with a clear structure and logical coherence for subsequent generation of dynamic early warning thresholds based on real-time change trends and evolution rates to collaboratively infer hydrological evolution trends, effectively ensuring the accurate advancement of flood forecasting and early warning work.

[0046] S4: Based on the real-time change trend and evolution rate in the time series feature data, evaluate the similarity and difference between the real-time hydrological features in the time series feature data and the historical critical hydrological features, so as to generate a dynamic early warning threshold adapted to the target area. In this embodiment of the invention, the step of evaluating the similarity and difference between real-time hydrological features and historical critical hydrological features in the time-series feature data based on the real-time change trend and evolution rate in the time-series feature data, in order to generate a dynamic early warning threshold adapted to the target area, includes: Extract the real-time change trend sequence and evolution rate sequence from the time-series feature data; Filter out historical critical hydrological feature sequences from the historical hydrological database of the target area that correspond to the geographical features and hydrological characteristics of the target area; The real-time trend sequence and the historical critical hydrological feature sequence are aligned on a time scale, and the degree of morphological matching between the real-time trend sequence and the historical critical hydrological feature sequence is identified. Assess the degree of deviation between the evolution rate sequence and the historical critical hydrological characteristic sequence; By integrating the matching degree and the deviation degree, and combining them with the real-time water level, rainfall and soil moisture data of the target area, a dynamic early warning threshold for the target area is obtained.

[0047] The assessment of the degree of difference between the evolution rate sequence and the historical critical hydrological characteristic sequence includes: Noise filtering is applied to the evolution rate sequence and the historical critical hydrological feature sequence to obtain a smoothed evolution rate sequence and a smoothed historical sequence for the target region. According to the corresponding time points, the smoothed evolution rate sequence and the smoothed history sequence are differentially processed to construct the residual sequence of the target region; The frequency of residual data in the residual sequence is statistically analyzed within different numerical intervals to obtain the frequency of the residual data. The frequency of the residual data is then structured and regularized to obtain the residual frequency distribution of the residual sequence. Based on the skewness and kurtosis characteristics in the residual frequency distribution, the degree of deviation between the evolution rate sequence and the historical critical hydrological characteristic sequence is comprehensively quantified.

[0048] The frequency of residual data in the residual sequence within different numerical intervals is statistically analyzed to obtain the frequency of the residual data. The frequency of the residual data is then structured and regularized to obtain the residual frequency distribution of the residual sequence, including: Based on the hydrological critical state reference standard corresponding to the target area, and combined with the hydrological evolution deviation characteristics reflected by the residual sequence, the residual classification intervals with clear hydrological significance in the residual sequence are divided to obtain the hydrological adaptation classification interval set of the residual sequence. Traverse the residual data in the residual sequence and, according to the interval definition rules of the hydrological adaptation classification interval set, assign the residual data to the corresponding classification interval to obtain the interval classification result of the residual data; By analyzing the residual data correlation information corresponding to different classification intervals in the interval classification results, the correlation characterization data of the interval classification results are obtained. The associated representation data is structured and organized to clarify the priority and interaction relationships of different classification intervals, thereby obtaining the residual frequency distribution of the residual sequence.

[0049] In the process of extracting real-time trend sequences and evolution rate sequences from time-series feature data, the time-series feature data is first defined as hydrological data collected by continuous monitoring equipment within the target area, covering raw information on continuous changes in water level, flow rate, and runoff over time. This data is naturally arranged in chronological order of collection to form a complete time-series dataset. When extracting the real-time trend sequence, the time-series feature data is sequentially analyzed according to time nodes. First, the direction of change of the value at each time node relative to the value at the previous time node is determined, including three possibilities: rising, falling, or remaining stable. Then, the direction of change of adjacent time nodes is correlated. For example, if the previous node shows an upward trend and the current node continues to rise, a continuous upward trend is formed; if the previous node rises and the current node falls, a reversal trend is formed. By analyzing the direction of change and correlations of all time nodes, a sequence continuously reflecting the trajectory of data change is formed; this sequence is the real-time trend sequence. When extracting the evolution rate sequence, the time order of the time series characteristic data is also used as the basis. The change in value between two adjacent time nodes is calculated. The absolute change is obtained by subtracting the value of the previous time node from the value of the later time node. At the same time, the time interval between two time nodes is determined. For example, if the interval between adjacent data collection is one hour, then the time interval is one hour. The calculated absolute change is divided by the corresponding time interval to obtain the data change rate per unit time. The unit time change rates corresponding to all adjacent time nodes are arranged in the order of collection time to form an evolution rate sequence that can reflect the speed of data change.

[0050] When selecting historical critical hydrological feature sequences from the historical hydrological database for a target area, corresponding to the area's geographical features and hydrological characteristics, a comprehensive review of the target area's geographical features and hydrological characteristics is first conducted. Geographical features include terrain types such as plains, mountains, and hills; drainage area; altitude range; and land cover types such as forests, cultivated land, and urban buildings. Hydrological characteristics include key information such as runoff coefficient, runoff duration, groundwater recharge ratio, and river gradient. Next, all historical hydrological feature sequences from the historical hydrological database are retrieved. Each historical hydrological feature sequence stored in this database is associated with the geographical features and hydrological characteristics of its corresponding collection area, ensuring that the background information for each sequence is complete and readily available. Subsequently, the geographical features of the target area were compared one by one with the geographical features associated with each historical hydrological feature sequence in the database. Historical hydrological feature sequences with completely identical topographic types, watershed area differences within a reasonable range determined based on common characteristics of similar regions, largely overlapping altitude ranges, and the same main categories of land cover types were selected. These initially selected sequences were then further compared with the associated hydrological characteristics of the target area to ensure similar runoff coefficients, matching confluence times, roughly equivalent groundwater recharge ratios, and river gradients within the same range. Finally, from the historical hydrological feature sequences after the second round of screening, those sequences corresponding to key hydrological changes before floods, waterlogging, or other hydrological disasters in the region or similar areas in history, or those formed when hydrological safety thresholds were reached, were extracted. These sequences are the historical critical hydrological feature sequences corresponding to the geographical features and hydrological characteristics of the target area.

[0051] When aligning the real-time trend sequence with the historical critical hydrological characteristic sequence on a time scale, the time interval standard for each sequence is first determined. The time interval for the real-time trend sequence is determined by the data acquisition frequency; for example, if data is collected once per hour, the time interval is one hour. The time interval for the historical critical hydrological characteristic sequence may vary depending on historical acquisition conditions, such as once every two hours or once per day. Using the time interval of the real-time trend sequence as a unified standard, time interpolation is performed on the historical critical hydrological characteristic sequence. For missing values ​​in the historical critical hydrological characteristic sequence corresponding to time nodes in the real-time sequence, a linear extension method is used to supplement them based on the value change trend of two adjacent existing data nodes in the historical sequence. For example, if two adjacent existing data nodes are the value 'a' at time A and the value 'b' at time B, and the value at time C is missing, and the change from A to B shows a uniform upward trend, then the supplementary value corresponding to time C is calculated based on the change amplitude from A to B and the time proportions from A to C and C to B. In this way, the time nodes of the historical critical hydrological characteristic sequence are adjusted to be completely consistent with the real-time trend sequence, achieving time scale alignment between the two. When identifying the degree of morphological matching, the two aligned sequences are compared one by one according to time nodes. First, it is determined whether the direction of data change at each corresponding time node is consistent, i.e., both are rising, both are falling, or both remain stable. Then, it is analyzed whether the fluctuation patterns formed by multiple consecutive time nodes are similar. For example, do they all show a trajectory of slow rise, then rapid rise, and finally stabilization, or do they all have the same number of key turning points from rising to falling, and from falling to rising? The proportion of time nodes with consistent change direction is counted out of the total number of time nodes. At the same time, it is determined whether the occurrence time of key turning points is close. The standard for closeness is determined based on the time response law of hydrological changes, i.e., the time difference of the occurrence of turning points does not exceed two data collection intervals. Based on the statistically obtained proportion and the matching of turning points, the degree of morphological matching is determined. A high degree of matching is defined as a proportion of time nodes with consistent change direction of more than 80% and a time difference of key turning points within a reasonable range. A moderate degree of matching is defined as a proportion between 60% and 80% and key turning points basically correspond. A low degree of matching is defined as a proportion below 60% and significant differences in turning points.

[0052] When assessing the deviation between the evolution rate sequence and the historical critical hydrological characteristic sequence, the historical critical hydrological characteristic sequence, after time-scale alignment, is first processed using the same method as the real-time evolution rate sequence. This involves calculating the numerical change between two adjacent time nodes in the historical critical hydrological characteristic sequence in chronological order, subtracting the previous node's value from the subsequent node's value to obtain the absolute change, and then dividing by the corresponding time interval to obtain the rate of change per unit time. All rate of change per unit time are arranged chronologically to form the historical critical evolution rate sequence. The real-time evolution rate sequence and the historical critical evolution rate sequence are then compared node by node, calculating the difference between the two rate values ​​at each time node. The absolute value of this difference is then compared with the rate value at the corresponding time node in the historical critical evolution rate sequence to obtain the relative deviation at each time node. For example, if the real-time rate is 5 and the historical critical rate is 4, the absolute difference is 1, and the relative deviation is the ratio of 1 to 4. The average relative deviation at all time points is calculated, and the degree of deviation is classified according to the magnitude of this average. If the average is less than 20% of the overall average of the historical critical evolution rate sequence, it is judged as a slight deviation; if the average is between 20% and 50% of the overall average of the historical critical evolution rate sequence, it is judged as a moderate deviation; if the average is greater than 50% of the overall average of the historical critical evolution rate sequence, it is judged as a severe deviation. This process is used to assess the degree of deviation between the evolution rate sequence and the historical critical hydrological characteristic sequence.

[0053] When integrating matching and deviation levels, the weights corresponding to different matching and deviation levels are first determined based on the historical hydrological disaster occurrence patterns in the target area. High matching levels are assigned higher weights, moderate matching levels are assigned medium weights, and low matching levels are assigned lower weights. Similarly, slight deviation levels are assigned higher weights, moderate deviation levels are assigned medium weights, and severe deviation levels are assigned lower weights. The core principle of weight allocation is that a higher risk of hydrological disasters and a slight deviation indicate a higher matching level, thus requiring a larger weight allocation. The weights corresponding to the matching and deviation levels are multiplied to obtain the integrated weight value. Then, a weight-threshold correspondence table based on historical hydrological disaster data and safety threshold standards for the target area is consulted. This table clearly defines the basic warning threshold ranges corresponding to different integrated weight values. Based on the calculated integrated weight value, the corresponding basic warning threshold is determined from the table. Subsequently, by combining real-time water level, rainfall, and soil moisture data of the target area, the basic warning threshold is dynamically adjusted. If the real-time water level is higher than the historical average for the same period in the area, the threshold is appropriately lowered based on the proportion by which the water level exceeds the average for the same period. If the cumulative real-time rainfall in the past three hours exceeds the historical average rainfall for the same period in the area, the threshold is further lowered. If the soil moisture reaches saturation, i.e., the soil water content reaches 100% of field capacity, the threshold is lowered again. If the real-time water level is lower than the historical average for the same period, the rainfall in the past three hours is low, and the soil moisture has not reached saturation, the threshold is appropriately raised based on the difference between the relevant data and the historical data for the same period. Through this process of fusion and dynamic adjustment, a dynamic warning threshold that accurately reflects the current hydrological safety status of the target area is finally obtained.

[0054] The formula for calculating the dynamic early warning threshold is as follows: ; In the formula, This indicates the dynamic early warning threshold. This represents the historical critical water level threshold in the historical hydrological database. This represents the preset matching degree weight coefficient. This indicates the degree of morphological matching between the real-time trend sequence and the historical critical hydrological characteristic sequence. This represents the average deviation of the evolution rate generated based on the residual sequence. This indicates that the real-time water level weights are dynamically allocated using a fixed interval. This indicates that the real-time rainfall weights are dynamically allocated using a fixed interval. This indicates that the real-time soil moisture weights are dynamically allocated using a fixed interval. This indicates the real-time water level. This indicates the amount of rainfall. This indicates the soil moisture data. This represents the regional hydrological characteristic coefficient preset based on the geographical and hydrological characteristics of the target area.

[0055] Historical critical water level thresholds are critical water level values ​​extracted from historical critical hydrological feature sequences corresponding to the geographical features and hydrological characteristics of the target area, extracted from historical hydrological databases. The matching degree weight coefficient is a fixed coefficient pre-set based on past experience in hydrological disaster early warning for the target area, combined with the impact of matching historical critical hydrological feature sequences with real-time sequences on early warning accuracy. Morphological matching degree is determined by comparing the direction of change, continuous fluctuation patterns, and key turning points at corresponding time nodes with the historical critical hydrological feature sequences aligned with the real-time trend sequence, statistically analyzing the proportion of consistent nodes and the matching of turning points. The average deviation of evolution rate is obtained by first calculating the relative deviation between the evolution rate sequence and the historical critical hydrological feature sequence at each time node, and then averaging the relative deviations of all nodes. Real-time water level weight, real-time rainfall weight, and real-time soil moisture weight are dynamically determined based on the geographical features and hydrological characteristics of the target area, combined with the degree of influence of different elements on hydrological safety status, within a fixed weight allocation range, according to the real-time impact priority of each element. Real-time water level is the current water level data continuously acquired by water level monitoring equipment deployed in the target area at a set collection frequency. Rainfall is the cumulative real-time rainfall value collected by rain gauge stations within the target area. Soil moisture data is the real-time soil moisture content data collected by soil moisture sensors deployed in the target area. Regional hydrological characteristic coefficients are fixed coefficients pre-set based on the geographical features of the target area, such as topography, watershed area, and land cover, as well as hydrological characteristics such as runoff coefficient and runoff time, reflecting the hydrological response characteristics of the area.

[0056] This formula uses historical critical water level thresholds as a foundation. It first adjusts the base threshold by considering the morphological matching degree and average deviation of evolution rate between the real-time trend sequence and the historical critical hydrological characteristic sequence. Then, it incorporates the influence of three current hydrological elements: real-time water level, rainfall, and soil moisture. Finally, it adjusts the threshold by combining regional hydrological characteristic coefficients, ultimately obtaining a dynamic early warning threshold that accurately reflects the current hydrological safety status of the target area. Its function is to integrate information from several dimensions: the reference value of historical critical states, the trend and rate characteristics of current hydrological changes, the actual situation of real-time hydrological elements, and the region's own hydrological characteristics. This ensures that the resulting early warning threshold references historical experience while also fitting current actual hydrological conditions, avoiding biases caused by relying solely on historical or real-time data.

[0057] When the morphological matching degree between the real-time trend sequence and the historical critical hydrological characteristic sequence is higher, and the average deviation of the evolution rate is lower, the adjustment result of the corresponding part in the formula will be closer to 1. In this case, the adjustment range of the basic historical critical water level threshold is smaller, and the dynamic warning threshold will be closer to the historical critical water level threshold. When any one or more values ​​of real-time water level, rainfall, and soil moisture are higher, the calculation result of the corresponding part of these factors will be larger, and the dynamic warning threshold will increase accordingly. When the regional hydrological characteristic coefficient of the target area is larger, the dynamic warning threshold will also increase accordingly, assuming other conditions remain unchanged. Conversely, if the morphological matching degree is lower and the average deviation of the evolution rate is higher, the adjustment result of the corresponding part will be less than 1, the basic threshold will be lowered, and the dynamic warning threshold will be lower than the historical critical water level threshold. At the same time, if the values ​​of real-time water level, rainfall, and soil moisture are lower, the corresponding calculation result will be smaller, and the dynamic warning threshold will also decrease accordingly.

[0058] When filtering noise from evolution rate sequences and historical critical hydrological feature sequences, a moving average method is used to process the evolution rate sequence. Several consecutive adjacent evolution rate data points are selected in chronological order as a sliding window. The arithmetic mean of all data points within the window is calculated, and this average value is used to replace the data in the middle of the window. Then, the sliding window is moved one data point forward in chronological order, and several new consecutive adjacent data points are selected. The operation of calculating the average value and replacing the data in the middle of the window is repeated until the entire evolution rate sequence is covered. For positions at the beginning and end of the sequence where a complete sliding window cannot be formed, the original data is directly retained. The sequence obtained after this processing is the smoothed evolution rate sequence. The same sliding window size and processing method are used for historical critical hydrological feature sequences. Similarly, the arithmetic mean of consecutive data points is calculated using a sliding window to replace the corresponding data points. The sequence obtained after this processing is the smoothed historical sequence.

[0059] When performing difference processing on the smoothed evolution rate sequence and the smoothed history sequence according to the corresponding time points, it is first confirmed that the time points of the smoothed evolution rate sequence and the smoothed history sequence are completely corresponding. Each time point contains the corresponding smoothed evolution rate data and smoothed history data. Then, the smoothed evolution rate data corresponding to each time point is extracted in chronological order. The smoothed history data corresponding to the same time point is subtracted from the data. The result is the residual data of that time point. The residual data corresponding to all time points are arranged in chronological order to form the residual sequence of the target region.

[0060] When calculating the frequency of residual data in a residual sequence within different numerical intervals, first, based on the minimum and maximum values ​​of the residual data in the residual sequence, several continuous and fixed numerical intervals are divided. Then, each residual data in the residual sequence is read one by one, and the numerical interval to which the data belongs is determined. For each interval to which a data belongs, the frequency of occurrence of the corresponding interval is increased by one. After all the residual data has been calculated, the frequency of occurrence of each numerical interval is recorded. This is the frequency of the residual data. Then, the frequency of the residual data is structured and regularized. All numerical intervals are arranged in ascending order, and the corresponding frequencies of occurrence are also arranged in the same order. At the same time, the range and corresponding frequency of each interval are defined. The resulting ordered set is the residual frequency distribution of the residual sequence.

[0061] When quantifying the degree of deviation based on the skewness and kurtosis characteristics of the residual frequency distribution, the skewness characteristic is analyzed first. The peak position of the residual frequency distribution is observed. If the peak is biased towards the smaller side of the numerical range, it indicates that the residual data is generally small. If the peak is biased towards the larger side of the numerical range, it indicates that the residual data is generally large. This determines the degree of deviation of the distribution from the symmetrical state. Then, the kurtosis characteristic is analyzed. If the peak of the residual frequency distribution is relatively steep, it indicates that most of the residual data is concentrated in a few numerical ranges. If the peak is relatively flat, it indicates that the residual data is scattered in multiple numerical ranges. This determines the steepness of the distribution. Then, these two characteristics are combined. If the skewness deviates greatly from the symmetry and the kurtosis is relatively flat, it indicates that the evolution rate sequence deviates greatly from the historical critical hydrological characteristic sequence. If the skewness is close to symmetry and the kurtosis is relatively steep, it indicates that the degree of deviation is small. This analysis completes the comprehensive quantification of the degree of deviation.

[0062] The reference standard for the hydrological critical state corresponding to the target area is retrieved. This standard clarifies the degree of deviation between the evolution rate and the historical critical sequence corresponding to different hydrological safety levels, as well as the reference range of the residuals corresponding to this degree of deviation. Simultaneously, the hydrological evolution deviation characteristics reflected by the residual sequence are analyzed, that is, the fluctuation of the hydrological state in the target area corresponding to changes in the magnitude of the residual data. The reference range of the residuals in the reference standard is matched with the actual deviation characteristics of the residual sequence to divide residual classification intervals with clear hydrological significance. One interval corresponds to the hydrological safety state, where the residual data indicates that the evolution rate deviates little from the historical critical sequence, and the hydrological state of the target area is stable. Another interval corresponds to the slightly deviated state, where the residual data indicates that the evolution rate begins to deviate from the historical critical sequence, and the hydrological state shows slight fluctuations. A third interval corresponds to the moderately deviated state, where the residual data indicates that the degree of deviation of the evolution rate increases, and the hydrological risk is somewhat enhanced. The last interval corresponds to the severely deviated state, where the residual data indicates that the degree of deviation of the evolution rate is large, and the hydrological state is close to the critical state. The set of these divided intervals constitutes the hydrological adaptation classification interval set.

[0063] First, define the interval definition rules of the hydrological adaptation classification interval set, that is, the upper and lower limits of the residual values ​​corresponding to each classification interval. Then, read each residual data in the sequence one by one according to the time order of the residual sequence. After obtaining a single residual data, compare it with the upper and lower limits of all intervals in the hydrological adaptation classification interval set to determine which interval the data falls into. For example, if the value of a residual data is greater than the upper limit of the safe interval but less than the upper limit of the slightly deviated interval, the data is assigned to the classification interval corresponding to the slightly deviated interval. After all residual data have completed such comparison and assignment operations, record the classification interval corresponding to each residual data according to the time order of the residual sequence. The resulting ordered result is the interval classification result.

[0064] For each classification interval in the interval classification results, firstly, the specific number of residual data contained in the interval is counted. Then, the time points corresponding to these residual data in the residual sequence are extracted. Next, through the hydrological monitoring records of the target area, the real-time hydrological data corresponding to each time point is retrieved, including real-time water level data, real-time rainfall data, etc. This information is associated and bound with the corresponding classification interval. For example, the safety classification interval will be associated with the number of residual data it contains, the list of time points corresponding to these data, and the real-time water level and rainfall data of each time point. All such association information corresponding to all classification intervals is integrated to form a complete set, which is the association representation data.

[0065] Based on the importance of each hydrological safety level in the reference standard for hydrological critical states of the target area, the priority of different classification intervals is determined. Among them, the classification interval corresponding to the severe deviation state has the highest priority, followed by the interval corresponding to the moderate deviation state, then the interval corresponding to the slight deviation state, and the interval corresponding to the safe state has the lowest priority. At the same time, the interaction relationship between different classification intervals is clarified. For example, when the residual data moves from the safe interval to the slight deviation interval, it means that the deviation of the evolution rate from the historical critical sequence begins to increase, and the hydrological risk gradually increases. Subsequently, the associated characterization data is structured and arranged in descending order of priority. The number of residual data contained in each interval is recorded below, that is, the frequency of occurrence of residual data in that interval. The ordered set formed after this arrangement is the residual frequency distribution of the residual sequence.

[0066] The beneficial effects are as follows: By accurately extracting real-time trend sequences and evolution rate sequences from time-series feature data, a comprehensive capture of the current hydrological changes in the target area is ensured, providing a reliable data foundation for subsequent early warning analysis; by screening historical critical hydrological feature sequences that highly match the geographical features and hydrological characteristics of the target area, the pertinence and effectiveness of comparative analysis are guaranteed, avoiding analytical biases caused by background differences; by eliminating time dimension differences between different sequences through time scale alignment processing, combined with detailed identification of morphological matching degree, the similarity between the current hydrological change trajectory and the historical critical state can be accurately grasped; by assessing the degree of deviation between the evolution rate sequence and the historical critical hydrological feature sequence, the difference between the current hydrological change rate and the safe critical state can be clearly understood; by scientifically integrating the degree of morphological matching degree and the degree of evolution rate deviation, and dynamically combining the real-time water level, rainfall, and soil moisture data of the target area, the final dynamic early warning threshold can accurately reflect the current hydrological safety status, effectively improving the timeliness and accuracy of hydrological disaster early warning, providing a scientific and reliable decision-making basis for hydrological disaster prevention and control in the target area, and minimizing the losses caused by disasters.

[0067] S5: Based on the core influencing factors, time-series evolution characteristics, and real-time change patterns of the time-series characteristic data, the hydrological evolution trend of the target area is synergistically deduced to obtain the preliminary flood forecast results for the target area; In this embodiment of the invention, the step of collaboratively deducing the hydrological evolution trend of the target area based on the core influencing factors, temporal evolution characteristics, and real-time change patterns of the time-series characteristic data to obtain preliminary flood forecast results for the target area includes: Principal component analysis was performed on the time-series feature data to select the features with the largest variance contribution rate from the time-series feature data, forming the core influencing factor set of the time-series feature data; Based on the real-time status of the core impact factor set, a status difference comparison is performed on the core impact factor set, and the recurring change paths in the status difference comparison results are extracted. Based on the aforementioned change path, the state transition pattern of the core influencing factor set over time is deduced to obtain the temporal evolution pattern of the target region; By identifying frequently co-occurring feature combinations in the time-series feature data, a set of real-time change patterns of the target region is obtained; Based on the core influencing factor set, time-series evolution pattern, and real-time change pattern set, the hydrological evolution trend of the target area is synergistically deduced; The hydrological evolution trend is de-standardized to obtain preliminary flood forecast results for the target area.

[0068] We comprehensively analyzed all features contained in the time series feature data, calculated the variance of each feature one by one. The calculation method was to first calculate the average value of all data values ​​for the feature, then subtract the average value from each data value to get the difference, square all the differences and sum them, and finally divide the sum by the total number of data values ​​to get the variance of the feature. The features were sorted from largest to smallest variance value, and the top five features with the highest variance contribution rate were selected. These features were then fully integrated to form the core influencing factor set of the time series feature data.

[0069] Collect the current real-time value, direction of change, and magnitude of change for each impact factor in the core impact factor set to form the real-time status of each factor. Retrieve the historical status data of the core impact factor set and compare the real-time status with the historical status one by one to find the differences between the two in terms of value, direction of change, and magnitude. Record the factor change path corresponding to each difference. Statistically count the change paths that appear more than three times in all difference comparison results. Extract these repeated change paths completely to form a clear set of change paths.

[0070] The extracted recurring change paths are systematically analyzed to sort out the transformation process of the core influencing factor set from the initial state to the intermediate state and then to the final state in each path. The common laws of factor state transition in different paths are summarized, and the influence of the interaction between factors on state transition is clarified. For example, after the value of a certain factor rises to a certain range, another factor will show a fixed trend of change. These common laws and state transition logics are organized into a standardized evolutionary framework to obtain the temporal evolution pattern of the target region.

[0071] The time-series feature data is divided into multiple consecutive time periods in chronological order, with each time period having a uniform length of 24 hours. The combination of all features within each time period is analyzed, and the frequency of different feature combinations in all time periods is counted. Feature combinations that appear more than 50% of the total number of time periods are identified as frequently co-occurring feature combinations. All such feature combinations are summarized and organized to obtain the real-time change pattern set of the target area.

[0072] Based on the core influencing factor set, the dominant role of each factor in hydrological changes is clarified. Combined with the pattern of factor state transition in the time-series evolution model, the future direction of factor change is predicted. At the same time, based on the characteristic combination of frequent co-occurrence of real-time change patterns, the reasonable range of evolution trend is constrained. The interaction of the three factors is comprehensively considered. For example, the real-time state of the core factor triggers a specific time-series evolution model, while the real-time change pattern limits the evolution amplitude under the model. Finally, the hydrological evolution trend of the target area in the future period is deduced.

[0073] The original data range and average values ​​used in the previous standardization of hydrological data are retrieved. The relative values ​​in the hydrological evolution trend obtained by deduction are restored by reversing the standardization process. That is, the relative value is first multiplied by the difference in the original data range, and then the original average value is added to obtain the corresponding actual hydrological data value. All the restored actual hydrological data are sorted in chronological order to form the preliminary flood forecast results for the target area.

[0074] The beneficial effects are that by systematically screening core influencing factors, extracting recurring change paths, deducing temporal evolution patterns, and identifying real-time change patterns, and then coordinating the three to accurately deduce hydrological evolution trends and perform de-standardization processing, the preliminary flood forecast results can closely match the hydrological characteristics and real-time changes of the target area. The accuracy and reliability of the forecast data are significantly improved, providing a high-quality core basis for the subsequent generation of flood warning information by combining dynamic warning thresholds, and ensuring the scientific nature and effectiveness of flood forecasting and warning work.

[0075] S6: Based on the dynamic early warning threshold, the preliminary flood forecast results that meet the early warning triggering conditions are integrated with the regional features in the consistent data from the same source to form the flood early warning information for the target area.

[0076] In this embodiment of the invention, the step of integrating the preliminary flood forecast results that meet the early warning triggering conditions with the regional features in the consistent data from the same source, based on the dynamic early warning threshold, into flood early warning information for the target area includes: The preliminary flood forecast results are compared point by point with the dynamic early warning threshold, and the time points in the preliminary flood forecast results that exceed the threshold are marked. The elevation distribution, river network, and land use type features in the consistent data are integrated into a regional topographic and water system feature set for the target area. Using the time point as a time reference, the inundation boundary and water depth changes of the topographic and water system feature set of the region are tracked to obtain the flood impact range and depth of the target area; Based on a preset warning level mapping library, and combined with the flood impact range and the depth, the warning level of the target area is divided; The warning level, the flood impact range, and the time point are integrated into flood warning information for the target area.

[0077] The hydrological data values ​​and dynamic warning thresholds in the preliminary flood forecast results are extracted at each time point. The two are compared one by one in chronological order. If the preliminary flood forecast data value at a certain time point is greater than the value of the dynamic warning threshold, the time point is directly marked as the threshold-exceeding time point. The specific time information of all threshold-exceeding time points is completely recorded to ensure that each threshold-exceeding time point is accurately marked without omission.

[0078] Elevation distribution data, river network data, and land use type data are extracted one by one from the consistent source data. The elevation distribution data contains the specific elevation values ​​and elevation gradient changes of each point in the target area. The river network data covers details such as the direction, branch width, depth, and connection relationships of rivers. The land use type data clarifies the distribution range and proportion of different land types such as cultivated land, forest land, building land, and water area in the region. The three types of data are linked and integrated according to their geographical coordinates to form a complete set of regional topographic and water system features that reflect the topography and water system of the target area.

[0079] Starting from the marked threshold time point, and combining the elevation distribution of the regional topographic and water system features, the area below the predicted water level at that time point is determined to establish the initial inundation boundary. Subsequently, the water level changes at fixed time intervals are extrapolated, and the expansion or contraction of the inundation boundary is adjusted according to the magnitude of water level rise and fall. At the same time, based on the water conveyance capacity of the river network and the topographic slope, the water depth of each inundation area is calculated. The water depth is determined by the difference between the current water level and the elevation of the area. The inundation boundary and corresponding water depth at different time points are recorded in a complete manner to obtain the flood impact range and depth of the target area.

[0080] The pre-set early warning level mapping database is a fixed correspondence database based on flood control standards and the degree of impact of historical disasters. It clearly defines the early warning levels corresponding to different combinations of flood impact range and water depth. The early warning levels are divided into four levels: blue, yellow, orange, and red. The larger the impact range and the deeper the water depth, the higher the corresponding early warning level. The actual flood impact range and water depth are compared one by one with the standard combinations in the mapping database to find the early warning level corresponding to the perfectly matching combination, thereby determining the early warning level of the target area.

[0081] Based on the chronological order of the time points exceeding the threshold, the flood impact range corresponding to each time point is integrated, and the information of each time point forms a complete record. The record clearly presents the details of the time warning level and the impact range. All records are arranged in chronological order to form a well-structured and complete flood warning information for the target area.

[0082] The beneficial effects are that by comparing each point to accurately mark the time points exceeding the threshold, integrating topographic and water system features to track the scope and depth of flood impact, and combining a preset mapping library to classify early warning levels, the final integrated flood early warning information includes key information such as time nodes, early warning levels, and scope of impact. The information is comprehensive, accurate, and logically clear, which can provide timely and effective reference for flood prevention and control decisions, help relevant departments accurately grasp the development trend of floods, take targeted prevention and control measures, and minimize the losses caused by flood disasters.

[0083] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0084] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A flood forecasting and early warning method based on artificial intelligence, characterized in that, The method includes: S1: Deviation data segments of the target area in the multi-source hydrological raw data, to obtain the same source consistent data of the target area; S2: Based on the inherent mapping relationship between different data in the same source consistent data, feature fusion is performed on the same source consistent data to obtain the fused feature data of the same source consistent data; S3: Integrate the trend information, periodic features and correlation patterns that evolve over time in the fused feature data into the time-series feature data of the target region; S4: Based on the real-time change trend and evolution rate in the time series feature data, evaluate the similarity and difference between the real-time hydrological features in the time series feature data and the historical critical hydrological features, so as to generate a dynamic early warning threshold adapted to the target area. S5: Based on the core influencing factors, time-series evolution characteristics, and real-time change patterns of the time-series characteristic data, the hydrological evolution trend of the target area is synergistically deduced to obtain the preliminary flood forecast results for the target area; S6: Based on the dynamic early warning threshold, the preliminary flood forecast results that meet the early warning triggering conditions are integrated with the regional features in the consistent data from the same source to form the flood early warning information for the target area.

2. The flood forecasting and early warning method based on artificial intelligence as described in claim 1, characterized in that, The deviation data segments of the target area in the multi-source hydrological raw data are corrected to obtain consistent data of the target area, including: Collect multi-source raw hydrological data related to flood forecasting in the target area; Noise filtering is performed on the multi-source hydrological raw data to obtain clean multi-source hydrological data for the target area. By unifying the storage format and description specifications of the multi-source hydrological clean data, standardized hydrological data for the target area can be obtained. By correcting the biased data segments in the standardized hydrological data, consistent data from the same source for the target area are obtained.

3. The flood forecasting and early warning method based on artificial intelligence as described in claim 1, characterized in that, The step of fusing features of the consistent data based on the inherent mapping relationship between different data in the consistent data to obtain fused feature data of the consistent data includes: Define the attribute descriptions and application scenarios corresponding to different data in the same source data to obtain a list of data attributes of the same source data; The data attribute list is mapped to preset element association rules to obtain a preliminary mapping relationship set of the consistent data from the same source. By removing false associations with no actual hydrological significance from the preliminary mapping relationship set, the effective intrinsic mapping relationship set of the consistent data from the same source is obtained; Based on the effective intrinsic mapping relationship set, the data that are related in the same source and consistent data are classified by features to obtain the related feature group of the same source and consistent data. The core attribute information in the associated feature group is integrated with the basic attribute data of the target region to form the fused feature data of the same source and consistent data.

4. The flood forecasting and early warning method based on artificial intelligence as described in claim 3, characterized in that, The process of integrating the core attribute information in the associated feature group with the basic attribute data of the target region to form the fused feature data of the same source and consistent data includes: The hydrological representation dimension of the core attribute information and the scenario adaptation dimension of the basic attribute data are decomposed, and the functional positioning of the core attribute information and the basic attribute data in the description of hydrological evolution is determined, generating a dimension function correspondence table of the homogeneous and consistent data. Based on the dimension function correspondence table, a bidirectional association mapping between the core attribute information and the basic attribute data is established to obtain the attribute association mapping set of the consistent data from the same source. Eliminate the differences in description perspective and expression standard of different attributes of the associated pairs in the attribute association mapping set to obtain the logically unified attribute association set of the consistent data from the same source; By mining the hidden hydrological synergistic patterns in the logically unified attribute association set, the association information that can jointly reflect the hydrological evolution characteristics of the target area is strengthened, and the synergistic strengthening attribute set of the consistent data from the same source is obtained. The core collaborative features of the collaborative enhancement attribute set are extracted and integrated into the fusion feature data of the homogeneous and consistent data.

5. The flood forecasting and early warning method based on artificial intelligence as described in claim 1, characterized in that, The process of integrating the trend information, periodic features, and correlation patterns that evolve over time from the fused feature data into the time-series feature data of the target region includes: By sorting out the temporal correlation clues of the fused feature data, clarifying the time record benchmarks corresponding to different types of features, and generating the temporal dimension identifier set of the fused feature data; Based on the time dimension identifier set, the feature content that shows a continuous change over time in the fused feature data is separated to obtain the trend feature subset of the fused feature data; Identify recurring feature patterns in the fused feature data to obtain a periodic feature subset of the fused feature data; The temporal relationship between the trend feature subset and the periodic feature subset is extracted to obtain the interaction relationship between the trend feature subset and the periodic feature subset; The interaction relationships are subjected to feature association logic condensation to obtain a subset of association patterns in the fused feature data; Using the time dimension identifier set as a unified benchmark, the time description specifications of the trend feature subset, the periodic feature subset, and the correlation pattern subset are calibrated to obtain the time-series feature data of the fused feature data.

6. The flood forecasting and early warning method based on artificial intelligence as described in claim 1, characterized in that, The step of evaluating the similarity and difference between real-time hydrological features and historical critical hydrological features in the time-series feature data based on the real-time change trend and evolution rate in the time-series feature data, in order to generate a dynamic early warning threshold adapted to the target area, includes: Extract the real-time change trend sequence and evolution rate sequence from the time-series feature data; Filter out historical critical hydrological feature sequences from the historical hydrological database of the target area that correspond to the geographical features and hydrological characteristics of the target area; The real-time trend sequence and the historical critical hydrological feature sequence are aligned on a time scale, and the degree of morphological matching between the real-time trend sequence and the historical critical hydrological feature sequence is identified. Assess the degree of deviation between the evolution rate sequence and the historical critical hydrological characteristic sequence; By integrating the matching degree and the deviation degree, and combining them with the real-time water level, rainfall and soil moisture data of the target area, a dynamic early warning threshold for the target area is obtained.

7. The flood forecasting and early warning method based on artificial intelligence as described in claim 6, characterized in that, The assessment of the degree of difference between the evolution rate sequence and the historical critical hydrological characteristic sequence includes: Noise filtering is applied to the evolution rate sequence and the historical critical hydrological feature sequence to obtain a smoothed evolution rate sequence and a smoothed historical sequence for the target region. According to the corresponding time points, the smoothed evolution rate sequence and the smoothed history sequence are differentially processed to construct the residual sequence of the target region; The frequency of residual data in the residual sequence is statistically analyzed within different numerical intervals to obtain the frequency of the residual data. The frequency of the residual data is then structured and regularized to obtain the residual frequency distribution of the residual sequence. Based on the skewness and kurtosis characteristics in the residual frequency distribution, the degree of deviation between the evolution rate sequence and the historical critical hydrological characteristic sequence is comprehensively quantified.

8. The flood forecasting and early warning method based on artificial intelligence as described in claim 7, characterized in that, The frequency of residual data in the residual sequence within different numerical intervals is statistically analyzed to obtain the frequency of the residual data. The frequency of the residual data is then structured and regularized to obtain the residual frequency distribution of the residual sequence, including: Based on the hydrological critical state reference standard corresponding to the target area, and combined with the hydrological evolution deviation characteristics reflected by the residual sequence, the residual classification intervals with clear hydrological significance in the residual sequence are divided to obtain the hydrological adaptation classification interval set of the residual sequence. Traverse the residual data in the residual sequence and, according to the interval definition rules of the hydrological adaptation classification interval set, assign the residual data to the corresponding classification interval to obtain the interval classification result of the residual data; By analyzing the residual data correlation information corresponding to different classification intervals in the interval classification results, the correlation characterization data of the interval classification results are obtained. The associated representation data is structured and organized to clarify the priority and interaction relationships of different classification intervals, thereby obtaining the residual frequency distribution of the residual sequence.

9. The flood forecasting and early warning method based on artificial intelligence as described in claim 1, characterized in that, The process involves collaboratively extrapolating the hydrological evolution trend of the target area based on the core influencing factors, temporal evolution characteristics, and real-time change patterns of the time-series characteristic data, to obtain preliminary flood forecast results for the target area, including: Principal component analysis was performed on the time-series feature data to select the features with the largest variance contribution rate from the time-series feature data, forming the core influencing factor set of the time-series feature data; Based on the real-time status of the core impact factor set, a status difference comparison is performed on the core impact factor set, and the recurring change paths in the status difference comparison results are extracted. Based on the aforementioned change path, the state transition pattern of the core influencing factor set over time is deduced to obtain the temporal evolution pattern of the target region; By identifying frequently co-occurring feature combinations in the time-series feature data, a set of real-time change patterns of the target region is obtained; Based on the core influencing factor set, time-series evolution pattern, and real-time change pattern set, the hydrological evolution trend of the target area is synergistically deduced; The hydrological evolution trend is de-standardized to obtain preliminary flood forecast results for the target area.

10. The flood forecasting and early warning method based on artificial intelligence as described in claim 1, characterized in that, The process of integrating the preliminary flood forecast results that meet the early warning triggering conditions with the regional characteristics in the consistent data based on the dynamic early warning threshold into flood early warning information for the target area includes: The preliminary flood forecast results are compared point by point with the dynamic early warning threshold, and the time points in the preliminary flood forecast results that exceed the threshold are marked. The elevation distribution, river network, and land use type features in the consistent data are integrated into a regional topographic and water system feature set for the target area. Using the time point as a time reference, the inundation boundary and water depth changes of the topographic and water system feature set of the region are tracked to obtain the flood impact range and depth of the target area; Based on a preset warning level mapping library, and combined with the flood impact range and the depth, the warning level of the target area is divided; The warning level, the flood impact range, and the time point are integrated into flood warning information for the target area.

Images