Hydrological information forecast data collection automatic processing method

Abstract

The invention relates to the technical field of big data processing, in particular to a hydrological information forecast data collection automatic processing method, which comprises the steps of multi-source data automatic acquisition and access, data quality intelligent diagnosis processing and abnormity early warning, multi-index comprehensive drought evaluation model construction and drought trend intelligent prediction. GIS spatial analysis and visual display are carried out, and drought evolution is dynamically reflected; and user-defined configuration and adaptive learning are carried out. The equipment state can be displayed, and related personnel can be prompted to maintain the observation station in time; existing hydrological data of the hydrological bureau are fully applied, rainfall, river flow, soil moisture content and other single items are normatively determined to determine the drought level, and meanwhile, according to multiple indexes, comprehensive drought judgment is performed according to the proportion, and based on the GIS technology, an interpolation method and a proximity association method are utilized to determine a Xinjiang whole administrative division plane drought map; and according to the analysis data, automatically sorting out soil moisture content brief report materials.

Description

Technical Field

[0001] This invention relates to the field of big data processing technology, and in particular to an automated processing method for hydrological information forecast data collection. Background Technology

[0002] Hydrological information forecasting refers to an applied science and technology that, based on the principle of hydrological cycle, comprehensively utilizes meteorological and hydrological monitoring data, and employs technologies such as hydrological models, mathematical statistics, and artificial intelligence, to monitor, comprehensively analyze, and predict the future trends of hydrological elements in rivers, lakes, and other water bodies in real time.

[0003] The existing technology involves sending real-time soil moisture data from 14 hydrological survey centers directly to the old platform, which only has data receiving and simple display functions. Because the Xinjiang Hydrological Bureau cannot understand the data reporting rate or data anomalies, the soil moisture data alone cannot provide useful drought decision-making analysis materials to the higher-level local governments. Summary of the Invention

[0004] The purpose of this invention is to propose an automated processing method for hydrological information forecasting data collection in order to solve the problems in the prior art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an automated processing method for hydrological information forecast data collection, comprising the following steps:

[0006] Step S1: Automated collection and access of multi-source data, collecting data from multiple sources including soil moisture, rainfall, river flow, equipment status, and vegetation information;

[0007] Step S2: Intelligent data quality diagnosis and anomaly warning. Automatically identify missing data, outliers, and duplicate data. Identify equipment anomalies based on historical data patterns and push warning information to relevant personnel in real time.

[0008] Step S3: Construction of dynamic drought assessment model. Combining soil moisture content, rainfall anomaly, and river flow interruption, a multi-index comprehensive drought assessment model is constructed to intelligently predict drought trends.

[0009] Step S4, GIS spatial analysis and visualization, uses spatial interpolation based on geographic information system to generate a drought distribution map of the whole region, an invalid rainfall distribution map, and a river flow interruption hotspot map, and supports time series animation display to dynamically reflect the evolution of drought.

[0010] Step S5, user-defined configuration and adaptive learning, allows users to dynamically configure site, crop type, growth stage, and evaluation parameters.

[0011] The beneficial effects of the technical solution provided by this invention include at least the following:

[0012] This invention uses intelligent data to diagnose and monitor the status of equipment, promptly reminding relevant personnel to perform station maintenance; it fully utilizes existing hydrological data from the hydrological bureau to standardize the determination of drought levels based on individual indicators such as rainfall, river flow, and soil moisture; it also comprehensively assesses drought conditions based on multiple indicators in proportion, and uses GIS technology with interpolation and proximity association methods to determine a drought map of the entire administrative region of Xinjiang Uygur Autonomous Region; based on the above analysis data, it automatically compiles soil moisture briefing materials.

[0013] This invention optimizes itself automatically, reducing manual user operations. It treats user configuration behavior as a correction signal for system settings, optimizes the drought prediction model based on a feedback model, establishes a feedback loop, records false alarm feedback, and stores all data features and user feedback labels corresponding to the warnings in a special dataset. This dataset is used periodically to incrementally train the prediction model, and the model will gradually learn the error-prone parts and correct its judgments in subsequent predictions. Attached Figure Description

[0014] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart illustrating the method steps provided in an embodiment of the present invention;

[0016] Figure 2 This is a flowchart of a method provided in an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an automated processing method for hydrological information forecasting data collection proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0020] The following description, in conjunction with the accompanying drawings, details the specific scheme of the automated processing method for hydrological information forecast data collection provided by this invention.

[0021] Example

[0022] Please see Figure 1 , Figure 2 The diagram illustrates a flowchart of an automated processing method for hydrological information forecast data collection according to an embodiment of the present invention. The method includes the following steps:

[0023] Step S1: Automated collection and access of multi-source data, collecting multiple data sources such as soil moisture, rainfall, river flow, equipment status, and vegetation information;

[0024] Step S1 further includes the following sub-steps:

[0025] S1-1, various data source acquisition methods: Automatic telemetry stations use miniature sensors based on FDR and TDR principles to periodically collect soil volumetric water content, gravimetric water content, and temperature data and report soil moisture data. Smart rain gauges use triggered reporting to collect rainfall information in a timely manner. Radar and pressure-type water level gauges comprehensively monitor water levels and convert water levels into flow rates. Local staff input vegetation information data such as crop type, planting area, and growth stage. Subsequently, vegetation index products from domestic and foreign satellites are obtained periodically through web crawling. Based on land use data, the system automatically delineates desert and arid areas, as well as relevant information such as woodlands, grasslands, and cultivated land in plains. Based on the zoning scope, plains and some grassy desert areas are identified as specific areas for drought prediction and analysis. The assessment focus can be freely adjusted to focus on specific areas, such as ecologically sensitive areas or agricultural activity areas.

[0026] S1-2, Equipment Status Data Detection: The data acquisition unit actively reports its own status information, such as battery voltage, signal strength, internal temperature, and operating status code, to obtain the status of the data acquisition equipment in complex and harsh environments. The equipment status data is correlated with the observation data. If the voltage of a certain station is too low, an abnormal status is reported.

[0027] Step S2: Intelligent data quality diagnosis and anomaly warning. Automatically identify missing data, outliers, and duplicate data. Identify equipment anomalies based on historical data patterns and push warning information to relevant personnel in real time.

[0028] Step S2 further includes the following sub-steps:

[0029] S2-1 performs high-speed, low-latency initial screening based on predefined rules. It applies hydrological knowledge and equipment parameters, sets absolute physical boundaries and simple logical rules, and is implemented based on a stream processing framework. In the stream processing, a heartbeat window of the last reported time is maintained for each station. If the current time minus the last reported time is greater than a preset threshold, a data missing warning is triggered. It judges whether the data value exceeds the possible physical range. If it does, a physical range anomaly is triggered. It judges whether the timestamps and values ​​of multiple consecutive data are exactly the same. If they are the same, duplicate data is triggered. It judges whether the instantaneous rate of change exceeds the reasonable physical process. If it triggers a sudden change anomaly, there may be equipment damage. It is necessary to eliminate the impact in time or start the backup acquisition equipment.

[0030] S2-2, analyze the data sequence within a short time window, capture abnormal patterns that do not conform to historical statistical regularities, use window functions for real-time calculation, use sliding windows to calculate the mean and standard deviation, if the current point exceeds the mean ± 3 times the standard deviation, it is considered an anomaly, and use median and absolute median deviation statistics to avoid multiple anomaly points from raising the σ value;

[0031] S2-3 identifies and detects abnormal equipment status by constructing a multi-dimensional feature vector, which is then fed into an isolated forest model for training and prediction. This model achieves unsupervised learning and excels at identifying highly scattered anomalies. Anomalies are defined as points that are easily isolated, requiring only a small amount of random partitioning to isolate them from the data space. It outputs an anomaly score for each data point, with scores closer to 1 indicating a higher probability of an anomaly. No labeled data is required, and it automatically discovers unknown types of equipment failure modes. An LSTM autoencoder is trained using a large amount of normal historical data. The LSTM autoencoder can perform supervised or unsupervised learning. The encoder compresses the input sequence into a low-dimensional code, and the decoder reconstructs the original sequence based on this code, training a model that minimizes the reconstruction error.

[0032] In the prediction phase, a new sequence is input and reconstructed using a trained autoencoder. The reconstruction error is calculated. If the reconstruction error is much higher than the normal level during training, it indicates that the pattern of the current sequence is something the model has never seen before, and therefore it is judged as an anomaly. It can capture extremely complex nonlinear spatiotemporal dependencies in the data and is extremely sensitive to gradual and trend anomalies.

[0033] The integrated decision center enables customized decision-making, employing weighted voting or fuzzy logic systems to calculate a comprehensive anomaly confidence level, adapting to judgments based on varying degrees of importance.

[0034] CAC=α*I r +β*I s +γ*I a

[0035] Where CAC is the comprehensive anomaly confidence score, α, β, and γ are the weights of each engine layer, which can be dynamically adjusted based on historical accuracy. r I s I a The outputs are from the rules engine, statistics engine, and AI engine, respectively, and intelligent hierarchical early warning pushes are made based on CAC and anomaly type.

[0036] Level 1 Warning (Alert): The CAC is low, which may be an occasional error. The message will be sent to the station-level maintenance personnel via WeChat and SMS.

[0037] Level 2 Warning (Alert): CAC is moderate, and the situation remains abnormal. In addition to being pushed to general warning personnel, it also includes municipal and prefecture-level management personnel and application system interfaces.

[0038] Level 3 Warning (Severe): High CAC, or involves key sites and parameters, pushes to not only Level 1 and Level 2 warning personnel, but also provincial decision-makers and triggers telephone voice alarm;

[0039] The early warning information not only includes specific details, but also the basis and the probability of accuracy of the prediction, such as "Site A, soil moisture sensor suspected of malfunction, accuracy rate 90%".

[0040] Basis: "The data remained unchanged for 12 hours, and the LSTM reconstruction error was 15 times the average value."

[0041] This solution achieves hierarchical detection, fusion decision-making, and intelligent early warning. It integrates traditional metrological methods, time series analysis, and machine learning algorithms at multiple levels, and finally outputs diagnostic results and early warning information through an interpretable decision engine. The weighted fusion of multiple detection results balances speed, breadth, and depth, resulting in higher reliability. It innovatively integrates equipment status data and observation data, and uses unsupervised learning to directly diagnose equipment anomalies. Anomaly detection based on deep time series prediction can discover complex and progressive anomaly patterns that traditional methods cannot identify. It is technologically advanced. The final early warning information is not a single conclusion, but a detailed report integrating evidence chains at each level, which improves the credibility and operability of the early warning. The system allows users to confirm or deny the early warning results. These feedback data will be recorded and used to periodically retrain the AI ​​model, making the system increasingly intelligent over time, forming a virtuous cycle, and constituting an intelligent data quality assurance system with self-evolution capabilities.

[0042] S2-4, Data Quality Report Generation and Monitoring: Automatically calculates data arrival rate, missing rate, anomaly rate, and other indicators for each site, generating daily, monthly, and annual reports. Based on a stream processing framework, it calculates the site's reporting heartbeat in real time. If the current time minus the last reporting time is greater than a threshold, a data missing warning is triggered and recorded in the report. A visual interface is provided, allowing users to query statistical information for different time periods. It supports exporting and printing reports, and the report data helps support equipment maintenance decisions, such as prioritizing maintenance for low-voltage sites.

[0043] Step S3: Construction of dynamic drought assessment model. Combining soil moisture content, rainfall anomaly, and river flow interruption, a multi-index comprehensive drought assessment model is constructed to intelligently predict drought trends.

[0044] Step S3 further includes the following sub-steps:

[0045] S3-1 transforms raw data into dimensionless, comparable standardized indicators, including soil relative humidity (SRH), rainfall anomaly percentage (PAP), and river flow interruption index (RDI). The dynamic comprehensive assessment classifies each indicator into different drought levels according to national and industry standards, including no drought, mild drought, moderate drought, severe drought, and extreme drought, and assigns a score to each level, ranging from 1 to 5 points. The higher the score, the more severe the drought. Users can customize the level thresholds, acknowledging that the importance of different indicators varies in different seasons and regions.

[0046] S3-2 uses the Analytic Hierarchy Process (AHP) combined with user-defined parameters to dynamically generate weights, based on a weight template derived from AHP and expert experience:

[0047] Spring or sowing season: The key to spring sowing is soil moisture, with SRH having the highest weight, followed by PAP, and RDI having the lowest.

[0048] Main flood season: Rainfall is the dominant factor during the flood season, with PAP having the highest weight;

[0049] Mountainous areas vs. plains: Mountainous areas rely more on rainfall, so PAP has a higher weight; Plains have irrigation, so SRH and RDI have higher weights.

[0050] Provincial and municipal users can drag sliders through a visual interface to adjust the weights of the three major indicators in real time. When the sum of the weight ratios equals 1, the system immediately recalculates the overall drought situation according to the new weights, which greatly enhances the model's adaptability and decision support capabilities.

[0051] S3-3, calculated based on the aforementioned Comprehensive Drought Index (CDI):

[0052] CDI = W s *S s +W p *S p +Wr *S r

[0053] Among them, S s S p S r These represent the drought severity scores corresponding to soil relative humidity, rainfall anomaly, and river flow interruption index, respectively, while W s W p W r The dynamic weights determined in the previous step are mapped back to the final comprehensive drought level based on the calculated CDI total score.

[0054] S3-4, Establish the soil moisture balance equation. By inputting weather forecast data for the next 72 hours, short-term predictions of SRH can be made.

[0055] θ t =θ {t-1} +P-ET-R

[0056] Where, θ t θ represents the soil moisture content at a future point in time. {t-1} P represents the current water content, ET represents the forecast rainfall, R represents the evapotranspiration, and R represents the runoff. Using a long short-term memory network time-series prediction model, drought assessment is treated as a time-series prediction problem. The model learns complex patterns in historical data and makes medium- to long-term intelligent predictions based on machine learning to predict future CDI or key indicators. The input historical time-series data includes SRH, PAP, RDI, CDI, temperature, evaporation, and other data for the past 90 days. The final output is the CDI value or drought level for the next 7 days and 15 days.

[0057] Unlike common indicator overlays, AI learns the entire dynamic pattern of drought occurrence, development, and decline from massive historical data, and can incorporate weather forecast information as future signals into the model, enabling it to output predictions with clear physical meaning and higher accuracy. This is a medium- to long-term intelligent prediction based on machine learning.

[0058] S3-5, Automatic Drought Report Generation: Based on the analysis results of the Comprehensive Drought Index (CDI), ineffective rainfall distribution, and river flow interruption hotspot maps, the system automatically generates a soil moisture report according to a preset template. The report includes drought level distribution, key areas of concern, and trend forecasts. It supports multi-dimensional statistics by administrative region, watershed, and crop type. Users can customize the report template, and the system automatically fills in the data according to the user configuration. The report is pushed to the specified user via API, and supports the scheduled generation of daily, monthly, and annual reports.

[0059] Step S4, GIS spatial analysis and visualization, uses geographic information system to perform spatial interpolation to generate a drought distribution map of the whole region, an invalid rainfall distribution map, and a river flow interruption hotspot map, and supports time series animation display to dynamically reflect the evolution of drought;

[0060] Step S4 further includes the following sub-steps:

[0061] S4-1 establishes a regression model describing the non-stationary relationship between the target variable and geographical environmental factors. Then, the residuals of the regression model, i.e., the portion unexplained by the model, are interpolated. Finally, the two are superimposed to overcome the spatial limitations of traditional interpolation methods, improve the understanding of the influence of geographical elements on hydrological and meteorological elements, and enhance the accuracy in sparsely populated areas of the stations.

[0062] Z(si)=β0(s i )+β1(s i )*Y1(s i )+β2(s i )*Y2(s i )+...+ε(s i )

[0063] Where Z(si) is the observation value at station i, and Y1(s) is the observation value at station i. i ),Y2(s i ... represents the geographical factors of station i, such as elevation, slope, NDVI, β1(s) i ),β2(s i ...indicates that the position s varies with spatial location. i The changing regression coefficients reflect the non-stationarity of the relationship; ε(s i ) represents the regression residuals. Interpolation is performed on the regression residuals, and ordinary kriging is used to interpolate the residuals ε(s). i Interpolation is performed to obtain the residual distribution surface ε(s0) for the entire region, and the final predicted surface is calculated:

[0064]

[0065] Wherein, [β0(s0)+β1(s0)*Y1(s0)+...] represents the trend term calculated using the GWR model, reflecting geographical control factors, and ε(s0) is the residual distribution surface of the entire region, reflecting local variations that the trend term cannot capture. Geographically weighted regression-based residual interpolation incorporates physical mechanisms into the interpolation process, making it more interpretable than purely mathematical interpolation methods and significantly improving accuracy in complex terrain areas.

[0066] S4-2 treats the interpolation result raster map of each day as a time slice, stacks them in chronological order to form a three-dimensional spatiotemporal cube, and the server renders the raster data of each time slice as a PNG image tile with corresponding numerical range metadata. The front end requests a time series tile and plays it in a fast loop according to the timestamp order to form an animation effect. The animation can be paused, the progress bar can be dragged, and the playback speed can be adjusted.

[0067] Users can draw lines on the map, and the system immediately generates a profile map of distance, time, and drought intensity, clearly showing the evolution of drought along a path. By comparing the differences between adjacent time slices, the system can automatically identify areas where drought is rapidly intensifying and mark them as "key areas of concern," pushing them to the administrator. Introducing GWR residual interpolation, which uses geographical environmental factors as driving variables, significantly improves the interpolation accuracy and physical rationality in sparse areas of the site, surpassing the limitations of traditional IDW and other methods. Through analysis tools such as heat maps and spatiotemporal profiles, the system guides users to discover patterns and identify problems from the data, achieving a leap from "being able to see" to "being able to understand."

[0068] S4-3, Thematic Map Generation and Representation: The interpolation results of the Comprehensive Drought Index (CDI) are used to output a drought distribution map for the entire region, rendered with gradient color bands, where dark to light colors represent extreme drought to no drought. An invalid rainfall distribution map is calculated and output, represented by a single color with varying shades; darker colors indicate more invalid rainfall and poorer drought resistance. Kernel density estimation is used to treat each interrupted river section as an "event point," calculating the density of interrupted flow events within a unit area around any location in the entire region to create a river interruption hotspot map, clearly displaying the most concentrated hotspot areas of interrupted flow. GIS is upgraded from a static electronic map system to a powerful, interactive, and future-oriented spatiotemporal analysis tool, transforming visualization from a post-event display tool to a real-time early warning tool, achieving a closed loop from analysis to application. Simultaneously, intelligent reports and briefings are statistically analyzed from multiple dimensions, including administrative regions, watersheds, and crop types, automatically generating daily, monthly, and annual reports based on preset templates, and automatically pushing different reports to designated users.

[0069] S4-4 provides a soil moisture information query interface, allowing users to select sites and time ranges to display real-time data on soil volumetric water content, gravimetric water content, temperature, and voltage. It generates a moisture content process line chart and, combined with rainfall data tables, displays historical trends. It also provides a function to compare soil moisture data from manual and automatic monitoring. Users can select points on the map or query through a form to display a distribution map showing the differences between manual and automatic monitoring data. Clicking on a map point will bring up a detailed data table supporting soil moisture point map display, invalid rainfall statistics, river flow interruption statistics, and rainfall anomaly distribution. Users can dynamically switch views using a time slider.

[0070] Step S5, User-defined configuration and adaptive learning, allows users to dynamically configure site, crop type, growth stage, and evaluation parameters;

[0071] Step S5 further includes the following sub-steps:

[0072] S5-1 allows users to manage monitoring stations and crops in a customizable manner, dynamically adding, editing, and deactivating monitoring stations. It enables users to dynamically add, edit, and deactivate monitoring stations, customize exclusive station data based on GIS map point selection and form filling, and establish a database of relationships between stations, crops, and growth stages. Each station is bound to the crop type and its current growth stage, combining ground observation with crop observation to lay the foundation for subsequent accurate drought assessment.

[0073] S5-2, the assessment parameters are customizable. It provides a drought assessment parameter template editor, allows for the customization of key thresholds based on different crop growth stage combinations, a visual rule editor, and the generation of configuration template files in JSON or XML format. Local knowledge and experience are quantified and solidified into the system, making the assessment standards more in line with reality.

[0074] S5-3, linked with the comprehensive drought index, provides an intuitive slider interface that allows users to adjust the weights of three indicators—soil moisture, rainfall anomaly, and river flow interruption—in real time during the comprehensive drought index calculation. The new weights are pushed to the evaluation model instantly via API, enabling users to make strategic adjustments based on different seasons and decision-making focuses.

[0075] S5-4 enables the system to learn from user configuration behavior, automatically optimize itself, reduce manual user operations, and treat user configuration behavior as a correction signal for system settings. Based on the feedback model, the drought prediction model is optimized and a feedback loop is established. If a user believes that a certain warning is inaccurate, they can click the false alarm or missed alarm button. The system records the feedback and stores all data features and user feedback labels corresponding to the warning in a special dataset. This dataset is used periodically to incrementally train the prediction model. The model will gradually learn the error-prone parts and correct its judgment in subsequent predictions.

[0076] In this way, an automated processing method for hydrological information forecasting data collection can be achieved.

[0077] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. An automated processing method for hydrological information forecast data collection, characterized in that, The method includes: Step S1: Automated collection and access of multi-source data, collecting data from multiple sources including soil moisture, rainfall, river flow, equipment status, and vegetation information; Step S2: Intelligent data quality diagnosis and anomaly warning. Automatically identify missing data, outliers, and duplicate data. Identify equipment anomalies based on historical data patterns and push warning information to relevant personnel in real time. Step S3: Construction of dynamic drought assessment model. Combining soil moisture content, rainfall anomaly, and river flow interruption, a multi-index comprehensive drought assessment model is constructed to intelligently predict drought trends. Step S4, GIS spatial analysis and visualization, uses spatial interpolation based on geographic information system to generate a drought distribution map of the whole region, an invalid rainfall distribution map, and a river flow interruption hotspot map, and supports time series animation display to dynamically reflect the evolution of drought. Step S5, user-defined configuration and adaptive learning, allows users to dynamically configure site, crop type, growth stage, and evaluation parameters.

2. The automated processing method for hydrological information forecast data collection as described in claim 1, characterized in that: Step S1 further includes the following sub-steps: S1-1, various data source acquisition methods: Automatic telemetry stations use miniature sensors based on FDR and TDR principles to periodically collect soil volumetric water content, gravimetric water content, and temperature data and report soil moisture data. Smart rain gauges use triggered reporting to collect rainfall information in a timely manner. Radar and pressure water level gauges comprehensively monitor water levels and convert water levels into flow rates. Local staff input vegetation information data such as crop type, planting area, and growth stage. Subsequently, vegetation index products from domestic and foreign satellites are obtained periodically through web crawlers. Based on land use data, the system automatically delineates desert and arid areas, as well as relevant information such as woodlands, grasslands, and cultivated land in plains. Based on the zoning scope, plains and some grassy desert areas are identified as specific areas for drought prediction and analysis. The system can freely adjust the focus on specific areas, such as ecologically sensitive areas and agricultural activity areas. S1-2, Equipment Status Data Detection: The data acquisition unit actively reports its own status information, such as battery voltage, signal strength, internal temperature, and operating status code, to obtain the status of the data acquisition equipment in complex and harsh environments. The equipment status data is correlated with the observation data. If the voltage of a certain station is too low, an abnormal status is reported.

3. The automated processing method for hydrological information forecast data collection as described in claim 1, characterized in that: Step S2 further includes the following sub-steps: S2-1 performs high-speed, low-latency initial screening based on predefined rules. It applies hydrological knowledge and equipment parameters, sets absolute physical boundaries and simple logical rules, and is implemented based on a stream processing framework. In stream processing, a heartbeat window of the last reported time is maintained for each station. If the current time minus the last reported time is greater than a preset threshold, a data missing warning is triggered. It judges whether the data value exceeds the possible physical range. If it does, a physical range anomaly is triggered. It judges whether the timestamps and values ​​of multiple consecutive data are exactly the same. If they are the same, duplicate data is triggered. It judges whether the instantaneous rate of change exceeds the reasonable physical process and triggers a sudden change anomaly. S2-2, analyze the data sequence within a short time window, capture abnormal patterns that do not conform to historical statistical regularities, use a sliding window to calculate the mean and standard deviation, and if the current point exceeds the mean ± 3 times the standard deviation, it is considered an anomaly; S2-3, Identify and detect abnormal equipment status, construct a multi-dimensional feature vector, and feed it into the Isolation Forest model for training and prediction. This model achieves unsupervised learning and is good at identifying high-scatter anomalies. Anomalies are defined as points that are easily isolated. They are isolated from the data space with a small amount of random splitting. An LSTM autoencoder is trained using a large amount of normal historical data. The LSTM autoencoder can perform supervised or unsupervised learning. The encoder compresses the input sequence into a low-dimensional code, and the decoder reconstructs the original sequence based on the code. The model that minimizes the reconstruction error is trained. The integrated decision-making center enables customized decision-making, employing weighted voting or fuzzy logic systems to calculate a comprehensive anomaly confidence level, adapting to guidance judgments of varying importance. CAC=α*I r +β*I s +γ*I a Where CAC is the comprehensive anomaly confidence score, α, β, and γ are the weights of each engine layer, which can be dynamically adjusted based on historical accuracy. r I s I a The outputs are from the rules engine, statistics engine, and AI engine, respectively, and intelligent hierarchical early warning pushes are performed based on CAC and anomaly type. S2-4, Data Quality Report Generation and Monitoring: Automatically calculates data arrival rate, missing rate, anomaly rate, and other indicators for each site, generating daily, monthly, and annual reports. Based on a stream processing framework, it calculates the site's reporting heartbeat in real time. If the current time minus the last reporting time is greater than a threshold, a data missing warning is triggered and recorded in the report. A visual interface is provided, allowing users to query statistical information for different time periods. It supports exporting and printing reports, and the report data helps support equipment maintenance decisions, such as prioritizing maintenance for low-voltage sites.

4. The automated processing method for hydrological information forecast data collection as described in claim 1, characterized in that: Step S3 further includes the following sub-steps: S3-1 transforms the raw data into dimensionless, comparable standardized indicators, including soil relative humidity (SRH), rainfall anomaly percentage (PAP), and river flow interruption index (RDI). The dynamic comprehensive assessment classifies each indicator into different drought levels according to national and industry standards, including no drought, mild drought, moderate drought, severe drought, and extreme drought, and assigns a score to each level, ranging from 1 to 5 points. The higher the score, the more severe the drought. Users can customize the level threshold. S3-2 uses the Analytic Hierarchy Process (AHP) combined with user-defined weights to dynamically generate weights. Based on the AHP and expert experience, different levels of management users can drag sliders through a visual interface to adjust the weights of the three major indicators in real time. The system immediately recalculates the overall drought situation according to the new weights, which greatly enhances the model's adaptability and decision support capabilities. S3-3, calculated based on the aforementioned Comprehensive Drought Index (CDI): CDI=W s *S s +W p *S p +W r *S r Among them, S s S p S r These represent the drought severity scores corresponding to soil relative humidity, rainfall anomaly, and river flow interruption index, respectively, while W s W p W r The dynamic weights determined in the previous step are mapped back to the final comprehensive drought level based on the calculated CDI total score. S3-4, Establish the soil moisture balance equation. By inputting weather forecast data for the next 72 hours, short-term predictions of SRH can be made. θ t =θ {t-1} +P-ET-R Where, θ t θ represents the soil moisture content at a future point in time. {t-1} P represents the current water content, ET represents the forecast rainfall, R represents the evapotranspiration, and R represents the runoff. Using a long short-term memory network time-series prediction model, drought assessment is treated as a time-series prediction problem. The model learns complex patterns in historical data and makes medium- to long-term intelligent predictions based on machine learning to predict future CDI or key indicators. The input historical time-series data includes past SRH, PAP, RDI, CDI, temperature, evapotranspiration, and other data. The final output is the future CDI value or drought level. S3-5 Drought Briefing Generation: Based on the analysis results of the Comprehensive Drought Index (CDI), ineffective rainfall distribution, and river flow interruption hotspot map, a drought briefing is automatically generated according to a preset template. The briefing content includes the drought level distribution, key areas of concern, and trend forecast.

5. The automated processing method for hydrological information forecast data collection as described in claim 1, characterized in that: Step S4 further includes the following sub-steps: S4-1 establishes a regression model describing the non-stationary relationship between the target variable and geographical environmental factors. Then, the residuals of the regression model, i.e., the portion unexplained by the model, are interpolated. Finally, the two are superimposed to overcome the spatial limitations of traditional interpolation methods, improve the understanding of the influence of geographical elements on hydrological and meteorological elements, and enhance the accuracy in sparsely populated areas of the stations. Z(si)=β0(s i )+β1(s i )*Y1(s i )+β2(s i )*Y2(s i )+...+ε(s i ) Where Z(si) is the observation value at station i, and Y1(s) is the observation value at station i. i ),Y2(s i ... represents the geographical factors of station i, such as elevation, slope, NDVI, β1(s) i ),β2(s i ...indicates that the position s varies with spatial location. i The changing regression coefficients reflect the non-stationarity of the relationship; ε(s i ) represents the regression residuals. Interpolation is performed on the regression residuals, and ordinary kriging is used to interpolate the residuals ε(s). i Interpolation is performed to obtain the residual distribution surface ε(s0) for the entire region, and the final predicted surface is calculated: Wherein, [β0(s0)+β1(s0)*Y1(s0)+...] represents the trend term calculated using the GWR model, reflecting geographic control factors, and ε(s0) is the residual distribution surface of the entire region, reflecting local variations that the trend term cannot capture. Geographically weighted regression-based residual interpolation incorporates physical mechanisms into the interpolation process, making it more interpretable than purely mathematical interpolation methods and significantly improving accuracy in complex terrain regions. S4-2 treats the interpolation result raster map of each day as a time slice, stacks them in chronological order to form a three-dimensional spatiotemporal cube, and the server renders the raster data of each time slice as a PNG image tile with corresponding numerical range metadata. The front end requests a time series tile and plays it in a fast loop according to the timestamp order to form an animation effect. The animation can be paused, the progress bar can be dragged, and the playback speed can be adjusted. Users can draw lines on the map, and the system will immediately generate a profile map of distance, time, and drought intensity, clearly showing the evolution of the drought along a path. By comparing the differences between adjacent time slices, the system can automatically identify areas where the drought is rapidly intensifying and mark them as "key areas of concern". S4-3, Thematic map generation and representation: The interpolation results of the comprehensive drought index (CDI) are used to output a drought distribution map of the whole region. Gradient color bands are used for rendering. Kernel density estimation is used to treat each interrupted river section as an event point. The density of interrupted events within a unit area around any location in the whole region is calculated to establish a river interruption hotspot map. S4-4 provides a soil moisture information query interface, allowing users to select sites and time ranges to display real-time data on soil volumetric water content, gravimetric water content, temperature, and voltage. It generates a moisture content process line chart and, combined with rainfall data tables, displays historical trends. It also provides a function to compare soil moisture data from manual and automatic monitoring. Users can select points on the map or query through a form to display a distribution map showing the differences between manual and automatic monitoring data. Clicking on a map point will bring up a detailed data table supporting soil moisture point map display, invalid rainfall statistics, river flow interruption statistics, and rainfall anomaly distribution. Users can dynamically switch views using a time slider.

6. The automated processing method for hydrological information forecast data collection as described in claim 1, characterized in that: Step S5 further includes the following sub-steps: S5-1 allows users to manage monitoring stations and crops in customizable dimensions, dynamically adding, editing, and deactivating monitoring stations. It uses GIS-based map point selection and form filling to establish a database of relationships between stations, crops, and growth stages. Each station is bound to its crop type and current growth stage. By combining ground observation with crop observation, it lays the foundation for subsequent accurate drought assessment. S5-2, the assessment parameters are customizable. It provides a drought assessment parameter template editor, allows for the customization of key thresholds based on different crop growth stage combinations, a visual rule editor, and the generation of configuration template files in JSON or XML format. Local knowledge and experience are quantified and solidified into the system, making the assessment standards more in line with reality. S5-3, linked with the comprehensive drought index, provides an intuitive slider interface that allows users to adjust the weights of three indicators—soil moisture, rainfall anomaly, and river flow interruption—in real time during the comprehensive drought index calculation. The new weights are pushed to the evaluation model instantly via API, enabling users to make strategic adjustments based on different seasons and decision-making focuses. S5-4 enables the system to learn from user configuration behavior, automatically optimize itself, reduce manual user operations, and treat user configuration behavior as a correction signal for system settings. Based on the feedback model, the drought prediction model is optimized and a feedback loop is established. If a user believes that a certain warning is inaccurate, they can click the false alarm or missed alarm button. The system records the feedback and stores all data features and user feedback labels corresponding to the warning in a special dataset. This dataset is used periodically to incrementally train the prediction model. The model will gradually learn the error-prone parts and correct its judgment in subsequent predictions.

Images