A method and system for constructing a short and medium term hydrological forecast knowledge base for a river basin

Abstract

The application provides a method and system for constructing a prediction knowledge base for short and medium term hydrological prediction of a river basin, belonging to the technical field of hydrological information prediction and hydrological information system construction. The method takes prediction knowledge units as the basic organizational units, collects and merges multiple sources of prediction related information, combines the river basin ontology model and multi-dimensional labels, completes data cleaning, structured modeling and text vectorization coding, constructs a hybrid index with structured, semantically similar and time series retrieval, and realizes dynamic maintenance of prediction knowledge through quality evaluation, version management and incremental updating. The system includes seven modules such as data acquisition and knowledge unit extraction, ontology modeling and label management. The application realizes unified extraction and rapid retrieval of prediction knowledge, improves the efficiency of case reuse and scheme comparison, and supports fine management and systematic decision-making of short and medium term hydrological prediction of a river basin. The application is suitable for knowledge management and decision support in the field of river basin hydrological prediction, and can be connected to external systems such as large language models to realize intelligent application.

Description

Technical Field

[0001] This invention belongs to the technical field of hydrological information forecasting and hydrological information system construction. Specifically, it relates to a method and system for constructing a forecast knowledge base for short- and medium-term hydrological forecasting in river basins. It is particularly suitable for knowledge management, case reuse, and decision support in short- and medium-term hydrological forecasting operations in river basins. It can be connected to intelligent external systems to realize intelligent retrieval and application of forecast knowledge. Background Technology

[0002] Short- and medium-term hydrological forecasting is a core technological foundation for river basin flood control and disaster reduction, joint operation of reservoir groups, optimal allocation of water resources, and safe and economical operation of hydropower stations. With the continuous advancement of water conservancy projects, large-scale water conservancy hubs and cascade hydropower stations have been built and put into operation. River basin operation targets are becoming increasingly diverse, and constraints are constantly increasing. Forecasting operations are characterized by multi-basin collaboration, multi-objective constraints, parallel operation of multiple models, and high-frequency rolling updates, placing higher technical demands on the refined management of forecasting schemes and the systematic utilization of experience and results.

[0003] The current management of short- and medium-term hydrological forecasting information in the basin suffers from numerous technical shortcomings. The core issues are multi-source and decentralized management, coarse-grained management, and a lack of evolutionary mechanisms. (1) Information is scattered and lacks connection: The measured data of the hydrological and rainfall automation system, the model operation results of the hydrological database, the forecast schemes and rainfall analysis in the reports / consultation minutes, the parameter calibration results and forecast accuracy assessment data, etc. are managed independently by different systems or personnel, lacking a unified organization and connection; the forecast schemes, scheduling procedures, etc. are mostly in natural language or semi-structured form, making it difficult to achieve rapid retrieval and horizontal comparison of multi-dimensional conditions, and lacking the connection of the whole chain of information around the "forecast task", which cannot reflect the internal connection of forecast knowledge.

[0004] (2) Coarse management granularity and inefficient retrieval: With the expansion of forecast objects and model systems, the number of forecast schemes and rainstorm cases has increased rapidly. The traditional coarse-grained management method based on watershed, time or folder cannot meet the needs of flexible retrieval and combination analysis based on multiple dimensions such as watershed level, business scenario, model applicability, and forecast effect. It is not conducive to the migration and reuse of existing experience knowledge in new watersheds and new working conditions.

[0005] (3) Insufficient dynamic evolution capability of knowledge: For the same watershed and control section, under different water inflow years and scheduling objectives, the model combination, parameter configuration and scheduling strategy need to be adjusted multiple times. However, the existing technology lacks a quality evaluation and version management mechanism for forecast knowledge, and there is no standardized knowledge writing method, which leads to untimely knowledge updates and the inability to form a traceable and iterative benign evolution.

[0006] In summary, existing technologies lack a dedicated knowledge base construction technology for short- and medium-term hydrological forecasting operations in watersheds. This makes it impossible to achieve unified extraction, structured modeling, and correlation management of multi-source forecast data, and also makes it difficult to support multi-dimensional retrieval, systematic review, and business-driven incremental updates. This restricts the refined management and systematic decision-making of short- and medium-term hydrological forecasting operations in watersheds. Summary of the Invention

[0007] The purpose of this invention is to address the problems in existing short- and medium-term hydrological forecasting operations in watersheds, such as the scattered storage of forecast schemes, rainfall and flood cases, scheduling procedures, parameter configurations, and evaluation results, the lack of unified modeling and correlation management, and the difficulty in supporting multi-dimensional conditional retrieval, systematic review, and continuous updates. This invention provides a method and system for constructing a forecast knowledge base for short- and medium-term hydrological forecasting in watersheds, enabling unified extraction of forecast knowledge, structured modeling, hybrid index management, and incremental updates. This provides a stable and reliable knowledge foundation for the refined management and comprehensive decision-making of short- and medium-term hydrological forecasting operations in watersheds.

[0008] In order to achieve the above-mentioned objectives, the present invention: The first aspect provides a method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting in watersheds, including the following steps: S1: Collect multi-source operational data related to short- and medium-term hydrological forecasts in the basin and group them into forecast knowledge units with unified identification and clear boundaries according to preset rules; S2: Construct a knowledge ontology and multi-dimensional tagging system based on the spatial hierarchy of the watershed, and associate forecast knowledge units with ontology entities and tags; S3: Standardize the raw operational data of the forecast knowledge unit, parse key information to generate structured records with a unified data model; S4: Standardize the textual information in the structured records and generate vector representations of the forecast knowledge units through the domain text vector model; S5: Construct structured retrieval indexes, vector retrieval indexes, and time-series retrieval indexes based on structured fields, vector representations, and time fields respectively, and establish the relationships between the indexes to form a hybrid retrieval forecast knowledge base; S6: Score the quality of forecast knowledge units based on multi-dimensional indicators, and record the version and mark the validity of updated knowledge units; S7: Map newly generated data from the short- and medium-term hydrological forecasting operations of the basin to the newly added or updated content of the forecasting knowledge unit, and execute the corresponding operations of steps S2 to S6 to realize online incremental updates and write-back of the knowledge base.

[0009] Preferably, step S1 specifically includes: S11: Collect at least one type of multi-source operational data related to at least one target watershed, including historical hydrological forecasting schemes, typical rainfall and flood process records, scheduling procedures, model parameter configurations, forecast accuracy evaluation results, and consultation minutes. S12: Based on the preset division rules, the operational data corresponding to the same forecast task, the same flood process, the same dispatching procedure, the same set of parameter calibration results, or the same forecast evaluation record are grouped into a forecast knowledge unit, and a unique identifier is assigned to each forecast knowledge unit.

[0010] Preferably, step S2 specifically includes: S21: Based on the spatial hierarchical relationship between watershed, river network, reservoir, control section, and hydrological station, define watershed entities, reservoir entities, control section entities, hydrological station entities, forecasting scheme entities, and stormwater event entities, as well as their interrelationships; S22: Design a multi-dimensional tag system that includes spatial dimension, time dimension, business scenario, model applicability and forecast effect, and associate at least one of the aforementioned tags with each forecast knowledge unit.

[0011] Preferably, step S3 specifically includes: S31: Standardize the format, complete the fields, and handle outliers of the original business data corresponding to the forecast knowledge unit; standardize the time information, spatial location, watershed name, and reservoir name. S32: Extract key information such as forecast objects, forecast period range, model type, scheduling measures and evaluation indicators from the forecast knowledge unit, and generate corresponding structured fields based on the forecast knowledge ontology to form a structured record that conforms to a unified data model.

[0012] Preferably, step S4 specifically includes: S41: Perform word segmentation, terminology standardization, and domain entity recognition on the text fields of scheme descriptions, consultation conclusions, and evaluation opinions contained in the structured records to obtain standardized text; S42: Using a pre-trained or fine-tuned domain text vector model, vector embedding is performed on the normalized text to generate a text vector representation corresponding to each forecast knowledge unit. If necessary, the vectors of multiple text fields within the same forecast knowledge unit are weighted and combined to obtain a comprehensive vector representation.

[0013] Preferably, step S5 specifically includes: S51: Based on the structured fields of forecast knowledge units, establish a structured retrieval index according to the dimensions of watershed, reservoir, control section, forecast type, flood type and time range; S52: Based on the text vector representation of the forecast knowledge unit, establish a vector retrieval index for similarity matching, and establish a time-series retrieval index based on the time field. Establish a correlation between the structured retrieval index, the vector retrieval index and the time-series retrieval index to form a hybrid retrieval forecast knowledge base for short- and medium-term hydrological forecasting scenarios in the watershed.

[0014] Preferably, step S6 specifically includes: S61: Calculate a quality score for each forecast knowledge unit based on data integrity, source credibility, forecast effectiveness evaluation indicators, and applicability timeliness. S62: When the same forecast knowledge unit is detected to be updated in subsequent operations, a new version is generated for it and the version chain record is retained. Versions that meet the preset conditions for quality scores are marked as valid versions, and expired or low-quality versions are marked as archived versions.

[0015] Preferably, step S7 specifically includes: S71: Map newly generated structured forecast task descriptions, forecast scheme configurations, operation logs, forecast results and evaluation indicators in the subsequent short- and medium-term hydrological forecasting operations of the watershed to new forecast knowledge units or updated content of existing forecast knowledge units. S72: For newly added or updated forecast knowledge units, perform the corresponding tag annotation, structured organization, vectorized encoding, index update and version management operations in steps S2 to S6 in sequence to realize online incremental update of forecast knowledge base.

[0016] Preferably, the forecast knowledge unit includes at least: The forecast scheme unit describes the watershed scope, target objects, model combinations, key parameter values, and areal rainfall weighting configuration used in a single forecast; the rainfall and flood event unit describes the natural rainfall process, inflow evolution process, flood peak characteristics, and corresponding hydrological situation; the dispatch procedure unit describes the dispatch rules, dispatch curves, and constraints under flood control, beneficial utilization, or comprehensive utilization conditions; the parameter configuration unit describes the parameter values ​​and calibration results under different models and operating conditions; and the evaluation record unit describes the forecast error, evaluation indicators, forecast applicability conditions, and empirical conclusions.

[0017] Preferably, the quality score is calculated based on at least the following indicators: data integrity indicator, used to measure whether the key fields in the forecast knowledge unit are complete; source credibility indicator, used to distinguish different data sources such as formal business system records, consultation minutes and manual supplementation; forecast performance indicator, used to reflect the degree of deviation between the forecast results and the measured data and the corresponding evaluation indicator level; and timeliness indicator, used to reflect whether the business scenario corresponding to the forecast knowledge unit is still within the validity period.

[0018] Preferably, the incremental update and write-back includes: generating corresponding forecast knowledge units from the structured forecast task description, model scheme configuration, operation log, forecast results, and consultation conclusions of each short- and medium-term hydrological forecast process in the watershed; when it is identified that the newly added operational data and the existing forecast knowledge unit meet the preset similarity conditions in terms of watershed, scenario, and model configuration, the newly added operational data is merged as a new version of the forecast knowledge unit; after the addition or merging of forecast knowledge units is completed, the structured retrieval index, vector retrieval index, and time-series retrieval index are automatically updated to ensure that subsequent retrieval can be based on the latest knowledge.

[0019] Preferably, the forecast knowledge base construction method is applicable to external system call scenarios that use natural language descriptions for querying and configuration, including large language model systems. The method reserves target object fields, forecast period fields, model constraint fields, and business scenario fields in the structured fields of the forecast knowledge units for external system identification and invocation. A unified query interface is set up on top of the hybrid retrieval forecast knowledge base, supporting joint retrieval based on structured query conditions and query vectors to obtain a set of forecast knowledge units matching the forecast requirements.

[0020] The second aspect of the present invention provides: A forecasting knowledge base construction system for short- and medium-term hydrological forecasting of watersheds, characterized in that it is used to execute the method described above, the system comprising: The data acquisition and knowledge unit extraction module is used to collect multi-source operational data related to short- and medium-term hydrological forecasts in the watershed and divide them into forecast knowledge units with unified identifiers and clear boundaries according to preset rules. The ontology modeling and tag management module is used to construct a knowledge ontology and multi-dimensional tag system based on the spatial hierarchy of the watershed, and to complete the association between forecast knowledge units and ontology entities and tags; The data cleaning and structuring module is used to standardize the raw business data of the forecast knowledge unit, parse key information and generate structured records with a unified data pattern. The text vectorization and feature encoding module is used to standardize textual information in structured records and generate vector representations of forecast knowledge units through a domain text vector model. The hybrid index building and retrieval module is used to build structured retrieval indexes, vector retrieval indexes, and time-series retrieval indexes, establish relationships between the various indexes, and provide hybrid retrieval capabilities. The quality assessment and version management module is used to score the quality of forecast knowledge units based on multi-dimensional indicators, and to record the version and mark the validity of updated knowledge units. The incremental update and knowledge write-back module is used to map newly generated data from hydrological forecasting operations to new or updated content in forecasting knowledge units, driving each module to complete online incremental updates of the knowledge base.

[0021] Preferably, the hybrid index construction and retrieval module includes an interface for natural language queries or semi-structured queries, used to receive query requests initiated by external systems, convert the query requests into structured retrieval conditions and query vectors, and jointly recall forecast knowledge units based on the structured retrieval index and the vector retrieval index.

[0022] Preferably, the data acquisition and knowledge unit extraction module interfaces with at least one of the following through a standard interface: a hydrological and rainfall database, a forecasting business system, an archive management system, and a dispatching procedure management system, to achieve automatic acquisition of multi-source business data.

[0023] Preferably, the quality assessment and version management module has a built-in quality scoring model. The input of the model is at least three parameters among data integrity, source credibility, forecast effect evaluation index, and applicability timeliness. The output is the quality score and version validity labeling result of the forecast knowledge unit.

[0024] Preferably, the incremental update and knowledge write-back module forms a closed loop with the basin short- and medium-term hydrological forecasting business system. After the forecasting task is completed, it automatically receives the incremental data pushed by the business system and triggers each module to complete the corresponding tag annotation, structured organization, vectorized encoding, index update and version management operations.

[0025] The third aspect of the present invention provides: A computer device includes a memory and a processor, wherein the memory stores a computer program, which, when executed by the processor, implements the method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed.

[0026] The fourth aspect of the present invention provides: A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed.

[0027] The present invention has the following beneficial effects: 1. Achieve unified organization and associated management of multi-source forecast knowledge: Using forecast knowledge units as the basic carrier, an ontology and multi-dimensional tag system are introduced to extract and structure multi-source forecast data scattered in different systems and forms into a unified database. Clear associations are established according to dimensions such as watershed, control section, and operational scenario, thereby achieving systematic organization of forecast knowledge and solving the problems of information dispersion and missing associations.

[0028] 2. Construct an efficient hybrid retrieval system to improve the efficiency of knowledge retrieval and reuse: The hybrid index system, which integrates structured retrieval, semantic similarity retrieval and time series retrieval, supports both precise conditional filtering retrieval and natural language-based similar case retrieval and time-dimensional process comparison. Forecasters can quickly locate matching historical cases, reduce the workload of manual searching and comparison, and significantly improve the efficiency of scheme preparation and selection.

[0029] 3. Achieve dynamic evolution and reliable management of forecast knowledge: Establish a quality assessment and version management mechanism, classify knowledge units based on multi-dimensional indicators, retain version chain records, and ensure the reliability of recommended knowledge; combine business-driven incremental updates and knowledge write-back processes to promptly write newly generated forecast business data into the knowledge base, so that the knowledge base keeps pace with actual forecast practice and forms a traceable and iterative knowledge evolution system.

[0030] 4. Excellent compatibility and intelligent scalability: It reserves interfaces for external system calls and sets standardized reserved fields in the structured fields, which can be connected to intelligent external systems such as large language models. It supports natural language queries and intelligent solution recommendations, laying a knowledge foundation for the intelligent upgrade of watershed short- and medium-term hydrological forecasts. At the same time, the system connects with existing hydrological forecasting systems through standard interfaces, without the need for large-scale modifications to existing systems, and has strong compatibility.

[0031] 5. Supporting refined management and systematic decision-making in watershed hydrological forecasting operations: This invention realizes full lifecycle management of forecasting knowledge, enabling forecasters to fully reuse historical successful experiences and review typical deviation cases, providing scientific knowledge support for the formulation of forecasting schemes, calibration of model parameters, and selection of scheduling strategies. It promotes the transformation of watershed short- and medium-term hydrological forecasting operations from "experience-driven" to "knowledge-driven", and improves the level of refined management and systematic decision-making in operations. Attached Figure Description

[0032] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0033] Figure 1 This is a flowchart illustrating the forecast knowledge base construction method described in Embodiment 1 of the present invention.

[0034] Figure 2 This is a schematic diagram of the functional module structure of the forecast knowledge base construction system for short- and medium-term hydrological forecasting of watersheds as described in Embodiment 2 of the present invention.

[0035] Figure 3 This is a geographic information layer of the upper reaches of the Jialing River in Embodiment 3 of the present invention.

[0036] Figure 4This is a schematic diagram of the application scenario of the present invention in the upper reaches of the Jialing River, as described in Embodiment 3 of the present invention. Detailed Implementation

[0037] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0038] Example 1: Method and Flow for Constructing a Forecast Knowledge Base This embodiment provides a method for constructing a forecast knowledge base based on the system of Embodiment 1, such as... Figure 1 As shown, the main steps include: Step S1, forecast knowledge unit extraction, is used to divide the scattered original forecast-related data into forecast knowledge units with unified identifiers and clear boundaries, providing a basis for subsequent unified management and retrieval, including: S11: Organize the existing hydrological forecasting system, hydrological and rainfall system and document archives of the watershed management unit, and export the original data such as forecasting task records, measured hydrological data, dispatch logs, meeting minutes, assessment reports and dispatching procedures during the flood season in recent years. S12 is divided according to preset granularity rules: taking a single forecast task as the core, the scheme configuration, forecast results, correction process, consultation conclusions, etc. corresponding to a single forecast task are merged into a forecast scheme unit; taking a single flood process as the core, the hydrological measured process line, characteristic water level, scheduling process, etc. are merged into a rainstorm event unit; taking a single procedure or operation diagram as the core, the flood control scheduling procedure, operation diagram, and discharge curve are organized into a scheduling procedure unit; taking the calibration results of a certain model in a specific watershed as the core, a parameter configuration unit is formed; taking the error analysis and evaluation opinions of a certain forecast as the core, an evaluation record unit is formed, and a unique identifier is assigned to each knowledge unit.

[0039] Step S2 involves constructing the predictive knowledge ontology and tagging system. This is used to establish a unified knowledge representation framework and tagging system, enabling knowledge units from different sources to be associated, managed, and retrieved within the same semantic space. This includes: S21. Based on the hydrological system and scheduling objects of the target watershed, determine the watershed division system and control section system, input basic information such as watershed code, sub-watershed code, section code, station code, and reservoir code, and construct entities such as watershed, river section, control section, hydrological station and reservoir and their upstream and downstream relationships. S22, around business objects such as forecasting schemes, rain and flood events, dispatching procedures, parameter configurations and evaluation records, designs spatial dimension labels (such as watershed level, main stream and tributary attributes, control section location), time dimension labels (such as year, flood season, flood number), business scenario labels (such as flood control, water benefit, water replenishment, ecology), model applicability labels (such as model type, forecast period range, data requirements), and forecast effect labels (such as error level, whether it meets the standard, whether it is a typical case), and associates the above labels with knowledge units.

[0040] Step S3, data cleaning and structured data import, is used to convert business data with inconsistent formats and structures into structured records conforming to a unified model, providing a standardized data foundation for index construction and quality evaluation, including: S31, preprocess the original documents and data records corresponding to the knowledge units, unify the time format, standardize the space name, map the stream domain name and site name of different systems to a unified directory, and mark or reasonably fill in missing values ​​and outliers. S32 analyzes key information such as forecast object, forecast period range, adopted model, key parameters, key points of dispatch curve, evaluation index and main conclusions for forecast scheme unit, rain and flood event unit, dispatch procedure unit, parameter configuration unit and evaluation record unit respectively. The analysis results are filled into standard field table, structured record is generated, and the original document path is recorded for traceability.

[0041] Step S4, text feature extraction and vector generation, is used to convert important textual information in structured records into vector representations that can be used for similarity retrieval, enabling the system to recall relevant cases based on semantic similarity, including: S41, select text fields such as scheme description, consultation conclusion, evaluation opinion, and typical event description, and perform word segmentation, terminology standardization and domain entity recognition based on the professional thesaurus and rule base of hydrology and reservoir scheduling, and merge synonyms and different expressions into unified terms; S42, the above-mentioned standardized text input domain text vector model is used to generate the corresponding vector representation; within the same knowledge unit, multiple text field vectors are weighted and averaged or concatenated according to the importance of the fields to generate the comprehensive vector representation of the knowledge unit, and stored in the vector index storage together with the unique identifier of the knowledge unit.

[0042] Step S5, Hybrid Index Construction, is used to build an index system that takes into account conditional filtering, semantic association, and time series analysis, supporting multiple retrieval methods and combined queries, including: S51. Establish a joint index in the database for structured fields such as watershed coding, cross-section coding, reservoir coding, forecast type, flood type, business scenario, and time range to achieve conditional filtering retrieval based on structured fields; S52, in the vector index engine, a vector retrieval index is established using a comprehensive vector representation of knowledge units to achieve similarity retrieval based on text description; at the same time, a time-series retrieval index is established in the time field by year, by flood season, and by flood number, and the relationship between the structured index, vector index, and time-series index is maintained.

[0043] Step S6, Quality Evaluation and Version Tagging, is used to classify and manage the evolution of knowledge units to ensure the reliability and interpretability of the recommendation results, including: S61. Based on information such as field completeness rate, data source type (e.g., formal business records, meeting minutes, manual supplementation, etc.), forecast error and evaluation index level, case occurrence time, etc., calculate the quality score of each knowledge unit, and mark low-scoring knowledge units as low quality or for reference only. S62. For multiple adjustment records of the same watershed, the same control section, and the same type of forecast scheme, establish a version chain, mark the latest record with a quality score that meets the preset threshold as the current recommended version, and mark other records as historical versions or archived versions.

[0044] Step S7, forecast service-driven incremental updates, are used to continuously write newly generated forecast process data back to the knowledge base, so that the knowledge base content is continuously improved as the service operates, including: S71. After each forecasting task is completed, the watershed hydrological forecasting business system will push the task configuration (including forecasting objects, lead time, model combination, initial parameters, etc.), operation log, forecast and observation comparison results, and consultation and evaluation opinions to the forecasting knowledge base construction system through the interface. S72, the forecast knowledge base construction system matches existing knowledge units based on information such as watershed, cross section, scene, and model. If the match is successful, a new version record is added for the knowledge unit; if the match is unsuccessful, a new knowledge unit is created. The system then performs the corresponding tag annotation, structured organization, vectorized encoding, index update, and quality evaluation operations in steps S2 to S6 on the newly added or updated knowledge units, so that the latest forecast knowledge can participate in retrieval and decision support in a short period of time.

[0045] Example 2: A system for constructing a forecast knowledge base for short- and medium-term hydrological forecasting in watersheds. This embodiment describes a forecast knowledge base construction system deployed in a watershed hydrological forecasting operational environment. This system is typically built within a watershed hydrological forecasting operational platform or the data center of a watershed management agency, and operates by interfacing with existing forecasting operational systems through standard interfaces. The system centrally accesses, organizes, and models dispersed data, focusing on forecast schemes, flood evolution processes, and assessment records generated during daily forecast production. It then provides forecasters with searchable and comparable typical forecast cases and knowledge services.

[0046] The system in this embodiment mainly includes the following parts: I. Data Sources and Business Systems: The system's data sources include hydrological and rainfall databases, forecasting systems, record management systems, and other related systems.

[0047] ① The hydrological and rainfall database is used to provide historical and real-time monitoring data such as rainfall, flow rate, and water level; ②The forecasting system has accumulated the forecasting task configurations, forecasting results, and comparisons with actual measurements from previous years; ③ The document management system stores various meeting minutes, forecast plan descriptions, scheduling summaries, and other documents; ④ Other related systems may include scheduling procedure management system, parameter calibration management system, and other external data sources related to watershed forecasting services.

[0048] II. Functional modules of the forecast knowledge base construction system (corresponding to construction method steps S1~S7): The system contains seven functional modules connected in series to complete the construction, indexing, quality management and incremental updating of forecast knowledge units.

[0049] M1: Data Acquisition and Knowledge Unit Extraction Module The data acquisition and knowledge unit extraction module connects to the aforementioned hydrological and rainfall databases, forecasting operational systems, record management systems, and other related systems via interfaces to collect raw operational data related to short- and medium-term hydrological forecasts for the watershed, including: ① Records of forecasting task configurations during historical flood seasons; ② Comparison sequence of forecast results and actual measurements; ③ Typical rainfall and flood process analysis report, meeting minutes, plan description, and dispatch summary document; ④ Scheduling procedures and operation diagrams, model parameter calibration results and applicable scope descriptions, etc.

[0050] Based on preset rules, this module uses a single forecast task, a single flood process, a single scheduling procedure, or a set of parameter calibration results of a model in a specific watershed scenario as the granularity to summarize the associated scheme configurations, forecast results, charts, and text descriptions into an initial forecast knowledge unit, and saves the original documents into a file storage.

[0051] M2: Ontology Modeling and Tag Management Module The ontology modeling and tag management module completes the configuration of the ontology and tag system for the watershed forecasting domain during the system initialization phase, and assigns unified semantics and tags to each knowledge unit during operation, specifically including: ① Define entities such as watersheds, sub-watersheds, river sections, control sections, hydrological stations, and reservoirs, as well as their upstream and downstream relationships, at the spatial level; ② Define entities such as forecast scheme, rainfall and flood event, dispatch procedure, parameter configuration, and evaluation record at the business level; ③ Configure spatial dimension labels (watershed, control section, etc.), time dimension labels (year, flood season, time window, etc.), business scenario labels (flood control, drought prevention, ecological scheduling, etc.), model applicability labels, and forecast effect labels at the label level.

[0052] After M1 generates the initial knowledge units, M2 performs ontology mapping and label assignment on them, providing a unified semantic foundation for subsequent structuring, vectorization, and retrieval.

[0053] M3: Data Cleaning and Structured Processing Module The data cleaning and structuring module receives knowledge units processed by M2 and performs standardization and structuring processing on the data, including: ① Standardize the format of time fields and check and correct the settings for time intervals and forecast periods; ② Standardize spatial information and map stream domain names and site names from different systems to a unified directory; ③ Perform content analysis and structure extraction on document-type materials to extract titles, abstracts, key paragraphs, table data, etc.; ④ According to the preset data pattern, fill the structured fields with information such as forecast object, forecast period, model used, main parameters, correction method, flood control constraints, key points of dispatch curve, forecast error index, and evaluation conclusion.

[0054] After cleaning and structuring, each forecast knowledge unit is written into a structured database in the form of structured records and associated with its corresponding tags and ontology entities.

[0055] M4: Text Vectorization and Feature Encoding Module The text vectorization and feature encoding module performs feature encoding on the text information in the knowledge unit to generate a vector representation suitable for similarity retrieval. The process includes: ①Based on the glossary of terms in the fields of hydrological forecasting and reservoir scheduling, the text is segmented, entity recognized, and synonyms are normalized; ② Call the domain text feature encoding model to map texts such as forecast scheme descriptions, consultation minutes, scheduling summaries, and evaluation conclusions into vectors of fixed dimensions; ③Weigh and combine or concatenate multiple text field vectors within the same knowledge unit to obtain a comprehensive vector representation of that knowledge unit.

[0056] Finally, the module writes the knowledge unit vector and its unique identifier into the vector index storage to support subsequent semantic similarity-based retrieval.

[0057] M5: Hybrid Index Building and Retrieval Module The hybrid index building and retrieval module is based on structured databases and vector index storage, and builds and maintains multiple types of indexes, including: ① Establish relational indexes based on structured fields such as watershed, control section, forecast object, forecast type, and time range; ② Create a vector index in the vector index engine based on the knowledge unit vector; ③ Create a time-series index based on the time field, and support recalling knowledge unit sets by flood season and time window.

[0058] Based on this, M5 enables combined query capabilities of conditional filtering and vector similarity retrieval. It can filter candidate knowledge units based on conditions such as watershed name, control section, forecast period, and flood control scenario, and then use vector similarity to rearrange the candidate set, prioritizing the return of typical cases that perform well in the target scenario.

[0059] The unified retrieval interface / forecast knowledge service is located on the left side of the M5 module, serving as the external service interface for M5: Forecasting business systems and forecasters initiate retrieval requests through the unified retrieval interface, submitting structured conditions and natural language descriptions; the unified retrieval interface passes the query conditions to M5, M5 executes the retrieval based on the hybrid index and returns a set of knowledge units and typical cases; the unified retrieval interface then encapsulates the results into a forecast knowledge service and outputs it to the caller.

[0060] M6: Quality Assessment and Version Management Module The quality assessment and version management module performs quality assessment and version management for each knowledge unit based on the prediction error indicators, case tags, and data source information in the structured database. ① The quality score is calculated by considering factors such as the completeness of the fields, the reliability of the data source, the magnitude of the forecast error, and whether it belongs to recent cases; ② Mark knowledge units with low quality scores as references or cases to be reviewed; ③ Establish a version chain for multiple adjustment records of the same watershed, the same control section, and the same type of forecast scheme, and automatically identify the current recommended version.

[0061] The results of quality assessment and version management are written back to the structured database, providing a basis for the unified search interface output result sorting and recommended case identification.

[0062] M7: Incremental Update and Knowledge Rewrite Module The incremental update and knowledge write-back module is used to form a closed loop with the forecasting operational system. When the short- and medium-term hydrological forecasting operational system generates new forecasting tasks and results during the production process, M7 is responsible for writing this new information back to the knowledge base: ①After each forecast mission is completed, the forecasting service system will provide the M7 with the forecast mission configuration, the comparison process between the forecast results and the actual measurements, and the consultation and evaluation opinions as incremental data; ②M7 combines the above data into new forecast knowledge units, or identifies them as subsequent versions of existing knowledge units, and calls M3~M4~M5 to complete cleaning, structuring, vector encoding and index updating; ③ For cases that are highly similar to the current recommended version but have better forecast performance, M7 triggers M6 to update their quality score and version status, automatically marking them as the new recommended version; ④ For cases with large forecast deviations, M7 marks them as cases that need to be reviewed, providing data support for subsequent technical analysis and model improvement.

[0063] III. Unified Search Interface and User Invocation Method The unified retrieval interface / forecast knowledge service serves as the external service entry point for this invention's system. Located on the left side of the M5 module, it provides unified knowledge retrieval and case recommendation capabilities to forecasting operational systems and forecasters. Forecasters can submit query requests through this interface within the existing forecasting operational system interface. The system returns typical historical forecast cases matching the current watershed, cross-section, lead time, and scenario, along with their model configurations, correction methods, and scheduling measures, to guide the selection and adjustment of current forecasting schemes.

[0064] The forecasting service system and its users are the direct users of the system of this invention: on the one hand, the forecasting service system, as one of the data sources, provides forecasting tasks and results data for M1 and M7; on the other hand, the forecasting service system and forecasters call M5 through a unified retrieval interface to obtain historical forecasting knowledge units and recommended schemes, realizing a closed-loop operation of feeding back past successful experiences to current forecasts.

[0065] Example 3: Application in a certain basin of the upper Yangtze River This embodiment, based on the forecast knowledge base construction system described in Embodiment 1 and the forecast knowledge base construction method described in Embodiment 2, presents a specific application scenario in the upper reaches of the Jialing River to illustrate the deployment method and usage process of the present invention in actual engineering, but does not limit the scope of protection of the present invention.

[0066] like Figure 3As shown, the upper reaches of the Jialing River are one of the important inflow areas of the upper Yangtze River. Water from the upper reaches flows into the Yangtze River via the main stream of the Jialing River. The basin has 10 control sections along the main stream and tributaries, 3 cascade reservoirs, and 49 rain gauge stations. Forecasts include the inflow and outflow of each reservoir, as well as hourly flow processes at key downstream control sections. The existing basin forecasting system (hereinafter referred to as the forecasting operational system) has long been responsible for flood season forecasting in the upper reaches of the Jialing River, accumulating over 10 years of flood season forecast records and a large amount of documents such as consultation minutes, forecast scheme descriptions, and scheduling summaries. However, these materials are scattered across different platforms such as the hydrological and rainfall database, the forecasting operational system, and the file management system, making it difficult to retrieve and reuse them in a timely manner according to flood type and scheduling scenario.

[0067] In this embodiment, the forecast knowledge base is deployed in the data center of the forecasting operational system in the upper reaches of the Jialing River, and it interfaces with the existing hydrological and rainfall database, forecasting operational system, and record management system. For example... Figure 4 As shown, in this embodiment, steps S1 to S4 of Embodiment 2 are aggregated into a "Historical Flood Case Import and Knowledge Unit Construction Module", steps S5 and S6 are aggregated into a "Hybrid Index Retrieval and Typical Case Recommendation Module", and steps S7 and S6 are aggregated into a "Forecast Result Write-back and Version Optimization and Update Module". The three modules, together with the forecast business system, constitute a closed loop for the construction and application of a forecast knowledge base for the upper reaches of the Jialing River.

[0068] (1) Introduction of historical flood cases and construction of knowledge units First, several representative flood events in the upper reaches of the Jialing River over the past decade were selected as samples, including small to medium-sized floods, floods approaching design levels, and floods approaching check levels. For each representative flood event, the forecasting operational system contains corresponding forecasting task configuration records, forecast results compared with measured data, and the archive management system contains corresponding consultation minutes, forecasting scheme descriptions, and scheduling summary documents.

[0069] In the historical flood case import and knowledge unit construction module, the system performs the following operations according to steps S1 to S4 described in Example 2: ① Through the data acquisition and knowledge unit extraction module, data such as forecast task configuration, forecast results and actual measurement comparison, dispatch procedures and consultation minutes related to a certain flood process are automatically collected from the hydrological and rainfall database, forecast business system and file management system. The "single forecast task + corresponding flood process + dispatch document" are integrated into the initial forecast knowledge unit. ② Through the ontology modeling and tag management module, the above-mentioned forecast knowledge units are associated with spatial entities such as the upper reaches of the Jialing River, specific control sections, and reservoirs, and are assigned business scenario tags and time tags such as "flood control as the main focus / water replenishment as a secondary consideration", "72-hour forecast period", and "rainstorm flood". ③ Through the data cleaning and structuring module, the time field, spatial field, model parameter field, forecast error field, etc. are uniformly formatted, and "forecast object, lead time, model combination used, key parameters, key points of scheduling curve, forecast error index, evaluation conclusion" are written into the structured database; ④ Through the text vectorization and feature encoding module, the texts such as forecast scheme descriptions, consultation minutes, and scheduling summaries are segmented into domain words, normalized with synonyms, and feature encoded to generate a vector representation of each forecast knowledge unit, which is then written into the vector index storage.

[0070] After the above processing, the system in this embodiment has accumulated dozens of typical flood events and hundreds of forecast knowledge units, covering common water inflow combinations and flood control / water replenishment scheduling scenarios during the main flood season in the upper reaches of the Jialing River, providing a structured knowledge foundation and semantic vector foundation for subsequent retrieval.

[0071] (2) Search and scheme recommendation during the new round of flooding When a regional rainstorm occurs again in the upper reaches of the Jialing River in a certain year, the meteorological department forecasts significant heavy rainfall in the next few days. The forecasting system automatically generates a flood forecast task for the upper reaches of the Jialing River (e.g., a forecast task numbered "JLJ-20XX-05"). In this embodiment, the forecasting system is integrated with the forecasting knowledge base construction system of this invention through a unified retrieval interface. The specific application process is as follows: ① In the forecasting system interface, forecasters select the "upper reaches of the Jialing River" and the target control section, set the forecast period to 72 hours, and check the scenario information such as "primarily for flood control, while also considering flood control in downstream cities" in the business scenario options; at the same time, they enter natural language descriptions in the text search box, such as "the reservoir water level is high in the early stage, the flood discharge capacity of the downstream river is limited, and it is necessary to control the peak flow of a certain control section while also considering the safety of downstream cities," etc.

[0072] ② The forecasting system submits the aforementioned structured conditions and textual descriptions to the hybrid index retrieval and typical case recommendation module through a unified retrieval interface. This module first uses the structured index to filter candidate forecasting knowledge units in the knowledge base that meet the criteria of "upper reaches of the Jialing River + target control section + forecast period of approximately 72 hours + dominant flood control scenario." Based on this, it performs semantic similarity retrieval on the candidate knowledge units using the vector index and, combined with the quality score provided by the quality assessment and version management module, comprehensively ranks the candidate cases.

[0073] ③ The hybrid index retrieval and typical case recommendation module returns several typical cases with high quality scores that are highly similar to the current scenario to the forecasting service system through a unified retrieval interface. In the trial operation of this embodiment, for rainstorm and flood events like "JLJ-20XX-05", the system can filter out nearly 10 candidate flood events from hundreds of historical cases within seconds and recommend 3 typical cases with high comprehensive quality scores.

[0074] ④ Forecasters can access the details of recommended cases in the forecasting system interface. They can directly view the model combination used in the case (e.g., "Xin'anjiang Model + Muskingen Confluence" or "Distributed Model + Statistical Correction Model"), key parameter configurations (including runoff generation parameters, confluence parameters, rainfall correction parameters, etc.), the combination of rainfall data used and their weighting settings, corresponding forecast error indicators (e.g., peak flow error, peak time error, process Nash efficiency, etc.), and actual dispatching measures taken (e.g., advance flood discharge, flow restriction at control sections, flood control level adjustment suggestions, etc.). With the support of the above information, forecasters can determine the initial forecast plan and initial parameter values ​​based on the recommended cases in a relatively short time. Compared to the previous method of relying on manual searching of historical data and personal memory, before the introduction of this invention, the forecast plan preparation time can be significantly shortened.

[0075] (3) Post-flood result write-back and version optimization After the flood event ends, the forecasting system automatically summarizes the forecasting task configuration, forecast results and actual measurement comparison process, and consultation and evaluation opinions for the "JLJ-20XX-05" flood event, forming a complete forecasting service record, and pushes it to the forecast result write-back and version optimization update module through the interface.

[0076] Based on information such as the basin name, control section, forecast period setting, and business scenario, this module identifies the current forecast process as a scenario highly similar to a certain existing knowledge unit, treats the current process as a new version of that knowledge unit, calls the data cleaning and structuring module to supplement the structured fields, calls the text vectorization and feature encoding module to update the vector representation, and the hybrid index building module to update the corresponding index.

[0077] Subsequently, the quality assessment and version management module calculates or updates quality scores for each version of the knowledge unit based on the forecast error indicators, deviation control, and consultation evaluation opinions of this process: if the overall error of this forecast on key indicators such as peak flow and peak occurrence time is better than the original recommended version, then the new version can be marked as the current recommended version; if this forecast has a large deviation under special conditions, then it can be marked as a "case requiring review" and explicitly marked in the knowledge base so that technical personnel can prioritize the analysis of this process in subsequent model improvement and parameter calibration work.

[0078] Example 4: This embodiment provides a forecasting knowledge base construction system for short- and medium-term hydrological forecasting of watersheds, the system comprising: The data acquisition and knowledge unit extraction module is used to collect multi-source operational data related to short- and medium-term hydrological forecasts in the watershed and divide them into forecast knowledge units with unified identifiers and clear boundaries according to preset rules. The ontology modeling and tag management module is used to construct a knowledge ontology and multi-dimensional tag system based on the spatial hierarchy of the watershed, and to complete the association between forecast knowledge units and ontology entities and tags; The data cleaning and structuring module is used to standardize the raw business data of the forecast knowledge unit, parse key information and generate structured records with a unified data pattern. The text vectorization and feature encoding module is used to standardize textual information in structured records and generate vector representations of forecast knowledge units through a domain text vector model. The hybrid index building and retrieval module is used to build structured retrieval indexes, vector retrieval indexes, and time-series retrieval indexes, establish relationships between the various indexes, and provide hybrid retrieval capabilities. The quality assessment and version management module is used to score the quality of forecast knowledge units based on multi-dimensional indicators, and to record the version and mark the validity of updated knowledge units. The incremental update and knowledge write-back module is used to map newly generated data from hydrological forecasting operations to new or updated content in forecasting knowledge units, driving each module to complete online incremental updates of the knowledge base.

[0079] Furthermore, the hybrid index construction and retrieval module includes an interface for natural language queries or semi-structured queries, used to receive query requests initiated by external systems, convert the query requests into structured retrieval conditions and query vectors, and jointly recall forecast knowledge units based on the structured retrieval index and the vector retrieval index.

[0080] Furthermore, the data acquisition and knowledge unit extraction module interfaces with at least one of the following through a standard interface: hydrological and rainfall database, forecasting business system, archive management system, and dispatching procedure management system, to achieve automatic acquisition of multi-source business data.

[0081] Furthermore, the quality assessment and version management module has a built-in quality scoring model. The input of the model is at least three parameters among data integrity, source credibility, forecast effect evaluation index, and applicability timeliness. The output is the quality score and version validity labeling result of the forecast knowledge unit.

[0082] Furthermore, the incremental update and knowledge write-back module forms a closed loop with the basin short- and medium-term hydrological forecasting business system. After the forecasting task is completed, it automatically receives the incremental data pushed by the business system and triggers each module to complete the corresponding tag annotation, structured organization, vectorized encoding, index update and version management operations.

[0083] Example 5: This embodiment provides a computer device, including a memory and a processor. The memory stores a computer program, which, when executed by the processor, implements the method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed.

[0084] Example 6: This embodiment provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed.

[0085] It should be noted that the above examples are merely specific embodiments of the present invention, and the present invention is obviously not limited to the above embodiments, with many similar variations. All modifications that can be directly derived or conceived by those skilled in the art from the content disclosed in this invention should fall within the protection scope of this invention.

Claims

1. A method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed, characterized in that, Includes the following steps: S1: Collect multi-source operational data related to short- and medium-term hydrological forecasts in the basin and group them into forecast knowledge units with unified identification and clear boundaries according to preset rules; S2: Construct a knowledge ontology and multi-dimensional tagging system based on the spatial hierarchy of the watershed, and associate forecast knowledge units with ontology entities and tags; S3: Standardize the raw operational data of the forecast knowledge unit, parse key information to generate structured records with a unified data model; S4: Standardize the textual information in the structured records and generate vector representations of the forecast knowledge units through the domain text vector model; S5: Construct structured retrieval indexes, vector retrieval indexes, and time-series retrieval indexes based on structured fields, vector representations, and time fields respectively, and establish the relationships between the indexes to form a hybrid retrieval forecast knowledge base; S6: Score the quality of forecast knowledge units based on multi-dimensional indicators, and record the version and mark the validity of updated knowledge units; S7: Map newly generated data from the short- and medium-term hydrological forecasting operations of the basin to the newly added or updated content of the forecasting knowledge unit, and execute the corresponding operations of steps S2 to S6 to realize online incremental updates and write-back of the knowledge base.

2. The method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed according to claim 1, characterized in that, Step S1 specifically includes: S11: Collect at least one type of multi-source operational data related to at least one target watershed, including historical hydrological forecasting schemes, typical rainfall and flood process records, scheduling procedures, model parameter configurations, forecast accuracy evaluation results, and consultation minutes. S12: Based on the preset division rules, the operational data corresponding to the same forecast task, the same flood process, the same dispatching procedure, the same set of parameter calibration results, or the same forecast evaluation record are grouped into a forecast knowledge unit, and a unique identifier is assigned to each forecast knowledge unit.

3. The method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed according to claim 1, characterized in that, Step S2 specifically includes: S21: Based on the spatial hierarchical relationship between watershed, river network, reservoir, control section, and hydrological station, define watershed entities, reservoir entities, control section entities, hydrological station entities, forecasting scheme entities, and stormwater event entities, as well as their interrelationships; S22: Design a multi-dimensional tag system that includes spatial dimension, time dimension, business scenario, model applicability and forecast effect, and associate at least one of the aforementioned tags with each forecast knowledge unit.

4. The method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed according to claim 1, characterized in that, Step S3 specifically includes: S31: Standardize the format, complete the fields, and handle outliers of the original business data corresponding to the forecast knowledge unit; standardize the time information, spatial location, watershed name, and reservoir name. S32: Extract key information such as forecast objects, forecast period range, model type, scheduling measures and evaluation indicators from the forecast knowledge unit, and generate corresponding structured fields based on the forecast knowledge ontology to form a structured record that conforms to a unified data model.

5. The method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed according to claim 1, characterized in that, Step S4 specifically includes: S41: Perform word segmentation, terminology standardization, and domain entity recognition on the text fields of scheme descriptions, consultation conclusions, and evaluation opinions contained in the structured records to obtain standardized text; S42: Using a pre-trained or fine-tuned domain text vector model, vector embedding is performed on the normalized text to generate a text vector representation corresponding to each forecast knowledge unit. If necessary, the vectors of multiple text fields within the same forecast knowledge unit are weighted and combined to obtain a comprehensive vector representation.

6. The method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed according to claim 1, characterized in that, Step S5 specifically includes: S51: Based on the structured fields of forecast knowledge units, establish a structured retrieval index according to the dimensions of watershed, reservoir, control section, forecast type, flood type and time range; S52: Based on the text vector representation of the forecast knowledge unit, establish a vector retrieval index for similarity matching, and establish a time-series retrieval index based on the time field. Establish a correlation between the structured retrieval index, the vector retrieval index and the time-series retrieval index to form a hybrid retrieval forecast knowledge base for short- and medium-term hydrological forecasting scenarios in the watershed.

7. The method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed according to claim 1, characterized in that, Step S6 specifically includes: S61: Calculate a quality score for each forecast knowledge unit based on data integrity, source credibility, forecast effectiveness evaluation indicators, and applicability timeliness. S62: When the same forecast knowledge unit is detected to be updated in subsequent operations, a new version is generated for it and the version chain record is retained. Versions that meet the preset conditions for quality scores are marked as valid versions, and expired or low-quality versions are marked as archived versions.

8. The method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed according to claim 1, characterized in that, Step S7 specifically includes: S71: Map newly generated structured forecast task descriptions, forecast scheme configurations, operation logs, forecast results and evaluation indicators in the subsequent short- and medium-term hydrological forecasting operations of the watershed to new forecast knowledge units or updated content of existing forecast knowledge units. S72: For newly added or updated forecast knowledge units, perform the corresponding tag annotation, structured organization, vectorized encoding, index update and version management operations in steps S2 to S6 in sequence to realize online incremental update of forecast knowledge base.

9. The method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed according to claim 1, characterized in that, The forecast knowledge unit includes at least: The forecast scheme unit describes the watershed scope, target objects, model combinations, key parameter values, and areal rainfall weighting configuration used in a single forecast; the rainfall and flood event unit describes the natural rainfall process, inflow evolution process, flood peak characteristics, and corresponding hydrological situation; the dispatch procedure unit describes the dispatch rules, dispatch curves, and constraints under flood control, beneficial utilization, or comprehensive utilization conditions; the parameter configuration unit describes the parameter values ​​and calibration results under different models and operating conditions; and the evaluation record unit describes the forecast error, evaluation indicators, forecast applicability conditions, and empirical conclusions.

10. The method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed according to claim 1, characterized in that, The quality score is calculated based on at least the following indicators: data integrity indicator, used to measure whether key fields in the forecast knowledge unit are complete; source credibility indicator, used to distinguish different data sources such as official business system records, consultation minutes and manual supplementation; forecast performance indicator, used to reflect the degree of deviation between forecast results and measured data and the corresponding evaluation indicator level; and timeliness indicator, used to reflect whether the business scenario corresponding to the forecast knowledge unit is still within the validity period.

11. The method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed according to claim 1, characterized in that, The incremental update and write-back process includes: generating corresponding forecast knowledge units from the structured forecast task description, model scheme configuration, operation log, forecast results, and consultation conclusions of each short- and medium-term hydrological forecast process in the watershed; when it is identified that the newly added operational data and the existing forecast knowledge unit meet the preset similarity conditions in terms of watershed, scenario, and model configuration, the newly added operational data is merged as a new version of the forecast knowledge unit; after the addition or merging of forecast knowledge units is completed, the structured retrieval index, vector retrieval index, and time-series retrieval index are automatically updated to ensure that subsequent retrieval can be based on the latest knowledge.

12. The method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed according to claim 1, characterized in that, The forecast knowledge base construction method is applicable to external system call scenarios that use natural language descriptions for querying and configuration, including large language model systems. In the structured fields of the forecast knowledge units, target object fields, forecast period fields, model constraint fields, and business scenario fields are reserved for external systems to identify and call. A unified query interface is set up on top of the hybrid retrieval forecast knowledge base, which supports joint retrieval based on structured query conditions and query vectors to obtain a set of forecast knowledge units that match the forecast requirements.

13. A forecasting knowledge base construction system for short- and medium-term hydrological forecasting of watersheds, characterized in that, The system for performing the method according to any one of claims 1-12, the system comprising: The data acquisition and knowledge unit extraction module is used to collect multi-source operational data related to short- and medium-term hydrological forecasts in the watershed and divide them into forecast knowledge units with unified identifiers and clear boundaries according to preset rules. The ontology modeling and tag management module is used to construct a knowledge ontology and multi-dimensional tag system based on the spatial hierarchy of the watershed, and to complete the association between forecast knowledge units and ontology entities and tags; The data cleaning and structuring module is used to standardize the raw business data of the forecast knowledge unit, parse key information and generate structured records with a unified data pattern. The text vectorization and feature encoding module is used to standardize textual information in structured records and generate vector representations of forecast knowledge units through a domain text vector model. The hybrid index building and retrieval module is used to build structured retrieval indexes, vector retrieval indexes, and time-series retrieval indexes, establish relationships between the various indexes, and provide hybrid retrieval capabilities. The quality assessment and version management module is used to score the quality of forecast knowledge units based on multi-dimensional indicators, and to record the version and mark the validity of updated knowledge units. The incremental update and knowledge write-back module is used to map newly generated data from hydrological forecasting operations to new or updated content in forecasting knowledge units, driving each module to complete online incremental updates of the knowledge base.

14. The forecasting knowledge base construction system for short- and medium-term hydrological forecasting of a watershed, as described in claim 13, is characterized in that... The hybrid index building and retrieval module includes an interface for natural language queries or semi-structured queries, which is used to receive query requests initiated by external systems, convert the query requests into structured retrieval conditions and query vectors, and jointly recall forecast knowledge units based on the structured retrieval index and the vector retrieval index.

15. The forecasting knowledge base construction system for short- and medium-term hydrological forecasting of a watershed according to claim 13, characterized in that, The data acquisition and knowledge unit extraction module interfaces with at least one of the following through a standard interface: hydrological and rainfall database, forecasting system, archive management system, and dispatching procedure management system, to achieve automatic acquisition of multi-source business data.

16. The forecasting knowledge base construction system for short- and medium-term hydrological forecasting of a watershed according to claim 13, characterized in that, The quality assessment and version management module has a built-in quality scoring model. The input of the model is at least three parameters from data integrity, source credibility, forecast effect evaluation index, and applicability timeliness. The output is the quality score and version validity labeling result of the forecast knowledge unit.

17. The forecasting knowledge base construction system for short- and medium-term hydrological forecasting of a watershed according to claim 13, characterized in that, The incremental update and knowledge write-back module forms a closed loop with the basin short- and medium-term hydrological forecasting business system. After the forecasting task is completed, it automatically receives the incremental data pushed by the business system and triggers each module to complete the corresponding tag annotation, structured organization, vectorized encoding, index update and version management operations.

18. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores a computer program, which, when executed by the processor, implements the method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed as described in any one of claims 1 to 12.

19. A computer-readable storage medium having a computer program stored thereon, the computer program, when executed by a processor, implementing the method for constructing a forecast knowledge base for short- and medium-term hydrological forecasting of a watershed as described in any one of claims 1 to 12.

Images