An AI-based hydrological intelligent management method and system, and a storage medium

Abstract

This invention discloses an AI-based intelligent hydrological management method, system, and storage medium, aiming to improve the intelligence and decision-making accuracy of hydrological management. The method receives and preprocesses multi-source heterogeneous hydrological operational data; performs semantic parsing and data fusion verification based on a hydrological semantic normalization framework; loads data and constructs a semantic association network through three-dimensional dynamic permission configuration; parses user natural language intent and matches the optimal hydrological operational scenario model; schedules AI agents to collaboratively process data; merges results and iteratively optimizes them before outputting decision support data; and feeds back relevant parameters to dynamically optimize the semantic framework and scenario model. The system includes corresponding functional modules, encrypted data transmission, and tamper-proof operation logs, with decision data accompanied by a traceability link. This invention achieves efficient hydrological data processing and intelligent decision-making, ensuring management security and reliability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hydrological intelligent management, and in particular to a hydrological intelligent management method and system based on AI and a storage medium. BACKGROUND

[0002] With the expansion of urban water supply scale and the improvement of hydrological management refinement requirements, hydrological enterprises have accumulated a large amount of structured and unstructured data from business systems such as water quality monitoring, water supply scheduling, equipment inspection, emergency response and customer service. However, these source data generally have problems such as inconsistent format, confusing naming, semantic ambiguity and scattered storage, which makes it difficult to effectively integrate and reuse the data. Traditional hydrological information systems mostly use isolated databases or simple document archiving methods, lack standardized management mechanisms for multi-source heterogeneous data, and make cross-departmental collaboration inefficient and knowledge assets difficult to deposit. At the same time, existing systems generally rely on manual report writing, compliance checking or scheduling decisions, which not only wastes time and is prone to errors, but also makes it difficult to trace the data source and poses compliance and security risks. Although some platforms have introduced search or automation tools, their functions are fragmented and their permissions are extensive, and they cannot provide context-aware intelligent services based on actual business scenarios. In addition, in terms of data security, the protection measures during static storage and transmission are often insufficient, making it difficult to meet the increasingly stringent industry regulatory requirements. Therefore, there is an urgent need for an intelligent processing method that can uniformly manage, hierarchically organize and support natural language-driven hydrological source data, to realize a hydrological intelligent management closed loop of data credibility, knowledge management and service traceability, and solve the core bottleneck problem of "having data, lacking knowledge and being difficult to intelligent" in the current hydrological digital transformation. SUMMARY

[0003] The present application aims to provide an AI-based hydrological intelligent management method and system to solve the technical problems of dispersed hydrological data sources, heterogeneous formats, non-uniform semantics, lack of standardized management and hierarchical control mechanisms for multi-source data in traditional information systems, disconnection between AI applications and business scenarios, untraceable processing results, rigid interaction methods, and the risk of unauthorized access and leakage in data security and permission management.

[0004] To achieve the above-mentioned purpose, in the first aspect of the present application, an AI-based hydrological intelligent management method is provided, comprising the following steps: receiving multi-source heterogeneous hydrological raw data and preprocessing to generate standardized input data stream with global unique identification and spatio-temporal attributes; based on a hydrological field semantic normalization framework, the standardized input data stream is differentiated analyzed according to the spatial density of the pipe network: high-density areas strengthen topological correlation verification, low-density areas focus on water quality and water source coupling calculation, and the hydrological fusion data set with credibility label is output after semantic compatibility degree grading fusion. loading the hydrological fusion dataset into a domain-isolated semantic resource pool through a three-dimensional permission mechanism and establishing a dynamic semantic link; Parsing user intent to calculate scene adaptation index, selecting the optimal model to generate scene task instruction data from the resource pool; Based on scene task instruction data driving adaptive collaboration cluster, according to the time sequence protocol chain collaborative processing of defect identification, risk rating to scheduling optimization, outputting sub-task processing data with traceable identification; Integrate sub-task processing data and calculate collaborative reliability, drive logical reconfiguration iteration when threshold is not reached, output hydrological decision support data, and feed back the hydrological decision support data and parameters to the semantic framework and model library, realize cognitive dynamic self-calibration.

[0005] Further, the generation of standardized input data stream with global unique identification and space-time attribute, specifically includes: Using edge computing nodes to perform localized cleaning on the multi-source heterogeneous hydrological raw data, unifying industrial device interfaces through multi-protocol adaptive conversion mechanism, mixing and packaging time series sensor data and non-time series business data; Starting local data preservation storage in the case of abnormal network interruption, and automatically triggering data resume based on two-way security authentication mechanism after network recovery; Through the hydrological data ingestion engine to perform real-time ingestion and dynamic subscription, using NLP processing pipeline to parse text logs, using improved defect identification model to extract pipe network inspection image features, using spectrum filtering algorithm to process pipe network leak audio, and converting the above unstructured data into structured intermediate state data; Based on AvroSchema definition data standard, using custom UDF function to adapt special format, through stream batch processing engine to process mixed data stream.

[0006] Further, the output of hydrological fusion dataset with credibility label, specifically includes: Based on OWL to build hydrology domain ontology tree, using sequence labeling model to extract hydrology exclusive entities and relationships from unstructured part of the standardized input data stream, and building triple storage structure; Through attention mechanism to identify context dependency, combined with regular expression matching library and hydrology domain exclusive dictionary for semantic disambiguation, to establish mapping matrix from heterogeneous data to ontology; The multi-modal correlation generation network is used to realize joint embedding expression of text, image and time series data: the text is converted into a semantic vector, the image is extracted into a defect feature vector, and the sensor data is standardized into a time series vector. The correlation weight between the vectors is dynamically adjusted according to the real-time water load, wherein the correlation proportion of the defect feature and the water supply pressure is increased during the peak water consumption period, and the coupling degree of the defect feature and the water quality index is strengthened during the water quality sensitive period. The hydrological time series cooperative alignment algorithm is used to unify the time stamp system of data with different sampling frequencies, perform time zone conversion and leap second compensation, and use a three-dimensional space alignment algorithm to unify the geographic spatial coordinate system. The ontology evolution strategy based on event triggering is used to dynamically access new concepts and update the linkage rules. After conflict detection and quality evaluation, the hydrological fusion data set is output.

[0007] Further, the establishment of semantic constraints between concepts specifically includes: The semantic similarity between concepts is calculated to establish a constraint relationship, and the semantic similarity is determined based on the ratio of the intersection number of the semantic feature set of two concepts to the total number of respective feature sets; When the semantic similarity is lower than a preset threshold, it is determined that there is a weak correlation or no correlation between the concepts, so that the ontology tree is pruned or marked as a to-be-verified relationship; When the semantic similarity is higher than a preset threshold and there is a logical contradiction, a conflict detection algorithm is triggered for correction to ensure the logical consistency of the ontology structure.

[0008] Further, in the hydrological time series cooperative alignment algorithm, the semantic similarity calculation method between time series is as follows: Wherein, is the semantic similarity between two time series, is the length of the time series, is the length of the time series, is the distance between the i-th point of the time series and the j-th point of the time series.

[0009] Further, the output of the sub-task processing data with traceable identification specifically includes: Access mobile terminals and IoT devices through a heterogeneous device adaptation interface, and use a hybrid networking protocol and a dynamic gradient compression method to transmit data; Security calculation based on homomorphic encryption technology for parameter updating, using a teacher-student network structure to transfer cross-modal feature representation, realizing data heterogeneous adaptation;​​​ The adaptive cooperative cluster captures data correlation patterns through a multi-head parallel attention layer, preferentially calculates the weight of abnormal data under working conditions, suppresses noise interference through an activation function, and generates a hydrological exclusive spatial attention heat map; A multi-objective collaborative optimization model is constructed, and the scene rapid adaptation of weight parameters is realized based on a meta-learning framework: different initial weight strategies are preset for different regions, and the weight distribution is dynamically corrected combined with real-time water load; Through adversarial sample disturbance test and gradient regularization constraint to monitor node health, trigger model rollback mechanism, output the sub-task processing data according to business logic time sequence.

[0010] Further, it also includes, The threshold secret sharing scheme is used to store the full-link data in fragments, and the secure Boolean logic operation is realized by integrating the garbled circuit technology; A consensus verification mechanism and zero-knowledge proof channel based on blockchain are established, and a composite encryption system combined with quantum encryption algorithm is used to protect the security of data transmission and storage; A trusted execution environment is deployed to realize physical-level isolation, a dynamic measurement trust chain is constructed to verify code integrity, and an attribute-based encryption strategy is implemented to divide the permissions into four categories: water quality monitoring, pipe network management, scheduling decision and operation and maintenance management. Different categories use different keys and perform dynamic rotation.

[0011] Further, the optimization output of the hydrological decision support data includes: Hydrological scene constraint reinforcement learning algorithm is used to interact with the environment to learn the optimal strategy, and the environmental state, system state and user state involved in the decision-making process are determined, and the model action and decision direction corresponding to the hydrological operation are determined; According to the hydrological business target, a special multi-objective reward function is designed, the reward weight is dynamically adjusted to adapt to the real-time working condition, and the model is allowed to learn by trial and error in the simulation environment through the reinforcement learning algorithm, and the strategy parameters are adjusted to maximize the cumulative reward; State distribution map and action selection probability map are used to realize the visualization of the decision-making process, real-time processing and analysis of input data, dynamic adjustment of parameters and strategies of the reinforcement learning model according to real-time data and environmental feedback; A feedback mechanism is established to incorporate the decision results and environmental feedback into the model continuous learning process, continuously optimize the decision scheme, determine the model action and decision direction through Q value, and the Q value calculation method is as follows: , Where, is the state , is the value of taking action A, is the immediate reward, is the reward discount factor, for the next state the optimal action that can be taken of value.

[0012] In a second aspect, the present application also provides an AI-based hydrological intelligent management system, comprising: A pre-processing module for receiving multi-source heterogeneous hydrological raw data and performing preprocessing, outputting standardized input data stream with global unique identification and spatio-temporal attributes; A semantic fusion module for performing spatial adaptive semantic analysis and fitness calculation on the standardized input data stream based on a hydrological field semantic normalization framework combined with pipe network topology real-time attributes, outputting a hydrological fusion data set with a dynamic credibility label after hierarchical alignment and compliance verification; A resource pool management module for loading the hydrological fusion data set into a domain-isolated semantic resource pool through a three-dimensional permission dynamic adaptation mechanism, and establishing a dynamic semantic link across business modules in the pool; A model adaptation module for analyzing user intent to calculate scene adaptation index, dynamically selecting the optimal model as the task reference, and calling data from the semantic resource pool to generate scene task instruction data; A collaborative scheduling module for driving an adaptive collaboration cluster based on scene task instruction data, processing in a time sequence protocol chain according to defect identification, risk rating, and scheduling optimization, and outputting task processing data with traceability identification; An integration optimization module for hierarchically integrating the task processing data and calculating collaborative reliability, driving logical reconstruction iteration if the preset threshold is not reached, and finally outputting hydrological decision support data; A feedback calibration module for feeding back the hydrological decision support data and adaptation parameters to the semantic framework and model library to realize cognitive dynamic self-calibration.

[0013] In a third aspect, the present application also provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the AI-based hydrological intelligent management method.

[0014] The present application has at least the following beneficial technical effects: The present application provides an AI-based hydrological intelligent management method and system, which has the following outstanding beneficial effects: Firstly, by obtaining water quality monitoring data, water supply scheduling logs, equipment inspection records, emergency plan documents and user service work orders from hydrological business systems, and implementing unified naming rules, field semantic alignment and metadata annotation on them, a hydrological standard data set with consistent format and traceable source is generated. The improved ResNet-W hydrological equipment defect identification model is designed specifically for the hydrological scene and can accurately identify common defects of core equipment such as pipes, valves and instruments. It is suitable for complex lighting and image blur in the inspection process, and supports lightweight deployment of edge nodes, solving the problem of insufficient adaptability and limited deployment of traditional general models in the hydrological scene. The standardized preprocessing mechanism effectively solves the knowledge fragmentation problem caused by data silos, inconsistent terminology and chaotic structure in traditional hydrological systems, enabling subsequent AI processing to be based on high-quality and consistent data, significantly improving the accuracy and reliability of intelligent services.

[0015] Secondly, the hydrological standard data set is organized according to the permission policy in the three-level hydrological knowledge base system (individual, team, shared), and fine-grained access control is implemented combined with user roles and business attributes. When responding to natural language instructions, AI agents only call data within their authorized range to perform retrieval, generation, review or decision-making tasks, and all output results are accompanied by original data source identifiers. This architecture realizes the safe sharing and precise reuse of knowledge assets, ensuring that sensitive data is not leaked beyond authorization, and supporting efficient cross-department collaboration, solving the contradiction between "data cannot be shared" and "intelligence cannot be landed" in existing systems.

[0016] Thirdly, the entire management process forms a closed-loop data intelligent link of "data access - standardized governance - hierarchical storage - intelligent application - security audit". When new business systems are connected or data specifications are updated, the system can automatically adapt to new source data and dynamically expand the hydrological data dictionary, ensuring the continuous evolution of the knowledge base. At the same time, all operation behaviors and data flow are encrypted and recorded, supporting full life cycle traceability. This mechanism significantly improves the automation, compliance and explainability of hydrological management. Test results show that the report generation efficiency is improved, the review accuracy is increased, and the system meets the third level of network security requirements, with good engineering practicability and promotional value. BRIEF DESCRIPTION OF DRAWINGS

[0017] The invention is further illustrated by the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the invention. For ordinary skilled persons in the art, other drawings can be obtained without creative labor based on the following drawings.

[0018] Figure 1 The working steps of the AI-based hydrological intelligent management method disclosed in one embodiment of the invention are shown in the figure. Figure 2This is a schematic diagram of an AI-based intelligent hydrological management system disclosed in one embodiment of the present invention. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0020] Example 1 refer to Figure 1 This invention provides an embodiment of an AI-based intelligent hydrological management method, comprising the following steps: S1. Receive and preprocess multi-source heterogeneous hydrological raw data to generate a standardized input data stream with a unique identifier and spatiotemporal attributes across the entire region; S2. Based on the semantic normalization framework in the hydrological field, the standardized input data stream is analyzed according to the spatial density difference of the pipeline network: the topological correlation verification is strengthened in the high-density area, and the water quality and water source coupling calculation is emphasized in the low-density area. After semantic fit hierarchical fusion, the hydrological fusion dataset with credibility label is output. S3. Load the hydrological fusion dataset into a domain-isolated semantic resource pool and establish dynamic semantic links through a three-dimensional permission mechanism; S4. Analyze user intent, calculate scene adaptation index, select the optimal model, and generate scene task instruction data from the resource pool. S5, based on scenario task instruction data-driven adaptive collaborative cluster, performs time-series protocol chain collaborative processing from defect identification, risk rating to scheduling optimization, and outputs sub-task processing data with traceability identifier; S6. Integrate the data from the sub-tasks and calculate the collaborative reliability. If the threshold is not reached, drive the logic to reconstruct and iterate, output hydrological decision support data, and feed the hydrological decision support data and parameters back to the semantic framework and model library to achieve cognitive dynamic self-calibration.

[0021] Furthermore, step S1 also includes preprocessing the multi-source heterogeneous hydrological raw data, the preprocessing steps of which include: Local data preprocessing is performed through edge computing nodes, a multi-protocol adaptive conversion mechanism is adapted to data access of various types of industrial equipment, mixed encapsulation of time-series and non-time-series data is realized, and a ring buffer dynamic expansion mechanism is adopted to dynamically adjust the storage capacity. In the event of an abnormal network outage, local data preservation and storage are initiated, and data transmission is automatically triggered after the network is restored. It supports compliant access of third-party systems through a two-way security authentication mechanism and is compatible with both relational and non-relational databases. A hydrological data ingestion engine is used to achieve real-time data ingestion and dynamic subscription. Unstructured data is parsed and processed. Text data is parsed through an NLP processing pipeline and log parsing tools. Pipeline inspection images are processed using an improved ResNet-W hydrological equipment defect identification model. Pipeline leakage audio is processed using an FFT spectrum hydrological noise filtering algorithm and a speech-to-text engine. In practical applications, the improved ResNet-W hydrological equipment defect identification model has been customized for the specific needs of hydrological inspection scenarios. In outdoor pipeline inspection scenarios, mobile inspection robots equipped with high-definition cameras and handheld inspection terminals collect images of equipment such as pipeline outer walls, valve interfaces, and instrument panels. After preliminary cropping and illumination equalization preprocessing by edge computing nodes, the images are input into the model for feature extraction. The model adds a dedicated convolutional kernel for hydrological equipment defects on the basis of the traditional ResNet network. It enhances the ability to extract subtle features for six high-frequency defects, including pipeline corrosion, valve leakage, instrument glass breakage, loose interfaces, flange corrosion, and missing manhole covers. It also focuses on key parts of the equipment through an attention mechanism to avoid interference from the background environment. Especially in a city's water supply network inspection project, for the low-light environment of underground tunnel inspection, the model has a built-in adaptive illumination compensation algorithm to improve the image signal-to-noise ratio and maintain a high level of defect recognition accuracy. For the dense equipment scene in the water plant area, the model supports simultaneous recognition of multiple targets. A single image can simultaneously detect multiple defects of multiple devices, and the processing latency meets the real-time feedback requirements of the inspection robot. For the complex corrosion morphology of old pipelines, the model incorporates the feature patterns of a large number of hydrological defect annotation samples through transfer learning, and has a high accuracy in distinguishing between normal corrosion and risky corrosion, effectively reducing the false alarm rate. The model output is synchronized to the semantic resource pool in a structured format. Each identification result includes the defect type, confidence level, unique device identifier, collection timestamp, and geographic coordinates. The subsequent semantic parsing module provides decision-making support to the agent collaboration module by mapping defect types to processing priorities in the ontology knowledge base. For example, upon identifying a valve leakage defect, it is automatically marked as a level-two response, triggering the pipeline fault diagnosis agent to extract historical maintenance data of the valve and pressure data of associated pipelines, generating collaborative handling suggestions.

[0022] AvroSchema is built to standardize data, custom UDF functions are used to adapt to special data formats, and mixed data streams are processed through real-time and offline dual channels using the Flink stream processing engine and Spark batch processing cluster respectively. RocksDB is used to implement window state persistence and breakpoint resume, and a priority scheduling queue for hydrological data and an adaptive back pressure control mechanism are configured.

[0023] Furthermore, the semantic parsing and data fusion include: Based on OWL, an ontology tree for the hydrological domain is constructed. A sequence labeling model is used to extract hydrological-specific entities and relationships from the unstructured part of the standardized input data stream, and a triplet storage structure is constructed. By identifying contextual dependencies through an attention mechanism, semantic disambiguation is performed by combining a regular expression matching library with a hydrology-specific dictionary, and a mapping matrix from heterogeneous data to ontology is established. A multimodal association generative network is used to achieve joint embedding and representation of text, images and time series data: text is converted into semantic vectors, images are extracted into defect feature vectors, sensor data is standardized into time series vectors, and the association weights between various vectors are dynamically adjusted according to real-time water load. In particular, the association ratio between defect features and water supply pressure is increased during peak water use periods, and the coupling degree between defect features and water quality indicators is strengthened during water quality sensitive periods. The timestamp system of data with different sampling frequencies is unified by the hydrological time series collaborative alignment algorithm, time zone conversion and leap second compensation are performed, and the geospatial coordinate system is unified by the three-dimensional spatial alignment algorithm. The event-triggered ontology evolution strategy dynamically incorporates new concepts and updates linkage rules. After conflict detection and quality evaluation, the hydrological fusion dataset is output.

[0024] Based on semantic parsing and data fusion, the method further includes a step of establishing semantic constraints between concepts. This step specifically includes: This also includes establishing semantic constraints between concepts, and the specific process is as follows: The semantic similarity between concepts is calculated to establish a constraint relationship. The semantic similarity is determined based on the ratio of the number of intersections of the semantic feature sets of the two concepts to the total number of their respective feature sets. When the semantic similarity is lower than a preset threshold, it is determined that there is a weak or no relationship between the concepts, and thus the relationship is pruned or marked as a relationship to be verified in the ontology tree. When the semantic similarity exceeds a preset threshold and there is a logical contradiction, a conflict detection algorithm is triggered to make corrections and ensure the logical consistency of the ontology structure.

[0025] Furthermore, in the hydrological time series collaborative alignment algorithm, the semantic similarity calculation method between time series is as follows: in, To determine the semantic similarity between the two time series, for The length of the time series, for The length of the time series, for The first time series points and The first time series The distance between points.

[0026] Furthermore, the intelligent agent is capable of extracting specific task data and performing processing, including: It supports the access of multiple types of nodes for mobile terminals and IoT devices through heterogeneous device adapter interfaces, adopts the ZigBee-Mesh hybrid networking protocol, and achieves dynamic gradient compression through Top-K sparsity and quantization coding methods. Secure computation of parameter update values ​​is performed based on Paillier homomorphic encryption. A mapping matrix between shared and private parameter spaces is constructed, which is compatible with local model structures of different modalities. Cross-modal feature representations are transmitted using teacher-student network structures to achieve heterogeneous data adaptation. A multi-head parallel attention layer is employed to capture the unique association patterns of various hydrological data types. Priority is given to calculating the association weights of abnormal operating data. A sigmoid function is used to dynamically activate key modal channels to suppress noise interference. Timestamp differences between multimodal data are aligned, and a 3D spatial attention map is used to generate a hydrological-specific spatial attention heatmap. A multi-objective collaborative optimization model for hydrology is constructed to balance the dynamic requirements of water quality compliance, energy consumption scheduling, and operation and maintenance costs. Based on the MAML framework, the weight parameters are rapidly adapted to different hydrological scenarios. During implementation, initial weights are preset for different hydrological scenarios: initial weights for areas surrounding water plants prioritize water quality compliance, initial weights for older residential areas prioritize operation and maintenance costs, and initial weights for commercial centers prioritize water supply stability. The weight allocation is then dynamically adjusted based on real-time water load to ensure the model adapts to different scenario requirements. The model will synchronously close... The system prioritizes real-time water load adjustment and optimization, ensuring water supply stability during peak water usage periods, focusing on optimizing water quality compliance rates during water quality-sensitive periods, and prioritizing cost reduction in aging pipeline areas. A dedicated multi-objective reward function is designed based on hydrological business objectives, with weights dynamically adjusted according to scenarios during implementation: the highest weight is given to rewards related to water quality compliance in drinking water protection zones, the increased weight is given to rewards related to energy consumption reduction in industrial water areas, and priority is given to rewards related to maintenance response in densely populated residential areas. The weight adjustment cycle is synchronized with the water load monitoring cycle to ensure the reward mechanism aligns with actual hydrological operating conditions. Specifically, the weight of rewards related to water quality compliance is increased in drinking water protection zones, the weight of rewards related to energy consumption reduction is increased during peak electricity usage periods, and the weight of rewards related to maintenance response is increased in densely populated pipeline areas. This scenario-based weight allocation adapts to actual hydrological needs. Adversarial sample perturbations are generated by simulating virtual adversarial attacks to disrupt multimodal data streams. Lipschitz continuity constraints are added using gradient regularization to monitor node health and evaluate client computing resource status in real time, triggering a model rollback mechanism. Backup feature vectors of core modes are retained through a hydrological core feature cache pool.

[0027] Furthermore, the method also includes a trusted computing process, specifically comprising: The Shamir threshold scheme is used for data sharding and storage, dynamic thresholds are set, and obfuscated circuit technology is integrated to realize the Boolean logic transformation of asymmetric computing tasks. A consensus verification mechanism based on blockchain is designed, and each computing node completes the Proof-of-Work (PoW). Establish a zero-knowledge proof verification channel to ensure the traceability of the calculation process, implement the CKKS scheme to support floating-point operations and control error precision, integrate the TFHE library to improve the speed of ciphertext operation, and construct a hydrological graded composite encryption system to achieve LWE-based post-quantum encryption in combination with national cryptographic algorithms. A trusted execution environment is deployed to achieve physical-level security isolation. A dynamic metric trust chain is constructed to verify the integrity of the executed code in real time. A hydrological business domain attribute-based encryption strategy is implemented to implement fine-grained access control according to the business domain. Specifically, the permissions are divided into four categories: water quality monitoring, pipeline management, scheduling decision-making, and operation and maintenance management. Among them, water quality monitoring permissions are only open to environmental protection specialists and water quality analysis intelligent agents, pipeline management permissions are restricted to operation and maintenance personnel and fault diagnosis intelligent agents, and scheduling decision-making permissions are only authorized to the scheduling center and optimization intelligent agents. Different encryption keys are used for different permission categories. The key rotation cycle is dynamically configured according to the sensitivity of business domain data. A dynamic key rotation mechanism is designed, and the cycle can be configured as needed. A hydrological key sharding storage scheme combined with a threshold signature mechanism is adopted to realize key storage, a key lifecycle management mechanism is deployed, the national cryptographic SM2 / SM4 algorithm is supported, the zk-SNARKs zero-knowledge proof framework is integrated to generate computational correctness proof, it is connected to the National Time Service Center NTP server, and the Merkle tree structure is used to solidify the hydrological full-link operation log.

[0028] Furthermore, the optimized output of the hydrological decision support data includes: We adopt a hydrological scenario-constrained reinforcement learning algorithm and interact with the environment to learn the optimal strategy, clarify the environmental state, system state, and user state involved in the decision-making process, and determine the model actions and decision directions corresponding to hydrological practice. Based on the hydrological business objectives, a dedicated multi-objective reward function is designed, and the weights of each reward are dynamically adjusted to adapt to real-time operating conditions. Through reinforcement learning algorithms, the model learns by trial and error in a simulated environment, and the strategy parameters are adjusted to maximize the cumulative reward. In the hydrological scenario-constrained reinforcement learning algorithm, the state space contains hydrological-specific pipeline topology parameters, such as pipe length, valve distribution, and water source distance, while the action space is bound to hydrological practical operations, such as zone pressurization amplitude, pump start-stop combination, and defect handling priority, ensuring that the learning strategy fits the actual hydrological operation and maintenance scenario rather than general scheduling logic. The decision-making process is visualized by using various charts such as state distribution diagrams and action selection probability diagrams. Input data is processed and analyzed in real time, and the parameters and strategies of the reinforcement learning model are dynamically adjusted based on real-time data and environmental feedback. A feedback mechanism is established to incorporate decision-making results and environmental feedback into the model's continuous learning process, constantly optimizing decision-making schemes. The Q-value is used to determine the model's actions and decision-making direction. The Q-value is calculated as follows: in, For state Take action A below value, For immediate reward, As a reward discount factor, For the next state The best possible action to take of value.

[0029] In some embodiments, source data is typically collected in real-time or periodically from hydrological business systems such as SCADA systems, work order platforms, inspection apps, and emergency management systems via API interfaces or message queues. Structured source data includes time-series data from water quality sensors (such as pH, turbidity, and residual chlorine) and dispatch instruction logs; unstructured source data includes emergency plans in PDF format, inspection reports in Word format, and service work orders in Excel format. All data is sent to the data standardization module after passing through a unified access gateway to generate a hydrological standard dataset, which is then mapped to a three-level knowledge base according to user identity and organizational structure. When a user inputs a natural language command such as "turbidity anomaly analysis of a certain water plant in the past week" via the web or mobile terminal, the NLU engine parses the intent and triggers the AI ​​agent, which retrieves data from the corresponding knowledge base, executes the task, and returns structured results (such as a JSON-formatted analysis summary or a Word report).

[0030] In its implementation, the system performs standardized preprocessing on raw data collected from various hydrological operational systems. First, addressing the differences in terminology, unit formats, and structural forms among data from different sources, the system maps heterogeneous descriptions of the same physical quantity or operational concept to consistent semantic identifiers based on pre-defined hydrological semantic standards, eliminating ambiguities caused by inconsistent naming. Second, the system performs quality verification on the data content, identifying and correcting outliers that significantly deviate from normal operating conditions. Simultaneously, considering the continuity of time series and the rationality of operational logic, missing or invalid data is imputed or marked. Based on this, structured metadata is added to each processed data record, including its original source system, collection time, operational category, and responsible unit, ensuring data traceability and verifiability. After these processes, a standardized hydrological dataset with uniform format, consistent semantics, reliable quality, and complete traceability information is formed, providing a high-quality data foundation for subsequent knowledge base construction and intelligent applications.

[0031] In some embodiments, a personal knowledge base space is automatically created for each registered user, with capacity quotas configured by the administrator. When a user uploads an equipment inspection report, the system converts it into a hydrological standard dataset and stores it in their personal knowledge base. If the user belongs to the "Pipeline Maintenance Group," the data can be optionally synchronized to the team knowledge base, where only that user can edit it, while other members of the group can read it. After approval by the department head, key contingency plans can be published to the shared knowledge base with a "preview only" policy set. All knowledge bases use a vector database (such as Milvus) at their underlying layer, embedding text blocks into 768-dimensional vectors to support subsequent semantic retrieval.

[0032] In some embodiments, when a user inputs "find records of the handling of last year's pipe burst incident," the system first determines the scope of the knowledge base that the user can access (e.g., only the shared knowledge base of this region + personal knowledge base) through the permission engine. The query statement is then encoded into a query vector by a BERT fine-tuned model. Calculate the embeddings of all data blocks in the vector library. The system calculates the cosine semantic similarity and returns the Top-K results. Each result includes the original file ID, the matched paragraph, the collection time, and the source system, with highlighted keywords. The system is deployed on an 8-core, 16GB server, and has an average response time of 2.4 seconds with 100,000 records.

[0033] The search score is calculated as follows: The first term represents the dense vector semantic similarity, the second term represents the sparse keyword matching score (BM25), and λ=0.3 is the fusion weight. This hybrid retrieval strategy balances semantic generalization and precise keyword recall, improving the precision in hydrological terminology scenarios.

[0034] In some embodiments, compliance auditing, decision support and security control functions are integrated into a unified intelligent hydrological processing flow. In a city's water supply network operation and maintenance scenario, the system achieves differentiated intelligent management through cross-module linkage: In the preprocessing stage, the improved ResNet-W model identifies pipeline corrosion defects in a certain pipe section and synchronizes structured data such as defect type, location, and confidence level to the semantic resource pool; the semantic normalization fusion module calls the water quality-pipeline association rules to match the water quality monitoring point data corresponding to the pipe section and detects abnormal turbidity; the pipeline fault diagnosis agent combines the pipeline topology attributes in the semantic framework to determine the leakage risk that corrosion defects may cause; the scheduling optimization agent, based on a hydrology-specific strategy using reinforcement learning, adjusts the water supply pressure in the area to a safe range, while simultaneously triggering the operation and maintenance agent to generate a work order; the entire process is mediated by hydrology... A dedicated semantic association link and hydrological scenario business flow protocol enable full-process automation: In the preprocessing stage, the improved ResNet-W model identifies pipeline corrosion defects and simultaneously outputs structured data such as the defect type, equipment unique identifier, and geographic coordinates to the semantic resource pool; the semantic normalization fusion module calls the hydrological dynamic association rules to match the water quality monitoring data corresponding to the pipe segment with the real-time attributes of the pipeline network topology, and marks high risk after detecting turbidity anomalies; the pipeline network fault diagnosis intelligent agent extracts the historical operation and maintenance data of the valve and the associated pipeline pressure data; the scheduling optimization intelligent agent adjusts the water supply pressure based on real-time water load; finally, the operation and maintenance intelligent agent is triggered to generate a work order, forming a hydrological-specific closed loop, which is different from the step-by-step processing logic of general data fusion systems. When the system receives documents to be reviewed (such as water supply engineering design plans, water quality anomaly handling records, etc.) and the review requirements specified by the user, it first retrieves relevant compliance data from the three-level hydrological knowledge base system, including currently effective legal provisions, technical standards, local normative documents, and historical approval cases. Subsequently, the AI ​​agent compares the content of the documents to be reviewed with the compliance data item by item, identifies whether there are problems such as missing key elements, exceeding numerical limits, or procedural violations, and marks the location of the problem, cites specific clauses, explains the nature of the risk, and proposes correction suggestions in the output results, forming a clear and well-founded review opinion. In terms of decision support, the system can automatically aggregate multi-source hydrological standard datasets within the scope of permissions according to the management needs proposed by the user (such as scheduling optimization, risk warning, or emergency response), and combine them with business rules and operational constraints to generate structured decision suggestions that include risk assessment, impact analysis, and disposal measures. All conclusions are accompanied by information on the original data source to ensure traceability. Meanwhile, the system implements strict security controls throughout the entire process: all data is encrypted using AES-256 during storage and uses the TLS 1.3 protocol during transmission; user access permissions for individuals, teams, and shared knowledge bases are dynamically allocated based on their roles, affiliated units, and business attributes, ensuring that only authorized data can be accessed; any data query, generation, download, or modification is recorded in an audit log, including the operation subject, time, object, and operation type. The log is encrypted and tamper-proof, meeting the hydrological industry's regulatory requirements for data security and compliance auditing. Through these mechanisms, a unified approach to precise auditing, intelligent decision-making, and secure management is achieved.

[0035] refer to Figure 2 Secondly, the present invention also provides an AI-based intelligent hydrological management system, comprising: The preprocessing module is used to receive multi-source heterogeneous hydrological raw data and perform preprocessing, outputting a standardized input data stream with a globally unique identifier and spatiotemporal attributes; The semantic fusion module is used to perform spatial adaptive semantic parsing and fit calculation on the standardized input data stream based on the semantic normalization framework in the hydrological field and combined with the real-time attributes of the pipeline network topology. After hierarchical alignment and compliance verification, it outputs a hydrological fusion dataset with dynamic credibility labels. The resource pool management module is used to load the hydrological fusion dataset into a domain-isolated semantic resource pool through a three-dimensional permission dynamic adaptation mechanism, and to establish a dynamic semantic link across business modules within the pool. The model adaptation module is used to parse user intent, calculate the scenario adaptation index, dynamically select the optimal model as the task benchmark, and call data from the semantic resource pool to generate scenario task instruction data. The collaborative scheduling module is used to drive an adaptive collaborative cluster based on scenario task instruction data. It performs time-series protocol chain-style collaborative processing from defect identification and risk rating to scheduling optimization, and outputs sub-task processing data with traceability identifiers. The integration and optimization module is used to hierarchically integrate the sub-task processing data and calculate the collaborative reliability. If the preset threshold is not reached, the logic reconstruction iteration is driven, and finally hydrological decision support data is output. The feedback calibration module is used to feed back the hydrological decision support data and adaptation parameters to the semantic framework and model library to achieve cognitive dynamic self-calibration.

[0036] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any of the AI-based intelligent hydrological management methods described in the present invention.

[0037] This embodiment provides an AI-based intelligent hydrological management method, system, and storage medium. By aggregating multi-source heterogeneous data from hydrological business systems, including water quality monitoring information, water supply scheduling records, equipment inspection documents, emergency plan texts, and user service work orders, and performing unified semantic standardization, structural alignment, and metadata enhancement processing on these data, a hydrological standard dataset with consistent format, clear semantics, and traceable source is formed. This standardized governance mechanism effectively overcomes the knowledge fragmentation problem caused by system fragmentation, terminology confusion, and structural disorder in traditional hydrological informatization, providing a high-quality and highly consistent data foundation for subsequent artificial intelligence applications, thereby significantly improving the accuracy, stability, and reliability of intelligent services. Based on a three-tiered knowledge base architecture of individuals, teams, and shared resources, the hydrological standard dataset is organized hierarchically according to preset permission policies, and fine-grained access control is implemented in conjunction with user roles and specific business scenarios. When responding to natural language commands, the AI ​​agent can only call relevant data within the authorized scope to perform tasks such as retrieval, content generation, compliance review, or decision support, and all output results are automatically associated with the original data source. This design achieves secure sharing and precise reuse of knowledge assets while ensuring that sensitive information is not accessed without authorization, effectively resolving the current contradiction between "unwillingness to share data" and "difficulty in implementing intelligence" in hydrological management. A complete closed-loop chain has been constructed, encompassing data access, standardized governance, hierarchical storage, intelligent applications, and security auditing. When new business systems are integrated or industry data standards are updated, the system can automatically identify and adapt to the new data types, dynamically expanding semantic mapping rules to ensure the continuous evolution of the knowledge system. Simultaneously, all user operations and data flow behaviors are encrypted and recorded, and included in the audit log, supporting full lifecycle traceability. Practical applications demonstrate that this method can improve the efficiency of automatic report generation, increase the accuracy of audit tasks, and fully meet the requirements of Level 3 Cybersecurity Protection, possessing strong engineering feasibility and industry promotion value.

[0038] In the description of this specification, the references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0039] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. An AI-based intelligent hydrological management method, characterized in that, Includes the following steps: Receive and preprocess multi-source heterogeneous hydrological raw data to generate a standardized input data stream with a unique identifier and spatiotemporal attributes for the entire region; Based on the semantic normalization framework in the field of hydrology, the standardized input data stream is analyzed according to the spatial density difference of the pipeline network: the topological correlation verification is strengthened in the high-density area, and the water quality and water source coupling calculation is focused on the low-density area. After semantic fit hierarchical fusion, the hydrological fusion dataset with credibility label is output. The hydrological fusion dataset is loaded into a domain-isolated semantic resource pool and a dynamic semantic link is established through a three-dimensional permission mechanism. Analyze user intent to calculate scenario adaptation index, select the optimal model to generate scenario task instruction data from resource pool; Based on scenario-based task instruction data-driven adaptive collaborative clusters, it performs time-series protocol chain-style collaborative processing from defect identification and risk rating to scheduling optimization, and outputs sub-task processing data with traceability identifiers. The system integrates data from different tasks and calculates collaborative reliability. When the threshold is not reached, it drives logical reconstruction iteration, outputs hydrological decision support data, and feeds back the hydrological decision support data and parameters to the semantic framework and model library to achieve cognitive dynamic self-calibration.

2. The AI-based intelligent hydrological management method according to claim 1, characterized in that, The generation of a standardized input data stream with a globally unique identifier and spatiotemporal attributes specifically includes: The multi-source heterogeneous hydrological raw data is cleaned locally using edge computing nodes, and the industrial equipment interface is unified through a multi-protocol adaptive conversion mechanism. Time-series sensor data and non-time-series business data are mixed and encapsulated. In the event of an abnormal network outage, local data preservation and storage are initiated, and data transmission is automatically resumed based on a two-way security authentication mechanism after the network is restored. Real-time data acquisition and dynamic subscription are performed through a hydrological data acquisition engine. NLP is used to process the pipeline parsing of text logs. An improved defect identification model is used to extract pipeline inspection image features. A spectrum filtering algorithm is applied to process pipeline leakage audio, transforming the above unstructured data into structured intermediate data. Based on AvroSchema, a data standard is defined. Custom UDF functions are used to adapt to special formats, and a stream-batch integrated processing engine is used to process mixed data streams.

3. The AI-based intelligent hydrological management method according to claim 1, characterized in that, The output hydrological fusion dataset with confidence labels specifically includes: Based on OWL, an ontology tree for the hydrological domain is constructed. A sequence labeling model is used to extract hydrological-specific entities and relationships from the unstructured part of the standardized input data stream, and a triplet storage structure is constructed. By identifying contextual dependencies through an attention mechanism, semantic disambiguation is performed by combining a regular expression matching library with a hydrology-specific dictionary, and a mapping matrix from heterogeneous data to ontology is established. A multimodal association generative network is used to achieve joint embedding and representation of text, images and time series data: text is converted into semantic vectors, images are extracted into defect feature vectors, sensor data is standardized into time series vectors, and the association weights between various vectors are dynamically adjusted according to real-time water load. In particular, the association ratio between defect features and water supply pressure is increased during peak water use periods, and the coupling degree between defect features and water quality indicators is strengthened during water quality sensitive periods. The timestamp system of data with different sampling frequencies is unified by the hydrological time series collaborative alignment algorithm, time zone conversion and leap second compensation are performed, and the geospatial coordinate system is unified by the three-dimensional spatial alignment algorithm. The event-triggered ontology evolution strategy dynamically incorporates new concepts and updates linkage rules. After conflict detection and quality evaluation, the hydrological fusion dataset is output.

4. The AI-based intelligent hydrological management method according to claim 3, characterized in that, This also includes establishing semantic constraints between concepts, and the specific process is as follows: The semantic similarity between concepts is calculated to establish a constraint relationship. The semantic similarity is determined based on the ratio of the number of intersections of the semantic feature sets of the two concepts to the total number of their respective feature sets. When the semantic similarity is lower than a preset threshold, it is determined that there is a weak or no relationship between the concepts, and thus the relationship is pruned or marked as a relationship to be verified in the ontology tree. When the semantic similarity exceeds a preset threshold and there is a logical contradiction, a conflict detection algorithm is triggered to make corrections and ensure the logical consistency of the ontology structure.

5. The AI-based intelligent hydrological management method according to claim 3, characterized in that, In the hydrological time series collaborative alignment algorithm, the semantic similarity between time series is calculated as follows: in, To determine the semantic similarity between the two time series, for The length of the time series, for The length of the time series, for The first time series points and The first time series The distance between points.

6. The AI-based intelligent hydrological management method according to claim 1, characterized in that, The output of task-specific processing data with traceability identifiers specifically includes: It connects to mobile terminals and IoT devices through heterogeneous device adapter interfaces, and transmits data using a hybrid networking protocol and dynamic gradient compression method. Secure computation for parameter updates is performed based on homomorphic encryption technology, and cross-modal feature representations are transmitted using a teacher-student network structure to achieve heterogeneous data adaptation. The adaptive collaborative cluster captures data correlation patterns through a multi-head parallel attention layer, prioritizes the calculation of weights for abnormal operating data, suppresses noise interference through activation functions, and generates a hydrological-specific spatial attention heatmap. Construct a multi-objective collaborative optimization model and achieve rapid scenario adaptation of weight parameters based on a meta-learning framework: preset differentiated initial weight strategies for different regions and dynamically adjust weight allocation in combination with real-time water load. By monitoring node health through adversarial sample perturbation testing and gradient regularization constraints, the model rollback mechanism is triggered, and the sub-task processing data is output according to the business logic sequence.

7. The AI-based intelligent hydrological management method according to claim 1, characterized in that, It also includes, A threshold secret sharing scheme is adopted to segment and store the data across the entire link, and obfuscation circuit technology is integrated to achieve secure Boolean logic operations; Establish a blockchain-based consensus verification mechanism and zero-knowledge proof channel, and adopt a composite encryption system combined with quantum encryption algorithms to protect data transmission and storage security; Deploy a trusted execution environment to achieve physical-level isolation, build a dynamic metric trust chain to verify code integrity, and implement an attribute-based encryption strategy to finely divide permissions into four categories: water quality monitoring, pipeline management, scheduling decision-making, and operation and maintenance management. Different categories use differentiated keys and perform dynamic rotation.

8. The AI-based intelligent hydrological management method according to claim 1, characterized in that, The optimized output of the hydrological decision support data includes: We adopt a hydrological scenario-constrained reinforcement learning algorithm and interact with the environment to learn the optimal strategy, clarify the environmental state, system state, and user state involved in the decision-making process, and determine the model actions and decision directions corresponding to hydrological practice. Design a dedicated multi-objective reward function based on hydrological business objectives, dynamically adjust the weight of each reward to adapt to real-time operating conditions, and use reinforcement learning algorithms to allow the model to learn through trial and error in a simulated environment, adjusting strategy parameters to maximize cumulative rewards. The decision-making process is visualized by using state distribution diagrams and action selection probability diagrams, and input data is processed and analyzed in real time. The parameters and strategies of the reinforcement learning model are dynamically adjusted based on real-time data and environmental feedback. A feedback mechanism is established to incorporate decision-making results and environmental feedback into the model's continuous learning process, constantly optimizing decision-making schemes. The Q-value is used to determine the model's actions and decision-making direction. The Q-value is calculated as follows: in, For state Take action A below value, For immediate reward, As a reward discount factor, For the next state The best possible action to take of value.

9. An AI-based intelligent hydrological management system, characterized in that, include: The preprocessing module is used to receive multi-source heterogeneous hydrological raw data and perform preprocessing, outputting a standardized input data stream with a globally unique identifier and spatiotemporal attributes; The semantic fusion module is used to perform spatial adaptive semantic parsing and fit calculation on the standardized input data stream based on the semantic normalization framework in the hydrological field and combined with the real-time attributes of the pipeline network topology. After hierarchical alignment and compliance verification, it outputs a hydrological fusion dataset with dynamic credibility labels. The resource pool management module is used to load the hydrological fusion dataset into a domain-isolated semantic resource pool through a three-dimensional permission dynamic adaptation mechanism, and to establish a dynamic semantic link across business modules within the pool. The model adaptation module is used to parse user intent, calculate the scenario adaptation index, dynamically select the optimal model as the task benchmark, and call data from the semantic resource pool to generate scenario task instruction data. The collaborative scheduling module is used to drive an adaptive collaborative cluster based on scenario task instruction data. It performs time-series protocol chain-style collaborative processing from defect identification and risk rating to scheduling optimization, and outputs sub-task processing data with traceability identifiers. The integration and optimization module is used to hierarchically integrate the sub-task processing data and calculate the collaborative reliability. If the preset threshold is not reached, the logic reconstruction iteration is driven, and finally hydrological decision support data is output. The feedback calibration module is used to feed back the hydrological decision support data and adaptation parameters to the semantic framework and model library to achieve cognitive dynamic self-calibration.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the AI-based intelligent hydrological management method as described in any one of claims 1 to 8.

Images