Intelligent question answering system of water affair service system

Abstract

The invention discloses an intelligent question-answering system of a water affair service system, which adopts a four-layer architecture of a sensing layer, a control layer, a computing layer and an application layer, realizes closed-loop intelligent management from data sensing to equipment execution, and ensures data quality from the source by the sensing layer through a self-checking water affair data acquisition device; and the calculation layer processes multi-source data by using a time sequence mechanism double-drive fusion algorithm, analyzes user questions by using an intelligent question and answer engine, and generates answers containing business decisions and equipment operation instructions in combination with a water affair knowledge base. The system has the prominent advantages that after a question and answer equipment linkage control unit of a deep linkage control layer receives an instruction, the real-time mechanical performance of field equipment is verified in advance, it is ensured that the instruction can be executed and then issued, the execution progress is monitored in the whole process, synchronous feedback is conducted, the data reliability and decision scientificity are fundamentally improved, and the system is suitable for popularization and application. Question and answer output is established on credible data, and decision predictability and accuracy are improved in emergency scheduling and other scenes.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of intelligent water affairs and industrial automation, and specifically relates to a water affairs business system integrating intelligent question answering, device linkage and self-evolution decision. BACKGROUND

[0002] Intelligent water affairs is a key path to promote the digital transformation of the water industry and realize fine and intelligent operation and management. The intelligent question answering module in the core business system aims to provide instant and professional decision support for operation and maintenance, dispatching and management personnel. Currently, the technical practice in this field mainly relies on traditional search engines, enterprise internal document libraries and general public large language models.

[0003] However, these existing technical solutions have significant limitations when dealing with highly specialized and strongly process-driven water business. First, the low knowledge integration leads to a serious lack of professional answering. General models lack understanding of water-specific terminology, specifications and enterprise implicit experience, and their answers are often based on generalized knowledge, which is disconnected from enterprise operation procedures and prone to misleading illusions, making them unable to provide professional business guidance. Second, the poor real-time performance of business data leads to delayed decision support. The question answering system is isolated from real-time production systems such as SCADA and GIS, and cannot access and integrate dynamic data such as pipe network pressure and device status, making the system unable to effectively respond to current water pressure anomaly analysis and other core problems that require real-time situational awareness, greatly reducing the value of decision-making. Third, the lack of system integration and operation capability, with single function. Existing technologies mostly stop at passive answering and cannot be linked with internal control systems to perform specific operations (such as starting and stopping water pumps), creating a gap between analysis and execution and hindering the realization of closed-loop automation. In addition, there are high data security risks and high costs of expansion and maintenance. Using public cloud models to process core business data poses a risk of leakage; at the same time, knowledge updating and system expansion are heavily dependent on manual and customized development, making the system rigid and difficult to adapt to the rapid evolution of business.

[0004] Therefore, in the face of the complex dilemma of insufficient professional knowledge, real-time decision-making disconnection, execution closed loop missing, and poor safety and cost controllability in existing technologies, it is an urgent and necessary task to develop a specialized intelligent question answering system that can deeply integrate field knowledge and real-time business data, and has the ability of safe closed-loop execution and flexible expansion, to promote the intelligent upgrading of the water industry. SUMMARY

[0005] The present application provides a water affairs business system intelligent question answering system, which aims to overcome the defects of existing water affairs system data distortion, decision-making and execution disconnection, lack of professional mechanism support, etc., and realize the leap from passive information answering to active closed-loop management and control through the construction of a full-closed-loop intelligent system of reliable perception, scientific decision-making, precise execution and continuous evolution.

[0006] The technical problem of the present application is solved by a water service business system intelligent question and answer system, comprising: a perception layer comprising a self-checking water service data acquisition device for acquiring multi-source data related to water service business; a control layer comprising a question and answer-equipment linkage control unit in communication with the perception layer for directly controlling water service field equipment and acquiring its real-time mechanical performance parameters; a calculation layer comprising a data processing module and an intelligent question and answer engine in communication with each other; the data processing module is in communication with the perception layer and the control layer for receiving the multi-source data and real-time mechanical performance parameters and performing fusion processing thereof using a time sequence-mechanism dual drive fusion algorithm; the intelligent question and answer engine is configured to: receive a water service business question raised by a user, analyze the user's question intention in combination with natural language processing technology, and determine the core business scene by matching the graph structure with a water service knowledge base; receive the fusion result of the data processing module, generate a target question and answer result containing business decision suggestions and corresponding equipment operation instructions; and send the equipment operation instructions to the control layer; an application layer in communication with the calculation layer for outputting the business decision suggestions; wherein the question and answer-equipment linkage control unit is further configured to: after receiving the equipment operation instructions, verify the device executability of the equipment operation instructions based on the acquired real-time mechanical performance parameters; if executable, the instructions are issued to the corresponding water service field equipment, and the instruction execution progress is monitored in real time, and the instruction execution progress data is fed back to the application layer for output.

[0007] Preferably, in communication with the calculation layer and the control layer; the system self-evolution module is configured to update the parameters of the water service professional mechanism model in the water service knowledge base and the time sequence-mechanism dual drive fusion algorithm through an online learning mechanism based on business effect data after instruction execution and user feedback.

[0008] Preferably, the core of the system self-evolution module is a digital twin driven by a reinforcement learning intelligent agent; the digital twin runs in parallel with a real water plant, the reinforcement learning intelligent agent tries various operation strategies in the digital twin, and compares the simulation results with real data to automatically correct the model parameters and find new operation strategies, thereby updating the water service knowledge base.

[0009] Preferably, the self-calibrating water data acquisition device is a non-powered auxiliary sampling device, comprising: an anti-clogging and scale removal component, including an elastic scraper and a flow guiding structure embedded in the inner wall of the sampling pipe, which uses the pressure of the water flow in the pipe network to drive the elastic scraper to reciprocate along the pipe wall to remove scale and impurities; an angle adaptive component, including a gravity-balanced ball joint and an arc-shaped limiting guide rail located at the bottom of the sampling probe, used to automatically adjust the probe angle according to the pipe diameter; and a data quality feedback component, including a built-in micro-differential pressure sensor, which mechanically triggers a signal rod to output a data abnormality signal when the sampling channel is blocked and the inlet and outlet pressure difference exceeds a preset threshold.

[0010] Preferably, the time-series-mechanism dual-drive fusion algorithm includes: a time-series calibration module for timestamping multi-source data and embedding mechanical action delay parameters as time-series compensation factors; and a water mechanism embedding module that integrates an activated sludge model (ASM) and a pipe network hydraulic loss model. The dynamic weight adjustment module is used to automatically adjust the weight ratio of data from different sources based on scenario priority.

[0011] Preferably, the time-mechanism dual-drive fusion algorithm further includes a slime mold biomimetic adaptive pipeline scheduling submodule, which is configured to: mimic the pathfinding behavior of slime mold, regard pipeline nodes as food sources, and use pressure and flow demand as stimulus signals to dynamically generate the optimal pump and valve scheduling scheme.

[0012] Preferably, the intelligent question-answering engine integrates multimodal fusion cognitive capabilities, enabling it to receive and parse images of on-site equipment or recordings of abnormal sounds uploaded by users, and to perform comprehensive analysis and fault diagnosis in conjunction with text questions.

[0013] Preferably, the application layer supports an augmented reality (AR) based interactive interface, which is configured to display the business decision suggestions and device operation instructions on the real device in the user's field of vision in a virtual overlay manner, and can automatically verify the task completion status through action recognition.

[0014] The beneficial effects of this invention are: fundamentally improved data reliability; through the self-verification acquisition device, data distortion caused by blockage and angular deviation is effectively curbed from the source; data quality signals provide a reliable basis for subsequent decision-making, so that the output of the question-and-answer system is built on a solid data foundation.

[0015] The scientific rigor and timeliness of decision-making are enhanced simultaneously. The time-series-mechanism dual-driven fusion algorithm ensures that decision recommendations not only conform to the scientific principles of water resources but also accurately match the actual action sequence of on-site equipment. Especially in emergency dispatch scenarios, the predictability and accuracy of decision-making have achieved a qualitative leap.

[0016] The system's closed-loop execution and enhanced reliability, along with the question-and-answer-equipment linkage control, break down information and control barriers, enabling immediate response and execution upon response. The pre-verification mechanism significantly improves the actual executability of operational suggestions, avoiding execution failures or secondary malfunctions due to poor equipment condition, thus significantly enhancing equipment operational safety.

[0017] The system's intelligence level continuously evolves. The self-evolution mechanism enables the system to learn from historical operations and actual results, automatically optimizing its knowledge base and decision-making model. As a result, the question-answering system's response capabilities become increasingly accurate and intelligent over time, giving it long-term value.

[0018] Human-computer interaction efficiency and experience optimization: Multimodal cognition and AR interaction present complex professional information in an intuitive way, reducing the technical threshold for operation and maintenance personnel, improving the efficiency and accuracy of on-site problem handling, and realizing efficient human-computer collaboration. Attached Figure Description

[0019] Figure 1 This is an overall module diagram of the intelligent question-and-answer system for water affairs business systems of the present invention; Figure 2 This is a technical architecture diagram of the intelligent question-and-answer system for water affairs business systems of the present invention; Figure 3 This is a flowchart of the timing mechanism dual-drive fusion algorithm; Figure 4 This is a complete workflow diagram of the system's intelligent question-and-answer and device linkage; Figure 5 This is a flowchart illustrating the implementation method of the intelligent question-and-answer system for water affairs business systems according to the present invention. Figure 6 This is a schematic diagram of a self-calibrating water data acquisition device; Figure 7 yes Figure 6 An explosion diagram; Figure 8 This is a block diagram illustrating the principle of the timing mechanism dual-drive fusion algorithm; Figure 9 This is a schematic diagram of the self-evolution mechanism of a system based on reinforcement learning digital twins; Figure 10 This is a flowchart of the internal logic verification process of the equipment linkage control unit.

[0020] Numbering in the diagram: 1-Pipeline network; 11-Pipe; 12-Pipe flange; 2-Sampling pipeline; 21-Pipe section; 22-Pipe section flange; 3-Elastic scraper assembly; 31-Scraper ring; 32-Central bushing; 33-Fixed plate; 34-Modible sleeve; 35-Modible plate; 36-Positioning sleeve; 4-Self-calibrating water data acquisition device; 41-Inner sleeve; 42-Radial shaft; 43-Central rod; 44-Tail wing; 5-Guide structure; 51-Guide strip; 52-Protrusion; 53-Z-shaped guide; 54-Guide block; 6-Base; 7-Micro differential pressure sensor. Detailed Implementation

[0021] The present invention will be further described in detail below with reference to the embodiments. It should be noted that the embodiments described in this section are preferred implementations of the present invention, but do not mean that the present invention can only be implemented in the following ways.

[0022] The intelligent question-and-answer system for water affairs provided in this embodiment is based on the core concept of building a closed-loop intelligent system that is reliable in data, makes scientific decisions, executes accurately, and is capable of self-optimization. This system ensures the quality of data sources through innovative devices in the perception layer, bridges the gap between information and the physical world through linkage units in the control layer, achieves scientific decision-making through advanced algorithms in the computing layer, and enables the entire system to continuously learn through a system self-evolution module. This system adopts a layered architecture design, such as... Figure 1 and Figure 2 As shown, from bottom to top, it includes a perception layer, a transmission layer, a platform layer, a computing layer, and an interaction layer. This layered architecture realizes a complete closed loop from data perception to intelligent decision-making and then to business execution.

[0023] The perception layer, as the data source of the system, is responsible for collecting multi-dimensional data, ensuring data quality from the source, and forming the basis for subsequent decision analysis. The core of the self-verifying water data acquisition device lies in its non-powered design, combining a clever mechanical structure with water flow dynamics to achieve automatic anti-clogging, angle self-adaptation, and data quality self-verification. The specific implementation includes the following three key components. (For example...) Figure 6 and Figure 7As shown, the first component is an anti-clogging and descaling assembly. This assembly consists of an elastic scraper assembly (3) and a guide structure (5), and is embedded in the inner wall of the sampling pipeline (2). The elastic scraper assembly (3) includes a scraper ring (31), a central bushing (32), a fixed plate (33), a movable sleeve (34), a movable plate (35), and a positioning sleeve (36); the guide structure (5) includes a guide strip (51) and a Z-shaped guide (53) provided on the inner wall of the pipe section (21), as well as a guide block (54) provided on the protrusion (52) of the scraper ring (31) and the movable plate (35). When the water in the pipe network (1) flows through at normal working pressure, the water pressure drives the entire elastic scraper assembly (3) to generate periodic reciprocating motion: under the action of the spring (8), the scraper ring (31) is driven to move towards the water-facing side. At this time, the guide block (54) rotates along the Z-shaped guide (53), which reduces the overlapping area between the fixed plate (33) and the movable plate (35), increases the water-blocking area, and reduces the spring force. When the water pressure is greater than the spring thrust, the assembly is pressed forward, the guide block (54) rotates in the opposite direction, the overlapping area increases, the water-blocking area decreases, and the spring force increases. This cycle is repeated to achieve mechanical scraping of impurities from the pipe wall and sampling port. At the same time, the unique guide structure (5) can change the flow field, guide the formation of a controllable vortex, and carry impurities away from the sampling area, thus ensuring smooth flow from both physical motion and flow field control aspects. Second, the angle adaptive assembly. This assembly is located on the base (6) of the sampling probe. The entire sampling probe, as part of the self-calibrating water data acquisition device (4), consists of an inner sleeve (41), a radial shaft (42), a central rod (43), and a tail fin (44). When the probe is installed in the central sleeve (32) of the elastic scraper assembly (3) through its radial shaft (42), the water flow balance ball joint at the rear of the probe can always maintain its axial direction under the action of water flow. At the same time, the sampling probe installed on the ball joint can automatically adjust its tilt angle through the constraint of the arc-shaped limiting ball seat, ensuring that the sensing surface of the probe is facing the main flow direction of the pipeline. This design effectively avoids the angular deviation caused by the reciprocating movement of the elastic scraper assembly (3), thereby solving the data distortion problems such as low pressure value and inaccurate flow monitoring caused by this. Third, the data quality feedback component. This component is the key to realizing the self-calibrating function. A micro differential pressure sensor (7) is built into the key position of the sampling channel formed by the self-calibrating water data acquisition device (4). This sensor continuously monitors the pressure difference between the inlet and outlet of the sampling channel. When the flow channel is unobstructed, the differential pressure value will remain stable in the low range; once the sampling channel becomes blocked due to impurity accumulation or biological adhesion, the increased flow resistance will cause the inlet and outlet differential pressure to exceed the preset threshold. At this time, the micro differential pressure sensor (7) will mechanically trigger a physical signal rod to generate a switching signal. This signal is interpreted by the system as a data anomaly indicator and synchronized to the computing layer through the transmission network.The computing layer can label the corresponding sensor data as low reliability, or trigger a data repair program based on the water mechanism model (such as using the pipeline network (1) hydraulic model) when the data is completely missing, so as to ensure the continuity and reliability of the data.

[0024] Deep water quality sensing involves adding a surface-enhanced Raman scattering (SERS) module for real-time sensing of micro-pollutants in water quality-sensitive areas such as water plant inlets and key nodes in the pipeline network. This module periodically allows water to flow through a built-in enhanced substrate chip, which, in conjunction with a laser to excite Raman signals, collects spectral data. This data is then analyzed in real-time by a neural network model within an embedded processor. This enables the identification and quantification of specific organic pollutant molecules (such as trace pesticide residues and industrial organic impurities) at extremely low concentrations (e.g., ppb levels) in the water, providing data dimensions that traditional detection methods cannot cover for subsequent precise water quality management decisions.

[0025] Multi-source data integration and acquisition, in addition to real-time monitoring data, involves the perception layer acquiring business data such as equipment operating status, pipeline spatial geographic coordinates, and user water consumption data from SCADA (Supervisory Control and Data Acquisition System), GIS (Geographic Information System), and revenue systems through system interfaces. It also gathers knowledge from water industry standards, internal enterprise operating procedures, technical manuals, and other fields, and accesses external data such as meteorological (precipitation, temperature) and map services, forming a multi-dimensional data input system that combines monitoring data, business data, domain knowledge, and external data.

[0026] The transport layer is the crucial link connecting the perception layer and the platform layer. It establishes dual transmission links through the enterprise intranet (fiber optic, Ethernet) and mobile communication networks (4G / 5G) to securely and reliably transmit various data collected by the perception layer, such as pressure, flow, water quality, equipment status, domain knowledge, and external weather data, to the platform layer using encrypted transmission protocols (such as MQTTs and HTTPS). This layer ensures the integrity, timeliness, and security of data during transmission through dynamic bandwidth allocation and data breakpoint resumption mechanisms, preventing data loss or leakage due to network fluctuations.

[0027] The platform layer, serving as the system's data foundation, undertakes the core functions of unified data storage and basic service integration, providing support for intelligent decision-making in the upper computing layer and user operations in the interaction layer. Multi-type data storage: A hybrid storage architecture is adopted, using relational databases (such as MySQL) to store structured basic business data such as device status and user information; unstructured data such as water quality spectral maps, equipment photos, and technical documents are stored through a distributed file system (such as HDFS); a specially constructed vector database is used to store embedded vectors of domain knowledge, providing efficient support for subsequent knowledge retrieval and augmentation generation. Core service integration: The SpringAI service framework is integrated as the core channel for interaction between the system and the large language model; RAG (Retrieval Augmentation Generation) services are provided, including knowledge base management (knowledge entry, update, and deletion), vector retrieval (rapid matching of similar domain knowledge), and document retrieval (precise location of technical document fragments), providing support for knowledge augmentation decision-making in the computing layer. General capability encapsulation: This layer implements general capabilities such as workflow engine (automatic decomposition and scheduling of business tasks), GIS services (display of pipeline spatial location and regional positioning), and visualization reports (data trend charts and equipment status dashboards), encapsulating complex technical functions into standardized interfaces for upper-layer modules to call.

[0028] The computing layer integrates domain knowledge, real-time data, and intelligent algorithms to achieve the core transformation from data input to decision output. Its core includes the following modules: Agent Framework Service: Based on the model context protocol, it enables dynamic discovery and scheduling of external intelligent agent services (such as water quality analysis agents and equipment diagnostic agents); simultaneously, through the function call mechanism of the Spring AI framework, it encapsulates the APIs of internal business systems (such as pump control APIs and valve regulation APIs) into executable tools for intelligent agents, achieving seamless linkage between intelligent agents and business systems. Workflow Orchestration Engine: For complex business problems such as emergency response to pipe bursts and abnormal water quality control in water plants, it automatically decomposes them into an automatically executable task sequence of data acquisition → anomaly location → solution generation → command issuance, improving the automation and standardization of business processing. The domain-specific model integrates a water affairs expert model deployed locally on the Ollama platform. This model uses the LLaMAFactory fine-tuning framework, based on a fundamental large language model (such as LLaMA2), and combines water affairs domain datasets (such as historical fault cases, pipeline hydraulic calculation data, and water quality treatment solutions) for efficient parameter fine-tuning. It possesses deep water affairs domain knowledge and can accurately understand professional issues such as abnormal water pressure and excessive water turbidity, avoiding the domain knowledge blind spots of general-purpose models. The intelligent routing strategy automatically selects the model based on the domain attributes of the user's question: for professional water affairs questions such as peak water usage prediction for a specific area and pipeline hydraulic loss calculation, it automatically calls the water affairs expert model to ensure decision accuracy; for general questions such as system operation guide queries and historical data export, it calls the lower-cost general-purpose model, balancing decision quality and operating costs. The intelligent question-answering engine, as the core of the decision output, receives user questions from the interaction layer, coordinates various modules to complete the analysis, and generates the final results. Natural Language Processing Module: Performs semantic parsing on user queries, extracting intent (such as cause analysis and handling suggestions) and entities (such as a specific area and abnormal water pressure) to clarify core business concerns. Data Processing Module: Based on data quality signals fed back from the perception layer (such as abnormal data from a certain sensor), filters currently available valid multi-source data (such as pipeline pressure, flow rate, pump station operating status, valve opening), eliminates low-reliability data, and provides clean input for algorithm analysis.

[0029] Timing mechanism dual-drive fusion algorithm such as Figure 3As shown, the core algorithm of the system's decision-making integrates two major capabilities: time-series data calibration and water mechanism simulation. The time-series calibration module aligns the original timestamps of all access data at the millisecond level, while embedding equipment mechanical action delay parameters obtained from the control layer (e.g., the full-stroke opening and closing time of a specific valve model is 15 seconds). For example, when calculating the impact of valve closure on downstream pressure, the actual valve action time is used as a time-series compensation factor to ensure that the decision recommendations are perfectly matched with the physical process on time. The water mechanism embedding module embeds a network hydraulic model, simulating the water flow state of the entire network based on real-time data (e.g., pressure and flow at each node), quickly locating the source of anomalies (e.g., abnormal valve closure causing low downstream water pressure, or a suspected pipe burst causing a sudden increase in flow). The dynamic weight adjustment module adjusts data weights according to core business scenarios. For example, in emergency dispatch scenarios, it automatically increases the weight ratio of equipment status data (e.g., pump vibration values, valve sealing status) and real-time pressure data, ensuring that decision recommendations prioritize equipment safety and water supply stability.

[0030] The interaction layer serves as the system-user interaction window, providing personalized interfaces for different user roles such as maintenance personnel and dispatchers. It supports multiple interaction methods, including a basic interaction terminal: supporting web, mobile app, and voice robot. The web interface is for dispatch center personnel, providing complete system operation and data monitoring functions; the mobile app is for on-site maintenance personnel, supporting data query and problem reporting; and the voice robot supports hands-free operation scenarios (such as when maintenance personnel are wearing gloves), obtaining information through voice commands (such as querying the operating status of pump group No. 3), covering different usage scenarios. Multimodal fusion cognitive interaction: The intelligent question-answering engine integrates multimodal fusion cognitive capabilities. In addition to asking questions via text, maintenance personnel can directly upload photos of equipment taken on-site (such as photos of pump group oil leaks) or recorded abnormal noises (such as valve noise audio). The engine extracts visual features such as oil leak marks and rust spots through image recognition models and acoustic features such as metal impact sounds and friction noises through audio analysis models. By combining unstructured image / audio information with text questions, it achieves comprehensive judgment of image, text, and audio, improving the accuracy of fault diagnosis. Augmented Reality (AR) Interaction: Supports efficient human-machine collaboration based on AR: When maintenance personnel wear AR glasses to inspect the site, the virtual control interface and information prompts are superimposed on the real equipment. When the system generates a suggestion to adjust the opening of a valve to 60%, the valve will be highlighted in the maintenance personnel's field of vision, and virtual operation buttons and rotation direction guide arrows will be displayed at the same time. After the maintenance personnel confirm the execution by gesture or voice, the AR system uses computer vision to identify the rotation angle of the valve wrench in real time, automatically verify whether the operation is in place (e.g., if it has been rotated to 60%, the operation is qualified), and feeds back the completion status to the system, forming a closed loop of human-machine collaboration of suggestion-execution-verification.

[0031] Based on the above layered architecture, the system achieves complete closed-loop control from user inquiry to business execution, such as... Figure 5 As shown, the specific steps are as follows: 1. User question initiation: Users can raise water-related business questions in natural language through the Web terminal, mobile App, voice robot or AR interface of the interaction layer (such as the abnormally low water pressure in the Chengdong area for the past hour, analyze the cause and give suggestions for handling).

[0032] 2. Problem Analysis and Data Preparation: After receiving the question, the intelligent question-answering engine in the computing layer calls the natural language processing module to parse out the core intent (cause analysis + handling suggestions) and entities (Chengdong area, low water pressure); then it calls the data processing module to filter out the available effective data (Chengdong area pipeline pressure, flow rate, surrounding pump station operating status and related valve opening) based on the data quality signals fed back by the perception layer, and inputs them into the time-series mechanism dual-drive fusion algorithm.

[0033] 3. Algorithm Analysis and Decision Generation: The time-series mechanism dual-drive fusion algorithm is activated. The time-series calibration module aligns all data timestamps to the millisecond level and embeds the mechanical delay parameter of 12 seconds for the full-stroke opening and closing time of valve No. 2 in the Chengdong area. The water mechanism embedding module, through hydraulic model simulation, identifies the abnormal closure of valve No. 2 (current opening degree only 30%) as the core reason for low water pressure. The dynamic weight adjustment module, due to the emergency response scenario, increases the weight of pump station status and real-time pressure data. The intelligent question-answering engine, based on the above analysis, generates the following decision result: Cause: Abnormal closure of valve No. 2 in the Chengdong area leads to increased downstream water supply resistance; Recommendation: Immediately activate the Chengdong backup booster pump and adjust the opening degree of valve No. 2 to 100%, while generating the corresponding equipment operation instruction set.

[0034] 4. Control Layer Linkage Verification and Command Issuance: The operation command set is sent to the question-and-answer device linkage control unit of the control layer. This unit does not immediately issue commands, but first reads the real-time mechanical performance parameters of the target equipment (standby booster pump, valve No. 2) through a multi-protocol compatible interface (supporting Modbus, OPCUA, etc.). These parameters include booster pump motor winding temperature of 45℃, bearing vibration value of 0.02mm / s, normal sealing status, and historical operating time of 3000 hours. Then, it calls the preset equipment capability pre-verification logic and, based on a deep learning model, predicts the instantaneous failure probability (e.g., booster pump failure probability of 0.05%) and remaining reliable life of the equipment when performing a new task.

[0035] If the verification is successful (e.g., the motor temperature is within the safe threshold and the failure probability is less than 0.1%), the unit will convert the operation command into a control signal that the device can recognize (e.g., a 4-20mA current signal) and send it out.

[0036] If the verification fails (e.g., the booster pump is found to have high cumulative fatigue damage or excessive vibration), the unit will immediately report the situation to the intelligent question-and-answer engine. The engine will then make dynamic adjustments (e.g., switch to starting the backup booster pump in the west of the city) and simultaneously generate a maintenance work order for the booster pump in the east of the city to prevent the equipment from operating with a fault.

[0037] 5. Execution Progress Feedback and Interface Display: After the command is issued, the linkage control unit collects execution progress data in real time through mechanical motion sensors on the equipment (such as valve stroke sensors and motor speed sensors). For example, valve No. 2 is opening, the current opening degree is 70%, and the countdown to the start of the backup booster pump is 3 seconds. This data is fed back to the interaction layer in real time and displayed to the user along with the initial decision suggestions, ensuring that the entire execution process is completely transparent and that the user can keep track of the progress at any time.

[0038] To prevent system decisions from becoming rigid, the system's self-evolution module is responsible for collecting feedback data after each decision execution loop and driving continuous system optimization, achieving a spiral of decision-execution-feedback-optimization. Feedback data collection includes two core types: first, business performance data (such as whether water pressure in the eastern part of the city returned to normal within 30 minutes, and whether water turbidity dropped below the standard value), objectively evaluating the effectiveness of decisions; and second, user subjective feedback (such as maintenance personnel's ratings of the clarity of operational suggestions and the ease of command execution), supplementing the subjective experience dimension. The collected data is used to drive a reinforcement learning agent-led digital twin, such as... Figure 9As shown, this digital twin is a high-fidelity virtual mapping of the real water plant and pipeline system, encompassing all core attributes such as pipeline hydraulic characteristics, equipment mechanical parameters, and water quality variation patterns. The reinforcement learning agent repeatedly tries new strategies (such as different pump combinations or step-by-step valve adjustment schemes) in the secure sandbox environment of the digital twin, without worrying about impacting the real system. The effects of the new strategies in the virtual environment (such as a 15% reduction in energy consumption for a certain pump combination) are compared with the actual effects in the real world. If discrepancies exist (such as a 15% reduction in virtual energy consumption versus a 12% reduction in real energy consumption), the system automatically corrects the local parameters of the water management mechanism model in the digital twin (such as adjusting the pipeline hydraulic loss coefficient) to make it infinitely closer to the real system. Simultaneously, through massive trial-and-error learning, the agent can discover new operational strategies beyond the scope of traditional experience (such as starting a backup pump one hour before peak water usage, which can reduce pipeline pressure fluctuations by 20%). After virtual verification and small-scale pilot testing in real-world scenarios, new operational strategies that prove effective will be updated as new knowledge entries in the knowledge base of the platform-level RAG service. Simultaneously, the training data of the water affairs expert model will be optimized, enabling subsequent decisions to directly utilize these new experiences and continuously improve the overall system's decision-making capabilities. In summary, this invention, through the deep synergy of the aforementioned hardware, software, and algorithms, achieves a fundamental transformation in intelligent question answering for water affairs from passive querying to proactive perception, scientific decision-making, precise execution, and self-evolution. Based on the above solution, the following sections illustrate the system's operation using two typical scenarios: emergency handling of pipe bursts and analysis of abnormal water pressure.

[0039] When dispatchers report a sudden drop in pressure on pipeline XX via the interaction layer, requesting analysis of the cause and appropriate action, the Smart Water Assistant (Agent) receives the question and passes it to the computation layer. The Agent framework service in the computation layer first invokes the local water expert model to understand the intent, identifying the issue as an emergency scenario requiring real-time data and operational execution. The workflow orchestration engine then breaks down this complex problem into a series of sub-tasks: acquiring real-time pressure data from the SCADA system, obtaining the pipeline topology from the GIS system, retrieving historical pipeline burst response plans, and generating control commands.

[0040] During the execution of this workflow, the RAG service at the platform layer first vectorizes the user's question and retrieves relevant burst pipe handling procedures from the vector database using similarity search as static knowledge background. Simultaneously, functions encapsulated by the Spring AI framework call the SCADA system API in real time to obtain the latest pressure readings and call GIS services to obtain the pipeline network topology. Meanwhile, the self-verifying data acquisition device deployed at the perception layer continuously uploads high-quality pressure monitoring data; its data quality signals are adopted by the system, ensuring the reliability of the decision-making basis.

[0041] The computational layer combines retrieved static regulations, real-time pressure data, and pipeline topology information into high-quality prompts, which are then submitted to the water expert model for reasoning. The model integrates all information and generates a decision suggestion and corresponding operational instructions for immediately closing upstream valve A if a suspected burst pipe is detected. This instruction is then sent to the question-and-answer-equipment linkage control unit, such as... Figure 10 As shown, this unit does not execute immediately. Instead, it first reads the real-time status parameters of valve A through the SpringAI channel and verifies the feasibility of the operation based on a pre-built mechanical performance library and health prediction model. Only after the verification is successful does the unit convert the instruction into a control signal that the equipment can recognize and send it out, and monitor the valve closing progress in real time through mechanical motion sensors.

[0042] Finally, the generated decision recommendations and dynamically updated execution progress are returned to the interaction layer and displayed to the dispatchers. Simultaneously, the system calls the map service via the MCP protocol to accurately mark the pipe burst point and valve location in the visual interface.

[0043] After the incident is resolved, the system's self-evolution module is activated. This module collects all data from the entire incident process, driving a reinforcement learning-based digital twin to perform debriefing and simulation in a virtual environment. By comparing the actual results with the simulation strategy, it automatically optimizes the parameters of the water management mechanism model and deposits the better handling strategy into the platform-level knowledge base, achieving continuous improvement in system capabilities. For knowledge-intensive problems such as analyzing the causes of abnormal water pressure in a residential area, the system focuses on leveraging its knowledge fusion and retrieval capabilities. The task flow planned at the computational layer prioritizes retrieving knowledge documents such as historical maintenance records and water quality reports, and combines them with real-time data for analysis. By combining static knowledge and dynamic data to form contextual clues, the system effectively guides the water management expert model to generate accurate and professional analysis reports, significantly reducing the illusion phenomenon commonly found in general-purpose models.

[0044] Through the above embodiments, this invention deeply integrates the general capabilities of a large language model, professional knowledge in the water sector, real-time business data, and specific equipment operations to construct a smart water affairs question-and-answer system capable of completing the entire process of perception-analysis-decision-execution-evolution, achieving a fundamental leap from passive question-and-answer to proactive closed-loop decision-making. The specific embodiments described above are merely illustrative or explanatory of the principles of this invention and do not constitute a limitation thereof. Therefore, any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of this invention should be included within the protection scope of this invention.

Claims

1. A smart question-and-answer system for water affairs business systems, characterized in that, include: The perception layer includes a self-verifying water data acquisition device for acquiring multi-source data related to water operations. The control layer includes a question-and-answer-device linkage control unit, which is communicatively connected to the perception layer and is used to directly control the water treatment field equipment and obtain its real-time mechanical performance parameters. The computing layer includes interconnected data processing modules and an intelligent question-answering engine; The data processing module is communicatively connected to the perception layer and the control layer, and is used to receive the multi-source data and real-time mechanical performance parameters, and to perform fusion processing on them using a time-series-mechanism dual-drive fusion algorithm; The intelligent question-answering engine is configured to: receive water-related business questions raised by users, analyze the user's question intent using natural language processing technology, and perform graph structure matching with the water knowledge base to determine the core business scenario; Receive the fusion results from the data processing module and generate target question-and-answer results that include business decision suggestions and corresponding device operation instructions; And send the device operation instructions to the control layer; The application layer, which communicates with the computing layer, is used to output the business decision suggestions; The question-and-answer-device linkage control unit is further configured to: upon receiving the device operation command, verify the device executability of the device operation command based on the acquired real-time mechanical performance parameters; if executable, send the command to the corresponding water affairs field equipment, monitor the command execution progress in real time, and synchronously feed back the command execution progress data to the application layer for output.

2. The system according to claim 1, characterized in that, It also includes a system self-evolution module, which is connected to the computing layer and the control layer; The system self-evolution module is configured to update the parameters of the water affairs knowledge base and the water affairs professional mechanism model in the time-series-mechanism dual-drive fusion algorithm based on the business effect data and user feedback after the instruction is executed, through an online learning mechanism.

3. The system according to claim 2, characterized in that, The core of the system's self-evolution module is a digital twin driven by a reinforcement learning agent. This digital twin operates in parallel with the real water plant. The reinforcement learning agent tries various operating strategies in the digital twin and compares the simulation results with real data to automatically correct model parameters and discover new operating strategies, thereby updating the water knowledge base.

4. The system according to claim 1, characterized in that, The self-calibrating water data acquisition device is a non-powered auxiliary sampling device, comprising: an anti-clogging and scale removal component, including an elastic scraper and a flow guide protrusion embedded in the inner wall of the sampling pipe, which uses the pressure of the water flow in the pipe network to drive the elastic scraper to reciprocate along the pipe wall to remove scale and impurities; an angle adaptive component, including a gravity balance ball joint and an arc-shaped limiting guide rail located at the bottom of the sampling probe, used to automatically adjust the probe angle according to the pipe diameter; and a data quality feedback component, including a built-in micro-differential pressure sensor, which mechanically triggers a signal rod to output a data abnormality signal when the sampling channel is blocked and the inlet and outlet pressure difference exceeds a preset threshold.

5. The system according to claim 1, characterized in that, The time-series-mechanism dual-drive fusion algorithm includes: a time-series calibration module, used to align timestamps of multi-source data and embed mechanical action delay parameters as time-series compensation factors; a water mechanism embedding module, which integrates an activated sludge model (ASM) and a pipeline hydraulic loss model; and a dynamic weight adjustment module, used to automatically adjust the weight ratio of data from different sources based on scenario priority.

6. The system according to claim 5, characterized in that, The time-mechanism dual-drive fusion algorithm also includes a slime mold biomimetic adaptive pipeline scheduling submodule, which is configured to: mimic slime mold pathfinding behavior, regard pipeline nodes as food sources, and use pressure and flow demand as stimulus signals to dynamically generate the optimal pump and valve scheduling scheme.

7. The system according to claim 1, characterized in that, The intelligent question-answering engine integrates multimodal fusion cognitive capabilities, enabling it to receive and parse user-uploaded images of on-site equipment or abnormal sound recordings, and perform comprehensive analysis and fault diagnosis in conjunction with text questions.

8. The system according to claim 1, characterized in that, The application layer supports an augmented reality (AR) based interactive interface, which is configured to display the business decision suggestions and device operation instructions on the real device in the user's field of vision in a virtual overlay manner, and can automatically verify the task completion status through action recognition.

Images