Urban water supply scheduling decision generation method and device, equipment and medium

Abstract

The application relates to a town water supply scheduling decision generation method, device, equipment and medium, the method comprising: receiving a water supply scheduling problem input by a user in natural language, retrieving a water supply scheduling knowledge fragment in a bimodal retrieval database based on a water supply scheduling question vector corresponding to the water supply scheduling problem, and fusing information of the water supply scheduling knowledge fragment, real-time water supply working condition data and the water supply scheduling problem to dynamically reconstruct a semi-structured enhanced prompt word through a lightweight small language model; inputting the semi-structured enhanced prompt word into a preset large language model, generating a water supply scheduling decision suggestion through deep semantic reasoning and logical deduction of the large language model; converting the water supply scheduling decision suggestion into a hydraulic simulation instruction, inputting the hydraulic simulation instruction into a preset lightweight hydraulic simulation model for hydraulic calculation verification, and outputting a water supply scheduling decision of a target town. The application can greatly improve the accuracy of water supply scheduling decisions under complex working conditions and emergency scenarios.

Description

Technical Field

[0001] This application relates to the field of water supply scheduling, and in particular to a method for generating urban water supply scheduling decisions, a corresponding device, electronic equipment, and a computer-readable storage medium. Background Technology

[0002] Water supply scheduling, as a core component of urban water supply system operation, directly relates to the safety, stability, and efficiency of water supply. Scientific and rational scheduling can effectively coordinate the operational status of water sources, water plants, pumping stations, and pipe networks, ensuring the water needs of residents and industrial and commercial users. Especially during peak water consumption periods or emergencies (such as pipe bursts, water pollution, equipment failures, etc.), rapid and accurate scheduling decisions are crucial to maintaining the normal operation of the city's "lifeline" project.

[0003] Currently, most water supply companies still rely heavily on the individual experience of dispatchers for scheduling decisions, lacking a unified and reusable basis for decision-making. This experience-driven model has obvious limitations. On the one hand, the decision-making process is highly subjective and difficult to adapt to complex and ever-changing water supply conditions. On the other hand, it is prone to delayed response, judgment bias, or even decision-making errors when facing sudden emergencies, which not only affects the quality of water supply services but may also increase risks.

[0004] Existing water supply scheduling decision-making mechanisms either rely excessively on a rule system based on solidified expert experience, resulting in lagging knowledge updates, poor flexibility, and difficulty in adapting to dynamic water supply scenarios; or they rely on high-precision, full-function hydraulic models, which not only have high initial setup costs and difficult maintenance in the later stages, but also require a professional technical team for continuous operation and maintenance, making them unaffordable for small and medium-sized water supply enterprises and extremely difficult to popularize.

[0005] In summary, in order to address the problems of response lag, judgment bias, and even decision-making errors caused by the reliance on individual experience in existing water supply scheduling decision generation methods, as well as the problems of lagging updates and poor flexibility in relying on expert experience and knowledge, the applicant has made corresponding explorations to solve these problems. Summary of the Invention

[0006] The purpose of this application is to solve the above-mentioned problems by providing a method for generating urban water supply scheduling decisions, a corresponding device, electronic equipment, and a computer-readable storage medium.

[0007] To achieve the various objectives of this application, the following technical solution is adopted:

[0008] A method for generating urban water supply scheduling decisions, proposed to meet one of the purposes of this application, includes:

[0009] Obtain the water supply scheduling knowledge base corresponding to the target town, slice the water supply scheduling knowledge base according to different chapters to obtain the slice content corresponding to each chapter, construct a special training dataset for the water supply scheduling domain based on the slice content, and use the training dataset to perform supervised fine-tuning of the general embedding model to obtain an embedding model adapted to the water supply scheduling domain.

[0010] Key entities and scheduling rule relationships between key entities in the water supply scheduling domain are extracted from the sliced ​​content to construct a water supply scheduling knowledge graph. The sliced ​​content is transformed into domain knowledge semantic vectors by adapting to the embedding model of the water supply scheduling domain, and a dual-modal retrieval database is constructed by combining the water supply scheduling knowledge graph.

[0011] The system receives water supply scheduling questions input by users in natural language and transforms the water supply scheduling questions into water supply scheduling question vectors. Based on the water supply scheduling question vectors, it retrieves matching water supply scheduling knowledge fragments from the bimodal retrieval database. After fusing the water supply scheduling knowledge fragments, real-time water supply condition data, and the water supply scheduling questions, it dynamically reconstructs semi-structured enhanced prompt words through a lightweight small language model.

[0012] The semi-structured enhanced prompt words are input into a preset large language model. Through deep semantic reasoning and logical deduction of the large language model, water supply scheduling decision suggestions containing the operation object, specific action instructions and operation basis are generated.

[0013] The water supply scheduling decision suggestions are converted into hydraulic simulation instructions, input into a preset lightweight hydraulic simulation model for hydraulic calculation verification, and output the water supply scheduling decision corresponding to the target town to complete the generation of town water supply scheduling decisions.

[0014] Optionally, the water supply scheduling knowledge base is sliced ​​according to different chapters to obtain the sliced ​​content corresponding to each chapter. A training dataset specifically for the water supply scheduling domain is constructed based on the sliced ​​content. The step of using the training dataset to perform supervised fine-tuning of the general embedding model includes:

[0015] Obtain the water supply scheduling knowledge base corresponding to the target town, and slice the water supply scheduling knowledge base according to different chapters to obtain the slice content corresponding to each chapter.

[0016] Based on the sliced ​​content, a training dataset specifically for the water supply scheduling domain is constructed. The training dataset is then used to perform supervised fine-tuning on the general embedding model to enhance its semantic understanding of water supply scheduling terminology, sentence structure, and contextual dependencies, thereby obtaining an embedding model adapted to the water supply scheduling domain.

[0017] Optionally, the training dataset is a triplet data constructed based on the query question generated from the slice content, positive examples of the slice content, and negative examples of the slice content;

[0018] The positive examples of the slice content represent slice content extracted from the slice content of the water supply scheduling knowledge base, which highly matches the core semantics of the query question generated based on the slice, and carries the target water supply scheduling knowledge and scheduling rules.

[0019] The negative examples of the slice content represent slice content from the slice content of the water supply scheduling knowledge base that are unrelated to the core semantics of the query question generated based on the slice, and that have similar professional terms or expression structures on the surface to the positive examples of the slice content but are essentially different in scheduling scenarios and operation logic from the same document manual slice content.

[0020] Optionally, the steps of extracting key entities and scheduling rule relationships between key entities from the sliced ​​content to construct a water supply scheduling knowledge graph, converting the sliced ​​content into domain knowledge semantic vectors by adapting to the embedding model of the water supply scheduling domain, and constructing a bimodal retrieval database in conjunction with the water supply scheduling knowledge graph include:

[0021] Extract key entities and scheduling rule relationships between key entities from the content slices of each chapter of the water supply scheduling knowledge base;

[0022] After deduplication, conflict detection, and logical consistency verification of the key entities and scheduling rule relationships, a water supply scheduling knowledge graph is constructed.

[0023] By adapting the embedding model of the water supply scheduling domain, the slice content corresponding to each chapter is vectorized and transformed into domain knowledge semantic vectors.

[0024] By integrating the structured relationships of the water supply scheduling knowledge graph with the semantic retrieval features of the domain knowledge semantic vectors, a dual-modal retrieval database is constructed that combines graph entity association reasoning with precise vector semantic retrieval capabilities.

[0025] Optionally, the steps of receiving a water supply scheduling question input by a user in natural language, converting the water supply scheduling question into a water supply scheduling query vector, and retrieving matching water supply scheduling knowledge fragments from the bimodal retrieval database based on the water supply scheduling query vector include:

[0026] The system acquires water supply scheduling questions input by users in natural language form and transforms these questions into water supply scheduling query vectors by adapting to an embedding model in the water supply scheduling domain.

[0027] Based on the water supply scheduling question vector, semantic matching and entity association reasoning retrieval are performed in the water supply scheduling bimodal retrieval database to obtain water supply scheduling knowledge fragments that match the water supply scheduling question.

[0028] Optionally, the step of fusing the water supply scheduling knowledge fragment, real-time water supply condition data, and the water supply scheduling problem, and then dynamically reconstructing semi-structured enhanced prompt words using a lightweight, small language model, includes:

[0029] The water supply scheduling knowledge fragments obtained from water supply scheduling retrieval, real-time water supply condition data, and water supply scheduling issues are fused together to determine multi-source fused information.

[0030] The multi-source fusion information is input into a lightweight small language model. The lightweight small language model then performs a structured reconstruction of the multi-source fusion information based on the characteristics of the water supply scheduling domain, generating semi-structured enhanced prompt words that are adapted to the reasoning of large language models.

[0031] Optionally, the steps for building a water supply scheduling knowledge base include:

[0032] Obtain the water supply dispatch manual corresponding to the target town, digitize the water supply dispatch manual, and organize the digitized water supply dispatch manual according to water supply dispatch knowledge categories to construct a water supply dispatch knowledge base corresponding to the target town. The water supply dispatch manual is a standardized professional document in the field of water supply in the target town, which contains core dispatch knowledge such as basic water supply parameters, routine dispatch strategies, emergency response plans and operation execution specifications.

[0033] A town water supply dispatching decision generation device provided for another purpose of this application includes:

[0034] The embedding model building module is configured to acquire the water supply scheduling knowledge base corresponding to the target town, slice the water supply scheduling knowledge base according to different chapters to obtain the slice content corresponding to each chapter, construct a special training dataset for the water supply scheduling domain based on the slice content, and use the training dataset to perform supervised fine-tuning of the general embedding model to obtain an embedding model adapted to the water supply scheduling domain.

[0035] The retrieval database construction module is configured to extract key entities and scheduling rule relationships between key entities from the sliced ​​content to construct a water supply scheduling knowledge graph. By adapting the embedding model of the water supply scheduling domain, the sliced ​​content is transformed into domain knowledge semantic vectors, and a dual-modal retrieval database is constructed in combination with the water supply scheduling knowledge graph.

[0036] The enhanced prompt word reconstruction module is configured to receive water supply scheduling questions input by users in natural language, and convert the water supply scheduling questions into water supply scheduling question vectors. Based on the water supply scheduling question vectors, it retrieves matching water supply scheduling knowledge fragments from the bimodal retrieval database. After fusing the water supply scheduling knowledge fragments, real-time water supply condition data and the water supply scheduling questions, it dynamically reconstructs semi-structured enhanced prompt words through a lightweight small language model.

[0037] The scheduling suggestion generation module is configured to input the semi-structured enhanced prompt words into a preset large language model, and generate water supply scheduling decision suggestions containing the operation object, specific action instructions and operation basis through deep semantic reasoning and logical deduction of the large language model.

[0038] The scheduling decision output module is configured to convert the water supply scheduling decision suggestions into hydraulic simulation instructions, input them into a preset lightweight hydraulic simulation model for hydraulic calculation verification, and output the water supply scheduling decision corresponding to the target town to complete the generation of town water supply scheduling decisions.

[0039] An electronic device provided for another purpose of this application includes a central processing unit and a memory, the central processing unit being configured to invoke and run a computer program stored in the memory to perform the steps of the urban water supply scheduling decision generation method of this application.

[0040] A computer-readable storage medium is provided for another purpose of this application, which stores, in the form of computer-readable instructions, a computer program implemented according to the urban water supply scheduling decision generation method, which, when invoked by a computer, executes the steps included in the corresponding method.

[0041] Compared to existing technologies, this application addresses the problems of response lag, judgment bias, and even decision-making errors caused by the reliance on individual experience in existing water supply scheduling decision generation methods, as well as the problems of delayed updates and poor flexibility in relying on expert experience and knowledge. This application includes, but is not limited to, the following beneficial effects:

[0042] Firstly, this application relies on a dedicated water supply scheduling knowledge base for the target town to construct standardized decision-making criteria, abandoning the subjective decision-making model driven by purely manual experience. It achieves accurate matching between scheduling knowledge and real-time operating conditions through a dual-modal retrieval database, and combines semi-structured enhanced prompt words to standardize the reasoning logic of the large language model, ensuring that scheduling decisions conform to domain rules and the actual water supply scenarios of the town. This can solve the problems of decision-making bias and response lag caused by individual experience differences, and significantly improve the accuracy of water supply scheduling decisions in complex operating conditions and emergency scenarios.

[0043] Secondly, this application constructs a domain-specific training dataset by processing the water supply scheduling knowledge base into slices, and fine-tunes the general embedding model to obtain an exclusive embedding model adapted to the water supply scheduling domain, which greatly improves the semantic accuracy of text vectorization. By combining the structured relationship of the knowledge graph with semantic vectors to construct a dual-modal retrieval database, it realizes dual matching of semantic similarity retrieval and entity association reasoning, which not only breaks through the limitations of literal retrieval, but also avoids invalid knowledge recall, greatly shortens the knowledge retrieval time, and improves the overall decision response speed.

[0044] Third, this application uses a semi-structured enhanced prompt word constraint large language model output to generate scheduling decision suggestions that clearly include the operation object, specific action instructions and operation basis, so as to realize the traceability of the decision process and the verifiability of the decision content; a lightweight hydraulic simulation model is introduced to simulate and verify the decision suggestions, and to identify potential engineering problems such as pipeline pressure imbalance and flow mismatch in advance, so as to ensure that the final output water supply scheduling decision can be directly implemented and solve the water supply risks and resource losses caused by ineffective decisions.

[0045] Furthermore, this application automates the entire process from knowledge base slicing, model fine-tuning, retrieval matching, prompt word reconstruction to decision generation and simulation verification, without requiring manual intervention. This reduces the workload of dispatchers and lowers the cost of manual on-duty personnel. It also reduces various resource losses such as pipeline losses and emergency response costs through precise decision-making, thereby achieving a dual improvement in water supply dispatch efficiency and economic benefits. Attached Figure Description

[0046] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0047] Figure 1 This is a flowchart illustrating the urban water supply scheduling decision generation method in the embodiments of this application;

[0048] Figure 2 This is an example diagram illustrating the intelligent question-and-answer mechanism for the conventional scheduling method between water plant A and water plant B in this application embodiment;

[0049] Figure 3 This is an example diagram of the intelligent question-and-answer system for the pressure monitoring point and normal pressure range in region C in this application embodiment;

[0050] Figure 4 This is an example diagram of the intelligent question-and-answer mechanism for scheduling measures when the pressure at monitoring point 11 is too low, as described in this application embodiment.

[0051] Figure 5 This is an example diagram of intelligent question-and-answer mechanism for emergency dispatch measures when the pressure at monitoring point 9 is insufficient, as described in this application embodiment.

[0052] Figure 6This is a schematic diagram of the urban water supply scheduling decision generation device in the embodiments of this application;

[0053] Figure 7 This is a schematic diagram of the structure of the computer device in the embodiments of this application. Detailed Implementation

[0054] The embodiments of this application 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 this application, and should not be construed as limiting this application.

[0055] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0056] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0057] Those skilled in the art will understand that the terms "client," "terminal," and "terminal device" as used herein include both devices that receive wireless signals, devices that only possess wireless signal receiver capabilities without transmission capabilities, and devices with receiving and transmitting hardware, devices that have receiving and transmitting hardware capable of bidirectional communication over a bidirectional communication link. Such devices may include: cellular or other communication devices such as personal computers or tablets, having single-line displays, multi-line displays, or cellular or other communication devices without multi-line displays; PCS (Personal Communications Service) that can combine voice, data processing, fax, and / or data communication capabilities; PDAs (Personal Digital Assistants) that may include radio frequency receivers, pagers, internet / intranet access, web browsers, notebooks, calendars, and / or GPS (Global Positioning System) receivers; and conventional laptops and / or handheld computers or other devices that have and / or include radio frequency receivers. As used herein, "client," "terminal," and "terminal device" can be portable, transportable, installed in a means of transportation (air, sea, and / or land), or suitable and / or configured to operate locally and / or in a distributed manner, operating in any other location on Earth and / or in space. "Client," "terminal," and "terminal device" as used herein can also be a communication terminal, an internet access terminal, or a music / video playback terminal, such as a PDA, a MID (Mobile Internet Device), and / or a mobile phone with music / video playback capabilities, or a smart TV, set-top box, etc.

[0058] The hardware referred to by the names "server," "client," and "service node" in this application is essentially an electronic device with the equivalent capabilities of a personal computer. It is a hardware device with the necessary components revealed by the von Neumann architecture, such as a central processing unit (including an arithmetic logic unit and a control unit), memory, input devices, and output devices. The computer program is stored in its memory, and the central processing unit loads the program stored in the secondary storage into the main memory to run it, execute the instructions in the program, and interact with the input and output devices to complete specific functions.

[0059] It should be noted that the concept of "server" used in this application can also be extended to the case of server clusters. Based on the network deployment principles understood by those skilled in the art, the servers should be logically divided. Physically, these servers can be independent of each other but accessible through interfaces, or they can be integrated into a single physical computer or a computer cluster. Those skilled in the art should understand this flexibility and should not use it to constrain the implementation of the network deployment method in this application.

[0060] One or more of the technical features of this application, unless explicitly specified herein, can be deployed on a server and accessed by a client remotely calling the online service interface provided by the server, or can be directly deployed and run on a client for access.

[0061] Unless otherwise specified, the neural network models referenced or potentially referenced in this application may be deployed on a remote server and invoked remotely on the client, or deployed on a client with the capability to invoke directly. In some embodiments, when running on the client, the corresponding intelligence may be acquired through transfer learning in order to reduce the requirements on the client's hardware resources and avoid excessive consumption of the client's hardware resources.

[0062] Unless otherwise specified, all data involved in this application may be stored remotely on a server or on a local terminal device, as long as it is suitable for use by the technical solution of this application.

[0063] Those skilled in the art will understand that although the various methods in this application are described based on the same concept and thus present commonality among them, they can be performed independently unless otherwise specified. Similarly, the various embodiments disclosed in this application are all based on the same inventive concept; therefore, concepts expressed in the same way, as well as concepts that are appropriately changed for convenience but are expressed differently, should be understood equivalently.

[0064] Unless otherwise expressly stated, the various embodiments disclosed in this application can be combined in a cross-cutting manner to flexibly construct new embodiments, as long as such combination does not depart from the inventive spirit of this application and can meet the needs of the prior art or solve a certain deficiency in the prior art. Those skilled in the art should be aware of such modifications.

[0065] Please see Figure 1 In one embodiment of the urban water supply scheduling decision generation method of this application, the method includes:

[0066] Step S10: Obtain the water supply scheduling knowledge base corresponding to the target town, slice the water supply scheduling knowledge base according to different chapters to obtain the slice content corresponding to each chapter, construct a special training dataset for the water supply scheduling domain based on the slice content, and use the training dataset to perform supervised fine-tuning of the general embedding model to obtain an embedding model adapted to the water supply scheduling domain.

[0067] The town water supply scheduling decision generation system in the terminal device can obtain the water supply scheduling knowledge base corresponding to the target town, slice the water supply scheduling knowledge base according to different chapters to obtain the slice content corresponding to each chapter, construct a special training dataset for the water supply scheduling domain based on the slice content, and use the training dataset to perform supervised fine-tuning of the general embedding model to obtain an embedding model adapted to the water supply scheduling domain.

[0068] In some embodiments, the step of constructing a water supply scheduling knowledge base includes:

[0069] Obtain the water supply dispatch manual corresponding to the target town, digitize the water supply dispatch manual, and organize the digitized water supply dispatch manual according to water supply dispatch knowledge categories to construct a water supply dispatch knowledge base corresponding to the target town. The water supply dispatch manual is a standardized professional document in the field of water supply in the target town, which contains core dispatch knowledge such as basic water supply parameters, routine dispatch strategies, emergency response plans and operation execution specifications.

[0070] Based on the electronic document processing of the water supply scheduling manual corresponding to the target town, a water supply scheduling knowledge base exclusive to the target town is constructed, which serves as the core knowledge source for the system to realize intelligent decision support.

[0071] The "Water Supply Dispatch Manuals" of urban water supply enterprises, regional dispatch centers, and water operation units with water supply dispatch functions in the target towns serve as the original knowledge source for building the water supply dispatch knowledge base. Urban water supply enterprises, regional dispatch centers, and water operation units should pre-establish a set of structured, unstructured, standardized, and continuously updated "Water Supply Dispatch Manuals" as the core knowledge source for building the knowledge base. The "Water Supply Dispatch Manuals" should meet the following requirements:

[0072] In terms of functional positioning, the "Water Supply Dispatch Manual" serves not only as an operational reference document, but also as a knowledge input source for the system's automatic decision-making. The content must be analyzable, executable, and traceable.

[0073] In terms of content structure, the "Water Supply Dispatch Manual" should have multiple logical functional modules, including a basic parameter module, a routine dispatch strategy module, an emergency response plan module, and an operation execution specification module, forming a complete closed loop of dispatch decision-making knowledge.

[0074] In terms of data management, the contents of the "Water Supply Dispatch Manual" are stored in common electronic document formats such as PDF, Word, and Excel. It supports data interaction with the dispatch automation system and establishes a version control mechanism. Each revision records the version number, revision summary, approval status, and effective time.

[0075] The "Water Supply Dispatch Manual" of urban water supply enterprises, regional dispatch centers and water operation units with water supply dispatch functions in the target town is obtained. The "Water Supply Dispatch Manual" is digitized and organized according to water supply dispatch knowledge categories to construct a water supply dispatch knowledge base corresponding to the target town. This provides basic data support for subsequent fine-tuning of embedded models, construction of knowledge graphs and establishment of a dual-modal retrieval database.

[0076] The urban water supply dispatching decision generation system uses the "Water Supply Dispatch Manual" as its core knowledge source. To ensure that the manual can be effectively parsed, matched, and executed by the system, it should include modules such as basic parameters, routine dispatching strategies, emergency response plans, and operational execution specifications. These are explained in detail below:

[0077] Firstly, the basic parameter module, which serves as the static data source for system operation, should be fully reflected in the "Water Supply Dispatch Manual." It should contain basic parameters for various equipment, compiled into quick reference tables, including: a quick reference table for generating units and water plants, which should include name, unit number, pump parameters, and water supply pressure zone; a quick reference table for clear water tanks and water plants, which should include name, number of clear water tanks, and effective volume; a quick reference table for elevated water tanks (booster stations), which should include name, design volume, generating unit, pump parameters, and remarks; the normal pressure range and representative area for each pressure measuring point, which should include area, pressure zone, point location, and normal range; and a table showing the relationship between water plants, booster stations (elevated water tanks), and each supply area, which should include area, main water supply plant, pressure zone, elevated water tank, and tank bottom elevation.

[0078] In a specific embodiment, the quick reference tables that should be included in the Water Supply Dispatch Manual are shown in Tables 1, 2, 3, 4, and 5.

[0079] Table 1 Units

[0080]

[0081] Table 2 Clear Water Pool

[0082]

[0083] Table 3. Elevated Water Tank (Boosting Station)

[0084]

[0085] Table 4 Normal pressure range and representative area for each pressure measurement point

[0086]

[0087] Table 5. Relationship between water plants, booster stations (elevated water tanks), and various supply areas.

[0088]

[0089] Secondly, the routine dispatching strategy module should be fully reflected in the water supply dispatching manual. It should include the normal operation modes of water plants and booster stations in each water supply area, as well as the corresponding water level control requirements. Periodic plans should also be available for different seasons and regions, including the operation modes and valve control of water plants and booster stations during the summer peak season, and the water level control requirements during the peak season; and the operation modes and valve control of water plants and booster stations under normal mode, and the water level control requirements under normal mode, during the non-summer peak season.

[0090] In one specific embodiment, the conventional scheduling strategy module is configured using the following alternative implementation schemes for different scheduling scenarios and real-time operating conditions:

[0091] In normal operation mode, the water plant in Area C operates as follows: Area C is supplied primarily by Water Plant D, supplemented by Water Plant C, with reasonable allocation based on the water intake permits of both plants. Pumping stations 1-5 of Water Plant C primarily supply the low-pressure area of ​​Area C and the booster station at monitoring point 6. Units 6 and 7, as well as units 1-4 at monitoring point 6, supplement the medium-pressure area and the water tank at monitoring point 12. Pumping stations 1-3 of Water Plant D primarily supply the medium-pressure area and the entire water supply area of ​​Area C, maintaining normal operation. Units 4 and 5 supplement the low-pressure area of ​​Area C and the water tank at monitoring point 6. During peak hours, four units can operate simultaneously in the low-pressure area (3 from Water Plant C + 1 from Water Plant D), and two units from Water Plant D + one unit from Water Plant C can operate simultaneously in the medium-pressure area.

[0092] During peak periods, the operation mode of the booster station at monitoring point 12 is as follows: The high-pressure zone units at monitoring point 12 can operate in alternating large and small configurations (ensuring the pressure at monitoring point 11 is not lower than 0.5 MPa). The ultra-high-pressure unit #4 is activated to supplement pressurization when the outlet pressure at monitoring point 12 is ≤0.7 MPa. At monitoring point 13, the medium-pressure zone can operate with units #1+#3 or #2+#4 when water volume at monitoring point 12 is insufficient. The high-pressure zone can operate with units #5+#7 or #6+#7 when water volume at monitoring point 16 is insufficient. At monitoring point 17, booster station #1... Unit 4 is a full-area ultra-high pressure unit, operating at a pressure of 0.60-0.70 MPa from 5:00 to 24:00, and approximately 0.50-0.55 MPa from 0:00 to 5:00 (the goal is to ensure that the pressure at monitoring point 18 water supply station is not lower than 0.20 MPa). The system automatically adjusts the number of units. Units 5-7 are high-pressure units at monitoring point 19, operating at a pressure of approximately 0.30 MPa from 5:00 to 24:00, and approximately 0.25 MPa from 0:00 to 5:00. The system automatically adjusts the number of units. The inlet water is controlled by an electric regulating valve, which can adjust the inlet water flow according to the outlet water flow to maintain liquid level balance.

[0093] During peak periods, the water level control requirements for Area C are as follows: the water level in the clear water tank of Water Plant D must reach at least 3.3 m³ (for tanks #1 and #2) by 8:30 AM, and the overflow water level must be 4.7 m. 3 Water Plant C Clear Water Pool 3000m 3 The overflow level is 4700m. 3 ; Night shift staff at the dispatch center handover (8:30) Water requirements (monitoring point 13, booster station 1800m) 3 Monitoring point 12 is 6000m 3 Monitoring point 16 is 1500m 3 Monitoring point 17 is 6000m 3 During shift changes between day and afternoon, each water tank should be maintained at a level above three-quarters to ensure water supply during peak nighttime hours. During the nighttime water storage period (after 24:00), the minimum water level in each water tank should not be lower than one-half.

[0094] Thirdly, the emergency response plan module is primarily designed to enable the system to quickly match plans and assist in decision-making during emergencies such as pipe bursts, power outages, and water pollution. This part should be fully reflected in the "Water Supply Dispatch Manual," which should include emergency water supply dispatch plans for each region. For example, in the event of sudden failures such as water plant shutdowns, main pipeline bursts, or power outages, there should be corresponding handling procedures: valve shut-off plans, upstream and downstream water plant support plans, and temporary pressure reduction plans. When writing this chapter, the detailed response mechanisms and operational procedures of the plans should be emphasized.

[0095] In a specific embodiment, when dealing with the specific scenario of "water outage at a certain water plant," the system retrieves the following contingency plan logic: At this time, the backup water plant needs to operate at full capacity, and its high-power units at the second pumping station are started to supply water to the low-level pipeline network in the core supply area. Based on the water level of the upstream regulating reservoir, the units operating at the original affected water source are reduced, and the units operating in the remote high-pressure area are increased. The inlet valves of the associated water tank and the outlet valves of the secondary low-pressure area are adjusted as appropriate (to keep the clear water tank operating at a high level to meet the water output). The connecting valves of key nodes are adjusted. If the water level of the regulating reservoir continues to drop, the units operating in the high-level area are reduced, and the emergency direct supply high-pressure units (two small units or a combination of one large and one small unit) are operated. The valves in non-critical areas are reduced or closed to ensure that the water tanks in the affected supply area operate at full capacity as much as possible. At the same time, the system predicts whether the water supply can be met based on the supply and demand data. If it cannot be guaranteed, the next level of emergency plan is activated.

[0096] Fourth, the operation execution specification module defines how to translate instructions into actual operations, ensuring that the generated scheduling suggestions can be executed safely and compliantly. This should be fully reflected in the water supply scheduling manual, which should include human-machine interaction specifications: definitions of buttons on the remote intelligent monitoring equipment interface, and the meanings of alarm colors.

[0097] In a specific embodiment, when executing a dispatch command for "emergency repair of a sudden pipe burst in a water supply area," the system automatically activates the corresponding operation process according to the operation execution specification module: First, on the remote intelligent monitoring equipment interface, the network elements in the affected area immediately switch to a flashing red state, and the relevant valve icons synchronously display real-time current and opening data. The alarm information bar scrolls out a Level 1 alarm stating "emergency valve closure and isolation required." After the dispatcher confirms the command, the system automatically pushes a standardized valve closure operation ticket, clearly listing the numbers of several key valves to be closed and the operation sequence of "downstream first, then upstream." At the same time, the system automatically generates upstream and downstream water source support plans, instructs the standby water supply plant to increase unit output to maintain the network pressure in unaffected areas, and pops up a simulated heat map of the temporary pressure reduction range on the interface, ensuring that the dispatcher can monitor pressure changes in real time during the execution process. All operation steps and equipment feedback data are completely recorded, forming a traceable electronic log.

[0098] The manual should establish a version control mechanism, and each revision should record the version number, a summary of the revision content, the effective date, and the approval process.

[0099] Through the above specifications, this application transforms the traditional paper-based "Water Supply Dispatch Manual" into an electronic document knowledge base that can be recognized, matched, and executed by computer systems, namely, a water supply dispatch knowledge base. This solves the technical problem of dispatch experience relying on manual memory in the prior art and realizes the standardization, systematization, and intelligent access to dispatch knowledge.

[0100] In some embodiments, the water supply scheduling knowledge base is sliced ​​according to different chapters to obtain sliced ​​content corresponding to each chapter. A training dataset specifically for the water supply scheduling domain is constructed based on the sliced ​​content. The steps of using the training dataset to perform supervised fine-tuning of a general embedding model include:

[0101] Step S101: Obtain the water supply scheduling knowledge base corresponding to the target town, and slice the water supply scheduling knowledge base according to different chapters to obtain the slice content corresponding to each chapter.

[0102] Step S102: Construct a training dataset specifically for the water supply scheduling domain based on the sliced ​​content. Use this training dataset to perform supervised fine-tuning of the general embedding model to enhance its semantic understanding of water supply scheduling terminology, sentence structure, and contextual dependencies, thus obtaining an embedding model adapted to the water supply scheduling domain. The training dataset consists of triples constructed from query questions generated from sliced ​​content, positive examples of sliced ​​content, and negative examples of sliced ​​content. Positive examples of sliced ​​content represent sliced ​​content extracted from the water supply scheduling knowledge base that highly matches the core semantics of the query questions generated from the slices and carries the target water supply scheduling knowledge and rules. Negative examples of sliced ​​content represent sliced ​​content extracted from the water supply scheduling knowledge base that is unrelated to the core semantics of the query questions generated from the slices and, while superficially similar to positive examples in terms of terminology or expression structure, differs fundamentally in scheduling scenarios and operational logic from the same document / manual slices.

[0103] Specifically, the water supply scheduling knowledge base is first sliced ​​according to different chapters; then, domain-adaptive data is constructed based on the sliced ​​content, and the general embedding model is fine-tuned; subsequently, a water supply scheduling knowledge graph is constructed, and the fine-tuned embedding model is used to convert unstructured text and structured tables in the slices into high-dimensional semantic vectors, which are then fused to construct a bimodal retrieval database. The slicing process refers to dividing the water supply scheduling knowledge base into different segments according to each chapter, obtaining the sliced ​​content corresponding to each chapter, which serves as the basis for subsequent model fine-tuning and knowledge extraction.

[0104] Fine-tuning the embedding model refers to constructing a dataset using sliced ​​content, and then optimizing and adjusting the internal parameters of a general embedding model through supervised methods to obtain an embedding model adapted to the water supply scheduling domain. The specific steps include:

[0105] Step S1001: Based on the content of the slice, a training dataset specifically for the water supply scheduling domain is constructed. The large language model is used to automatically generate several corresponding query questions for each text / table slice. Other similar but mismatched slices in the same document are combined as negative samples to form a triplet data of "(query, positive example, negative example)".

[0106] Step S10002: Using the selected general embedding model as a base, supervised fine-tuning is performed on it using the above triplet data to enhance the model's understanding of terminology, sentence structure and contextual dependencies in the field of water supply network scheduling, thereby obtaining an embedding model adapted to the field of water supply scheduling.

[0107] Step S20: Extract key entities and scheduling rule relationships between key entities from the sliced ​​content to construct a water supply scheduling knowledge graph. Transform the sliced ​​content into domain knowledge semantic vectors by adapting to the embedding model of the water supply scheduling domain, and construct a dual-modal retrieval database by combining the water supply scheduling knowledge graph.

[0108] A water supply scheduling knowledge base corresponding to the target town is obtained. The knowledge base is then sliced ​​according to different chapters to obtain the sliced ​​content for each chapter. A training dataset specifically for the water supply scheduling domain is constructed based on the sliced ​​content. The general embedding model is then supervisedly fine-tuned using the training dataset to obtain an embedding model adapted to the water supply scheduling domain. Key entities and scheduling rule relationships between key entities in the water supply scheduling domain are extracted from the sliced ​​content to construct a water supply scheduling knowledge graph. The sliced ​​content is then transformed into domain knowledge semantic vectors by adapting the embedding model to the water supply scheduling domain, and a bimodal retrieval database is constructed by combining the water supply scheduling knowledge graph.

[0109] In some embodiments, the steps of extracting key entities and scheduling rule relationships between key entities from the sliced ​​content to construct a water supply scheduling knowledge graph, converting the sliced ​​content into domain knowledge semantic vectors by adapting to the embedding model of the water supply scheduling domain, and constructing a bimodal retrieval database in conjunction with the water supply scheduling knowledge graph include:

[0110] Step S201: Extract key entities and scheduling rule relationships between key entities from the sliced ​​content of each chapter of the water supply scheduling knowledge base.

[0111] Step S202: After deduplication, conflict detection and logical consistency verification of the key entities and scheduling rule relationships, a water supply scheduling knowledge graph is constructed.

[0112] Step S203: Vectorize the sliced ​​content corresponding to each chapter by adapting the embedding model of the water supply scheduling domain, and transform it into domain knowledge semantic vectors;

[0113] Step S204: Integrate the structured relationships of the water supply scheduling knowledge graph with the semantic retrieval features of the domain knowledge semantic vectors to construct a dual-modal retrieval database that combines graph entity association reasoning with vector semantic precise retrieval capabilities.

[0114] Specifically, knowledge graph construction refers to automatically identifying key entities in the water supply scheduling domain and the scheduling rule relationships between these entities from data slices, forming a structured knowledge network. The extracted triplet data undergoes deduplication, conflict detection, and logical consistency verification to ultimately form a high-quality domain knowledge graph. The cleaned triplet data is then stored in a graph database, establishing node labels, edge types, and attribute indexes to support efficient relation traversal and multi-hop queries, providing structured support for subsequent reasoning. The specific steps include:

[0115] Step S2001: Using a large language model or a dedicated information extraction model, identify key entities (such as "water plant", "Unit 1", "DN500 main pipeline", "pressure threshold 0.4MPa", etc.) and scheduling rule relationships between entities (such as "control", "connection", "influence", "belong to", "dependence", etc.) from each slice, and generate preliminary triples (head entity, relationship, tail entity);

[0116] Step S2002: Deduplication, conflict detection and logical consistency verification are performed on the extracted triples to finally form a high-quality domain knowledge graph;

[0117] Step S2003: Store the cleaned triples into a graph database, establish node labels, edge types and attribute indexes to support efficient relation traversal and multi-hop queries, and provide structured support for subsequent reasoning.

[0118] Furthermore, the vectorization process refers to inputting unstructured and structured slices into a fine-tuned embedding model adapted to the water supply scheduling domain after completing the slicing process of the water supply scheduling knowledge base. The embedding model outputs vector representations in a high-dimensional semantic space to obtain domain knowledge semantic vectors. Simultaneously, entities and relations in the knowledge graph are mapped to graph embedding vectors and combined with the original vectors to achieve deep encoding and semantic computability of knowledge. Finally, a dual-modal retrieval database integrating semantic vectors and structured relations is constructed to support retrieval, artificial intelligence reasoning, and context-aware decision-making in subsequent steps.

[0119] Step S30: Receive the water supply scheduling question input by the user in natural language, and convert the water supply scheduling question into a water supply scheduling question vector. Based on the water supply scheduling question vector, retrieve and match water supply scheduling knowledge fragments in the bimodal retrieval database. After fusing the water supply scheduling knowledge fragments, real-time water supply condition data and the water supply scheduling question, dynamically reconstruct semi-structured enhanced prompt words through a lightweight small language model.

[0120] Key entities and scheduling rule relationships between key entities in the water supply scheduling domain are extracted from the sliced ​​content to construct a water supply scheduling knowledge graph. The sliced ​​content is transformed into domain knowledge semantic vectors by adapting to the embedding model of the water supply scheduling domain. After constructing a bimodal retrieval database in conjunction with the water supply scheduling knowledge graph, water supply scheduling questions input by users in natural language are received and transformed into water supply scheduling question vectors. Based on the water supply scheduling question vectors, water supply scheduling knowledge fragments are retrieved and matched in the bimodal retrieval database. After information fusion of the water supply scheduling knowledge fragments, real-time water supply condition data and the water supply scheduling questions, semi-structured enhanced prompt words are dynamically reconstructed through a lightweight small language model.

[0121] The urban water supply scheduling decision generation system includes a Retrieval Enhanced Generation (RAG) module, a scene-adaptive prompt reconstruction module, and a large language model inference module. These components work collaboratively to complete the entire process from user question input to the generation of scheduling decision suggestions. The RAG module is an external knowledge retrieval system used to dynamically retrieve the most relevant water supply scheduling knowledge fragments from a bimodal retrieval database before the LLM (Large Language Model) generates a response. This provides factual evidence for the model and addresses the inherent static and illusionary nature of knowledge in LLM. The scene-adaptive prompt reconstruction module receives relevant knowledge fragments retrieved by the RAG module and the user's original question. It combines real-time operating data (such as real-time pressure or flow at monitoring points, equipment operating status, etc.) and introduces a lightweight small language model to dynamically construct semi-structured enhanced prompts, ensuring that the output of the large language model conforms to safety regulations and operational habits. The large language model reasoning module, as the core of the system's natural language understanding and response generation, receives enhanced prompts output by the scene adaptive prompt reconstruction module, performs deep semantic reasoning and logical deduction, and finally generates water supply scheduling decision suggestions that are semantically coherent, logically rigorous, compliant with operating procedures, and executable.

[0122] In some embodiments, the steps of receiving a water supply scheduling question input by a user in natural language, converting the water supply scheduling question into a water supply scheduling query vector, and retrieving matching water supply scheduling knowledge fragments from the bimodal retrieval database based on the water supply scheduling query vector include:

[0123] Step S301: Obtain the water supply scheduling question input by the user in natural language form, and transform the water supply scheduling question into a water supply scheduling question vector by adapting the embedding model of the water supply scheduling domain.

[0124] Please see Figure 2 as well as Figure 3 ,in, Figure 2 This is an example diagram illustrating the intelligent question-and-answer mechanism for the conventional scheduling method between water plant A and water plant B in this application embodiment; Figure 3 This is an example diagram illustrating the intelligent question-and-answer system for pressure monitoring points and normal pressure ranges in Area C, as described in an embodiment of this application. The dispatcher describes the current operating conditions or questions in natural language on the terminal interface, such as, "What are the standard scheduling methods for water plants A and B?". Upon receiving this text, the system first performs preprocessing, including removing irrelevant characters, identifying key entities (such as "water plant A," "water plant B," and "standard scheduling method"), and classifying the intent. Further, by adapting to an embedding model specific to water supply scheduling, the system transforms this filtered and extracted natural language text into a high-dimensional numerical vector, i.e., the water supply scheduling question vector.

[0125] Step S302: Based on the water supply scheduling question vector, perform semantic matching and entity association reasoning retrieval in the water supply scheduling bimodal retrieval database to obtain water supply scheduling knowledge fragments that match the water supply scheduling question.

[0126] Based on the generated water supply scheduling query vector, the system retrieves scheduling rules and relevant specification fragments that highly match the current scenario from a bimodal retrieval database, ensuring the accuracy of the retrieval results. Specifically, after obtaining the water supply scheduling query vector, the system quickly filters out the knowledge fragments with the highest relevance to the current question by calculating the similarity between the query vector and the fragment vectors contained in the database; simultaneously, based on the key entities identified in the question, the system performs retrieval matching in the knowledge graph, automatically associating their corresponding control units and relevant scheduling rules. This retrieval method can break through the limitations of literal expression. For example, when a user asks, "How should the water supply be managed when the pressure at monitoring point 11 is too low?", the system can not only automatically associate the semantic content of the "low pressure" handling section in the vector library, but also locate the basic information, operating status, and corresponding basic control rules of the control units (such as XX water pump, XX valve) related to the monitoring point through the graph. This ensures that no semantically similar content is missed, and that precise entries containing specific professional terms are captured. Furthermore, it deduces the implicit causal chain and operational dependencies, thereby ensuring the accuracy of the retrieval results and obtaining water supply scheduling knowledge fragments that match the water supply scheduling problem.

[0127] In a further embodiment, the step of fusing the water supply scheduling knowledge fragment, real-time water supply condition data, and the water supply scheduling problem, and then dynamically reconstructing semi-structured enhanced prompt words using a lightweight, small language model, includes:

[0128] Step S3001: Integrate the water supply scheduling knowledge fragments and real-time water supply status data obtained from the water supply scheduling retrieval with the water supply scheduling problem to determine the multi-source fusion information;

[0129] The urban water supply scheduling decision generation system integrates retrieved water supply scheduling knowledge fragments, real-time water supply condition data, and the user's original question to determine multi-source fusion information.

[0130] Step S3002: Input the multi-source fusion information into a lightweight small language model, and use the lightweight small language model to perform semi-structured reconstruction of the multi-source fusion information according to the characteristics of the water supply scheduling domain, and generate semi-structured enhanced prompt words that are adapted to the reasoning of large language models.

[0131] The multi-source fusion information is input into a lightweight small language model, which performs semantic reconstruction on the input multi-source fusion information to generate semi-structured enhanced prompt words. The lightweight small language model can automatically determine the type of the current question (such as "basic question and answer type", "emergency response type", "multi-step deduction type", etc.) and dynamically construct a suitable prompt word template structure accordingly. Subsequently, the module uses water supply scheduling knowledge fragments and real-time water supply condition data provided by the Retrieval Enhancement Generation (RAG) module as contextual knowledge injection templates, and embeds constraints corresponding to the current condition (such as "must refer to the content of Section X", "risk contingency plan needs to be explained", etc.), reasoning path guidance (such as "first locate the root cause, then give the operation steps") and output format requirements, and finally generates semi-structured enhanced prompt words adapted to the reasoning of the large language model.

[0132] Step S40: Input the semi-structured enhanced prompt words into the preset large language model. Through deep semantic reasoning and logical deduction of the large language model, generate water supply scheduling decision suggestions containing the operation object, specific action instructions and operation basis.

[0133] The system receives water supply scheduling questions input by users in natural language and transforms these questions into a water supply scheduling question vector. Based on this question vector, it retrieves matching water supply scheduling knowledge fragments from the bimodal retrieval database. After fusing the water supply scheduling knowledge fragments, real-time water supply condition data, and the water supply scheduling question, it dynamically reconstructs semi-structured enhanced prompts using a lightweight small language model. These semi-structured enhanced prompts are then input into a preset large language model. Through deep semantic reasoning and logical deduction by the large language model, a water supply scheduling decision suggestion containing the operation object, specific action instructions, and operation basis is generated.

[0134] Specifically, please refer to Figure 4 as well as Figure 5 ,in, Figure 4 This is an example diagram of the intelligent question-and-answer mechanism for scheduling measures when the pressure at monitoring point 11 is too low, as described in this application embodiment. Figure 5 This is an example diagram illustrating intelligent question-and-answer mechanisms for emergency dispatch measures when monitoring point 9 experiences insufficient pressure, as described in this embodiment. The Large Language Model (LLM) uses semi-structured enhanced prompts output by the scene-adaptive prompt reconstruction module to generate dispatch suggestions in natural language form. The output clearly includes the target of the operation, specific action instructions, and operational basis, assisting the dispatcher in decision-making.

[0135] After receiving the semi-structured enhanced prompts from the output of step S3002, the Large Language Model (LLM) begins to perform the final inference and generation tasks. The model first performs deep analysis of the information in the prompts to identify the core of the current problem, available resources, and constraints that must be followed.

[0136] Furthermore, the Large Language Model (LLM), based on its internalized general logic and knowledge of water supply scheduling, performs multi-step reasoning, ultimately generating a semi-structured, executable set of scheduling instructions. The output strictly adheres to a preset format, including: the operation object, such as the specific facilities / areas to be regulated, such as "monitoring point 12 high-pressure zone unit" or "XX valve"; the specific action instructions, such as directly executable operation requirements like "maintain pressure at 0.5MPa and above" or "first open valve #2, then adjust the frequency of unit #1"; and the operational basis, such as rules and data supporting the operation, such as "according to the emergency response plan in Chapter X of the Water Supply Scheduling Manual" or "based on the real-time operating condition that the pressure at monitoring point 11 is below 0.3MPa".

[0137] At this time, the urban water supply dispatch decision generation system provides specific operational steps that can be directly executed, such as clearly telling the dispatcher to "first open valve #2 to 50% opening, and then adjust the frequency of unit #1 to 45Hz", so that each operation is verifiable and can be implemented.

[0138] Step S50: Convert the water supply scheduling decision suggestion into a hydraulic simulation instruction, input it into a preset lightweight hydraulic simulation model for hydraulic calculation verification, and output the water supply scheduling decision corresponding to the target town to complete the generation of town water supply scheduling decision.

[0139] The semi-structured enhanced prompts are input into a preset large language model. After deep semantic reasoning and logical deduction by the large language model, water supply scheduling decision suggestions containing operation objects, specific action instructions and operation basis are generated. The water supply scheduling decision suggestions are then converted into hydraulic simulation instructions and input into a preset lightweight hydraulic simulation model for hydraulic calculation verification. The water supply scheduling decision corresponding to the target town is then output to complete the generation of town water supply scheduling decision.

[0140] Specifically, the water supply scheduling decision suggestions generated by the Large Language Model (LLM) are converted into hydraulic simulation instructions recognizable by the lightweight hydraulic simulation model, realizing the mapping from natural language decisions to executable simulation parameters (e.g., converting "maintain the pressure of Unit #1 at 0.5 MPa and above" into boundary condition setting instructions recognizable by the model). The converted hydraulic simulation instructions are input into the preset lightweight hydraulic simulation model to complete the hydraulic calculation verification of the water supply network. The core verification content includes node pressure, pipe flow, pump station, valve control response, etc., further verifying the engineering feasibility of the water supply scheduling decision suggestions given by the LLM, avoiding engineering biases caused by pure knowledge reasoning. The water supply scheduling decision suggestions that have passed the verification by the lightweight hydraulic simulation model are output as the water supply scheduling decision for the target town and returned to the user; if the verification fails, it can be fed back to the upstream link to regenerate or optimize the scheduling decision suggestions, and finally complete the generation of the town's water supply scheduling decision.

[0141] In some embodiments, the urban water supply dispatch decision generation system continuously optimizes the water supply dispatch decision suggestions generated by the Large Language Model (LLM) and presents them in the form of "natural language text" on the dispatch terminal, supporting dispatchers to perform adoption, modification, or rejection operations. The system automatically records the dispatcher's feedback behavior, actual execution results, and subsequent changes in operating conditions to form closed-loop feedback data. This closed-loop feedback data is used to periodically update the knowledge base and optimize the embedded model, thereby achieving continuous evolution of system performance.

[0142] As can be seen from the above embodiments, compared with the prior art, this application addresses the problems of response lag, judgment bias, and even decision-making errors caused by the reliance on individual experience in the water supply scheduling decision generation method, as well as the problems of delayed updates and poor flexibility caused by the reliance on expert experience and knowledge. This application includes, but is not limited to, the following beneficial effects:

[0143] Firstly, this application relies on a dedicated water supply scheduling knowledge base for the target town to construct standardized decision-making criteria, abandoning the subjective decision-making model driven by purely manual experience. It achieves accurate matching between scheduling knowledge and real-time operating conditions through a dual-modal retrieval database, and combines semi-structured enhanced prompt words to standardize the reasoning logic of the large language model, ensuring that scheduling decisions conform to domain rules and the actual water supply scenarios of the town. This can solve the problems of decision-making bias and response lag caused by individual experience differences, and significantly improve the accuracy of water supply scheduling decisions in complex operating conditions and emergency scenarios.

[0144] Secondly, this application constructs a domain-specific training dataset by processing the water supply scheduling knowledge base into slices, and fine-tunes the general embedding model to obtain an exclusive embedding model adapted to the water supply scheduling domain, which greatly improves the semantic accuracy of text vectorization. By combining the structured relationship of the knowledge graph with semantic vectors to construct a dual-modal retrieval database, it realizes dual matching of semantic similarity retrieval and entity association reasoning, which not only breaks through the limitations of literal retrieval, but also avoids invalid knowledge recall, greatly shortens the knowledge retrieval time, and improves the overall decision response speed.

[0145] Third, this application uses a semi-structured enhanced prompt word constraint large language model output to generate scheduling decision suggestions that clearly include the operation object, specific action instructions and operation basis, so as to realize the traceability of the decision process and the verifiability of the decision content; a lightweight hydraulic simulation model is introduced to simulate and verify the decision suggestions, and to identify potential engineering problems such as pipeline pressure imbalance and flow mismatch in advance, so as to ensure that the final output water supply scheduling decision can be directly implemented and solve the water supply risks and resource losses caused by ineffective decisions.

[0146] Furthermore, this application automates the entire process from knowledge base slicing, model fine-tuning, retrieval matching, prompt word reconstruction to decision generation and simulation verification, without requiring manual intervention. This reduces the workload of dispatchers and lowers the cost of manual on-duty personnel. It also reduces various resource losses such as pipeline losses and emergency response costs through precise decision-making, thereby achieving a dual improvement in water supply dispatch efficiency and economic benefits.

[0147] Please see Figure 6A town water supply scheduling decision generation device provided for one of the purposes of this application includes an embedded model building module 1100, a retrieval database building module 1200, an enhanced prompt word reconstruction module 1300, a scheduling suggestion generation module 1400, and a scheduling decision output module 1500. The embedding model construction module 1100 is configured to acquire a water supply scheduling knowledge base corresponding to the target town, slice the water supply scheduling knowledge base according to different chapters to obtain slice content corresponding to each chapter, construct a water supply scheduling domain-specific training dataset based on the slice content, and use the training dataset to perform supervised fine-tuning of a general embedding model to obtain an embedding model adapted to the water supply scheduling domain; the retrieval database construction module 1200 is configured to extract key entities and scheduling rule relationships between key entities from the slice content to construct a water supply scheduling knowledge graph, convert the slice content into domain knowledge semantic vectors by adapting the embedding model to the water supply scheduling domain, and construct a bimodal retrieval database by combining the water supply scheduling knowledge graph; the enhanced prompt word reconstruction module 1300 is configured to receive water supply scheduling questions input by the user in natural language, and convert the water supply... The scheduling problem is transformed into a water supply scheduling question vector. Based on the water supply scheduling question vector, a water supply scheduling knowledge fragment is retrieved and matched in the bimodal retrieval database. After fusing the water supply scheduling knowledge fragment, real-time water supply condition data, and the water supply scheduling problem, a semi-structured enhanced prompt word is dynamically reconstructed through a lightweight small language model. The scheduling suggestion generation module 1400 is configured to input the semi-structured enhanced prompt word into a preset large language model. Through deep semantic reasoning and logical deduction of the large language model, a water supply scheduling decision suggestion containing the operation object, specific action instructions, and operation basis is generated. The scheduling decision output module 1500 is configured to convert the water supply scheduling decision suggestion into a hydraulic simulation instruction, input it into a preset lightweight hydraulic simulation model for hydraulic calculation verification, and output the water supply scheduling decision corresponding to the target town to complete the generation of the town's water supply scheduling decision.

[0148] Based on any embodiment of this application, please refer to Figure 7 Another embodiment of this application also provides an electronic device, which can be implemented by a computer device, such as... Figure 7The diagram shows the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. The computer-readable storage medium stores an operating system, a database, and computer-readable instructions. The database may store a sequence of control information. When the computer-readable instructions are executed by the processor, the processor can implement a method for generating urban water supply scheduling decisions. The processor of the computer device provides computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can execute the urban water supply scheduling decision generation method of this application. The network interface of the computer device is used for communication with a terminal. Those skilled in the art will understand that… Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0149] In this embodiment, the processor is used to execute... Figure 6 The specific functions of each module are defined within the device, and the memory stores the program code and various types of data required to execute these modules. A network interface is used for data transmission between user terminals and the server. In this embodiment, the memory stores the program code and data required to execute all modules in the urban water supply scheduling decision generation device of this application, and the server can call the server's program code and data to execute the functions of all modules.

[0150] This application also provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the urban water supply scheduling decision generation method described in any embodiment of this application.

[0151] This application also provides a computer program product, including a computer program / instructions that, when executed by one or more processors, implement the steps of the urban water supply scheduling decision generation method described in any embodiment of this application.

[0152] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. This computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0153] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for generating urban water supply scheduling decisions, characterized in that, include: Obtain the water supply scheduling knowledge base corresponding to the target town, slice the water supply scheduling knowledge base according to different chapters to obtain the slice content corresponding to each chapter, construct a special training dataset for the water supply scheduling domain based on the slice content, and use the training dataset to perform supervised fine-tuning of the general embedding model to obtain an embedding model adapted to the water supply scheduling domain. Key entities and scheduling rule relationships between key entities in the water supply scheduling domain are extracted from the sliced ​​content to construct a water supply scheduling knowledge graph. The sliced ​​content is transformed into domain knowledge semantic vectors by adapting to the embedding model of the water supply scheduling domain, and a dual-modal retrieval database is constructed by combining the water supply scheduling knowledge graph. The system receives water supply scheduling questions input by users in natural language and transforms the water supply scheduling questions into water supply scheduling question vectors. Based on the water supply scheduling question vectors, it retrieves matching water supply scheduling knowledge fragments from the bimodal retrieval database. After fusing the water supply scheduling knowledge fragments, real-time water supply condition data, and the water supply scheduling questions, it dynamically reconstructs semi-structured enhanced prompt words through a lightweight small language model. The semi-structured enhanced prompt words are input into a preset large language model. Through deep semantic reasoning and logical deduction of the large language model, water supply scheduling decision suggestions containing the operation object, specific action instructions and operation basis are generated. The water supply scheduling decision suggestions are converted into hydraulic simulation instructions, input into a preset lightweight hydraulic simulation model for hydraulic calculation verification, and output the water supply scheduling decision corresponding to the target town to complete the generation of town water supply scheduling decisions.

2. The urban water supply scheduling decision generation method according to claim 1, characterized in that, The water supply scheduling knowledge base is sliced ​​according to different chapters to obtain the sliced ​​content corresponding to each chapter. A special training dataset for the water supply scheduling domain is constructed based on the sliced ​​content. The steps of supervising fine-tuning the general embedding model using the training dataset include: Obtain the water supply scheduling knowledge base corresponding to the target town, and slice the water supply scheduling knowledge base according to different chapters to obtain the slice content corresponding to each chapter. Based on the sliced ​​content, a training dataset specifically for the water supply scheduling domain is constructed. The training dataset is then used to perform supervised fine-tuning on the general embedding model to enhance its semantic understanding of water supply scheduling terminology, sentence structure, and contextual dependencies, thereby obtaining an embedding model adapted to the water supply scheduling domain.

3. The urban water supply scheduling decision generation method according to claim 2, characterized in that, The training dataset consists of triples constructed from query questions, positive examples of slice content, and negative examples of slice content generated based on slice content. The positive examples of the slice content represent slice content extracted from the slice content of the water supply scheduling knowledge base, which highly matches the core semantics of the query question generated based on the slice, and carries the target water supply scheduling knowledge and scheduling rules. The negative examples of the slice content represent slice content from the slice content of the water supply scheduling knowledge base that are unrelated to the core semantics of the query question generated based on the slice, and that have similar professional terms or expression structures on the surface to the positive examples of the slice content but are essentially different in scheduling scenarios and operation logic from the same document manual slice content.

4. The urban water supply scheduling decision generation method according to claim 1, characterized in that, The steps include: extracting key entities and scheduling rule relationships between key entities from the sliced ​​content to construct a water supply scheduling knowledge graph; transforming the sliced ​​content into domain knowledge semantic vectors by adapting to the embedding model of the water supply scheduling domain; and constructing a bimodal retrieval database by combining the water supply scheduling knowledge graph. Extract key entities and scheduling rule relationships between key entities from the content slices of each chapter of the water supply scheduling knowledge base; After deduplication, conflict detection, and logical consistency verification of the key entities and scheduling rule relationships, a water supply scheduling knowledge graph is constructed. By adapting the embedding model of the water supply scheduling domain, the slice content corresponding to each chapter is vectorized and transformed into domain knowledge semantic vectors. By integrating the structured relationships of the water supply scheduling knowledge graph with the semantic retrieval features of the domain knowledge semantic vectors, a dual-modal retrieval database is constructed that combines graph entity association reasoning with precise vector semantic retrieval capabilities.

5. The urban water supply scheduling decision generation method according to claim 1, characterized in that, The steps of receiving a water supply scheduling question input by a user in natural language, converting the water supply scheduling question into a water supply scheduling query vector, and retrieving matching water supply scheduling knowledge fragments from the bimodal retrieval database based on the water supply scheduling query vector include: The system acquires water supply scheduling questions input by users in natural language form and transforms these questions into water supply scheduling query vectors by adapting to an embedding model in the water supply scheduling domain. Based on the water supply scheduling question vector, semantic matching and entity association reasoning retrieval are performed in the water supply scheduling bimodal retrieval database to obtain water supply scheduling knowledge fragments that match the water supply scheduling question.

6. The urban water supply scheduling decision generation method according to claim 1, characterized in that, The steps of fusing the water supply scheduling knowledge fragment, real-time water supply condition data, and the water supply scheduling problem, and then dynamically reconstructing semi-structured enhanced prompt words using a lightweight, small language model, include: The water supply scheduling knowledge fragments obtained from water supply scheduling retrieval, real-time water supply condition data, and water supply scheduling issues are fused together to determine multi-source fused information. The multi-source fusion information is input into a lightweight small language model, which then reconstructs the multi-source fusion information based on the characteristics of the water supply scheduling domain, generating semi-structured enhanced prompt words that are adapted to the reasoning of large language models.

7. The urban water supply scheduling decision generation method according to any one of claims 1 to 6, characterized in that, The steps to build a water supply scheduling knowledge base include: Obtain the water supply dispatch manual corresponding to the target town, digitize the water supply dispatch manual, and organize the digitized water supply dispatch manual according to water supply dispatch knowledge categories to construct a water supply dispatch knowledge base corresponding to the target town. The water supply dispatch manual is a standardized professional document in the field of water supply in the target town, which contains core dispatch knowledge such as basic water supply parameters, routine dispatch strategies, emergency response plans and operation execution specifications.

8. A device for generating urban water supply dispatching decisions, characterized in that, include: The embedding model building module is configured to acquire the water supply scheduling knowledge base corresponding to the target town, slice the water supply scheduling knowledge base according to different chapters to obtain the slice content corresponding to each chapter, construct a special training dataset for the water supply scheduling domain based on the slice content, and use the training dataset to perform supervised fine-tuning of the general embedding model to obtain an embedding model adapted to the water supply scheduling domain. The retrieval database construction module is configured to extract key entities and scheduling rule relationships between key entities from the sliced ​​content to construct a water supply scheduling knowledge graph. By adapting the embedding model of the water supply scheduling domain, the sliced ​​content is transformed into domain knowledge semantic vectors, and a dual-modal retrieval database is constructed in combination with the water supply scheduling knowledge graph. The enhanced prompt word reconstruction module is configured to receive water supply scheduling questions input by users in natural language, and convert the water supply scheduling questions into water supply scheduling question vectors. Based on the water supply scheduling question vectors, it retrieves matching water supply scheduling knowledge fragments from the bimodal retrieval database. After fusing the water supply scheduling knowledge fragments, real-time water supply condition data and the water supply scheduling questions, it dynamically reconstructs semi-structured enhanced prompt words through a lightweight small language model. The scheduling suggestion generation module is configured to input the semi-structured enhanced prompt words into a preset large language model, and generate water supply scheduling decision suggestions containing the operation object, specific action instructions and operation basis through deep semantic reasoning and logical deduction of the large language model. The scheduling decision output module is configured to convert the water supply scheduling decision suggestions into hydraulic simulation instructions, input them into a preset lightweight hydraulic simulation model for hydraulic calculation verification, and output the water supply scheduling decision corresponding to the target town to complete the generation of town water supply scheduling decisions.

9. An electronic device comprising a central processing unit and a memory, characterized in that, The central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implemented according to any one of claims 1 to 7, which, when invoked by a computer, executes the steps included in the corresponding method.

Images