Constraint-aware reservoir scheduling decision method, device and equipment
By combining a large language model and a constraint knowledge base, and introducing an expert collaborative revision mechanism, the problem of poor scalability of existing reservoir scheduling methods is solved, enabling flexible and adaptive reservoir scheduling decisions and improving the accuracy and interpretability of the decisions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SOUTH CHINA UNIV OF TECH
- Filing Date
- 2026-03-16
- Publication Date
- 2026-07-14
AI Technical Summary
Existing reservoir scheduling methods are based on rule bases or expert systems. The rules are rigid and have poor scalability, making it difficult to adapt to complex and ever-changing scheduling scenarios, resulting in insufficient adaptability to dynamic scheduling requirements.
A constraint-aware reservoir scheduling decision-making method is adopted, which combines a large language model and a structured constraint knowledge base. The scheduling scheme is generated through a constraint retrieval mechanism, and an expert collaborative revision mechanism is introduced to form a closed loop of generation-revision-decision, ensuring that the scheme conforms to professional standards and expert experience.
The generated scheduling strategies are accurate, feasible, and optimized, improving the flexibility and adaptability of scheduling decisions, reducing the risk of constraint violations, and enhancing the interpretability and traceability of the decision-making process.
Smart Images
Figure CN122390261A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of smart water conservancy and intelligent decision-making technology, and more specifically, to a reservoir scheduling decision-making method, device and equipment based on constraint perception. Background Technology
[0002] Water conservancy scheduling is a core means of ensuring flood control safety and optimizing water resource allocation in river basins. Current scheduling methods mainly rely on rule bases or expert system-assisted decision-making tools to solidify some scheduling procedures, flood storage and detention area activation conditions, water supply priorities, etc., into rules to achieve recommended operations under specific operating conditions. However, the rules are rigid and have poor scalability, making it difficult to adapt to complex and ever-changing scenarios, resulting in insufficient adaptability to dynamic scheduling needs. Summary of the Invention
[0003] This application provides a constraint-aware reservoir scheduling decision-making method, apparatus, and equipment to at least solve the technical problem of difficulty in adaptively generating reservoir scheduling strategies in related technologies.
[0004] According to one aspect of the embodiments of this application, a reservoir scheduling decision-making method based on constraint awareness is provided, including: Based on the scheduling type, water conservancy project characteristic parameters, and current project operation status and hydrological situation, the corresponding constraints are automatically retrieved from the preset constraint knowledge base to form a scenario constraint set; The system receives the scheduling target description text input by the dispatcher through the human-computer interaction interface, and inputs the description text, the scenario constraint set, and preset water conservancy domain knowledge into the big language model. Receive one or more candidate scheduling schemes generated by a large language model, which include time-segmented outflow, reservoir water level control strategies, downstream river flow evolution and inundation risk analysis; The system receives revision opinions from experts via an interactive interface for the candidate scheduling scheme, transforms these revision opinions into new constraints, and drives the large language model to generate a revised scheduling scheme again. After responding to the expert confirmation signal, the system outputs the final scheduling decision scheme.
[0005] In one implementation, before automatically retrieving the corresponding constraints from a preset constraint knowledge base based on the scheduling type, water conservancy project characteristic parameters, and the current project operation status and hydrological situation, the method further includes: Access to real-time weather forecasts, engineering condition data, and hydrological monitoring data; Based on the real-time weather forecast, engineering condition data, and hydrological monitoring data, the scheduling type, water conservancy project characteristic parameters, and current project operation status and hydrological situation are determined.
[0006] In one implementation, before inputting the descriptive text, the scenario constraint set, and preset water conservancy domain knowledge into the large language model, the method further includes: Acquire knowledge from multiple sources, including watershed planning, scheduling procedures, preset historical typical flood events, and expert experience. Perform structured processing on the knowledge in these domains to construct a knowledge graph for water conservancy scheduling. Based on the scheduling type, water conservancy project characteristic parameters, current project operation status, and hydrological situation, relevant entities are matched in the knowledge graph; Starting with the matched entity, a graph traversal algorithm is used to retrieve related knowledge. The retrieved related knowledge is then converted into natural language descriptions to form knowledge prompt text in the water conservancy field. This knowledge prompt text in the water conservancy field is then input into a large language model.
[0007] In one implementation, after receiving one or more candidate scheduling schemes generated by the large language model, which include time-segmented outflow, reservoir water level control strategies, and downstream river flow evolution and inundation risk analysis, the method further includes: The constraint knowledge base is invoked to perform multi-dimensional constraint verification on the candidate scheduling schemes, thereby obtaining candidate scheduling schemes that satisfy the constraints. Based on the candidate scheduling schemes that satisfy the constraints, and the multi-objective optimization strategy in the domain knowledge, a differentiated scheduling scheme including flood control priority, power generation priority and ecological priority is generated.
[0008] In one implementation, converting the revised opinion into a new constraint includes: The revised opinions are semantically parsed to extract the constraint object, constraint type, applicable time period, and threshold parameter to obtain structured constraint items; A consistency check is performed on the structured constraint items. If there is a conflict with the scenario constraint set, the conflict is resolved or the conflict item is marked according to the preset priority rules. Write the structured constraints after the consistency check into the scenario constraint set.
[0009] In one implementation, it further includes: After the scheduling plan is implemented, actual hydrological monitoring data, engineering operation data, and scheduling effect data are collected and compared with the expected process in the plan generation stage to obtain the deviation of key indicators. Based on the deviation of the key indicators, the parameters and domain knowledge rules in the constraint knowledge base are dynamically updated.
[0010] In one implementation, it further includes: When the scheduling type is joint scheduling of reservoir groups, constraints on the joint action of multiple reservoirs, constraints on inter-basin water transfer, and constraints on multi-objective benefits are constructed. The constraints of the combined effect of multiple reservoirs, the constraints of inter-basin water transfer, and the constraints of multi-objective benefits are added to the scenario constraint set. Receive the output data of the pre-trained joint scheduling optimization model and the multi-node hydrodynamic model, and construct model prediction prompt information based on the output data; The description text, the set of scene constraints, the preset knowledge of the water conservancy field, and the model prediction prompts are input into the large language model.
[0011] In one implementation, before receiving the output data of the pre-trained joint scheduling optimization model and the multi-node hydrodynamic model, the method further includes: Real-time weather data, engineering condition data, and hydrological monitoring data of each reservoir are input into a pre-trained joint scheduling optimization model, which outputs time-segmented scheduling strategies for each reservoir. Real-time weather data, engineering condition data, and hydrological monitoring data of each reservoir are input into a pre-trained multi-node hydrodynamic model to obtain simulation data of the future outflow water level and downstream water level changes of each reservoir.
[0012] According to another aspect of the embodiments of this application, a reservoir scheduling decision-making device based on constraint perception is provided, comprising: The constraint construction module is used to automatically retrieve corresponding constraints from a preset constraint knowledge base based on the scheduling type, water conservancy project characteristic parameters, and the current project operation status and hydrological situation, and form a scenario constraint set. The large language model human-computer interaction module is used to receive the scheduling target description text input by the dispatcher through the human-computer interaction interface, and input the description text, the scenario constraint set and the preset water conservancy domain knowledge into the large language model. The constraint-aware decision generation module is used to receive one or more candidate scheduling schemes generated by the large language model, which include time-segmented outflow, reservoir water level control strategies, downstream river flow evolution and inundation risk analysis. The expert collaboration module is used to receive the revision opinions of experts on the candidate scheduling scheme through the interactive interface, transform the revision opinions into new constraints, drive the large language model again to generate the revised scheduling scheme, and output the final scheduling decision scheme after responding to the expert confirmation signal.
[0013] According to another aspect of the embodiments of this application, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the above-described constraint-aware reservoir scheduling decision-making method through the computer program.
[0014] The technical solutions provided in this application embodiment may include the following beneficial effects: This application constructs an intelligent scheduling system that integrates constraint awareness and expert collaboration by deeply coupling a large language model with a structured constraint knowledge base and expert decision-making suggestions. Specifically, it utilizes a constraint knowledge base to uniformly model multi-source constraints, and a constraint retrieval mechanism enables the large language model to possess constraint awareness and compliance capabilities, ensuring that the generated solutions conform to professional standards. Simultaneously, it incorporates expert revision opinions and feedback on candidate solutions, forming a generation-revision-decision closed loop. This allows the large language model to dynamically integrate expert experience and generate accurate, feasible, and optimized scheduling strategies for complex scheduling scenarios. Attached Figure Description
[0015] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a constraint-aware reservoir scheduling decision-making method according to an embodiment of this application; Figure 2 This is a schematic diagram of a reservoir scheduling decision-making method based on constraint perception according to an embodiment of this application; Figure 3 This is a schematic diagram of a reservoir scheduling decision-making device based on constraint perception according to an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an optional electronic device according to an embodiment of this application. Detailed Implementation
[0016] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.
[0017] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0018] In the field of water conservancy scheduling, basin flood control scheduling decision support systems and reservoir group joint scheduling systems based on hydrodynamic and optimization models can be used. These systems provide reservoir water level and downstream water level processes through model calculations, but their ability to explain whether they fully comply with legal and regulatory constraints is weak. Alternatively, document question-and-answer and writing assistance tools based on general large language models can provide question-and-answer and writing analysis and explanations for regulations and reports. However, when directly used for scheduling decisions, they lack explicit modeling of constraints such as engineering safety, flood control, ecology, and regulations, and are not deeply coupled with hydrological and hydrodynamic models and optimization models.
[0019] This application unifies the modeling of multi-source constraints through a constraint knowledge base and constraint engine, and introduces a constraint retrieval mechanism before solution generation. This enables the large language model to have constraint awareness and constraint compliance capabilities in scheduling scenarios, thereby reducing the risk of constraint violation suggestions. By combining the large language model, physical model, and expert decision-making process, a closed loop of "generation-revision-decision-feedback" is formed. The system's capabilities can evolve continuously with use, rather than being fixed once. At the same time, the automatically generated decision explanations and log records significantly improve the interpretability and traceability of the decision-making process, which is beneficial for future review, responsibility definition, and experience review.
[0020] The constraint-aware reservoir scheduling decision-making method of this application, as described below with reference to the accompanying drawings, will be described in detail. Figure 1 As shown, the method mainly includes the following steps: S101 automatically retrieves the corresponding constraints from the preset constraint knowledge base based on the scheduling type, water conservancy project characteristic parameters, current project operation status, and hydrological situation, forming a scenario constraint set.
[0021] In one implementation, the system of this application is deployed on the server or cloud platform of the reservoir dispatch center, and first accesses real-time weather forecasts, engineering condition data and hydrological monitoring data.
[0022] Specifically, it can access real-time weather forecasts through standard API interfaces, and also includes acquiring engineering operating data, such as reservoir water level, outflow, gate opening, unit operating status, and engineering safety monitoring status. It also includes acquiring hydrological monitoring data, such as real-time rainfall, cumulative rainfall, river flow, and reservoir inflow from upstream and the reservoir area. The above data undergoes time alignment and missing value interpolation, and necessary statistical indicators are calculated before being stored in the database.
[0023] Furthermore, based on real-time weather forecasts, engineering condition data, and hydrological monitoring data, the scheduling type, water conservancy project characteristic parameters, and current project operation status and hydrological situation are determined. First, based on real-time rainfall forecasts, inflow data, and reservoir water level data, the scheduling type is automatically identified through rule matching. Scheduling types include single reservoir flood control scheduling, water supply scheduling, power generation scheduling, ecological scheduling, and joint beneficial scheduling of reservoir groups. In an optional implementation, the scheduling type can also be specified by the user through an interactive interface. Second, engineering characteristic parameters, such as reservoir size, functional positioning, and design parameters, are extracted from the engineering basic database. Finally, based on real-time hydrological monitoring data and engineering condition data, the current project operation status and hydrological situation are determined, such as current reservoir water level, inflow, outflow, gate opening, unit operating status, and external environmental conditions.
[0024] It also includes building a constraint knowledge base, which determines the reservoir's scheduling boundary conditions from design documents and scheduling procedures, including upper and lower limits of water level, upper and lower limits of discharge flow, safety threshold of downstream control section, engineering operation capacity constraints, necessary initial operating states, laws and regulations, and other constraint parameters, and stores them in the constraint knowledge base in a unified structure.
[0025] Taking the flood control scheduling of a single large reservoir as an example, when the flood season arrives, the constraint engine automatically retrieves the engineering safety, flood control, water supply, power generation, ecological and legal constraints that the scheduling needs to meet from the constraint knowledge base based on the scheduling type, water conservancy project characteristic parameters, current project operation status and hydrological situation, thus forming a scenario constraint set.
[0026] Specifically, the constraint engine automatically retrieves and matches multi-source constraints that the current scheduling needs to meet based on real-time determined scheduling types, such as flood control scheduling; water conservancy project characteristic parameters, such as reservoir size and design flood level; and the current project operation status, such as reservoir water level, outflow, and gate opening. These constraints include engineering safety constraints, such as check flood level limits; flood control constraints, such as downstream safe discharge; water supply constraints, such as minimum water demand guarantee; power generation constraints, such as unit output range; ecological constraints, such as minimum ecological flow; and regulatory constraints, such as scheduling procedure clauses. This ultimately forms a scenario constraint set highly adapted to the current scenario.
[0027] S102 receives the scheduling target description text input by the dispatcher through the human-computer interaction interface, and inputs the description text, scenario constraint set and preset water conservancy domain knowledge into the large language model.
[0028] In this embodiment, dispatchers can directly describe their dispatching goals and concerns in natural language through the dialogue interface provided by the system. For example, they can input "Flood control dispatching in the next 24 hours, prioritizing the safety of downstream towns while also taking into account ecological flow needs." The system uses natural language processing technology to parse the description, extract key dispatching intentions, and integrate them with scenario constraint sets and domain knowledge to construct input prompts for a large language model, thereby driving the model to generate candidate dispatching schemes that fit actual needs.
[0029] In one implementation, before inputting the descriptive text, scenario constraint set, and preset water conservancy domain knowledge into the large language model, the method further includes acquiring multi-source domain knowledge such as watershed planning, scheduling procedures, preset flood cases, and expert experience, performing structured processing on the domain knowledge, and constructing a water conservancy scheduling domain knowledge graph.
[0030] Specifically, water conservancy project entities, such as reservoirs and dams; functional entities, such as flood control and water supply; and their relationships are extracted from watershed planning documents, such as Reservoir A undertaking flood control functions; scheduling rule entities, such as flood control limit water level control rules, and their applicable conditions, such as water level thresholds and flow thresholds, are extracted from scheduling procedures; case scenario entities, such as the 1998 flood, scheduling action entities, and effect evaluation entities, are extracted from preset historical typical flood events; and empirical rule entities, such as avoiding large-scale discharge at night, are extracted from expert experience. The extracted entities and relationships are then integrated to form a unified knowledge graph model. The knowledge graph is stored in a graph database, where nodes represent entities and edges represent relationships between entities, supporting efficient graph traversal and relational queries.
[0031] Based on the scheduling type, water conservancy project characteristic parameters, current project operation status, and hydrological situation, relevant entities are matched in the knowledge graph, such as project type, gate / unit capacity, control section, scheduling rules, etc. Then, using these nodes as the starting point, graph traversal / graph query is used to convert the retrieved knowledge into natural language descriptions, forming water conservancy knowledge prompt text, which is then input into the large language model.
[0032] Specifically, based on the current scheduling type, water conservancy project characteristic parameters, and the current project operation status and hydrological situation, relevant entities are matched in the knowledge graph. For example, in a flood control scheduling scenario, engineering entities, rule entities, and historical case entities related to "flood control function" are matched. Starting from the matched entities, related knowledge is retrieved using a graph traversal algorithm. For example, scheduling rules, historical cases, and expert experience associated with the current engineering entity are retrieved. The retrieved knowledge is prioritized, such as legal requirements taking precedence over empirical rules, and cases highly relevant to the current scenario taking precedence over general cases. The prioritized knowledge is then transformed into natural language descriptions, such as domain knowledge hints based on preset templates. These hints include rule clauses, case references, and expert experience, serving as input components of the large language model.
[0033] In the embodiments of this application, general-purpose large language models with strong natural language understanding and generation capabilities, such as the GPT series, LLaMA series, and ChatGLM series, are preferentially selected to ensure the accuracy of their understanding of natural language inputs such as scheduling objectives and expert revision opinions. This application does not specifically limit the specific implementation of the large language model.
[0034] S103 receives one or more candidate scheduling schemes generated by a large language model, which include time-segmented outflow, reservoir water level control strategies, downstream river flow evolution, and inundation risk analysis.
[0035] The large language model generates one or more candidate scheduling schemes based on the input scheduling objective, scenario constraint set, and domain knowledge. Specifically, the candidate scheduling schemes include: first, outputting time-segmented outflow sequences, such as hourly flood discharge flow, clearly defining the flow control values for each time period; second, providing reservoir water level control strategies; and finally, generating textual descriptions of the downstream impacts, such as the estimated peak downstream flow after the scheme's implementation being 1500 m³ / s, lower than the flow corresponding to the guaranteed water level, but requiring attention to the risk of inundation in local low-lying areas. This provides comprehensive and interpretable candidate schemes for expert decision-making.
[0036] After receiving one or more candidate scheduling schemes generated by the large language model, which include time-segmented outflow, reservoir water level control strategies and downstream impact analysis, the process also includes calling the constraint knowledge base to perform multi-dimensional constraint verification on the candidate scheduling schemes to obtain candidate scheduling schemes that meet the constraints.
[0037] For example, the constraint engine is invoked to review each candidate scheduling scheme item by item based on preset thresholds such as flood level, flood limit level, downstream warning / guarantee level, and minimum ecological flow. Scheduling schemes that do not meet the constraints are eliminated, and candidate scheduling schemes that meet the constraints are obtained.
[0038] Optionally, it also includes candidate scheduling schemes based on satisfying constraints, as well as multi-objective optimization strategies in domain knowledge, to generate differentiated scheduling schemes that prioritize flood control, power generation, and ecology.
[0039] Optionally, based on candidate scheduling schemes that meet the constraints, and combined with multi-objective optimization strategies from domain knowledge, differentiated scheduling schemes including flood control priority, power generation priority, and ecological priority are generated. The flood control priority scheme reduces flood risk by pre-releasing water from the reservoir, sacrificing some power generation or water supply benefits; the power generation priority scheme maintains a high reservoir water level to maximize the power generation head and controls the outflow to not exceed the downstream safe discharge; the ecological priority scheme ensures the continuous release of ecological base flow and appropriately adjusts flood control and power generation plans, thereby providing expert decision-making with a set of alternative schemes covering different scheduling objectives.
[0040] S104 receives the experts' revision opinions on the candidate scheduling scheme through the interactive interface, transforms the revision opinions into new constraints, and drives the large language model to generate the revised scheduling scheme again. After responding to the expert confirmation signal, it outputs the final scheduling decision scheme.
[0041] In the embodiments of this application, collaborative decision-making with experts is possible. Experts provide revision suggestions for candidate solutions through the system, and these revision suggestions are transformed into new constraints. This includes: performing semantic parsing on the revision suggestions to extract constraint objects, constraint types, applicable time periods, and threshold parameters to obtain structured constraint items; performing consistency checks on the structured constraint items, and if there are conflicts with the scenario constraint set, resolving the conflicts or marking the conflicting items according to preset priority rules; and writing the consistency-checked structured constraint items into the scenario constraint set.
[0042] For example, if the proposed revision involves adjusting the timing or flow rate of flood discharge, the system will convert the revision into new constraints, such as "adjust the outflow rate at 8:00 to 1200 m³ / s," or preference parameters, such as "increase the flood control weight to 0.8." This will then drive the large language model to generate a revised plan. Finally, after the plan is confirmed by experts, the final scheduling decision will be output.
[0043] In one alternative implementation, the scheduling decision scheme is a scheduling recommendation document that includes a constraint list, scheduling strategy, key points of the scheduling curve, risk description, and source of the terms.
[0044] In one implementation, the method further includes collecting actual hydrological monitoring data, engineering operation data, and scheduling effect data after the scheduling plan is implemented, comparing and analyzing them with the expected process during the plan generation stage to obtain the key indicator deviation; and dynamically updating the parameters and domain knowledge rules in the constraint knowledge base based on the key indicator deviation.
[0045] Specifically, after the scheduling plan is implemented, actual hydrological monitoring data such as reservoir water level and downstream flow, as well as engineering operation data and scheduling effect data, such as the degree of reservoir water level decline and ecological flow satisfaction rate, are collected. These are compared and analyzed with the expected process during the plan generation phase to calculate deviations in key indicators, such as peak flow deviation and water level control deviation. Based on these deviations, parameters and domain knowledge rules in the constraint knowledge base are dynamically updated, such as converting effective scheduling strategies into rules, thereby achieving continuous evolution of system capabilities.
[0046] In another implementation scenario of this application, this application supports joint beneficial scheduling of reservoir groups by extending the constraint knowledge base and model interface. For example, in the joint scheduling of reservoir groups, constraints on the joint action of multiple reservoirs, constraints on inter-basin water transfer, and constraints on multiple objectives of benefits are constructed; and the constraints on the joint action of multiple reservoirs, constraints on inter-basin water transfer, and constraints on multiple objectives of benefits are added to the scenario constraint set.
[0047] Specifically, in the scenario of joint operation of multiple reservoirs, constraints on the joint action of multiple reservoirs are constructed. These constraints refer to the collaborative rules set in the scenario of joint operation of a group of reservoirs to coordinate the operation of each reservoir and ensure the overall safety and benefits of the basin. Examples include total outflow constraints and reservoir capacity allocation constraints. Inter-basin water transfer constraints are also constructed. These constraints refer to the limiting conditions set for inter-basin water resource allocation projects to ensure the sustainable use of both the sending and receiving basins. Examples include water conveyance capacity constraints and water source allocation constraints. Finally, multi-objective benefit constraints are constructed. These constraints refer to the quantitative trade-off rules set in water conservancy operation to balance conflicting objectives such as flood control, water supply, power generation, and ecology. Examples include power generation benefit constraints, water supply priority constraints, and ecological benefit constraints.
[0048] Based on engineering safety, flood control, water supply, power generation, ecology and legal constraints, constraints on the joint action of multiple reservoirs, inter-basin water transfer, and multi-objective benefit are added to the scenario constraint set.
[0049] Furthermore, by integrating the existing joint scheduling optimization model for the reservoir group and the multi-node hydrodynamic model, real-time weather data, engineering condition data, and hydrological monitoring data of each reservoir are input into the pre-trained joint scheduling optimization model, which outputs time-segmented scheduling strategies for each reservoir. Real-time weather data, engineering condition data, and hydrological monitoring data of each reservoir are then input into the pre-trained multi-node hydrodynamic model to obtain simulation data of future outflow water levels and downstream water level changes for each reservoir.
[0050] Furthermore, the system receives output data from a pre-trained joint scheduling optimization model and a multi-node hydrodynamic model, and constructs model prediction prompts based on the output data. The descriptive text, scenario constraint set, pre-defined hydraulic domain knowledge, and model prediction prompts are input into a large language model. In this implementation, a specialized model first calculates quantitative predictions, and then the large language model performs scheme organization and semantic interpretation. Downstream impact analysis can be based on the hydrodynamic model output; for single reservoirs not connected to the hydrodynamic model, prompts are generated based on rules / domain knowledge.
[0051] Specifically, dispatchers input the target in natural language, and the system integrates the target, the set of scenario constraints, the theoretical optimal solution output by the optimization model, and the downstream evolution data provided by the hydrodynamic model into a large language model prompt, generating multiple candidate solutions, including information such as the time-segmented outflow of each reservoir, reservoir water level control strategies, and joint dispatch instructions.
[0052] Furthermore, the generated candidate solutions can be constrained and revised collaboratively by experts. The system then transforms the revised opinions into new constraints, driving the large language model to regenerate solutions. The solutions include information such as multi-reservoir joint scheduling strategies, revenue distribution, and risk descriptions.
[0053] This application only adds constraints on the joint action of multiple reservoirs, inter-basin water transfer constraints, and multi-objective benefit constraints to the constraint knowledge base, and integrates the corresponding joint scheduling optimization model and multi-node hydrodynamic model. This allows for constraint-aware human-machine collaborative decision-making for joint output and water supply of multiple reservoirs, following the same process as single-reservoir scheduling. This implementation demonstrates that, by simply coupling constraint extension with the model interface, this application can achieve constraint-aware, large-scale language model-based human-machine interactive expert collaborative decision-making in water conservancy scheduling scenarios while maintaining the computational accuracy of professional models.
[0054] By using the existing joint scheduling optimization model and multi-node hydrodynamic model as the basic computing engine, and deeply coupling them with the constraint knowledge base and large language model, the optimization model outputs the theoretically optimal scheduling strategy based on the input data, and the hydrodynamic model provides physical process simulation data. Together, they constitute the quantitative benchmark and feasibility verification basis for the large language model generation scheme.
[0055] To facilitate understanding of the methods in the embodiments of this application, the following description is provided in conjunction with the appendix. Figure 2 Further description. For example... Figure 2As shown, the system first integrates real-time weather data and engineering condition data, and initializes the scheduling scenario based on the current time and scheduling cycle. Then, the constraint engine automatically retrieves the engineering safety, flood control, water supply, power generation, ecological, and regulatory constraints that the current scheduling must meet from the constraint knowledge base, based on the scheduling type, engineering attributes, and current conditions, forming a scenario constraint set. Scheduling personnel describe their scheduling objectives and concerns in natural language through a dialog interface. The system organizes this description, along with the scenario constraint set and relevant domain knowledge, as input to the large language model. Based on this, the large language model generates one or more candidate scheduling schemes, including time-segmented outflow, reservoir water level control strategies, and textual descriptions of downstream impacts, assisting experts in decision-making. Experts can submit revision suggestions for the schemes through the system, which transforms these into new constraints or preference parameters, driving the large language model to generate revised schemes again. Finally, after the scheme is confirmed by experts, the system automatically generates a scheduling recommendation document containing a constraint list, key points of the scheduling curve, risk descriptions, and the source of the clauses. This document is used for scheduling commands, and the actual process is compared with the expected process afterward to update some constraint parameters and empirical rules.
[0056] This application unifies the modeling of multi-source constraints, including engineering safety, flood control, water supply, power generation, ecology, and regulations, through a constraint knowledge base. A constraint retrieval mechanism is introduced before scheme generation, enabling the large language model to possess constraint awareness and compliance capabilities in scheduling scenarios, thereby effectively reducing the risk of constraint violations. By deeply coupling the large language model with existing hydrological and hydrodynamic models and scheduling optimization models, and introducing expert revision and feedback mechanisms, a closed-loop collaborative decision-making process of "generation-revision-decision-feedback" is formed. This process can generate accurate, feasible, and optimized scheduling strategies for complex and ever-changing scheduling scenarios. Simultaneously, automatically generated decision explanations and full-process log recording significantly improve the interpretability and traceability of the decision-making process, providing reliable evidence for review, responsibility definition, and experience review. The system's capabilities can evolve continuously with use, rather than being fixed once, achieving a comprehensive improvement in the scientific nature, flexibility, and adaptability of water conservancy scheduling decisions.
[0057] According to another aspect of the embodiments of this application, a constraint-aware reservoir scheduling decision-making apparatus for implementing the above-described constraint-aware reservoir scheduling decision-making method is also provided. For example... Figure 3 As shown, the device includes: The constraint construction module 301 is used to automatically retrieve corresponding constraints from the preset constraint knowledge base based on the scheduling type, water conservancy project characteristic parameters, and the current project operation status and hydrological situation, and form a scenario constraint set. The human-computer interaction module 302 of the large language model is used to receive the scheduling target description text input by the scheduling personnel through the human-computer interaction interface, and input the description text, the scenario constraint set and the preset water conservancy domain knowledge into the large language model. The constraint-aware decision generation module 303 is used to receive one or more candidate scheduling schemes generated by the large language model, which include time-segmented outflow, reservoir water level control strategy, downstream river flow evolution and inundation risk analysis. The expert collaboration module 304 is used to receive the revision opinions of experts on the candidate scheduling scheme through the interactive interface, transform the revision opinions into new constraints, drive the large language model to generate the revised scheduling scheme again, and output the final scheduling decision scheme after responding to the expert confirmation signal.
[0058] It should be noted that the constraint-aware reservoir scheduling decision-making device provided in the above embodiments is only illustrated by the division of the above functional modules when executing the constraint-aware reservoir scheduling decision-making method. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the constraint-aware reservoir scheduling decision-making device and the constraint-aware reservoir scheduling decision-making method embodiments provided in the above embodiments belong to the same concept, and the implementation process is detailed in the method embodiments, which will not be repeated here.
[0059] According to another aspect of the embodiments of this application, an electronic device corresponding to the constraint-aware reservoir scheduling decision-making method provided in the foregoing embodiments is also provided, so as to execute the above-described constraint-aware reservoir scheduling decision-making method.
[0060] Please refer to Figure 4 This illustrates a schematic diagram of an electronic device provided by some embodiments of this application. For example... Figure 4 As shown, the electronic device includes: a processor 400, a memory 401, a bus 402, and a communication interface 403. The processor 400, the communication interface 403, and the memory 401 are connected via the bus 402. The memory 401 stores a computer program that can run on the processor 400. When the processor 400 runs the computer program, it executes the constraint-aware reservoir scheduling decision-making method provided in any of the foregoing embodiments of this application.
[0061] The memory 401 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 403 (which can be wired or wireless), such as the Internet, wide area network, local area network, or metropolitan area network.
[0062] Bus 402 can be an ISA bus, PCI bus, or EISA bus, etc. Buses can be divided into address buses, data buses, control buses, etc. Memory 401 is used to store programs. After receiving execution instructions, processor 400 executes the program. The constraint-aware reservoir scheduling decision-making method disclosed in any of the aforementioned embodiments of this application can be applied to processor 400, or implemented by processor 400.
[0063] The processor 400 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 400 or by instructions in software form. The processor 400 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 401. The processor 400 reads the information in memory 401 and, in conjunction with its hardware, completes the steps of the above method.
[0064] The electronic device provided in this application embodiment and the reservoir scheduling decision-making method based on constraint perception provided in this application embodiment are based on the same inventive concept and have the same beneficial effects as the methods they adopt, operate or implement.
[0065] According to another aspect of the embodiments of this application, a computer-readable storage medium corresponding to the constraint-aware reservoir scheduling decision method provided in the foregoing embodiments is also provided, wherein a computer program (i.e., a program product) is stored thereon, and when the computer program is run by a processor, it executes the constraint-aware reservoir scheduling decision method provided in any of the foregoing embodiments.
[0066] It should be noted that examples of computer-readable storage media may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.
[0067] The computer-readable storage medium provided in the above embodiments of this application and the reservoir scheduling decision-making method based on constraint perception provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.
[0068] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0069] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.
Claims
1. A reservoir scheduling decision-making method based on constraint perception, characterized in that, include: Based on the scheduling type, water conservancy project characteristic parameters, and current project operation status and hydrological situation, the corresponding constraints are automatically retrieved from the preset constraint knowledge base to form a scenario constraint set; The system receives the scheduling target description text input by the dispatcher through the human-computer interaction interface, and inputs the description text, the scenario constraint set, and preset water conservancy domain knowledge into the big language model. Receive one or more candidate scheduling schemes generated by a large language model, which include time-segmented outflow, reservoir water level control strategies, downstream river flow evolution and inundation risk analysis; The system receives revision opinions from experts via an interactive interface for the candidate scheduling scheme, transforms these revision opinions into new constraints, and drives the large language model to generate a revised scheduling scheme again. After responding to the expert confirmation signal, the system outputs the final scheduling decision scheme.
2. The method according to claim 1, characterized in that, Before automatically retrieving the corresponding constraints from a preset constraint knowledge base based on the scheduling type, water conservancy project characteristic parameters, and current project operation status and hydrological conditions, the process also includes: Access to real-time weather forecasts, engineering condition data, and hydrological monitoring data; Based on the real-time weather forecast, engineering condition data, and hydrological monitoring data, the scheduling type, water conservancy project characteristic parameters, and current project operation status and hydrological situation are determined.
3. The method according to claim 1, characterized in that, Before inputting the description text, the scenario constraint set, and the preset water conservancy domain knowledge into the large language model, the following steps are also included: Acquire knowledge from multiple sources, including watershed planning, scheduling procedures, preset historical typical flood events, and expert experience. Perform structured processing on the knowledge in these domains to construct a knowledge graph for water conservancy scheduling. Based on the scheduling type, water conservancy project characteristic parameters, current project operation status, and hydrological situation, relevant entities are matched in the knowledge graph; Starting with the matched entity, a graph traversal algorithm is used to retrieve related knowledge. The retrieved related knowledge is then converted into natural language descriptions to form knowledge prompt text in the field of water conservancy. This knowledge prompt text in the field of water conservancy is then input into a large language model.
4. The method according to claim 1, characterized in that, After receiving one or more candidate scheduling schemes generated by the large language model, which include time-segmented outflow, reservoir water level control strategies, and downstream river flow evolution and inundation risk analysis, the process further includes: The constraint knowledge base is invoked to perform multi-dimensional constraint verification on the candidate scheduling schemes, thereby obtaining candidate scheduling schemes that satisfy the constraints. Based on the candidate scheduling schemes that satisfy the constraints, and the multi-objective optimization strategy in the domain knowledge, a differentiated scheduling scheme including flood control priority, power generation priority and ecological priority is generated.
5. The method according to claim 1, characterized in that, Transforming the aforementioned revisions into new constraints includes: The revised opinions are semantically parsed to extract the constraint object, constraint type, applicable time period, and threshold parameter to obtain structured constraint items; A consistency check is performed on the structured constraint items. If there is a conflict with the scenario constraint set, the conflict is resolved or the conflict item is marked according to the preset priority rules. Write the structured constraints after the consistency check into the scenario constraint set.
6. The method according to claim 1, characterized in that, Also includes: After the scheduling plan is implemented, actual hydrological monitoring data, engineering operation data, and scheduling effect data are collected and compared with the expected process in the plan generation stage to obtain the deviation of key indicators. Based on the deviation of the key indicators, the parameters and domain knowledge rules in the constraint knowledge base are dynamically updated.
7. The method according to claim 1, characterized in that, Also includes: When the scheduling type is joint scheduling of reservoir groups, constraints on the joint action of multiple reservoirs, constraints on inter-basin water transfer, and constraints on multi-objective benefits are constructed. The constraints of the combined effect of multiple reservoirs, the constraints of inter-basin water transfer, and the constraints of multi-objective benefits are added to the scenario constraint set. Receive the output data of the pre-trained joint scheduling optimization model and the multi-node hydrodynamic model, and construct model prediction prompt information based on the output data; The description text, the set of scene constraints, the preset knowledge of the water conservancy field, and the model prediction prompts are input into the large language model.
8. The method according to claim 7, characterized in that, Before receiving the output data from the pre-trained joint scheduling optimization model and the multi-node hydrodynamic model, the following steps are also included: Real-time weather data, engineering condition data, and hydrological monitoring data of each reservoir are input into a pre-trained joint scheduling optimization model, which outputs time-segmented scheduling strategies for each reservoir. Real-time weather data, engineering condition data, and hydrological monitoring data of each reservoir are input into a pre-trained multi-node hydrodynamic model to obtain simulation data of the future outflow water level and downstream water level changes of each reservoir.
9. A reservoir scheduling decision-making device based on constraint perception, characterized in that, include: The constraint construction module is used to automatically retrieve corresponding constraints from a preset constraint knowledge base based on the scheduling type, water conservancy project characteristic parameters, and the current project operation status and hydrological situation, and form a scenario constraint set. The large language model human-computer interaction module is used to receive the scheduling target description text input by the dispatcher through the human-computer interaction interface, and input the description text, the scenario constraint set and the preset water conservancy domain knowledge into the large language model. The constraint-aware decision generation module is used to receive one or more candidate scheduling schemes generated by the large language model, which include time-segmented outflow, reservoir water level control strategies, downstream river flow evolution and inundation risk analysis. The expert collaboration module is used to receive the revision opinions of experts on the candidate scheduling scheme through the interactive interface, transform the revision opinions into new constraints, drive the large language model again to generate the revised scheduling scheme, and output the final scheduling decision scheme after responding to the expert confirmation signal.
10. An electronic device, characterized in that, It includes a processor and a memory storing program instructions, the processor being configured to execute, when executing the program instructions, the constraint-aware reservoir scheduling decision method as described in any one of claims 1 to 8.