A Physics AI Engine-Based Intelligent Decision-Making System and Method for Nuclear Power Plant Water Intake and Discharge

The intelligent decision-making system based on a physics AI engine solves the problem of real-time perception and assessment of temperature changes in the nuclear power plant's intake and discharge systems, enabling refined control of the operation of nuclear power plant intake and discharge outlets, improving safety and economy, and enhancing environmental sustainability.

CN123069950APending Publication Date: 2026-11-10HANGZHOU MOSI INTELLIGENT TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611168518.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-03
Publication Date
2026-11-10

Smart Images

  • Figure CN123069950A_ABST
    Figure CN123069950A_ABST
Patent Text Reader

Abstract

This invention discloses a nuclear power plant water intake and discharge intelligent decision-making system and method based on a physics AI engine, relating to the field of nuclear power plant marine water intake and discharge thermal emission technology. It includes a multi-source sensing module, an intelligent simulation module, a model calibration module, a strategy optimization module, and a closed-loop evolution module. Addressing the challenges in nuclear power plant water intake and discharge operation, such as the difficulty in timely identification of water intake temperature rise risks, accurate prediction of the discharge temperature rise diffusion range, difficulty in quantifying the impact of temperature rise, and the lack of safety, ecological, and economic synergistic optimization in scheduling schemes, this invention develops a nuclear power plant water intake and discharge intelligent decision-making system and method based on a physics-sensing AI agent. It integrates multi-source physical monitoring, mechanism numerical simulation, historical operating condition calibration, and intelligent optimization decision-making technologies to establish an intelligent management system for nuclear power plant water intake and discharge. This system can provide refined control schemes under different sea conditions and operating conditions, improving water intake safety assurance capabilities and discharge compliance control capabilities, and reducing the risk of localized overheating.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of nuclear power plant marine water intake and discharge heat emission technology, specifically to a nuclear power plant water intake and discharge intelligent decision-making system and method based on a physics AI engine. Background Technology

[0002] The intake and drainage system of a nuclear power plant plays a crucial role in the safe and stable operation of the unit. In particular, when seawater or mixed river and sea water is used as a cold source, the temperature of the water at the intake, the temperature rise at the discharge outlet, and the temperature diffusion in the nearshore sea area will directly affect the cooling efficiency, the heat exchange effect of the condenser, and the compliance of emissions.

[0003] Existing technologies for managing the impact of water intake and drainage temperatures still have the following problems:

[0004] First, there is a lack of real-time perception and continuous prediction capabilities for changes in the marine temperature field, making it difficult to detect high-temperature sea conditions, weak exchange sea conditions, or temperature accumulation effects in a timely manner.

[0005] Secondly, there is a lack of unified modeling for local temperature rise at water intake and the impact range of drainage, making it impossible to accurately assess the impact of increased water intake temperature on cold source efficiency.

[0006] Third, the existing scheduling methods rely heavily on experience and cannot make dynamic decisions based on future changes in ocean temperature, tidal exchange conditions and unit load changes.

[0007] Fourth, there is a lack of intelligent optimization methods oriented towards temperature constraints, making it difficult to balance safety, economy, and environmental protection requirements and carry out collaborative optimization.

[0008] Therefore, a physical perception AI intelligent agent system is needed for temperature-affected scenarios to realize the perception, prediction, optimization, and closed-loop decision-making of temperature changes at water intakes and in the sea area. Summary of the Invention

[0009] This invention provides a nuclear power plant water intake and discharge intelligent decision-making system and method based on a physics AI engine, which can effectively solve the problems mentioned in the background art.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a nuclear power plant water intake and discharge intelligent decision-making system based on a physics AI engine. This system, based on multi-source monitoring, mechanism simulation, historical calibration, and optimization decision-making technologies, constructs an intelligent decision-making system according to a technology chain of perception, modeling, identification, optimization, and execution, and includes the following modules:

[0011] Multi-source sensing module, intelligent simulation module, model calibration module, strategy optimization module, and closed-loop evolution module;

[0012] The multi-source sensing module provides basic input for subsequent simulation, calibration and optimization; the intelligent simulation module constructs a physical sensing AI agent; the model calibration module provides a high-precision simulation base for subsequent risk identification and optimization decision-making; the strategy optimization module realizes joint optimization; and the closed-loop evolution module realizes the continuous self-evolution of the water intake and drainage agent.

[0013] According to the above technical solution, the multi-source sensing module is designed for real-time thermal environment monitoring of the sea area where nuclear power plants take off and discharge water, and is uniformly connected to satellite remote sensing temperature field, nearshore in-situ hydrological observation, plant circulating water system operating conditions and meteorological data;

[0014] The intelligent simulation module is based on the high-precision flow field, temperature field, and key point state sequence output by the model calibration module to construct a physical perception AI intelligent agent for nuclear power plant water intake and drainage scenarios.

[0015] According to the above technical solution, the model calibration module uses historical operating condition data, real-time monitoring data and intelligent simulation module output results to perform inversion calibration on the drag coefficient, thermal diffusivity coefficient and local mixing coefficient of the ocean flow field;

[0016] The strategy optimization module takes the risk probability, comprehensive risk index, future temperature sequence, and high-risk period output by the intelligent prediction module as input.

[0017] According to the above technical solution, the closed-loop evolution module receives the recommended and alternative solutions output by the strategy optimization module, generates field-executable instructions and sends them to the operators and control system, and continuously tracks the environmental feedback and equipment response data after execution to verify and evaluate the effectiveness of the solution.

[0018] According to the above technical solution, a smart decision-making method for nuclear power plant water intake and discharge based on a physics AI engine includes the following steps:

[0019] S1: Real-time acquisition of multi-source heterogeneous data, including satellite remote sensing temperature field, in-situ hydrological observation and nuclear power operation status, through the integrated air-ground-water monitoring network; use spatiotemporal registration algorithm to perform multi-dimensional feature alignment and standard optimization preprocessing to build a basic physical data environment for water intake and drainage intelligent agents;

[0020] S2: Embed physical constraints into the numerical simulation model, and achieve deep integration of mechanism-driven and data features through an industrial physics AI engine to simulate the impact of nuclear power plant water intake and discharge in complex marine environments;

[0021] S3: Enhance the physical perception capabilities of AI agents, use real-time monitoring data from near and far sea areas to drive adaptive updates of simulation results, and calibrate key parameters, including flow field resistance and thermal diffusivity, through a backpropagation mechanism.

[0022] S4: Based on the high-precision simulated flow field constructed in step S3, establish an evaluation system for the influence factor of temperature rise intensity, conduct a pre-simulation of intake and drainage conditions, and use an optimization algorithm to search for and obtain the optimal intake and drainage decision strategy.

[0023] S5: The intake and drainage intelligent agent automatically generates and outputs the optimal operation instructions to guide the adjustment of the actual working conditions, continuously tracks the environmental feedback data after execution to verify the effect, and stores the experience in the closed-loop process into the expert knowledge base to realize the continuous self-evolution of the system's decision-making capabilities.

[0024] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0025] 1. To address the challenges in timely identification of intake water temperature rise risks, accurate prediction of the temperature rise diffusion range, quantitative assessment of temperature rise impact, and lack of safety, ecological, and economic synergy optimization in dispatching schemes during nuclear power plant intake and discharge operations, this paper develops a nuclear power plant intake and discharge intelligent decision-making system and method based on a physical perception AI agent. This method integrates multi-source physical monitoring, mechanism numerical simulation, historical operating condition calibration, and intelligent optimization decision-making technologies to establish an intelligent management system for nuclear power plant intake and discharge. It can provide refined control schemes under different sea conditions and operating conditions, improve intake water safety assurance capabilities and discharge compliance control capabilities, reduce the risk of local overheating, and be used for monitoring and predicting nearshore marine temperature fields, as well as evaluating and optimizing the operation schemes of nuclear power plant intake and discharge outlets.

[0026] Meanwhile, by visually predicting and demonstrating the effects of the marine temperature field, intake temperature, and thermal impact range under future operating conditions, a scientific basis is provided for the scheduling, operation, and environmental management of nuclear power plants.

[0027] 2. By integrating industrial operation data such as sea background temperature, intake temperature, forebay temperature, discharge temperature, tidal flow field, meteorological conditions, unit load, and circulating water flow, we can achieve collaborative perception of marine environmental information and plant operating conditions, improve the real-time characterization capability of intake and discharge thermal processes under complex marine conditions, establish marine hydrodynamic models, temperature transport and diffusion models, and local temperature response models at the intake, and calibrate key parameters of the models using historical operating data and historical measured data, so that the simulation results are more consistent with the actual marine thermal processes and improve the accuracy of temperature field simulation under complex operating conditions.

[0028] The calibrated mechanistic model output, real-time monitoring data, and historical operational data are input into the industrial physics AI engine to identify and predict the risks of water intake temperature rise, drainage temperature rise exceeding limits, and high temperature risks in ecologically sensitive areas. Taking into account the consistency of mechanisms, the accuracy of predictions, and the interpretability of engineering, the ecological thermal impact, water intake safety margin, marine area use costs, and operation and regulation costs are all incorporated into the temperature evaluation system. The number of water intake and drainage outlets, flow allocation, and combination of operation modes are used as optimization variables to search for the optimal water intake and drainage decision. Combined with regulations and experience, an executable scheduling scheme is output to form a closed-loop optimization mechanism.

[0029] 3. Based on multi-source physical monitoring data, a simulation model is established to perform marine hydrodynamic calculations on the nuclear power plant's water intake area, simulating the ocean current movement process under the influence of tides, waves, and wind. Simultaneously, a temperature diffusion module is coupled to calculate the sea temperature field under the waste heat discharge scheme at the discharge outlet. By leveraging AI intelligent agents to aggregate real-time monitoring data from near and far sea areas, the physical perception capability of the real sea area is enhanced, and key parameters of the simulation model are calibrated. Then, using the obtained high-precision simulated flow field, a temperature rise impact assessment system is established to evaluate the impact on marine ecological security and the water intake scheme at the intake outlet. Finally, combined with optimization algorithms, the optimal water intake and discharge operation scheme is searched to precisely control the water intake and discharge equipment, achieving real-time physical perception, accurate forecasting, and intelligent decision-making for nuclear power plant water intake and discharge scenarios, thereby improving the economic benefits and environmental sustainability of nuclear power plants.

[0030] In summary, based on multi-source monitoring, mechanism simulation, historical calibration, and optimization decision-making technologies, intelligent management of the temperature impact in nuclear power plant water intake and discharge scenarios can be achieved throughout the entire process. This is suitable for the construction of intelligent decision-making platforms and related digital twin systems for nuclear power plant water intake and discharge. The intelligent decision-making system is constructed according to the technology chain of perception, modeling, identification, optimization, and execution. The data transmission relationship is clear, the evaluation indicators are closely aligned with the business needs of nuclear power plant water intake and discharge, the system modules are relatively independent with low coupling, and it supports flexible configuration under different sea conditions and unit operating conditions. It has strong robustness and a wide range of applications. Attached Figure Description

[0031] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0032] In the attached diagram:

[0033] Figure 1 This is a schematic block diagram of the intelligent decision-making system of the present invention;

[0034] Figure 2 This is a schematic diagram of the 24-hour scheduling scheme for the water intake and discharge outlets of a nuclear power plant according to the present invention;

[0035] Figure 3This is a flowchart of the intelligent decision-making method of the present invention. Detailed Implementation

[0036] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0037] Example: Figure 1-2 As shown, this invention provides a technical solution: a nuclear power plant water intake and discharge intelligent decision-making system based on a physics AI engine. Based on multi-source monitoring, mechanism simulation, historical calibration, and optimization decision-making technologies, it constructs an intelligent decision-making system according to the technology chain of perception, modeling, identification, optimization, and execution, realizing intelligent management of the entire process of temperature impact in nuclear power plant water intake and discharge scenarios. The system includes the following modules:

[0038] Multi-source sensing module, intelligent simulation module, model calibration module, strategy optimization module, and closed-loop evolution module;

[0039] The multi-source sensing module provides basic input for subsequent simulation, calibration and optimization; the intelligent simulation module constructs a physical sensing AI agent; the model calibration module provides a high-precision simulation base for subsequent risk identification and optimization decision-making; the strategy optimization module realizes joint optimization; and the closed-loop evolution module realizes the continuous self-evolution of the water intake and drainage agent.

[0040] Based on the above technical solution, the multi-source sensing module is designed for real-time thermal environment monitoring of the nuclear power plant's intake and discharge sea areas. It integrates satellite remote sensing temperature field, nearshore in-situ hydrological observation, plant circulating water system operating conditions and meteorological data to build an integrated data chain covering "outer sea - nearshore - inlet - forebay - plant". It updates sea temperature, tide level, current velocity and direction, wind field, unit load and circulating water flow rate at three time scales of 6h, 12h and 24h, providing basic input for subsequent simulation, calibration and optimization.

[0041] Taking the intake and discharge system of a coastal nuclear power plant with two units as the object, the monitoring area covers 12km along the coast and 8km offshore, with a focus on the intake front 1.5km, the discharge outlet diffusion zone 2.0km, and the potential heat plume backflow channel between the two. The background temperature of the open sea is obtained by satellite remote sensing and nearshore buoys. The spatial resolution of the satellite sea surface temperature is 1km, with an update cycle of 6h. The update cycle of the surface temperature of the nearshore buoys is 10min, with a temperature accuracy of ±0.05℃. The satellite temperature field is used to characterize the background temperature distribution of the large-scale sea area, while the buoys and fixed stations are used to characterize the local temperature changes near the shore.

[0042] The monitoring deployment includes buoys, ADCP, fixed temperature points, and tide gauges. All data are uniformly registered in time and space after entering the module. The DCS data in the plant is updated at 10-second levels and averaged according to a 1-minute window. The ADCP data is aligned according to the 5-minute center time. Bilinear interpolation and deep hierarchical mapping methods are used to complete spatial matching.

[0043] Data quality is controlled by setting the effective temperature range to 0℃-45℃, the effective tide range to -2.5m-4.5m, and the effective flow velocity range to 0m / s-3.0m / s. A sudden alarm is triggered when the difference between two adjacent 1-minute temperature samples exceeds 1.2℃. When the distance between adjacent measuring points is less than 300m and the temperature difference exceeds 2.0℃, abnormalities are automatically marked and removed.

[0044] Based on the high-precision flow field, temperature field, and key point state sequence output by the model calibration module, the intelligent simulation module constructs a physical perception AI agent for nuclear power plant water intake and discharge scenarios. It identifies the risks of water intake temperature rise, water discharge temperature rise exceeding limits, heat plume backflow, and high temperature in ecologically sensitive areas, and predicts the temperature evolution trend in future periods. It outputs the key risk probability, comprehensive risk index, and temperature trend results for three time windows of 6 hours, 12 hours, and 24 hours, respectively, with a rolling step of 15 minutes.

[0045] The input to the intelligent agent consists of five categories of variables: the first category is the environmental state at the current moment, including background sea temperature, tide level, near-field surface velocity, near-field mainstream direction, and 10m wind speed; the second category is the plant operating conditions, including the total load of the two units, the total flow rate of circulating water, the outlet temperature, and the operating status of the intake and outlet; the third category is the model calibration results, including the predicted temperature of the intake, the predicted temperature of the forebay, the area of ​​2℃ temperature rise, and the intake water regeneration coefficient; the fourth category is the time series characteristics, including the change sequence of intake temperature, forebay temperature, and near-field velocity sampled every 15 minutes in the past 2 hours; and the fifth category is the diurnal periodic characteristics, including hourly, tidal phase, and day / night indicators.

[0046] A single input vector contains a total of 36 features, and the predicted output includes 4 types of risk probabilities, 1 comprehensive risk index, and the intake temperature and forebay temperature sequences for the next 24 time points.

[0047] Under typical operating conditions, the temperature trend forecast for the next 6 hours shows that the intake temperature will rise from the current 30.6℃ to 31.0℃, and the forebay temperature will rise from 30.9℃ to 31.4℃. The maximum temperature rise will occur at 19:15, which corresponds to 1.5 hours after high tide. The table below shows the forecast results for the next 4 key moments.

[0048] Table 1: Temperature Forecast for Key Future Moments

[0049]

[0050] In terms of output format, the intelligent simulation module provides three types of results to the next module: the first type is the 24-hour temperature prediction sequence of the water intake and forebay, the second type is the 24-hour comprehensive risk index sequence and its level, and the third type is the list of high-risk triggering periods, risk types and key influencing factors.

[0051] Specifically, the output results state that "there is a high risk of thermal plume backflow from 18:30 to 20:00. It is recommended to optimize the flow distribution of the drainage outlet and reduce the drainage temperature difference during this period." The above results serve as the direct input to the strategy optimization module, which is used to search for the optimal water intake and drainage strategy that takes into account safety, ecology, and economic goals.

[0052] Based on the above technical solution, the model calibration module uses historical operating data, real-time monitoring data and intelligent simulation module output results to perform inversion calibration on the drag coefficient, thermal diffusion coefficient and local mixing coefficient of the ocean flow field, so that the model calculation results are close to the real intake and discharge thermal process. The module takes the intake temperature, forebay temperature, near-zone temperature rise of the discharge outlet and thermal plume diffusion boundary as calibration objects, performs parameter updates every 6 hours and performs a full-day playback verification every 24 hours, providing a high-precision simulation base for subsequent risk identification and optimization decision-making.

[0053] Three consecutive days of summer high-temperature operation data were selected as the calibration sample. The sample included various typical operating conditions such as dual-unit full power operation, single-unit operation, low wind speed at night, strong sea breeze during the day, high tide backflow, and low tide outward expansion. The data used for calibration included intake temperature, forebay temperature, outlet temperature, near-zone surface velocity, tide level, dual-unit load, and total circulating water flow rate. Specifically, the measured range for intake temperature was 28.6℃-31.4℃, forebay temperature was 28.9℃-31.9℃, near-zone surface velocity was 0.08m / s-0.93m / s, tide level was -0.64m-2.31m, and total circulating water flow rate was 42.1m³ / s-87.6m³ / s. The key parameters adjusted during model calibration included the bottom friction coefficient. Horizontal thermal diffusivity Local mixing coefficient of drainage outlet The objective function for calibration is a weighted sum of the intake temperature error, the forebay temperature error, and the thermal plume area error:

[0054]

[0055] in, Let be the parameter vector to be calibrated. and The first Simulated and measured temperatures at the water intake points of each sample. and These are the simulated temperature and the measured temperature of the forebay, respectively. and These are the simulated and measured inversion values ​​of the isopleth envelope area at a temperature rise of 2℃, respectively, with weights set as follows: Prioritize ensuring the accuracy of temperature predictions for the water intake and forebay;

[0056] The parameter search employs a two-stage approach: a coarse search followed by local iterations. First, a coarse search is performed within the empirical interval to determine the minimum friction coefficient. Discretized in steps of 0.0002 between 0.0018 and 0.0035, the horizontal thermal diffusivity is... Discretize within the range of 6 m² / s to 30 m² / s with a step size of 2 m² / s, local mixing coefficient Discretize the parameter between 0.65 and 1.25 with a step size of 0.05, and then perform local iterations using a quasi-Newton method within the parameter neighborhood where the error is minimized. The parameter update expression is as follows:

[0057]

[0058] in, For the first The parameters for the next iteration. Step size factor This is an approximation of the Hessian matrix inverse. As the gradient of the objective function, the intelligent simulation module is called again after each iteration to calculate the 24-hour temperature field and calculate the new objective function value. When the objective function decreases by less than 1% for three consecutive iterations, the calibration cycle ends.

[0059]

[0060] The mean absolute error of the forebay temperature is:

[0061]

[0062] The relative error of the thermal plume area is:

[0063]

[0064] The initial parameters used before calibration are: After calibration, the parameters converge to Compared with the initial parameters, the simulated peak nearshore velocity is closer to the actual ADCP measurement, the temperature rise decay slope near the outlet is closer to the actual temperature chain profile, and the local heat accumulation phenomenon in the forebay is also significantly corrected. The error comparison results before and after calibration are shown in the table below.

[0065] Table 2 Comparison of errors before and after parameter calibration:

[0066]

[0067] The strategy optimization module takes the risk probability, comprehensive risk index, future temperature sequence and high-risk period output by the intelligent prediction module as input. Around the three objectives of water intake safety, ecological constraints and operation economy, it jointly optimizes the combination of water intake and drainage outlets, circulating water flow allocation and necessary load adjustment to generate the optimal water intake and drainage decision under the current operating conditions. The module solves once every 15 minutes, and the time domain of each optimization is the next 6 hours. It outputs 1 set of recommended solutions and 2 sets of alternative solutions.

[0068] Optimization variables include the number of water intakes in operation. Number of drainage outlets in operation Flow rate at each gate and Circulating water pump start-stop combination and unit load fine-tuning Under dual-unit operation conditions, the number of water intakes in operation is 2-4, and the number of drainage outlets in operation is 1-2. The flow rate of a single outlet is 8.0 m³ / s-24.0 m³ / s, and the single load adjustment range is limited to 0 MW-80 MW. The optimization objective function consists of ecological objectives, safety objectives, and economic objectives.

[0069]

[0070] in, For the decision variable vector, The weights for ecological, security, and economic objectives are respectively taken as follows: Prioritizing water safety, the ecological target is defined as the weighted value of the average temperature rise in ecologically sensitive areas and the area with a 2°C temperature rise over the next 6 hours:

[0071]

[0072] The safety target is defined as the weighted average of the highest temperature at the intake, the highest temperature in the forebay, and the peak value of the intake water regeneration coefficient:

[0073]

[0074] The economic objective consists of the cost of marine thermal impact, pump set regulation cost, and load loss cost:

[0075]

[0076] in, The unit is thousands of yuan; The cost is 0.8 thousand yuan per pump start-up and shutdown, and 0.12 thousand yuan for each additional 1 m³ / s flow meter. Calculated at 0.35 thousand yuan per 1MW load reduction for 1 hour, optimized time period For the next 6 hours, there are 24 discrete time points;

[0077] The optimization process also satisfies rigid constraints, with the intake temperature constraint being:

[0078]

[0079] The temperature constraint of the forebay is:

[0080]

[0081] The average temperature rise constraint for ecologically sensitive areas is:

[0082]

[0083] In addition, the difference between the total water intake flow and the total drainage flow shall not exceed 1.5%, the flow rate change rate of a single gate shall not exceed 2.5 m³ / s·15 min, and the single unit load reduction shall not exceed 80 MW. Candidate schemes that do not meet the constraints shall be directly eliminated in the search phase.

[0084] A two-stage optimization strategy was adopted. The first stage used a surrogate model to quickly screen 300 candidate schemes. The input of the surrogate model was the scheme variables and future tides, wind fields, and background sea surface temperature. The output was the highest temperature at the water intake and the area with a temperature rise of 2℃. The calculation time for a single scheme was 0.08s. The second stage used a calibrated high-precision simulation model to perform a precise evaluation on the top 20 candidate schemes. The NSGA-II algorithm was used for 40 iterations with a population size of 30. After testing, the total time for a single rolling optimization was 132s, which met the 15min scheduling cycle requirement.

[0085] The optimal solution is as follows: maintain dual-unit operation, reduce the load of Unit 1 by 40MW, keep Unit 2 unchanged, put all four water intakes into operation with flow rates of 20.8m³ / s, 21.1m³ / s, 21.0m³ / s and 22.0m³ / s respectively, put the two drainage outlets into operation simultaneously with flow rates of 48.6m³ / s and 35.4m³ / s, add one circulating water pump on the east side, and deflect the main axis of high-temperature drainage towards the southeast offshore. The comparison results between the optimal solution and the baseline solution are shown in the table below.

[0086] Table 3 Comparison of the effects of the optimal solution and the baseline solution:

[0087]

[0088] The optimal solution, through the combined effect of "dual-outlet diversion of drainage outlets + local load fine-tuning + uniform distribution of water intake outlets," resulted in a 0.5℃ decrease in the highest temperature at the water intake outlet and a 0.6℃ decrease in the highest temperature at the forebay within the next 6 hours. The peak regeneration coefficient decreased from 0.121 to 0.082, the peak comprehensive risk index decreased from 0.73 to 0.48, and the risk level decreased from "high risk" to "concern." Although the additional 6 hours of comprehensive cost amounted to 46,800 yuan, it significantly improved the safety margin of the cold source and kept the average temperature rise in the ecologically sensitive area below the constraint of 1.0℃. The evaluation results of the candidate solutions are shown in the table below.

[0089] Table 4: Optimization Results of Candidate Solutions

[0090]

[0091] Based on the above technical solution, the closed-loop evolution module receives the recommended and alternative solutions output by the strategy optimization module, generates on-site executable instructions and sends them to operators and control systems. At the same time, it continuously tracks the environmental feedback and equipment response data after execution, verifies and evaluates the effectiveness of the solution, and writes the current working conditions, execution actions and feedback results into the expert knowledge base to realize the continuous self-evolution of the water intake and drainage intelligent agent. The module tracks the execution process with a sampling period of 1 minute and completes closed-loop archiving and experience updates in units of 1 scheduling event.

[0092] The optimal solution output by the strategy optimization module is executed after being confirmed by the shift leader at 17:00. The system breaks down the solution into four types of on-site actions: the first type is the gate switching action, including opening the No. 2 drainage outlet and redistributing the flow of the four water intake outlets; the second type is the pump group action, including starting the circulating water pump on the east side; the third type is the load adjustment action, including reducing the active load of Unit 1 from 1040MW to 1000MW; and the fourth type is the monitoring action, including increasing the temperature monitoring frequency of the water intake outlet, forebay, and drainage outlet near the area. Sequential constraints are set between each action, requiring that the water intake redistribution can only be performed after the drainage outlet has been switched to prevent instantaneous hydraulic imbalance.

[0093] Actual results showed that Drainage outlet No. 2 was opened at 17:03, 3 minutes later than planned; Unit 1 load dropped to 1002MW at 17:16, 2MW higher than the target; flow rate adjustments at all four intakes were completed at 17:24, with individual intake flow rate deviations all less than 0.4m³ / s. The overall deviation was calculated as follows: If the value is within the preset allowable range of 0.15, it is determined to be a high-consistency execution.

[0094] After the scheme is implemented, the system continuously collects data on the intake temperature, forebay temperature, outlet temperature, near-zone flow velocity, and temperature rise in the sensitive area, and compares these data hourly with the predicted values ​​from the strategy optimization module. Let the actual feedback vector be:

[0095]

[0096] The corresponding prediction vector is:

[0097]

[0098] The effect deviation vector is then:

[0099]

[0100] Within 2 hours of the implementation of the plan, the maximum measured temperature at the water intake was 30.54℃, the predicted value was 30.50℃, and the deviation was 0.04℃; the maximum measured temperature in the forebay was 30.87℃, the predicted value was 30.80℃, and the deviation was 0.07℃; the maximum measured area of ​​temperature rise of 2℃ was 2.28km², the predicted value was 2.21km², and the deviation was 0.07km²; the peak measured value of the regeneration coefficient was 0.086, the predicted value was 0.082, and the deviation was 0.004. Overall, the results of the implementation of the plan are consistent with the predicted results.

[0101] To comprehensively evaluate the effectiveness of the plan, an effectiveness index is constructed:

[0102]

[0103] in, Rate the temperature control effectiveness. Scoring the effectiveness of ecological impact control. To perform consistency scoring, take ,in, Calculated based on the temperature decrease at the water intake and forebay. Calculated based on the average temperature rise in the sensitive area and the reduction in area due to a 2℃ temperature rise. According to movement deviation Reverse calculation shows that in this scheduling event, The final comprehensive score is obtained. It was judged as "excellent effect".

[0104] like Figure 3 As shown, a smart decision-making method for nuclear power plant water intake and discharge based on a physics AI engine includes the following steps:

[0105] S1: Real-time acquisition of multi-source heterogeneous data, including satellite remote sensing temperature field, in-situ hydrological observation and nuclear power operation status, through the integrated air-ground-water monitoring network; use spatiotemporal registration algorithm to perform multi-dimensional feature alignment and standard optimization preprocessing to build a basic physical data environment for water intake and drainage intelligent agents;

[0106] S2: Embed physical constraints into the numerical simulation model, and achieve deep integration of mechanism-driven and data features through an industrial physics AI engine to simulate the impact of nuclear power plant water intake and discharge in complex marine environments;

[0107] S3: Enhance the physical perception capabilities of AI agents, use real-time monitoring data from near and far sea areas to drive adaptive updates of simulation results, and calibrate key parameters, including flow field resistance and thermal diffusivity, through a backpropagation mechanism.

[0108] S4: Based on the high-precision simulated flow field constructed in step S3, establish an evaluation system for the influence factor of temperature rise intensity, conduct a pre-simulation of intake and drainage conditions, and use an optimization algorithm to search for and obtain the optimal intake and drainage decision strategy.

[0109] S5: The intake and drainage intelligent agent automatically generates and outputs the optimal operation instructions to guide the adjustment of the actual working conditions, continuously tracks the environmental feedback data after execution to verify the effect, and stores the experience in the closed-loop process into the expert knowledge base to realize the continuous self-evolution of the system's decision-making capabilities.

[0110] Based on the above technical solution, S1, the specific steps are as follows:

[0111] S11: Construct a sensing object and monitoring index system for nuclear power plant water intake and discharge scenarios. The sensing objects include sea background temperature, intake temperature, forebay temperature, discharge outlet temperature, discharge diffusion zone temperature, tide level, tidal phase, ocean current velocity, ocean current direction, wind speed, wind direction, air temperature, humidity, solar radiation, rainfall, unit load, circulating water flow, water intake and discharge outlet operating status, pump start-up and shutdown status, and valve opening. By uniformly collecting the above objects, a comprehensive perception of the nuclear power plant's marine thermal environment, cold source status, and heat discharge status can be achieved.

[0112] Among them, the sea background temperature is mainly used to characterize the sea background thermal state outside the influence of heat discharge; the intake temperature and forebay temperature are used to reflect the safety margin of the cold source; the discharge outlet temperature and the temperature of the discharge diffusion zone are used to characterize the local temperature rise distribution and heat plume evolution characteristics after the waste heat is discharged; the tide level, ocean current and wind field parameters are used to characterize the external dynamic boundary in the heat diffusion process; and the unit load, circulating water flow and pump valve status are used to characterize the heat source intensity and the changes in the intake and discharge operating conditions.

[0113] Based on this, a monitoring indicator system is further constructed, which is divided into environmental boundary indicators, water intake safety indicators, emission impact indicators, and operating condition indicators. Among them, environmental boundary indicators include marine background temperature, tide level, flow velocity and direction, wind speed and direction, and meteorological parameters; water intake safety indicators include water intake temperature, forebay temperature, water intake temperature rise amplitude, and temperature rise change rate; emission impact indicators include discharge outlet temperature, discharge temperature difference, heat plume diffusion range, and local overheating area; and operating condition indicators include unit load, circulating water flow, valve start / stop status, and pump and valve regulation status. Through the above hierarchical classification, subsequent model calls and decision analysis have a clear data organization structure.

[0114] S12: Set the sampling frequency, accuracy requirements, spatial coverage and update cycle for monitoring indicators, and establish data standards suitable for nuclear power plant water intake and discharge scenarios. Since different sensing objects have different requirements for real-time performance, differentiated configurations are made according to decision-making needs. The intake temperature, forebay temperature, discharge temperature, circulating water flow and unit load are sampled at a high frequency of seconds, while the background temperature of the sea area, meteorological conditions and information on large-scale ocean current fields are updated at a rate of minutes.

[0115] To ensure consistency in subsequent model calculations and multi-source analysis, it is necessary to establish unified dimensions, time bases, and coordinate bases for all types of data. Parameters should be standardized through unit unification, coordinate mapping, and standardization. for:

[0116]

[0117] in, Indicates the first Class data source at time Spatial location The original observations at that location, and They represent the first For data of this type, the mean and standard deviation in historical samples are used. For data that is not suitable for standard deviation normalization, min-max normalization is used.

[0118]

[0119] Furthermore, to achieve a unified representation of heterogeneous data, at the same time... The following spatiotemporal state vector is constructed for the marine water intake and discharge scenario. It is used to represent the comprehensive spatiotemporal state of the current marine environment, water intake and discharge status, and unit operating conditions, and serves as the standard input for subsequent models and decision-making modules. The spatiotemporal state vector is:

[0120]

[0121] in, Indicates time The unified state vector of multi-source sensing. Indicates the total number of connected metrics;

[0122] S13: Deploy multi-source heterogeneous monitoring equipment in nearshore waters, intakes, forebays, drainage outlets, and heat-affected zones to form a data acquisition network covering "sea area - intake - forebay - plant process system". The marine monitoring equipment includes sea surface temperature buoys, temperature chains, tide gauges, ADCP current velocity and direction meters, wave monitoring equipment, shore-based meteorological stations, and infrared thermometers. The plant process side accesses the unit load, circulating water flow, pump status, valve status, and drainage process parameters in real time through the data interfaces of the DCS system, SCADA system, environmental monitoring system, and operation management system.

[0123] For key locations such as water intake, forebay and drainage outlet, high-reliability online temperature sensors are prioritized for deployment, and redundant measuring points are configured according to site conditions to ensure that key decision variables are continuously observable. When a core measuring point fails, the redundant measuring point serves as an alternative data source to reduce the impact of single-point failure on subsequent decision-making links.

[0124] During implementation, marine forecast data, remote sensing sea surface temperature data, and historical marine observation databases can be further integrated to enhance the perception of large-scale marine background and future boundary conditions. By introducing data sources that combine the open ocean and near-shore, external and internal, and environmental and operational conditions, a more complete physical perception foundation can be provided for subsequent intake and drainage thermal environment modeling.

[0125] S14: Unify and standardize the access of multi-source heterogeneous data from different monitoring devices, business systems, and external platforms. Since the data usually differs in sampling frequency, timestamp, spatial coordinates, data format, and units of measurement, it is necessary to first perform data format conversion, interface protocol adaptation, time alignment, spatial coordinate mapping, and unit normalization to unify the time axis. and unified spatial grid As a benchmark for data reconstruction, the original discrete monitoring data Through time interpolation function and space mapping function To obtain the reconstructed values ​​under a unified grid and a unified time:

[0126]

[0127] in, One of the following methods is used: linear interpolation, spline interpolation, or sliding window interpolation. One of the following methods is adopted: nearest neighbor mapping, inverse distance weighted interpolation, and grid cell-based region weighted mapping. In this way, data from different measurement points and different frequencies are uniformly mapped to a standardized spatiotemporal framework that can be called by subsequent models.

[0128] After unified access is completed, the raw data also needs to be cleaned and preprocessed. Because marine monitoring equipment operates under conditions of high salinity, high humidity, strong corrosion, and complex sea conditions, it is prone to problems such as instantaneous jumps, signal drift, communication interruptions, random noise, and outliers. Therefore, outlier identification, noise filtering, missing value imputation, drift correction, and outlier diagnosis are required for time-series data. Anomalies can be identified in the following ways:

[0129]

[0130] in, The mean of the sliding window. The standard deviation of the sliding window. This is the anomaly detection threshold coefficient. When an observation exceeds the threshold, it is marked as an outlier and corrected or removed.

[0131] For missing values, time-nearest interpolation is used for estimation:

[0132]

[0133] in, and These represent the nearest valid observation times before and after the missing time, respectively. Through the above processing, the accuracy, continuity, and availability of monitoring data are significantly improved.

[0134] Establish a data quality assessment and perceived reliability assurance mechanism, specifically setting credibility labels and quality scores for various types of monitoring data. The scoring criteria include data completeness, continuity, fluctuation stability, communication status, equipment health status, and consistency with adjacent monitoring points. (The text then abruptly shifts to a different topic: "Setting the first...") Each sensor at time The perceived credibility is Then it can be expressed as:

[0135]

[0136] in, This represents the data integrity rate metric. Indicates stability index, Indicates the health status of the equipment. This indicates the consistency index of neighboring measuring points. Let be the weighting coefficient, and satisfy:

[0137]

[0138] When the system detects anomalies in key sensing data, it constructs alternative estimates based on a confidence-weighted approach, and then fuses these estimates with those from similar redundant measurement points. It can be represented as:

[0139]

[0140] in, To reduce the number of redundant measurement points participating in the fusion, the aforementioned quality assessment and credibility weighting mechanism is used to maintain data link stability under complex sea conditions and local sensor anomalies, thereby improving the robustness of subsequent numerical simulation and intelligent decision-making results.

[0141] S15: Extract key physical features reflecting the marine thermal state and heat diffusion behavior from preprocessed multi-source monitoring data, and construct an input organization method suitable for subsequent modeling and intelligent analysis. The physical features mainly include the intake temperature change rate. Background sea surface temperature gradient Local heat accumulation index The calculation formula is as follows:

[0142]

[0143]

[0144]

[0145] in, Indicates time The water intake temperature, As a reference temperature threshold, Used to characterize within a time window The degree of high temperature accumulation in local sea areas, based on the statistical summary of historical monitoring data, is a numerical expression of the physical processes in the actual sea area. On the one hand, it serves as the parameter input for the mechanism-driven numerical model, and on the other hand, it can also serve as an important input variable for the subsequent physical perception AI agent to carry out risk identification and trend prediction.

[0146] The processed data is further organized into a boundary condition input set, a local operating condition input set, and a state feature input set, among which the boundary condition input set... Local operating condition input set State feature input set It can be represented as:

[0147]

[0148]

[0149]

[0150] in, represents the tide level, Represents the ocean current vector. Represents the wind field vector. Represents a set of meteorological parameters; Indicates the unit load. Indicates the circulating water flow rate. and These represent the number of water intakes and the number of drainage outlets in operation, respectively. This represents a set of pump and valve states. Through this structured organization, multi-source sensing data can be efficiently transmitted to subsequent models and decision-making processes.

[0151] Based on the above technical solution, the specific steps of S2 are as follows:

[0152] S21: Establish a spatial geometric model and computational domain for the intake and discharge sea area of ​​the nuclear power plant. Based on the topography, water depth distribution, shoreline morphology, intake location, discharge location, dike structure, forebay layout, and boundary conditions of adjacent sea areas, construct an integrated spatial geometric model of the sea area, intake, and forebay. The spatial geometric model is used to describe the influence of water depth fluctuations, shoreline constraints, and artificial structures on the propagation of flow and temperature fields within the sea area.

[0153] The sea area is divided into far-field sea area, near-field sea area, local area of ​​intake and discharge outlet, and local area of ​​forebay. The far-field sea area is mainly used to describe tidal propagation and background flow field evolution, the near-field sea area is mainly used to describe heat plume diffusion and backflow behavior, and the intake and discharge outlet and forebay areas are used to finely characterize local temperature response and heat accumulation phenomena. In order to balance computational accuracy and computational efficiency, a hierarchical grid division method is adopted, with a relatively sparse grid for the far-field area and a denser grid for the areas near the intake, discharge outlet and forebay.

[0154] Let the entire computational domain be denoted as:

[0155]

[0156] in, Indicates the calculation area of ​​the far-field ocean. This indicates the near-field ocean calculation area. Indicates a localized area of ​​the water intake. Indicates a localized area of ​​the drain outlet. The forepool region is represented by the above-mentioned regional division, forming a spatial modeling foundation that balances large-scale environmental driving forces with fine-grained local responses. The grid nodes in the computational domain... Represented as:

[0157]

[0158] in, Indicates the first The position coordinates of each grid node This represents the total number of grid nodes. Based on the grid system, the background temperature, tide level, flow velocity and direction, wind field and operating condition data formed in S1 are mapped to the corresponding computing nodes, serving as the driving input for the subsequent mechanism model.

[0159] S22: Construct a marine hydrodynamic numerical model to describe the seawater movement process under the combined action of tides, ocean currents, wind-driven currents, and local structural boundaries. The hydrodynamic numerical model is established based on the mass conservation equation and the momentum conservation equation. By solving the changes in free liquid surface, velocity field distribution, and flow direction, the time-varying flow field under complex marine environment is obtained.

[0160] Under the two-dimensional shallow water approximation, the continuity equation can be expressed as:

[0161]

[0162] in, Indicates the elevation of the free surface. Indicates the total water depth. It indicates that still water is deep. and They represent direction and The momentum equation corresponding to the depth-averaged velocity component in the direction can be expressed as:

[0163]

[0164]

[0165] in, Represents gravitational acceleration. Represents the Coriolis parameter. and These represent the wind stress terms, and , respectively, represent the bottom friction term. The above equation can characterize the influence of tidal level fluctuations, flow field evolution, and external dynamic forcing on the flow state of the sea area.

[0166] Regarding boundary condition settings, the open boundary in the open sea is input with tidal level process, current velocity process, and tidal harmonic constant; the land boundary adopts an impenetrable boundary condition; and the intake and discharge outlet boundary can be set with flow boundary parameters combined with the circulating water system operating parameters. Let the discharge outlet boundary flow rate be... The boundary flow rate of the water intake is Then the local boundary normal velocity can be expressed as:

[0167]

[0168] in, Indicates the cross-sectional area of ​​the outlet. This represents the boundary flow at the corresponding moment. In this way, the actual intake and discharge conditions of the nuclear power plant are embedded into the marine flow field calculation process.

[0169] S23: Based on the hydrodynamic model, a numerical model of temperature transport and heat diffusion is constructed to simulate the evolution of the marine temperature field after waste heat discharge. The temperature transport model describes the migration, mixing, diffusion and attenuation of waste heat in the marine area through the coupling of convection, diffusion and source terms.

[0170] The temperature transport control equation can be expressed as:

[0171]

[0172] in, Indicates water temperature, and These represent the velocity components obtained from the hydrodynamic model. and They represent direction and Equivalent diffusion coefficient in the direction, The temperature source term is used to describe the heat input from the drain outlet and necessary heat exchange correction terms. When the drain outlet is used as a local heat source boundary, its heat flux can be expressed as:

[0173]

[0174] in, Indicates the density of seawater. This indicates the specific heat capacity of seawater. Indicates drainage flow rate. Indicates the temperature of the drain outlet. This represents the background sea surface temperature. This heat flux determines the intensity of residual heat entering the sea area under drainage conditions. To characterize the diffusion range of the heat plume and the degree of temperature rise impact, the temperature rise field can also be defined as:

[0175]

[0176] in, Indicates spatial location place, time The temperature rise value is obtained by calculating the temperature rise field to obtain the changes in the position of the heat plume axis, the distribution of local high temperature areas, and the range of thermal influence under different working conditions;

[0177] The background sea surface temperature, drainage temperature, drainage flow rate, wind field, and tide level boundary inputs in the temperature transport model all come from the multi-source sensing dataset formed in step S1, thereby achieving deep coupling between marine feature data and the mechanism model, enabling the numerical model to more realistically represent the actual marine thermal processes.

[0178] S24: Construct a numerical model of the local temperature response of the intake and forebay to refine the simulation of key phenomena such as local heat accumulation and intake water temperature rise. Since the intake and forebay areas are typically affected by local structural boundaries, gate arrangement, flow distribution methods, and tidal current direction, using only a sea-scale model for coarse-grained simulation is insufficient to accurately reflect the detailed temperature changes in the cold source area. Therefore, based on the sea-scale model, a local response model of the intake and forebay is further constructed. This local response model represents the temperature change near the intake as a coupled result of the sea temperature field, drainage boundary conditions, and local hydraulic conditions. Let the local temperature at the intake be... Then it can be expressed as:

[0179]

[0180] in, Indicates the location of the water intake. Indicates water intake flow rate. Indicates drainage flow rate. This represents the local velocity vector near the water intake. represents the tide level, This indicates the local temperature response relationship;

[0181] The local temperature of the forecourt is further expressed as:

[0182]

[0183] in, Indicates the forebay control volume. The average temperature of the forebay is represented by the volume average method described above. The overall heat accumulation degree and temperature rise response trend of the forebay are calculated.

[0184] To identify the reflux behavior of hot plumes, a water intake reflux coefficient can also be defined:

[0185]

[0186] in, This indicates the relative extent to which emitted heat re-affects the water intake, when When the value increases, it indicates that the impact of the drainage heat plume on the backflow of the water intake area is enhanced. The local response model provides key numerical support for subsequent heat plume backflow identification, water intake safety analysis and optimized scheduling.

[0187] S25: Construct a multi-condition numerical simulation result library to form a basic sample set covering different sea states, different operating conditions and different intake and discharge combinations. The multi-condition numerical simulation result library is not a parameter calibration result library, but a set of numerical results obtained by batch simulation under preset typical conditions based on the mechanism model established in S21-S24. It is used to form a numerical characterization basis for the thermal impact of nuclear power intake and discharge.

[0188] The preset operating conditions include different seasonal background sea surface temperature conditions, different tidal types, different tidal levels, different wind speeds and directions, different unit load levels, different combinations of circulating water flow rates, different numbers of water intakes in operation, and different numbers of drainage outlets in operation. For each operating condition, the numerical model outputs the corresponding flow field results and temperature field results, forming corresponding data and storing them in the database. By systematically combining and batch simulating typical operating conditions, a numerical simulation result library covering multiple seasons, multiple tides, multiple loads, and multiple outlet combinations can be formed. This result library is used for parameter calibration and model calibration under subsequent historical operating condition data, and can also serve as training samples for physical perception AI intelligent agents, rapid scheme retrieval samples, and initial samples for optimization calculations, thereby improving the efficiency and accuracy of subsequent risk identification, trend prediction, and scheduling decisions.

[0189] Based on the above technical solution, the specific steps of S3 are as follows:

[0190] S31: Construct a historical operating condition dataset, a historical measured dataset, and a historical simulation comparison sample set. The historical operating condition dataset includes operational data such as unit load, circulating water flow, intake operation status, discharge outlet operation status, pump start / stop status, valve opening, discharge temperature, and discharge flow during historical periods. The historical measured dataset includes background sea temperature, intake temperature, forebay temperature, discharge outlet temperature, tide level, flow velocity and direction, wind speed and direction, and other environmental and meteorological data during the same period. By performing temporal alignment, spatial matching, and unified coding on the two types of data, an integrated sample foundation of historical operating conditions, historical environment, and historical temperature response is formed.

[0191] S32: Based on historical sample sets, parameter calibration and verification of the mechanistic numerical model are carried out. The historical operating condition inputs and environmental boundary inputs constructed in S31 are substituted into the mechanistic-driven numerical model in S2 to obtain the simulated flow field and simulated temperature field for the corresponding historical time period. These are then compared and analyzed with historical measured results for the same period. By adjusting key parameters in the model, the model output is made to approximate historical measured data, improving the model's ability to reproduce real marine thermal processes. Let the numerical model parameter vector be:

[0192]

[0193] in, This indicates the parameters related to bottom friction. and Represents the equivalent diffusion coefficient. Indicates heat exchange parameters, The input represents the local mixing parameters, and the remaining terms represent other parameters to be calibrated. After substituting into the numerical model, the simulated response can be obtained:

[0194]

[0195] in, This indicates that the mechanism-driven numerical model established by S2, in order to make the simulation results fit the historical measured results, constructs the objective function as follows:

[0196]

[0197] in, Indicates the first Simulated temperature at the water inlet of each sample. This indicates the simulated temperature of the forebay. This represents a characteristic quantity representing the simulated thermal influence range. Assuming the weights of each error term, the parameter calibration process is transformed into the following optimization problem:

[0198]

[0199] The optimal parameter set after calibration is obtained by using least squares optimization. The calibrated model can more accurately characterize the ocean current field, temperature field, and local temperature rise process under historical conditions, providing a highly reliable simulation basis for subsequent online prediction and scheme optimization. The root mean square error (RMSE) is further used to evaluate the calibration effect. Let the RMSE of the intake temperature be:

[0200]

[0201] When the calibrated error index meets the preset threshold requirements, the numerical model of the determination mechanism is deemed to be ready for subsequent AI agent construction and online application.

[0202] S33: Construct the overall architecture of the physical perception AI agent, couple the calibrated mechanism model with the data-driven model, so that the AI ​​agent has both physical consistency and data adaptability. The physical perception AI agent takes real-time monitoring data, historical operating condition data, calibrated numerical simulation results and short-term environmental forecast information as joint inputs. By embedding the constraint of marine heat transport mechanism, it improves the ability to identify and predict temperature risks under complex sea conditions. The feature set of simulation results includes the current flow field distribution, temperature field distribution, temperature rise distribution, intake response value and forebay response value.

[0203] Structurally, the physical perception AI agent adopts a layered architecture of mechanism model layer, feature fusion layer, risk identification layer, and trend prediction layer. The mechanism model layer is responsible for providing physically consistent prior output, the feature fusion layer is responsible for jointly encoding the monitoring data features and mechanism simulation features, the risk identification layer is responsible for outputting the risk category and risk level, and the trend prediction layer is responsible for outputting the changing trend of key temperature indicators within the future time window. This architecture can avoid the problem of pure black box models deviating significantly from physical laws in extrapolation scenarios.

[0204] S34: Construct a temperature risk identification model to classify and identify risks of water intake temperature rise, excessive discharge temperature rise, thermal plume backflow, and ecological thermal impact. The risk identification model is based on supervised learning and its goal is to map the current state of the sea area to the corresponding risk label.

[0205] The temperature risk identification model is used to output the probability of occurrence of various risks based on the current system state input, and further form a comprehensive risk index, which serves as a unified quantitative basis for subsequent early warning triggering and optimization decisions. Specifically, it first determines the probability of occurrence based on the time... The system state input constructs a risk probability output vector:

[0206]

[0207] The risk probability output vector can be represented as:

[0208]

[0209] Based on this, the probabilities of various risks are weighted and summed according to preset weights to obtain a comprehensive risk index:

[0210]

[0211] in, Indicates time System status input, This represents a risk identification model. Indicates time The risk probability output vector, Indicates the first Risks at all times The probability of occurrence, Indicates the total number of risk types. This represents the overall risk index. Indicates the first The weighting coefficients for risk categories include the risk of water intake temperature rise, the risk of excessive temperature rise in drainage, the risk of thermal plume backflow, and the risk of high temperature in ecologically sensitive areas.

[0212] S35: Construct a temperature trend prediction and early warning model to predict the evolution of key temperature indicators within a preset time window and output risk warning results. The prediction objects include water intake temperature, forebay temperature, drainage heat plume diffusion range, location of local high temperature zone and temperature rise level of ecologically sensitive area. The prediction model integrates time series model and introduces the output of mechanism model as prior constraint, so as to take into account both time series fitting ability and physical rationality.

[0213] Let the prediction time window length be... Then the predicted output for the future time period can be expressed as:

[0214]

[0215] in, This represents a trend prediction model. This indicates a review of historical timelines. Indicates from arrive The prediction sequence of the forecast results at each time point, and the prediction of the future temperature sequence at the water intake, can be expressed as:

[0216]

[0217] To enhance the physical consistency of the prediction results, a physical constraint loss term is introduced during the training process, assuming the data fitting loss is... The physical constraint loss is The total loss function can then be expressed as:

[0218]

[0219] in, Indicates the physical constraint weight coefficient. This is used to constrain the prediction results to meet the requirements of temperature continuity, smoothness of change, and consistency with the output of the mechanistic model;

[0220] After obtaining the forecast results for the future period, an early warning is triggered based on a preset threshold. Let the predicted value for a certain type of risk indicator be... The warning threshold is The warning trigger condition can be expressed as:

[0221]

[0222] When the above conditions are met, the system automatically generates a warning signal of the corresponding level and outputs the risk type, expected time of occurrence, scope of impact and key causes to the subsequent optimization decision module. Through the processing of S31-S35, the system realizes the integrated construction from historical working condition matching and mechanism model calibration to AI risk identification and trend prediction, providing highly reliable input for subsequent optimization scheduling that takes into account safety, ecology and economic goals.

[0223] Based on the above technical solution, the specific steps of S4 are as follows:

[0224] S41: Construct a temperature impact assessment system for nuclear power plant water intake and discharge scenarios. The temperature assessment system comprehensively evaluates different water intake and discharge schemes from ecological, safety, and economic dimensions to achieve quantifiable, comparable, and optimizable thermal impacts during nuclear power plant water intake and discharge processes. The ecological dimension is mainly used to evaluate the impact of discharge temperature rise on marine ecological environment and aquatic organism habitat conditions. The safety dimension is mainly used to evaluate the impact of intake temperature, forebay temperature, and heat plume return flow on the safe operation of cold source. The economic dimension is mainly used to evaluate the comprehensive costs of different water intake and discharge regulation schemes in terms of marine use, operation regulation, and power generation load changes.

[0225] From an ecological perspective, the ecological thermal impact assessment is expressed as indicators including the area affected by high temperatures, the degree of temperature exceedance in sensitive areas, and the duration of high temperatures. A certain temperature rise threshold is set as... Then at time The area affected by the ecological high temperature can be expressed as:

[0226]

[0227] in, Represents the characteristic function, Indicates spatial location The temperature rise value at a certain location is used to measure the spatial extent of high-temperature sea areas caused by heat dissipation.

[0228] If there are ecologically sensitive areas in the sea area The average temperature rise in the sensitive area can then be expressed as:

[0229]

[0230] in, This indicates the area of ​​ecologically sensitive regions, and this indicator is used to characterize the degree of direct impact of thermal emissions on key ecological protection areas.

[0231] From a safety perspective, the evaluation focuses on the intake temperature, forebay temperature, and intake temperature rise margin, with a safety upper limit set at [value missing]. The water safety margin can then be expressed as:

[0232]

[0233] in, The smaller the value, the higher the risk to water intake safety. Similarly, the safety margin of the forebay can be expressed as:

[0234]

[0235] in, This indicates the upper limit of the allowable temperature of the forebay. At the same time, the water intake heat recovery coefficient defined in S24 can still be used. This is used to measure the impact of hot plume recirculation on the water intake system;

[0236] From an economic perspective, it is necessary to comprehensively consider the marine area usage costs, operational adjustment costs, and power generation load loss costs arising from different water intake and drainage schemes. Marine area usage costs can be understood as the cost of occupying marine space due to the expansion of the affected marine area caused by water intake and drainage. Especially under certain management models, the larger the area affected by heat discharge, the higher the corresponding marine area usage cost. Let the unit marine area usage cost coefficient be... The heat-affected area is The cost of using the sea area can then be expressed as:

[0237]

[0238] Operating adjustment costs can be expressed as the additional energy consumption and equipment adjustment costs caused by different pump set combinations, flow regulation, and switching actions. If the... The cost coefficient corresponding to this type of adjustment action is The amount of movement is The operating adjustment cost can then be expressed as:

[0239]

[0240] in, This represents the total number of adjustment actions. If the plan involves reducing the load on generating units, the economic loss from power generation must also be taken into account. Let the load offset be... The unit load loss cost coefficient is Then the cost of load loss can be expressed as:

[0241]

[0242] This results in the total economic cost:

[0243]

[0244] The above three categories of indicators—ecological, safety, and economic—form a temperature impact assessment system for comparing multiple scenarios, providing a quantitative basis for the subsequent construction of the objective function for optimization.

[0245] S42: Define the optimization variables and feasible solution space of the water intake and drainage scheme. The optimization variables are used to characterize the operation combination mode of the water intake and drainage under different scheduling schemes. They are the basis of the independent variables for multi-objective optimization solutions. The optimization variables include the number of water intakes in operation, the number of drainage outlets in operation, the flow distribution of each outlet, the total flow of circulating water, the pump operation combination, the discharge sequence, and the unit load adjustment when necessary.

[0246] Let the decision vector for water intake and drainage schemes be:

[0247]

[0248] in, Indicates the number of water intakes in operation. Indicates the number of drainage outlets in operation. Indicates the first The flow rate at each water intake point Indicates the first Flow rate of each drain outlet Indicates the pump set operating combination status. It represents the unit load adjustment amount. The variables can include both continuous and discrete variables, so it belongs to the set of mixed decision variables.

[0249] To ensure the optimization results meet the requirements for project feasibility, constraints need to be set for the above decision variables. Specifically, the total water intake flow and the total drainage flow must satisfy the system equilibrium relationship:

[0250]

[0251] At the same time, the flow rate at each gate must meet the equipment capacity boundary constraints:

[0252]

[0253]

[0254] The pump unit status, the number of valves in operation, and the load adjustment must also meet the unit operation procedures and equipment interlocking logic constraints, thus forming the feasible solution space for the water intake and drainage scheme:

[0255]

[0256] This feasible solution space provides a constraint range for subsequent multi-objective optimization searches;

[0257] S43: Construct a multi-objective optimization function that takes into account ecological, safety and economic goals, and embed key indicators in the temperature evaluation system into the optimization objectives. The optimization objectives are not simply to minimize a certain temperature value, but to comprehensively balance the three directions of minimizing ecological impact, minimizing economic cost and maximizing operational safety margin under the premise of meeting water intake safety and environmental protection boundary conditions.

[0258] The ecological objective function can be expressed as:

[0259]

[0260] in, Indicates the evaluation time window, Represents a given scheme The area affected by the high temperatures is [not specified]. This indicates the average temperature rise in the sensitive area. and These are the weighting coefficients; the objective function is used to measure the degree of adverse impact of the plan on the ecological environment.

[0261] The security objective function can be expressed as:

[0262]

[0263] in, This represents the water intake temperature under a given scheme. Indicates the temperature of the forebay. This represents the hot plume recirculation coefficient, and the objective function reflects the degree of impact of different schemes on the safety of the cold source.

[0264] The economic objective function can be expressed as:

[0265]

[0266] Each component corresponds to the cost of sea area use, the cost of operation and adjustment, and the cost of load loss, respectively. A single optimization objective is constructed using a weighted average approach.

[0267]

[0268] in, Let be the weight coefficient of the overall objective, and satisfy:

[0269]

[0270] Pareto front approach is used for multi-objective collaborative optimization to preserve the non-dominated solution set among different objectives, allowing operators to choose according to their preferences;

[0271] In addition to the objective function, rigid constraints must also be set, including ensuring that the water intake temperature does not exceed the safe upper limit:

[0272]

[0273] The temperature of the forebay must not exceed the upper limit of permissible limits:

[0274]

[0275] Temperature rise in sensitive areas must not exceed ecological constraint thresholds:

[0276]

[0277] By transforming evaluation indicators into optimization objectives and constraints, a unified modeling process is achieved, from temperature evaluation to finding the optimal solution.

[0278] S44: Based on the calibrated mechanism model and trend prediction results, rapid simulation evaluation and multi-objective optimization solutions are performed for different water intake and drainage schemes. Specifically, for any candidate scheme... The results of changes in the flow field, temperature field, intake temperature, forebay temperature and thermal impact range after the implementation of the scheme are obtained by inputting the current environmental boundary conditions, predicted sea conditions and unit operating conditions into the calibrated numerical model. Based on this, the corresponding ecological, safety and economic evaluation indicators are calculated.

[0279] Given a scheme The model response is as follows:

[0280]

[0281] The corresponding objective function value is:

[0282]

[0283] Regarding the solution algorithm, a genetic algorithm is used, and when Pareto optimization is employed, the set of non-dominated solutions is obtained:

[0284]

[0285] in, This represents the Pareto optimal solution set. Let represent the number of non-dominated alternatives. When using single-objective weighted optimization, the final optimal solution can be expressed as:

[0286]

[0287] Through the above methods, the optimal water intake and drainage decision that takes into account ecological impact, operational safety and economy can be found among multiple water intake and drainage scheme combinations. In order to improve the efficiency of online solution, the multi-condition result library formed by S2 and the AI ​​approximate prediction model trained by S3 can be combined to quickly pre-screen the candidate schemes, and then a high-precision mechanism model is called to finely verify a small number of preferred schemes, thereby realizing a hierarchical optimization strategy of rapid screening + fine verification.

[0288] S45: Sort and optimize the multi-objective optimization results, outputting the optimal water intake and drainage decision and a set of alternative decisions. For non-dominated solutions obtained using Pareto optimization, further construct decision ranking criteria based on current operational preferences to select the most suitable recommended solution from the non-dominated solution set for the current operating conditions. For single-objective weighted optimization, directly use the optimal solution... As a recommendation decision, a comprehensive scoring function is constructed:

[0289]

[0290] in, These represent the normalized evaluation values ​​for ecological, security, and economic objectives, respectively. The ranking weight coefficients are used to prioritize the candidate solutions based on the scoring results.

[0291]

[0292] Among them, symbols This indicates that the priority is higher, and the first option after sorting is the recommended optimal decision, while the subsequent options are output as alternative decisions.

[0293] The output results include the recommended scheme's corresponding intake and drainage combination, flow distribution method for each outlet, circulating water system operation mode, load adjustment amount involved, as well as the corresponding highest predicted temperature of the intake, highest predicted temperature of the forebay, ecological high temperature impact area, and comprehensive economic cost. Through the above forms, the optimal intake and drainage decision has a clear quantitative basis, providing an input basis for step S5 to further form an executable on-site scheduling scheme.

[0294] Based on the above technical solution, the specific steps of S5 are as follows:

[0295] S51: Perform procedure verification and empirical rule verification on the optimal water intake and drainage decision results obtained in step S4, and screen out schemes that do not meet the on-site execution conditions and do not meet the operation and management requirements of nuclear power plants. Although the multi-objective optimization results are optimal in a mathematical sense, they still need to meet the requirements of equipment start-up and shutdown sequence, minimum operating interval, pump and valve interlocking relationship, operating authority, environmental protection limit boundary and duty management system in the engineering implementation. Therefore, a second verification needs to be carried out before the official issuance.

[0296] Let the candidate optimal solution output by S4 be The corresponding rule verification function is denoted as:

[0297]

[0298] in, Indicates the first The verification results of the rules, This represents the total number of rules. When all rules are satisfied, it is denoted as:

[0299]

[0300] If a rule is not met, the corresponding check result is 0. Therefore, the feasibility determination function of the solution is defined as follows:

[0301]

[0302] when When the solution meets all procedures and rules of experience, it enters the implementation plan generation stage. When this happens, it indicates that the plan has obstacles to implementation and needs to be modified or downgraded.

[0303] The rule set includes upper limit constraints on intake water temperature, upper limit constraints on forebay temperature, boundary constraints on discharge temperature rise, minimum interval constraints on pump start-up and shutdown, valve switching sequence constraints, single flow rate adjustment range constraints, and manual operation permission constraints. Specifically, the minimum interval constraints on pump start-up and shutdown can be expressed as:

[0304]

[0305] in, Indicates the next pump start-up time. Indicates the last time the pump stopped. This indicates the minimum start-stop time interval. By verifying the above rules, we can avoid triggering equipment protection, operational conflicts, and management violations when the optimization results are executed on-site.

[0306] Regarding empirical rules, historical successful handling cases, operational preference strategies under specific tidal patterns, and the experience of on-duty experts are introduced. Specifically, during a specific high tide phase, historical data shows that a certain combination of outlets is more likely to form thermal plume recirculation. Therefore, empirical rules are used to lower the priority of this combination, with the empirical correction term set as follows: The final comprehensive score of the scheme can be expressed as:

[0307]

[0308] in, The original comprehensive score obtained in step S4 is used to introduce an empirical correction term to make the final recommended scheme more in line with the long-term operation of nuclear power plants.

[0309] S52: Transform the optimal decision results verified by the rules into a clear and executable on-site scheduling plan. The on-site scheduling plan directly guides the front-line operators to complete the actual adjustment of the water intake and drainage system. Therefore, it is necessary to transform the optimization variables and model results into specific operation actions, execution order, target parameters and expected effect descriptions.

[0310] The verified plan is recorded as follows: Map it to a sequence of scheduling actions:

[0311]

[0312] in, Indicates the first A scheduling action, This represents the total number of actions, and each scheduled action... This includes the action object, action type, target setting value, execution time, and execution preconditions. A single action can be represented as:

[0313]

[0314] in, Indicates the object of the action. Indicates the action type. Indicates the target set value. Indicates the execution time. This indicates that the preconditions must be met.

[0315] The action objects are a water intake, a drainage outlet, a circulating water pump group, and a valve. The action types are start / stop, flow increase, flow decrease, switching, and hold. The target setpoints are target flow rate, target opening degree, and target load adjustment amount. Through this structured expression, the abstract optimization results are transformed into standardized scheduling instructions.

[0316] The on-site dispatch plan includes the following: the number and number of water intakes, the number and number of drainage outlets, the target flow distribution for each outlet, the start-stop combination of circulating water pumps, the valve switching sequence, load adjustment suggestions when necessary, the expected start time, the expected duration, and the expected recovery conditions. For complex dispatching that requires multiple steps, the order of actions is further given to avoid instantaneous impact caused by the simultaneous switching of different equipment.

[0317] In addition, to enhance the understandability of the plan, the expected implementation effect corresponding to the scheduling plan can also be output. The corresponding key effect vector is:

[0318]

[0319] in, This indicates the expected decrease in water intake temperature. This indicates the expected decrease in the forebay temperature. This indicates the projected reduction in the area affected by the high temperatures. This indicates the change in overall economic costs brought about by the plan. By outputting the above effect information, on-site personnel can intuitively understand the necessity and expected benefits of implementing the plan.

[0320] S53: Track and evaluate the operational effects after the implementation of the scheme, and compare and analyze the actual feedback results with the optimized prediction results to verify the effectiveness of the scheme and identify the source of deviation. The tracking objects include the intake temperature, forebay temperature, drainage heat plume diffusion range, temperature rise in sensitive areas, pump and valve operation status and relevant unit operating parameters.

[0321] Let the time after the plan is executed be... The actual observed response is:

[0322]

[0323] The corresponding predicted response is:

[0324]

[0325] Construct an implementation effect deviation vector:

[0326]

[0327] in, This is used to reflect the degree of deviation between the actual execution results and the system's prediction results. If the key deviation items exceed the preset threshold, it indicates that changes in the on-site environment, equipment response delays, and model errors have had a significant impact on the implementation effect, requiring further analysis and the definition of solution effectiveness indicators:

[0328]

[0329] in, Indicators representing the evaluation of temperature control effectiveness. Indicators representing the evaluation of the effectiveness of ecological impact control. Indicates the consistency evaluation index of the implementation plan. As a weighting coefficient, this indicator comprehensively evaluates whether the scheduling plan has achieved the expected goals. Specifically, if the water intake temperature drops significantly and the ecological high temperature impact range shrinks effectively, and the equipment operates without abnormalities, then the implementation effect of this plan is considered to be good.

[0330] S54: The execution feedback results are sent back to the model parameter library, the experience rule library, and the strategy knowledge base to realize the continuous learning and closed-loop optimization of the system. The sent-back content includes the environmental status before the implementation of the plan, the optimization decision results, the manual adjustment records, the actual execution trajectory, the post-implementation monitoring results, and the deviation analysis results. By archiving and reusing the above information, the accuracy and applicability of decision-making in complex sea conditions, special tidal patterns, and multi-unit combination scenarios are gradually improved.

[0331] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A nuclear power plant water intake and discharge intelligent decision-making system based on a physics AI engine, characterized in that: Based on multi-source monitoring, mechanism simulation, historical calibration, and optimization decision-making technologies, an intelligent decision-making system is constructed according to the technology chain of perception, modeling, identification, optimization, and execution, including the following modules: Multi-source sensing module, intelligent simulation module, model calibration module, strategy optimization module, and closed-loop evolution module; The multi-source sensing module provides basic input for subsequent simulation, calibration and optimization; the intelligent simulation module constructs a physical sensing AI agent; the model calibration module provides a high-precision simulation base for subsequent risk identification and optimization decision-making; the strategy optimization module realizes joint optimization; and the closed-loop evolution module realizes the continuous self-evolution of the water intake and drainage agent.

2. The intelligent decision-making system for nuclear power plant water intake and discharge based on a physics AI engine according to claim 1, characterized in that: The multi-source sensing module is designed for real-time thermal environment monitoring of the nuclear power plant's intake and discharge sea areas, and integrates satellite remote sensing temperature field, nearshore in-situ hydrological observation, plant circulating water system operating conditions, and meteorological data. The intelligent simulation module is based on the high-precision flow field, temperature field, and key point state sequence output by the model calibration module to construct a physical perception AI intelligent agent for nuclear power plant water intake and drainage scenarios.

3. The intelligent decision-making system for nuclear power plant water intake and discharge based on a physics AI engine according to claim 2, characterized in that: The model calibration module uses historical operating condition data, real-time monitoring data, and the output results of the intelligent simulation module to perform inversion calibration on the drag coefficient, thermal diffusivity coefficient, and local mixing coefficient of the ocean flow field. The strategy optimization module takes the risk probability, comprehensive risk index, future temperature sequence, and high-risk period output by the intelligent prediction module as input.

4. The intelligent decision-making system for nuclear power plant water intake and discharge based on a physics AI engine according to claim 3, characterized in that: The closed-loop evolution module receives the recommended and alternative schemes output by the strategy optimization module, generates field-executable instructions and sends them to operators and the control system, while continuously tracking environmental feedback and equipment response data after execution to verify and evaluate the effectiveness of the scheme.

5. A nuclear power plant water intake and discharge intelligent decision-making method based on a physics AI engine, used in the decision-making method of a nuclear power plant water intake and discharge intelligent decision-making system based on a physics AI engine according to any one of claims 1-4, characterized in that: Includes the following steps: S1: Real-time acquisition of multi-source heterogeneous data, including satellite remote sensing temperature field, in-situ hydrological observation and nuclear power operation status, through the integrated air-ground-water monitoring network; use spatiotemporal registration algorithm to perform multi-dimensional feature alignment and standard optimization preprocessing to build a basic physical data environment for water intake and drainage intelligent agents; S2: Embed physical constraints into the numerical simulation model, and achieve deep integration of mechanism-driven and data features through an industrial physics AI engine to simulate the impact of nuclear power plant water intake and discharge in complex marine environments; S3: Enhance the physical perception capabilities of AI agents, use real-time monitoring data from near and far sea areas to drive adaptive updates of simulation results, and calibrate key parameters, including flow field resistance and thermal diffusivity, through a backpropagation mechanism. S4: Based on the high-precision simulated flow field constructed in step S3, establish an evaluation system for the influence factor of temperature rise intensity, conduct a pre-simulation of intake and drainage conditions, and use an optimization algorithm to search for and obtain the optimal intake and drainage decision strategy. S5: The intake and drainage intelligent agent automatically generates and outputs the optimal operation instructions to guide the adjustment of the actual working conditions, continuously tracks the environmental feedback data after execution to verify the effect, and stores the experience in the closed-loop process into the expert knowledge base to realize the continuous self-evolution of the system's decision-making capabilities.