Multi-objective constraint solving method for fire-fighting monitoring and generator set collaborative optimization

Abstract

The invention discloses a fire-fighting monitoring and generator set collaborative optimization multi-target constraint solving method, which comprises the following steps: carrying out real-time acquisition and structured processing on multi-source data, acquiring safety monitoring and equipment operation data through a sensor network, and storing the data in a distributed real-time database after preprocessing; carrying out dynamic thermodynamic field modeling and risk area division, discretizing a plant area as a grid unit, solving a heat conduction equation in combination with heat source modeling and boundary conditions, and dividing three levels of risk areas; building a multi-objective optimization model, defining three types of core constraints, building a normalized objective function, and realizing multi-objective decoupling through dynamic weight adjustment; according to an asymmetric load migration strategy, differential load adjustment is implemented according to an optimization result, an auxiliary system is linked, constraints are verified in real time, and closed-loop control is formed. According to the technology, the collaborative goals of safety risk controllability, system efficiency maintenance and equipment redundancy optimization are achieved, and the method is suitable for fire prevention and efficient operation in an energy production scene.

Description

Technical Field

[0001] The present invention relates to the technical field of energy production safety and intelligent control, and particularly relates to a multi-objective constraint solving method for collaborative optimization of fire monitoring and generator sets. Background Art

[0002] In the operation and management system of traditional gas power plants, the fire monitoring system and the generator set control system have been in an independent operation technical state for a long time. The existing fire protection system mainly relies on independently deployed devices such as smoke detectors and temperature sensors to achieve fire early warning. Its alarm signal can only trigger the activation of fire extinguishing devices or manual emergency response processes, and is always in a data island state with the operating parameters of power generation equipment such as unit load, steam temperature, and cooling water flow, and no dynamic correlation mechanism has been formed. The existing generator set optimization control algorithms focus on improving power generation efficiency and economic dispatching. Taking the unit collaborative optimization method proposed in the patent "Unit Collaborative Optimization Method Based on Multiple Application Scenarios" with the publication number CN117557929A as an example, its constraint conditions are limited to conventional parameters such as equipment physical limits and grid dispatching requirements, and real-time fire risk indicators have never been incorporated into the optimization model system.

[0003] In recent years, the application of industrial Internet of Things technology in the power plant monitoring field has realized the centralized collection of equipment status data, but there are still the following bottlenecks in the deep integration of fire protection data and power generation control data. Data response lag, the traditional centralized data processing architecture leads to a significant lag in data response. The fire alarm signal needs to go through multi-level system interactions to trigger equipment adjustment, which is difficult to meet the real-time control requirements of high-temperature risk areas. For example, the regional division dynamic monitoring mechanism proposed in the patent "Dynamic Fire Safety Risk Monitoring Method and Platform" with the publication number CN119918926A, but the real-time problem of cross-system collaboration has not been solved; Static weight defect, when the existing optimization algorithms process multi-objective constraints of safety and efficiency, they generally use the weighted summation method for simplified processing. However, due to the spatio-temporal mutation characteristics of fire risks, the static weight allocation mode cannot achieve a fast response mechanism when the safety threshold is dynamically broken through. For example, the patent "A Smart Fire Monitoring Method, Medium and Terminal Based on Intelligent Algorithm" with the publication number CN118658121A introduces an AI algorithm, but the cross-system collaborative optimization problem has not been effectively solved; Load transfer strategy blank, the existing load transfer strategies are mainly based on grid demand or equipment balanced loss principles, and no asymmetric load transfer mechanism has been established between the fire risk area and the non-risk area. The patent "Fire Monitoring Method and System Combining Lora Wireless Transmission and Neural Network Algorithm" with the publication number CN118890567B has optimized the data transmission technology, but does not involve the power generation control collaborative strategy. Summary of the Invention

[0004] To solve the problems raised in the above background art, the present invention provides a multi-objective constraint solving method for collaborative optimization of fire monitoring and generator sets, which realizes the dynamic balance between safety and efficiency through real-time data fusion, dynamic thermodynamics modeling, multi-objective optimization and asymmetric load transfer strategy.

[0005] To solve the above technical problems, the technical solution proposed by the present invention is as follows:

[0006] A multi-objective constraint solving method for collaborative optimization of fire monitoring and generator sets, comprising the following steps:

[0007] S1. Construct a data interaction architecture through distributed computing devices, and rely on the data interaction architecture to collect safety monitoring data and equipment operation data in real time through a sensor network. The data is accessed into a distributed real-time database through an industrial communication network and a standardized protocol, and then the safety monitoring data and equipment operation data are preprocessed to form structured data and stored in the distributed real-time database, providing data input in a unified format for the subsequent steps;

[0008] S2. Based on the structured data in the distributed real-time database, according to the three-dimensional layout of the plant area and the distribution of physical equipment, discretize the plant area into grid units including core heat-generating equipment, fluid transmission components, and power transmission components. Then, calculate the heat source intensity of each grid unit through a heat source modeling method. The heat source intensity combines boundary conditions including environmental heat dissipation, cooling system efficiency, and heat resistance of the thermal insulation structure, and uses a numerical simulation method to solve the three-dimensional unsteady heat conduction equation to generate a temperature field cloud map. Based on the temperature field cloud map, combined with the preprocessed safety monitoring data and equipment heat tolerance parameters, divide the risk area, and the risk area division result is stored in the distributed real-time database in real time, providing risk assessment data in the spatial dimension for subsequent multi-objective optimization;

[0009] S3. Based on the risk area division result and the preprocessed equipment operation data, first define three types of core constraint parameters, including a safety threshold reflecting the thermal safety critical value of the equipment or area, a minimum efficiency index for meeting the external scheduling requirements, and an equipment redundancy for the load adjustment margin that the equipment can withstand. Then, construct a normalized objective function, and perform weight allocation on the safety risk ratio, efficiency loss ratio, and redundancy retention ratio in the normalized objective function through a dynamic weight factor adjustment mechanism associated with the risk level and scheduling requirements, realizing the dynamic decoupling of multi-objective constraint conditions, generating a structured optimization result including load adjustment instructions and effect prediction data, and finally storing the optimization result in the distributed real-time database;

[0010] S4. Based on the optimization results and risk area division results, the asymmetric load migration strategy is first implemented. Through the generation of load adjustment instructions for the main equipment, the auxiliary system linkage control is synchronized, and then the execution status of the control instructions is collected and fed back in real time, triggering the triple constraint verification. At the same time, combined with the quantitative evaluation of the control effect and the cross-step data association storage, the collaborative control goals of controllable safety risks, maintenance of system operation efficiency and optimization of equipment redundancy are achieved.

[0011] Preferably, the safety monitoring data includes regional temperature values output by temperature sensors, gas concentration parameters detected by smoke concentration sensors, and image data collected by flame monitoring equipment; the equipment operation data covers key parameters reflecting the equipment operating status including generator set load, steam system pressure, cooling system flow, and electrical system current.

[0012] Preferably, the preprocessing of the safety monitoring data includes: using a noise reduction algorithm to process regional temperature values and gas concentration parameters, and at the same time establishing a sensor spatial position mapping table, and interpolating single-point monitoring data into regional grid temperature distribution and concentration distribution data in combination with the three-dimensional coordinates of the plant area to form a continuous spatial thermal safety parameter field; for image data, processing is performed through an image feature extraction algorithm, first using a convolutional neural network model to extract key feature parameters including flame area, spread speed, and temperature gradient, and then using a threshold segmentation and edge detection algorithm to identify the flame boundary and calculate the geometric center coordinates, providing spatial coordinate data for subsequent risk area positioning.

[0013] Preferably, the preprocessing of the equipment operation data includes: for the key parameters reflecting the equipment operation status, calibrating the clock deviation of each sensor through the global clock synchronization protocol to ensure that the timestamp error of multi-source data is less than 1ms, using linear interpolation to fill in the missing data points, and eliminating invalid data that exceeds the physical limit of the equipment based on the equipment operation logic. At the same time, the heterogeneous data generated by different manufacturers and different types of sensors, including analog, digital, and protocol data, are converted into a unified industrial data model, standardized data labels are defined, and a data quality assessment mechanism is established to add "valid, invalid, and suspicious" quality stamps to the preprocessed data to provide reliability identification for subsequent modeling.

[0014] Preferably, the data quality assessment mechanism constructs a multi-dimensional assessment system for safety monitoring data and equipment operation data, and realizes intelligent screening and hierarchical management of data through completeness, accuracy, consistency and timeliness assessment.

[0015] Preferably, in the integrity assessment, the system counts the proportion of missing data points within the data collection period, and marks the sensor data with a missing rate exceeding the preset threshold as "suspicious" to avoid modeling deviations caused by data missing; for the accuracy assessment, the system checks the physical limit threshold of the device and analyzes the correlation of adjacent sensor data, and marks the data beyond the reasonable range as "invalid" to ensure that the input data conforms to the physical laws of the device operation; for the consistency assessment, the system compares the real-time values of the same parameter from different data sources, and triggers a manual verification process and marks it as "suspicious" when the deviation exceeds 3% to solve the problem of multi-source data conflicts; for the timeliness assessment, the system sets the inspection data update period according to the device type to check whether it meets the regulations, and marks the data with a delay exceeding 2 cycles as "invalid" to ensure the real-time requirement of the data.

[0016] Preferably, based on the above assessment results, the data is assigned three quality stamps: "valid", "invalid", and "suspicious": "Valid" data is complete, accurate, consistent, and meets the real-time standard, and can be directly used for subsequent modeling and optimization calculations; "Invalid" data has serious errors, including exceeding the physical limit and not being updated for a long time. It is frozen in the distributed real-time database and triggers a sensor failure alarm. At the same time, the system calls the historical valid values or adjacent sensor data for replacement; "Suspicious" data has minor anomalies, including a missing rate < 5% and a cross-source deviation of 3% - 5%. It is allowed to participate in the calculation limitedly but with a warning sign attached, so that the edge computing node can automatically reduce its weight during modeling to balance the data availability and reliability.

[0017] Preferably, the preprocessed safety monitoring data and device operation data form structured data, which is stored in the distributed real-time database according to the three-dimensional index structure of "time series - spatial location - device type", supporting millisecond-level data read and write responses; a data version management mechanism is established to retain the historical data traces of key parameters, including temperature and load, for at least 72 hours for subsequent model calibration and fault tracing.

[0018] Preferably, the distributed real-time database provides a unified data access entry for the subsequent steps through a standardized interface to ensure the consistency, timeliness, and traceability of the whole-process data. During the data interaction process, the subscription-publish mechanism is adopted between the distributed computing device and the database to synchronize the data changes in real time, providing a reliable data basis for subsequent modeling and control, and adopting encryption verification measures during the data interaction process to ensure the integrity and security of data transmission.

[0019] Preferably, the discretization of the grid cells adopts a hierarchical adaptive strategy: First, according to the plant equipment topology and process flow, the plant is divided into functional modules including the generator set area, steam pipeline area, and cable trench area, and each module is further subdivided into basic grid cells; for core heat-generating equipment including generator stators and boiler furnaces, and heat-prone connecting components including flanges and valves, millimeter-level fine grids are used to capture local thermal gradients, and meter-level coarse grids are used for non-critical areas to optimize the calculation efficiency; when the equipment load suddenly increases by more than 10% or the safety monitoring data shows abnormal fluctuations, the system automatically triggers the grid local encryption algorithm to refine the target area grid to 0.05 m, realizing dynamic adjustment of the calculation accuracy.

[0020] Preferably, the heat source modeling method includes heat source type classification and heat source intensity calculation. The heat source type classification divides the plant heat sources into three categories: electric conversion heat sources, fluid transmission heat sources, and contact loss heat sources according to the equipment function and heat generation mechanism, and clarifies the physical models and calculation bases of different types of heat sources.

[0021] Preferably, for the heat source intensity calculation, calculation models are established for various heat sources based on the real-time operation data of the equipment and physical laws: for electric conversion heat sources, the heat generation amount is calculated in real time through the electromagnetic loss formula where is the real-time load collected in step S1, and is the rated power generation efficiency stored in the equipment ledger data; for magnetic loss heat sources, the copper loss and iron loss are integrated, depending on the parameters preprocessed in step S1, including current I, winding resistance R, magnetic flux density B, and frequency f; for fluid transmission heat sources, the heat dissipation amount is calculated through the fluid energy equation Combined with the data including mass flow , specific heat capacity , and temperature drop stored in step S1, it reflects the energy loss of the fluid during transmission.

[0022] Preferably, the boundary condition setting integrates the environmental physical characteristics including environmental temperature, wind speed, and equipment surface material, and the equipment heat dissipation parameters including cooling medium flow rate, insulation layer thickness, and thermal conductivity.

[0023] Preferably, the environmental heat dissipation is quantified using Newton's cooling formula where is the equipment surface temperature, which is the regional temperature value collected by the temperature sensor in step S1 or calculated through the heat conduction equation. The convective heat transfer coefficient h is preset based on the equipment surface material and environmental wind speed, is the real-time environmental temperature collected in step S1.

[0024] Preferably, the efficiency of the cooling system calculates the heat exchange amount through the flow rate and temperature parameters of the cooling medium , reflecting the forced heat dissipation capacity of the cooling system for the equipment body, where: V is the volume flow rate of the cooling medium, which is the cooling system flow rate parameter from the equipment operation data in step S1, and is stored in the distributed real-time database after the noise reduction filtering and timestamp synchronization processing in step S1; 、 are the inlet and outlet temperatures of the cooling medium respectively, which are collected and preprocessed by the temperature sensors in step S1, reflecting the actual heat dissipation capacity of the cooling system; is the density of the cooling medium, is the specific heat capacity, which comes from the medium physical property data in the equipment ledger.

[0025] Preferably, the thermal resistance of the heat insulation structure is determined according to Fourier's law where is calculated from the thickness of the heat insulation layer and the thermal conductivity , and the above parameters are all stored in the physical property table of the distributed real-time database.

[0026] Preferably, the numerical simulation method is realized through spatial and temporal domain discretization, algebraic equation system construction and large-scale matrix solution.

[0027] Preferably, the spatial discretization methods include but are not limited to the finite element method, the finite volume method and the finite difference method, which are adaptively selected according to the grid type and accuracy requirements. The temporal domain discretization uses implicit or explicit difference formats to discretize the time partial derivative terms to balance the calculation stability and solution efficiency; the algebraic equation system construction is based on the energy conservation principle and the discretization format, and converts the heat conduction equation into a linear algebraic equation system , where the heat conduction matrix K, the heat capacity matrix C, and the heat source vector Q are all calculated and generated from the grid cell heat source intensity and boundary condition parameters obtained by heat source modeling; for the large-scale matrix solution, an iterative solution algorithm is used to accelerate the calculation, and the domain decomposition parallel algorithm is enabled in the distributed scenario to distribute the calculation tasks to multiple computing nodes to ensure that the calculation time for a single time step meets the requirements of industrial real-time monitoring.

[0028] Preferably, for the risk area division, it is first necessary to construct a risk threshold system including equipment-level thermal safety thresholds and area-level thermal safety thresholds. The equipment-level thermal safety thresholds are calibrated by material thermal tolerance parameters and historical failure data; the area-level thermal safety thresholds are set based on industry standards and historical fire accident statistical data.

[0029] Preferably, based on the above risk threshold system, combining the real-time temperature field cloud map with the preprocessed safety monitoring data including smoke concentration and image data, the plant area is divided into three levels of risk areas, including high-risk areas, medium-risk areas, and safe areas.

[0030] Preferably, in the high-risk area, when the temperature of the grid cell exceeds 120% of the corresponding equipment-level threshold and the duration is ≥ 30 s, or the flame monitoring module detects the flame area , it is determined as an area that directly threatens the safety of equipment and the lives of personnel. The system immediately triggers a red warning, marks this area as the priority processing object for load adjustment, and forcibly starts the pre-assessment process for load reduction.

[0031] Preferably, the flame area is the data extracted from the image data in step S1.

[0032] Preferably, in the yellow medium-risk area, when the temperature is in the range of 100% - 120% of the equipment-level threshold, or the smoke concentration ≥ 5% LEL, it is determined as an area that requires dynamic monitoring. This area is included in the candidate adjustment range of the load migration strategy, and a comprehensive evaluation is carried out in combination with the remaining life of the equipment and the system efficiency.

[0033] Preferably, the LEL is the lower explosion limit, which is the gas concentration parameter from step S1.

[0034] Preferably, in the green safe area, when the temperature is lower than 100% of the equipment-level threshold and there are no abnormal flames and smoke, it is determined as a safe area. This area serves as the target receiving area for the load migration of the high-risk area. The system preferentially selects equipment with a longer remaining life and a larger load margin to receive the migrated load, ensuring the system power balance and equipment loss balance during the load adjustment process.

[0035] Preferably, to accurately reflect the dynamic evolution process of the risk area over time, based on the real-time temperature field cloud map and the risk level determination result, the level set method is used to extract the boundary of the temperature gradient field. The risk boundary position is automatically identified by calculating the temperature difference between adjacent grid cells, and a continuous regional contour line is generated in combination with the risk level distribution. When the temperature of a certain area rises rapidly (the heating rate or the smoke concentration suddenly increases, the boundary algorithm updates the contour line in real time based on the latest grid data, ensuring that the boundary of the risk area continuously evolves with the time step and accurately reflecting the spatio-temporal diffusion characteristics of the fire risk.

[0036] Preferably, the risk area division results are stored in a distributed real-time database in the form of a spatio-temporal data cube containing temperature values, risk levels, heating rates, and change trends. Each grid cell is associated with three-dimensional coordinates, equipment ledger information, and the risk evolution history, supporting multi-dimensional applications, including three-dimensional visual monitoring, spatio-temporal trajectory tracing, and intelligent warning triggering.

[0037] Preferably, the three-dimensional visual monitoring uses WebGL technology to render the temperature field cloud map and the risk area contour into a real-time 3D model. The operator can accurately locate high-risk equipment including overheated cable joints and overheated turbine bearings through interactive operations such as zooming, rotating, and cutting.

[0038] Preferably, the spatiotemporal trajectory tracing supports the retrieval of historical risk data by time range and space range, and reproduces the complete evolution process of abnormal temperature rise events including the temperature rise path caused by cooling water pump failure through the animation of temperature curve and risk level change.

[0039] Preferably, the intelligent early warning triggers the built-in risk change monitoring algorithm. When the high-risk area expands by more than , or when a grid unit with a temperature exceeding 110% of the device-level threshold appears in a medium-risk area for the first time, early warning information containing the specific location, risk level, and list of affected devices is automatically pushed to the edge computing node, triggering the multi-objective optimization pre-assessment process of step S3 5-10 seconds in advance to gain golden response time for the load adjustment strategy.

[0040] Preferably, the safety threshold constraint in the three types of core constraint parameters directly refers to the risk threshold system established in step S2, including equipment-level thermal safety thresholds and regional-level thermal safety thresholds, whose values come from the material temperature resistance parameters and industry standards provided by the equipment manufacturer. These thresholds serve as critical constraints on the safety risk ratio to ensure that the equipment and regional temperatures do not exceed the thermal safety bottom line during the optimization process; the minimum efficiency index constraint sets the lower limit of the overall operating efficiency of the system according to the external power grid dispatching instructions and the equipment load efficiency characteristic curve, including the power generation efficiency of the whole plant ≥ 90%. This index is used as the lower limit of the efficiency loss ratio constraint, and is dynamically calibrated with the real-time efficiency data of the equipment collected in step S1 and the dispatching requirements to avoid energy inefficient operation caused by excessive safety adjustments; the equipment redundancy constraint is based on the rated capacity of the equipment, defines the maximum load rate of a single device, and ensures that each device retains at least 15% of the load adjustment margin. This parameter comes from the rated parameter in the equipment ledger, which serves as the adjustment space for the redundancy retention ratio to reserve emergency capacity for load migration under sudden working conditions.

[0041] Preferably, based on the above constraints, the normalized objective optimization function is constructed by converting the safety risk, efficiency loss, and redundancy into dimensionless normalized indicators. , where SRR is the safety risk ratio, ELR is the efficiency loss ratio, and RRR is the redundancy retention ratio.

[0042] Preferably, the safety risk ratio focuses on thermal safety control, based on the risk threshold system and real-time temperature field data of step S2: , where is the real-time temperature of the device surface calculated by the heat conduction equation in step S2, is the device-level safety threshold, is the ignition point limit of the material, is the importance weight of the device. By the ratio of the weighted sum of molecular overtemperature to the weighted sum of denominator limit temperature difference, it quantifies the overall thermal runaway risk of the system. The smaller the value, the higher the safety, and it is directly related to the safety threshold constraint.

[0043] Preferably, the efficiency loss ratio measures the impact of load adjustment on power generation efficiency and is related to the device operation efficiency parameter in step S1: , where is the system efficiency before optimization, is the efficiency under the current load distribution, is the minimum efficiency index required by the external dispatching. While ensuring that the denominator is positive, the smaller the value, the better the efficiency retention, and it is directly restricted by the minimum efficiency index constraint.

[0044] Preferably, the redundancy retention ratio evaluates the emergency response ability of the system based on the difference between the real-time load of the device and the rated capacity: , where the numerator is the actually reserved redundant capacity, is the rated capacity of the device, is the real-time load, and the denominator is the maximum redundant capacity that the device can provide. is the minimum load rate. The closer the value is to 1, the stronger the anti-disturbance ability of the system. It is directly defined by the device redundancy constraint to calculate the space and avoid the risk of equipment overload after load adjustment.

[0045] Preferably, the dynamic weight factor adjustment mechanism adopts a weight adjustment algorithm that is linked in real time with the risk level, dispatching requirements, and redundancy status, corresponding to the weight factors of the safety risk ratio, efficiency loss ratio, and redundancy retention ratio , , , satisfying . The adjustment logic of each weight factor is directly related to the input parameters, forming a working condition adaptive target priority switching mechanism.

[0046] Preferably, the safety weight adjustment linked to the risk level is based on the risk area division result in step S2. When the proportion of the high-risk area exceeds the preset threshold, the safety weight increases linearly to ensure that the safety risk ratio obtains a higher optimization priority during temperature anomalies. The calculation formula is: , where, is the basic weight, is the risk sensitivity coefficient.

[0047] Preferably, the efficiency weight adjustment linked to the dispatching requirements is based on the emergency degree index of the external power grid dispatching instruction The index ranges from 0 to 1, with 1 for peak period and 0.3 for valley period. Dynamically adjust according to the scheduling priority to balance safety control and economic benefits. The calculation formula is: ,in, is the basic weight, is the dispatch response factor.

[0048] Preferably, the redundant weight adjustment of the redundancy state linkage is performed when the overall redundancy of the system is maintained at , that is, when the actual redundant capacity is less than 60% of the rated redundant capacity, the redundant weight Automatically increase the load adjustment range of a single device, including not exceeding 80% of the rated capacity, and retain emergency margin. No independent formula is required, directly from Determined dynamically to ensure that the total weight is always 1.

[0049] Preferably, after completing the definition of constraint parameters, the construction of normalized objective function and the adjustment of dynamic weight factors, the dynamic decoupling of multi-objective constraints is performed through the non-dominated sorting genetic algorithm II. First, the algorithm parameters are configured. Based on the historical data test, the population size is set to 100. This scale can achieve a balance between computational efficiency and solution set diversity. The number of iterations is 50 generations, the crossover probability is 0.8 to retain the high-quality characteristics of the parent generation, and the mutation probability is 0.1 to avoid premature convergence. The Pareto optimal solution set is generated by simulating natural selection; then the constraint condition is checked. In each generation of evolution, a triple constraint check is performed on the individual, including whether the equipment temperature exceeds the safety threshold, whether the system efficiency is lower than the minimum index, and whether the single equipment load exceeds the redundancy upper limit, and invalid solutions that do not meet the conditions are eliminated; the result screening mechanism is randomly adopted, and the optimization schemes with a safety risk ratio decrease of ≥15%, an efficiency loss ratio of ≤0.15, and a redundancy retention ratio of ≥0.6 are given priority to form a feasible solution set.

[0050] Preferably, the optimization results include: load adjustment instructions, effect prediction data, and cross-step data associations. The load adjustment instructions are generated based on the comprehensive weights of SRR, ELR, and RRR in the normalized objective function F output by the multi-objective optimization model, and each instruction is associated with the equipment ID, load before and after adjustment, and constraint verification results; the effect prediction data is based on the risk area, and simulates the change in the temperature field after the load of equipment in the high-risk area is reduced by superimposing load adjustment instructions. It is expected that the temperature in the high-risk area will drop by 20°C, the system efficiency loss will be controlled at 9%, and the redundancy will be increased to 0.72, forming a mapping relationship of "adjustment plan-risk change-efficiency redundancy"; the cross-step data association associates the optimization results with the risk area coordinates and the equipment health status, and pushes them to step S4 through the database standardized interface, providing precise instructions with constraints for equipment collaborative control.

[0051] Preferably, the asymmetric load transfer strategy is executed according to the risk level and real-time redundancy of the area where the device is located, and the real-time redundancy is the difference between the rated capacity of the device defined in step S3 and the actual load.

[0052] Preferably, for the devices in the high-risk area, with the core goal of quickly reducing the thermal safety risk, based on the real-time temperature of the device in step S2 and the safety threshold defined in step S3 calculate the load reduction amount , and the formula for calculating the load reduction amount is: , where is the temperature tolerance limit of the device material from the device ledger in step S1, ensuring that the device temperature drops below the safety threshold plus a 10% margin after load reduction, and at the same time meeting the device redundancy constraint defined in step S3, retaining a rated capacity margin of ≥15%, and sorting according to the device importance weight defined in step S3 for core devices , give priority to quickly reducing the load of high-weight devices such as generator sets and main transformers, and the adjustment rate ≤ 10% of the rated capacity per minute to avoid chain failures caused by overheating of key devices.

[0053] Preferably, for the devices in the safe area and medium-risk area, evenly distribute the load migrated out of the high-risk area according to the redundancy ratio distribution method, and the amount of the migrated-in load is allocated proportionally according to the real-time redundancy of the device : Ensure that the load rate of a single device does not exceed the maximum load rate defined in step S3, that is, 85% of the rated capacity, and the rated capacity data comes from the rated parameter field of the device ledger in step S1. Combining with the real-time efficiency data of the device collected in step S1, preferentially allocate the migrated-in load to the devices with a load rate in the efficient range of 60% to 80%, and dynamically calculate the optimal allocation combination through the efficiency characteristic curve to reduce the system efficiency loss caused by load migration.

[0054] Preferably, during the main device load adjustment, control instructions including device ID, target load rate, and adjustment rate are extracted from the optimization result. The adjustment rate is restricted by the mechanical stress limit of the device to ensure the safe operation of the device, and a real-time constraint verification mechanism is implemented. During the load adjustment process, the temperature field cloud map and constraint conditions are called for verification: if the device temperature breaks through the safety threshold , trigger an emergency load reduction; when the efficiency is lower than the minimum efficiency index, automatically reduce the migrated-in load of low-efficiency devices; when the load rate of a single device exceeds 85% of the rated capacity, stop migrating in. If a constraint is triggered, automatically roll back to the valid state and mark an exception code.

[0055] Preferably, to match the synchronous triggering of the auxiliary system linkage control for the main equipment load adjustment, for the equipment with increased load, the cooling medium flow rate is automatically adjusted according to the load increment ratio. The cooling medium flow rate comes from the data of the pre-treated cooling system flowmeter, and the real-time matching of the cooling efficiency and the load change is achieved through a PID controller. For the equipment in the high-risk area, when the boundary of the risk area is updated, the detachable heat insulation baffle is automatically closed, and the opening and closing state of the baffle is triggered in real time by the risk area division result, improving the local heat insulation performance by more than 30% and reducing the heat conduction rate of the equipment to the surrounding area.

[0056] Preferably, the execution status of the control instruction is collected in real time through the sensor network. After being processed by the data quality evaluation mechanism defined in step S1, it is stored in the distributed real-time database to form a closed-loop feedback data interaction. Based on the execution feedback data, the dynamic weight factor and constraint parameters in the multi-objective optimization model are automatically calibrated to achieve the adaptive correction of the equipment redundancy constraint.

[0057] Preferably, the control effect is quantitatively evaluated through key performance indicators. The load migration balance degree reflects the evenness of the load distribution between the medium-risk area and the equipment in the safe area. The formula is , calculated through the migrated load data and the redundancy constraint parameters. The closer the value is to 1, the more balanced the distribution is. The compliance rate of the constraint compliance evaluates the degree of compliance of the control instruction with the constraint conditions. The formula is , which is the core index of the system reliability.

[0058] Preferably, the control instruction and the execution data are indexed by the time stamp, and are associated with information such as the risk area coordinates, the optimization plan ID, and the equipment importance weight from the spatio-temporal data cube for cross-step data association storage, specifically including instruction association information, execution status data, effect evaluation data, and exception records, forming a complete fault traceability chain.

[0059] Preferably, the instruction association information includes the optimization plan ID, the equipment ID, the target load rate, and the adjustment rate; the execution status data includes the actual load rate, the real-time temperature, and the cooling system flow rate; the effect evaluation data includes the risk level change, the efficiency change amount, and the redundancy change; and the exception record includes recording the exception type, the occurrence time, and the fallback strategy if the constraint fallback is triggered.

[0060] The beneficial effects of the present invention compared with the prior art are as follows:

[0061] 1. The present invention integrates the safety monitoring data and the equipment operation data through the distributed real-time database, constructs a three-dimensional dynamic temperature field cloud map, realizes the spatial visualization and real-time quantitative evaluation of the thermal safety risk, and solves the problem of data islands in the traditional system;

[0062] 2. The present invention establishes a normalized objective function that includes safety, efficiency, and redundancy, and through dynamic weight factors , , associates the risk level with the scheduling requirements in real time, breaks through the adaptability limitations of existing static weights, automatically raises the priority of the safety objective during high-temperature anomalies, and ensures the dynamic balance between safety and efficiency;

[0063] 3. The present invention adopts asymmetric load migration and cooperative control, designs a gradient load reduction formula for equipment in high-risk areas, quickly reduces the risk of thermal runaway while retaining emergency redundancy; distributes the load to the safe area based on the redundancy ratio, preferentially selects equipment in the efficient interval, and reduces efficiency losses. Synchronously link the cooling system and the heat insulation structure to form a "monitoring - optimization - execution" closed loop, and solve the problem of the separation of safety and economy in traditional load transfer strategies;

[0064] 4. The present invention ensures that the calculation time per single time step is ≤ 500 ms through hierarchical adaptive grid discretization, data quality assessment mechanism, and distributed parallel computing, meeting the requirements of industrial real-time monitoring; the constraint verification and exception fallback mechanism improves the system robustness and reduces the accident risk caused by parameter overrun. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is the method flow chart of the present invention;

[0066] Figure 2 is the flow chart of dynamic thermodynamic field modeling and risk area division of the present invention;

[0067] Figure 3 is the architecture diagram of multi-objective optimization model construction and dynamic weight adjustment of the present invention;

[0068] Figure 4 is the execution flow chart of the asymmetric load migration strategy of the present invention;

[0069] Figure 5 is the schematic diagram of the data quality assessment mechanism of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0070] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0071] Embodiment 1

[0072] Refer to Figure 1, a multi-objective constraint solving method for coordinated optimization of fire monitoring and generator sets, the specific implementation process is as follows:

[0073] Specifically, a distributed sensor network is deployed in key areas of the factory to collect safety monitoring data and equipment operation data in real time. The safety monitoring data includes regional temperature values, gas concentration parameters and image data. The equipment operation data includes generator set load, steam system pressure, cooling system flow, and electrical system current, which are transmitted to distributed computing devices in real time through the Modbus / TCP protocol.

[0074] Specifically, the regional temperature value is output in real time by temperature sensors deployed in the generator stator, transformer oil pillow, and cable trench. The gas concentration parameters are detected and output by smoke concentration sensors installed in the distribution room and cable interlayer. The image data is extracted by the flame monitoring camera aimed at the boiler furnace and the heat-prone flange interface through the image recognition algorithm to extract characteristic parameters such as flame area and spread speed.

[0075] Specifically, in the data preprocessing stage, a 5-point sliding average filtering algorithm is used to eliminate high-frequency noise for temperature and gas concentration data, a sensor spatial position mapping table is constructed based on the three-dimensional coordinates of the plant area, and the Kriging interpolation method is used to interpolate single-point data into a continuous temperature field and concentration field of a 1m×1m×1m grid to form a spatial thermal safety parameter field; the flame image is extracted using the ResNet-50 convolutional neural network to extract features including edges and textures, the flame boundary is identified using the Otsu threshold segmentation algorithm, and the three-dimensional coordinates of the geometric center are calculated to provide spatial coordinate data for risk area positioning.

[0076] Specifically, during data preprocessing, the sensor timestamp is calibrated through the IEEE 1588 global clock synchronization protocol to ensure that the timestamp error of multi-source data is less than 1ms; linear interpolation is used to fill in missing data points, and invalid data that exceeds the physical limit of the equipment is eliminated, such as load > 110% of the rated capacity; heterogeneous data including Modbus and OPC UA are converted into a unified industrial data model, and the standardized label of "device ID-parameter type-timestamp" is defined to establish a data quality assessment mechanism.

[0077] refer to Figure 5 The data quality assessment mechanism includes completeness, accuracy, consistency and timeliness assessment. In the completeness assessment, the missing data points within 1 minute are counted, and the sensor data with a missing rate of >10% is marked as "suspicious"; the accuracy assessment is verified through the physical limit threshold of the equipment, and the data exceeding ±20% is marked as "invalid"; the consistency assessment compares the real-time values of the same parameter in different data sources. When the deviation is >3%, the manual verification process is triggered and marked as "suspicious"; the timeliness assessment checks whether the data update cycle is >2s according to the device type setting, and the data with a delay of more than 2 cycles is marked as "invalid".

[0078] Specifically, based on the preprocessed data, a hierarchical adaptive grid strategy is used to discretize the factory area: first, the factory area is divided into three grid units: core heat-generating equipment, fluid transmission components, and power transmission components according to the equipment topology. The core heat-generating equipment includes the generator stator, boiler furnace, and heat-prone connecting components including flanges and valves, which use a 0.05m millimeter-level fine grid. Non-critical areas include ordinary walls and channels, which use a 1m coarse grid. When the equipment load suddenly increases by more than 10% or the abnormal temperature rise rate is ≥5℃ / min, the local encryption algorithm is automatically triggered to refine the target area grid to 0.02m.

[0079] refer to Figure 2 , the heat source intensity of each grid unit is calculated through heat source modeling, and the three-dimensional non-steady-state heat conduction equation is solved by numerical simulation method in combination with boundary conditions to generate temperature field cloud map. The risk area is divided by combining safety monitoring data and equipment thermal tolerance parameters, and the risk area division results are stored in real time in a distributed real-time database.

[0080] Specifically, the heat source modeling divides the heat sources in the plant into electric conversion heat sources, fluid transmission heat sources, and contact loss heat sources according to the equipment functions and heating mechanisms. For the electric conversion heat sources, the heat generation is calculated by the electromagnetic loss formula Calculate heat value in real time, The current real-time load of the unit is 240MW. Rated power generation efficiency ; Aiming at the magnetic loss heat source, integrate the copper loss Iron loss Real-time calculation, where current I=1500A, winding resistance R=0.01Ω, magnetic flux density B=1.2T, frequency f=50Hz; for fluid transmission heat source, the heat dissipation is calculated by the fluid energy equation Calculation, where the steam mass flow rate m = 200t / h, specific heat capacity , temperature drop .

[0081] Specifically, the boundary condition setting integrates the physical characteristics of the environment including the ambient temperature, wind speed, and equipment surface material, and the heat dissipation parameters of the equipment including the cooling medium flow rate, the thickness of the insulation layer, and the thermal conductivity.

[0082] Specifically, the ambient heat dissipation adopts Newton's cooling formula , equipment surface convection heat transfer coefficient , real-time ambient temperature ; The efficiency of the cooling system is determined by Calculation, cooling medium water density , volume flow , inlet and outlet temperature , ; The thermal resistance of the heat insulation structure is determined according to Fourier's law , , the thickness of the heat insulation layer , and the thermal conductivity .

[0083] Specifically, the numerical solution of the three-dimensional unsteady heat conduction equation is achieved by numerical simulation methods, through discretization in the spatial and time domains, construction of algebraic equations, and solution of large-scale matrices. Based on the law of conservation of energy, the three-dimensional unsteady heat conduction equation is established as follows: , where is the material density (kg / m³), and the density of the generator stator copper winding in the equipment inventory , transformer oil , is the specific heat capacity at constant pressure (J / (kg・K)), which is dynamically corrected with temperature. For example, for the copper winding , k is the thermal conductivity (W / (m・K)), k = 0.15 for the equipment surface coating and k = 0.2 for the cable insulation layer. Q is the heat source intensity (W / m³) defined in step S2, including heat generated by electrical conversion, magnetic loss heat, and fluid transfer heat.

[0084] Specifically, the spatial discretization method adaptively selects the discretization method according to the grid type and accuracy requirements. The finite volume method is used for the core equipment, and the finite difference method is used for non-critical areas. The finite volume method divides the target area into control volumes, and integrates the control equation for each grid cell i to obtain the discrete form: , where is the control volume, is the interface area, is the interface spacing, is the temperature of adjacent grids. For complex geometric structures such as the generator stator, unstructured tetrahedral grids are used, and the grids are automatically refined at locations with large curvatures. The minimum grid size is 0.02m, and the grid quality is verified by the Jacobian determinant ≥ 0.6; for regular areas such as cable trenches, Cartesian grids are used, and the gradient term is discretized using second-order central differences: , with a grid size of 1m×1m×1m, and local refinement to 0.05m is triggered during abnormal fluctuations.

[0085] Specifically, the time domain is discretized using an implicit difference scheme to handle the time partial derivative, balancing stability and solution efficiency: , with an initial time step , which is automatically reduced to 0.2s when the heating rate ; The Picard iteration method is used to linearize the non-linear terms, and the iteration convergence criterion is .

[0086] Specifically, after the discretization in the spatial and temporal domains, the algebraic equations are constructed and the large-scale matrix is solved. The discrete equations are organized into the form of linear algebraic equations. , where the coefficient matrix includes diffusion terms, convection terms, and source terms. The sparse storage technology is used to control the proportion of non-zero elements below 3%. For large-scale grid systems, the domain decomposition parallel algorithm is enabled in the distributed scenario. First, the computational domain is divided into 8 computational sub-domains through the METIS graph partitioning technology. Each node is responsible for the matrix operations of the independent sub-domain, and the boundary data is exchanged through a high-speed network to achieve load balancing.

[0087] Specifically, the preconditioned conjugate gradient method is adopted in the solution process. First, the coefficient matrix is subjected to incomplete LU decomposition. By retaining the main diagonal elements and some sub-diagonal elements, the convergence is accelerated while maintaining the sparsity of the matrix. During the iterative process, the calculation is terminated when the residual norm drops to or below. The time consumption for solving a single time step is controlled within 200 ms for a grid scale of 100,000.

[0088] Specifically, in the numerical solution process of the three-dimensional unsteady heat conduction equation, it is divided into three categories according to the physical characteristics of the device and environmental parameters: For the scenario where the surface temperature of the device is known, such as the components directly controlled by the cooling system, the first type of boundary condition is adopted, and the measured temperature value is directly assigned to the boundary grid cells. For example, the surface temperature of the stator cooling water jacket of the generator is measured in real time by a thermocouple as , and this temperature value is directly used as the boundary node temperature of the corresponding grid , without iterative calculation; for the area covered by the heat insulation structure, the second type of boundary condition is adopted to define the heat flux density , and according to Fourier's law , combined with the physical parameters of the heat insulation layer, the boundary heat flux is calculated: the thermal resistance of the heat insulation layer , where the thickness , the thermal conductivity , and the boundary heat flux density is converted to , where is the real-time acquisition value of the environmental temperature ; for the surface of the device exposed to the environment, such as the flange interface without active cooling, the third type of boundary condition is adopted to describe the natural convection heat dissipation, and the convective heat transfer is quantified through Newton's cooling formula , where the convective heat transfer coefficient is preset based on the surface material of the device and the environmental wind speed: for a smooth metal surface, take , for a rough surface, take , and the environmental wind speed is calibrated in real time by a wind speed sensor, with a default windless condition .

[0089] Specifically, the risk area division strictly follows the threshold system: when the temperature of the grid cell exceeds 120% of the equipment-level threshold and lasts for 30 seconds or the detected flame area is determined as a high-risk area, triggering a red warning and marked as the priority area for load adjustment; when the temperature is between 100% - 120% of the equipment-level threshold or the smoke concentration ≥ 5% LEL, it is a medium-risk area and included in the candidate range for load migration; when the temperature is below 100% of the threshold and there is no abnormal flame or smoke, it is a safe area and serves as the target receiving area for load migration.

[0090] Specifically, the risk area division results are stored in a distributed real-time database in the form of a spatio-temporal data cube containing temperature values, risk levels, heating rates, and change trends. Each grid cell is associated with three-dimensional coordinates, equipment ledger information, and risk evolution history, supporting multi-dimensional applications, including three-dimensional visual monitoring, spatio-temporal trajectory tracing, and intelligent warning triggering.

[0091] Reference Figure 3 , based on the risk area division results in the distributed real-time database and the preprocessed equipment operation data, three core constraint parameters including safety thresholds, minimum efficiency indicators, and equipment redundancy are defined, a normalized objective function is constructed, and through a dynamic weight factor adjustment mechanism associated with risk levels and scheduling requirements, the dynamic decoupling of multi-objective constraint conditions is achieved, that is, the value logic of the weight factor is directly associated with the real-time threshold offset of the constraint parameters, and the optimization results are stored in the distributed real-time database.

[0092] Specifically, the safety thresholds directly refer to the risk area threshold system, including 130°C for the generator stator winding, 85°C for the transformer oil, and 70°C for the cable trench; the minimum efficiency indicator is set to ≥ 90% for the plant's power generation efficiency according to the grid scheduling requirements and dynamically calibrated with real-time efficiency data and the scheduling curve, and the current efficiency is 92%; the equipment redundancy requires that the single-device load rate ≤ 85% of the rated capacity, that is, a load adjustment margin of ≥ 15% is reserved.

[0093] Specifically, the normalized objective function quantifies multiple objectives through the safety risk ratio SRR, efficiency loss ratio ELR, and redundancy retention ratio RRR. The safety risk ratio , where the equipment importance weight is set as: for the generator set w = 0.8, for the transformer w = 0.7, for the auxiliary equipment w = 0.3, and the material tolerance limit takes the ignition point of the insulating material; the efficiency loss ratio efficiency before optimization , current efficiency is calculated in real-time with load adjustment, and the minimum index ; the redundancy retention ratio minimum load rate , the current total system redundancy is 45MW, and the rated total capacity is 300MW.

[0094] Specifically, the dynamic weight factor adjustment mechanism adopts a weight adjustment algorithm that is real-time linked with the risk level, scheduling requirements, and redundancy status, corresponding to the safety risk ratio , efficiency loss ratio , redundancy retention ratio of the weight factor: when the area ratio of the high-risk area exceeds 5%, the safety weight increases; the emergency degree index of scheduling requirements takes 1 during the peak period and 0.3 during the valley period, and the efficiency weight ; when , the redundancy weight automatically increases, restricting the single-device load adjustment to ≤ 80% of the rated capacity.

[0095] Specifically, after completing the definition of constraint parameters, the construction of the normalized objective function, and the adjustment of the dynamic weight factor, the multi-objective constraint conditions are dynamically decoupled through the non-dominated sorting genetic algorithm II. Set the population size to 100, the number of iterations to 50, the crossover probability to 0.8, and the mutation probability to 0.1. Select the solutions with a safety risk ratio decrease ≥ 15%, an efficiency loss ratio ≤ 0.15, and a redundancy retention ratio ≥ 0.6. Finally, generate the load adjustment instructions: for example, the load of Generator No. 3 decreases from 80% to 65%, and the load of Unit No. 4 increases from 60% to 75%.

[0096] Reference Figure 4 , based on the optimization results and risk area division results in the distributed real-time database, implement the asymmetric load migration strategy. Through the generation of main equipment load adjustment instructions, the linkage control of auxiliary systems, the real-time acquisition and feedback of the execution status of control instructions, combined with the quantitative evaluation of control effects and cross-step data associated storage, achieve the collaborative control goal of controllable safety risks, maintaining system operation efficiency, and optimizing equipment redundancy. The relevant control instructions, execution data, and evaluation results are interacted throughout the process through the distributed real-time database.

[0097] Specifically, the asymmetric load migration strategy implements gradient load reduction for the equipment in the high-risk area. Taking Generator No. 3 with a temperature of 150°C as an example, its safety threshold is 130°C, and its tolerance limit is 180°C. Calculate the load reduction amount according to the formula , that is, the load decreases from 240MW to 192MW, 64% of the rated capacity, and the adjustment rate is controlled at 10% of the rated capacity per minute, that is, 30MW / min, and the load reduction is given priority to the auxiliary equipment; for the equipment in the medium-risk area of the safe area, the incoming load is allocated according to the redundancy ratio. For example, the real-time redundancy of Transformer No. 5 , allocate the incoming load according to the redundancy ratio: , after the migration load, the load rate increases from 70% to 74%, which is less than the 85% threshold. Equipment with an efficiency characteristic curve in the 60%-80% high-efficiency area is preferentially selected.

[0098] Specifically, to match the synchronous triggering of the auxiliary system linkage control with the load adjustment of the main equipment, the flow rate of the cooling system of the No. 5 transformer with an increased load automatically increases by 15%, from rises to , and the real-time matching of the cooling efficiency and the load change is achieved through a PID controller; the No. 3 generator in the high-risk area triggers the closing instruction of the heat insulation baffle, improving the local heat insulation performance by 35%, and the heat conduction rate drops from drops to .

[0099] Specifically, the real-time acquisition of the execution status of the control instruction is through the deployed sensor network, which captures the main equipment parameters, the status of the auxiliary system, and the environmental parameters in real time. Main equipment parameters: the actual load of the No. 3 generator is 192MW, the target is 192MW, and the deviation is 0%; the stator temperature is 145°C, which drops by 5°C compared with before the adjustment; Auxiliary system status: the cooling flow rate of the No. 5 transformer , target value , PID control accuracy ±2%, heat insulation baffle status "closed", triggered by the boundary of the risk area; Environmental parameters: the temperature in the cable trench is 68°C, below the safety area threshold of 70°C, the steam pipeline pressure is 12MPa, the rated value is 15MPa, and the redundancy is 20%.

[0100] Specifically, the cross-step data association storage associates the verification results with the instruction association information, the effect evaluation data, and the exception records and stores them in the distributed real-time database. Instruction association information: optimization plan ID-20250513-001, equipment ID-003, target load rate 64%, adjustment rate 10% / min; Effect evaluation data: change in risk level: the area of the No. 3 generator drops from medium risk to the safety area, change in efficiency: +0.5%, change in redundancy: the total system redundancy increases from 45MW to 57MW; Exception record: If a constraint rollback is triggered, the exception type, occurrence time, and rollback strategy are detailedly recorded to form a complete traceability chain, such as the historical record shows that there is no abnormal event at 14:00 on May 13, 2025.

[0101] Specifically, based on the execution feedback data, the multi-objective optimization model is automatically calibrated: if the actual temperature of the No. 3 generator drops by 5°C every 10 minutes and the model predicts 8°C, and the temperature drop rate after load reduction is lower than expected, the system will correct the heat conduction coefficient of this equipment, from adjusted to , optimize the load reduction calculation for the next time step; if the redundancy calculation deviation of the No. 5 transformer > 5%, the rated capacity in the ledger is 300MW, and the actual detection is 310MW, trigger the equipment rated parameter verification process and update the equipment ledger data in the distributed database.

[0102] Embodiment 2

[0103] Reference Figure 1 , this embodiment is for the nuclear power supporting generator set. On the basis of Embodiment 1, this embodiment strengthens the data accuracy and abnormal response mechanism to meet the collaborative control requirements of high safety level scenarios.

[0104] In the original monitoring system, the temperature sensor uses a point-type temperature sensor to obtain the equipment temperature data. The point-type temperature sensor can measure the temperature of a specific position of the equipment in real time, but its monitoring range is limited, and it can only provide discrete single-point temperature information, making it difficult to intuitively present the overall temperature distribution of the equipment. To make up for the deficiencies of the traditional point-type temperature sensor, this embodiment adds a variety of sensors.

[0105] Specifically, the added variety of sensors include an infrared thermal imager, a vibration monitoring system, and an optical fiber distributed temperature sensing system. The infrared thermal imager performs a full-coverage scan of the equipment surface every 15 minutes, and the generated three-dimensional thermal imaging map can intuitively present the temperature distribution on the equipment surface and obtain the temperature data of each point on the equipment surface; the vibration monitoring system, which collects bearing vibration data in real time, including vibration frequency and amplitude data, to achieve early fault warning; an optical fiber distributed temperature sensing system is laid along the cable trench, which, as a supplement to the traditional temperature sensor, can realize continuous distributed monitoring of the cable temperature and provide the temperature data along the cable.

[0106] Specifically, in the data preprocessing stage, the temperature data sources include the temperature data provided by the original temperature sensor, as well as the temperature data provided by the added infrared thermal imager and optical fiber distributed temperature sensing system. To obtain more accurate and comprehensive equipment surface temperature information, the Kalman filter algorithm is used to fuse the infrared thermal imaging data with the point-type temperature sensor data to generate a higher-precision equipment surface temperature field.

[0107] Specifically, for the vibration data, wavelet transform is used for frequency domain analysis to extract 12 characteristic frequencies in the range of 0.5~10kHz, and then the support vector machine classifier is used to process the characteristic frequencies and identify the fault types including bearing wear and misalignment.

[0108] After data preprocessing, it enters the comprehensive data quality assessment link. The data quality assessment is strengthened from four dimensions, and the four dimensions include timeliness, integrity, accuracy, and consistency.

[0109] Specifically, in terms of timeliness assessment, the cycle is shortened to 1 second. For temperature data, vibration data, or other equipment operation data, once the data delay exceeds 2 seconds, it will be marked as "invalid" and trigger the replacement with backup sensor data.

[0110] Specifically, in terms of integrity assessment, the data missing situation within 1 minute is counted. If the missing rate exceeds 10%, it will be marked as "suspicious".

[0111] Specifically, the accuracy assessment is achieved through the verification of the physical limit threshold of the equipment and the analysis of the correlation between adjacent sensor data. Data beyond the reasonable range will be marked as "invalid" to ensure that the data conforms to the physical laws of equipment operation.

[0112] Specifically, for consistency assessment, the real-time values of the same parameter in different data sources are compared. When the deviation of vibration data obtained by different sensors exceeds 3%, the manual verification process is triggered and marked as "suspicious".

[0113] Specifically, after the four-dimensional assessment, the data is given quality stamps of "valid, invalid, suspicious". Valid data is directly used for dynamic thermodynamics field modeling, invalid data triggers an alarm and calls for historical replacement, and suspicious data is attached with a warning sign and the weight of dynamic thermodynamics field modeling is reduced, which is consistent with the data quality assessment mechanism of Embodiment 1.

[0114] Reference Figure 2 , in the dynamic thermodynamics field modeling, for the core equipment areas including the reactor coolant pump and the generator stator, a 0.01m ultra-fine grid modeling is adopted to capture the local thermal gradient at the level of, and the grid accuracy of non-critical areas is maintained at 0.05m.

[0115] Specifically, during the equipment operation, when it is monitored that the equipment load fluctuation > 5% or the temperature change rate ≥ 3℃ / min, the local grid encryption algorithm is triggered, and the grid will be automatically encrypted to 0.005m. At the same time, in order to ensure the calculation efficiency, the computing node resource allocation will be increased according to the grid encryption situation.

[0116] Specifically, by solving the three-dimensional unsteady heat conduction equation, combining the heat source intensity of each grid unit calculated by the heat source modeling and the boundary conditions including environmental heat dissipation, cooling system efficiency, and thermal resistance of the heat insulation structure, the temperature field data is finally generated.

[0117] Specifically, the temperature field data is based on grid units and details the real-time temperature information at different positions, reflecting the thermal state distribution of the equipment and the area.

[0118] Specifically, the risk warning mechanism is adjusted according to the temperature field data. When the temperature in the medium-risk area first exceeds 110% of the threshold, the pre-assessment process will be triggered 10 seconds in advance to calculate the temperature change trend under different load adjustment schemes.

[0119] Specifically, when dividing risk areas, the "thermal stress accumulation index" is introduced, and its calculation formula is , where represents the temperature of the device at time t, is the reference temperature, is the maximum temperature that the device material can withstand. By integrating the deviation degree of the device temperature from the safety threshold, when TSCI > 0.8, even if the temperature does not reach the traditional threshold, a warning will be triggered, thus effectively avoiding fatigue damage of the material caused by long-term thermal stress.

[0120] Reference Figure 3 , and the constraint parameters are extended and strengthened in the construction of the multi-objective optimization model.

[0121] Specifically, the expansion of the constraint parameters is based on the fused temperature data. On the basis of the three types of core constraint parameters in Embodiment 1, a "thermal stress safety threshold" is newly added, which specifically includes that the bearing temperature gradient ≤ 10°C / m and the stator winding temperature change rate ≤ 5°C / min.

[0122] Specifically, the strengthening of the constraint parameters is achieved by modifying the redundancy constraint to improve the safety redundancy. The maximum load rate of a single device is reduced from 85% to 80%, the total system redundancy requirement ≥ 25%, and a safety margin of bearing temperature gradient ≤ 10°C / m is newly added.

[0123] Based on the above constructed multi-objective optimization model, the dynamic weight adjustment mechanism introduces scenario adaptive logic in this embodiment.

[0124] Specifically, when the redundancy retention ratio is , the redundancy weight is automatically increased to 0.4, and the "device health protection mode" is activated, restricting the load adjustment range of a single device not to exceed 5% of the rated capacity.

[0125] Specifically, for the safety weight a time factor is introduced, and the calculation formula becomes , where represents the area of the high-risk area, represents the total area of the plant area, is the duration of the high-risk state, realizing the non-linear growth of the risk weight.

[0126] After the dynamic weight adjustment mechanism is optimized, the solution algorithm adopts the improved non-dominated sorting genetic algorithm II.

[0127] Specifically, the Non-dominated Sorting Genetic Algorithm II increases the population size to 200, raises the number of iterations to 70 generations, sets the crossover probability to 0.9, adjusts the mutation probability to 0.05, and at the same time introduces the "elitist strategy", retaining 5% of the optimal individuals in each generation of iteration.

[0128] Specifically, in response to the requirements of the nuclear power scenario, a new index of "thermal stress balance degree" is added to the optimization objective: , and the constraint condition is , ensuring uniform distribution of the thermal stress of the equipment.

[0129] In terms of the asymmetric load transfer strategy, based on the optimization scheme output by the multi-objective optimization model, when implementing the load reduction strategy for the equipment in the high-risk area, the equipment collaborative control mechanism is started synchronously.

[0130] Refer to Figure 4 , based on the load adjustment instruction generated by the multi-objective optimization model, when reducing the load of the equipment in the high-risk area, strictly limit the load reduction rate of the core equipment, requiring it not to exceed 5% of the rated capacity per minute to avoid failures caused by sudden changes in mechanical stress.

[0131] Specifically, when multiple devices need to reduce the load simultaneously, they are sorted according to the "safety risk ratio × equipment importance coefficient", and the priority is main pump > generator > auxiliary equipment, ensuring the safety of key equipment is guaranteed first.

[0132] Specifically, while implementing the above load transfer strategy, the bearing temperature and vibration data are collected in real time as the key feedback parameters for equipment collaborative control. If the vibration value is greater than 8 mm / s after the flow rate of the cooling system is adjusted, automatically roll back 10% of the load adjustment amount and mark the "vibration abnormal" code.

[0133] Specifically, the index of "control energy consumption" is introduced , and on the premise of meeting the safety constraints, the adjustment scheme with the minimum control energy consumption is preferentially selected.

[0134] Specifically, on the basis of equipment collaborative control, for extreme scenarios where the proportion of the high-risk area exceeds 15% and the duration is greater than 5 minutes, a three-level emergency response mechanism is established.

[0135] Specifically, the first-level response is to start the standby cooling system and increase the coolant flow rate by 30% to quickly reduce the surface temperature of the equipment; the second-level response is to cut off the non-critical load and give priority to ensuring the power supply of the safety system to ensure the normal operation of key equipment in case of emergency; the third-level response is to trigger the emergency shutdown procedure, release the carbon dioxide fire extinguishing system at the same time, link the fire control monitoring system to lock the fire source location, and record the parameter changes of the whole process through the distributed real-time database to provide a complete data chain for post-fault traceability.

[0136] Embodiment 3

[0137] Reference Figure 1 , in this embodiment, for the hybrid energy scenario with photovoltaic and wind power access, the load migration strategy is optimized on the basis of Embodiment 1 to adapt to the fluctuations of intermittent power sources.

[0138] Specifically, the newly added data sources include real-time data of photovoltaic power plants, operation parameters of wind farms, and grid-side data. The real-time data of photovoltaic power plants include inverter efficiency, photovoltaic array temperature, and DC / AC conversion efficiency; the operation parameters of wind farms include wind turbine speed, converter temperature, and gearbox oil temperature; the grid-side data include real-time electricity price, grid frequency, and voltage fluctuation.

[0139] Specifically, in a photovoltaic power plant, the inverter efficiency is measured by a high-precision power sensor, the photovoltaic array temperature is measured by a thermistor, and the DC / AC conversion efficiency is obtained by a power quality analyzer. These sensors are respectively installed inside the inverter, on the surface of the photovoltaic array, and at key circuit nodes.

[0140] Specifically, in a wind farm, the wind turbine speed is measured by a speed sensor, and the converter temperature and gearbox oil temperature are measured by thermal resistors. The thermal resistors are respectively installed at the position of the converter heat sink close to the power module and at the middle of the gearbox oil and close to the gear meshing position.

[0141] Specifically, the grid-side data is collected through a grid monitoring terminal. The real-time electricity price is obtained from the power market trading platform. The measurement accuracies of the grid frequency and voltage fluctuation are ±0.01Hz and ±0.5% respectively.

[0142] Specifically, the collected data is transmitted to a distributed real-time database through an industrial data transmission network. At the same time, a three-dimensional index structure of "time series - data source - data type" is used to store the data.

[0143] Specifically, a new energy power prediction module is added in the data preprocessing stage. This module is based on a long short-term memory artificial neural network to implement 15-minute prediction data as input parameters for dynamic weight adjustment. The historical power data of nearly one year is used for training, and the data scale reaches more than 100,000 pieces. The number of training rounds is set to 500 rounds, and the mean square error is selected as the loss function.

[0144] Specifically, in addition to the historical power data, the input parameters also include meteorological data such as light intensity and wind speed. These data are obtained through real-time collection by sensors of meteorological stations and new energy power stations. At the same time, to eliminate the impact of power sudden changes on load migration, the real-time data of photovoltaic power plants, the operation parameters of wind farms, and the grid-side data are subjected to fluctuation smoothing processing using a moving average filter with a window size of 15 minutes.

[0145] Specifically, after the model training is completed, the model is evaluated by calculating metrics including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE).

[0146] Specifically, in terms of risk assessment, the device-level thermal safety threshold adds the thermal safety threshold for new energy devices. The thermal safety threshold for new energy devices includes a photovoltaic inverter temperature threshold of 60°C, and a high-risk warning is triggered when it exceeds 80°C; a wind power converter temperature threshold of 70°C, and a gearbox oil temperature threshold of 75°C; the State of Charge (SOC) of the energy storage battery pack is maintained in the range of 20% - 80%, and a risk warning is triggered when the range is exceeded.

[0147] Specifically, the safety zone division is expanded by including the redundancy of new energy devices in the safety zone assessment indicators. It is required that the photovoltaic power station retains at least 20% of the rated capacity as a backup; when the wind power prediction power fluctuation > 15%, the safety zone range is automatically expanded to increase the reserve of backup capacity.

[0148] Reference Figure 2 , according to the device thermal safety threshold, the redundancy of new energy devices, and the real-time operation data, the plant area is divided into three risk areas: high-risk area, medium-risk area, and safety area.

[0149] Specifically, the high-risk area is the area where the device temperature exceeds the thermal safety threshold, the SOC of the energy storage battery pack is outside the normal range, and the wind power prediction power fluctuation amplitude exceeds 30%; the medium-risk area is the area where the device temperature is close to the thermal safety threshold, the redundancy of new energy devices is less than 30%, or the wind power prediction power fluctuation amplitude is between 15% - 30%; the safety area is the area where the device operates normally, the redundancy of new energy devices is sufficient, and the power fluctuation amplitude is less than 15%.

[0150] Specifically, for different risk areas, differential response measures are formulated, including immediately starting load transfer and equipment protection measures in the high-risk area, strengthening monitoring and preparing to adjust the load in the medium-risk area, and maintaining the normal operation state in the safety area.

[0151] Reference Figure 3 , when constructing the multi-objective optimization model, new constraint conditions are added, including system reserve capacity ≥ 20%, including the redundancy of new energy devices; the charge and discharge depth of the energy storage system ≤ 70% to avoid battery life attenuation; the new energy consumption rate ≥ 90%, and the output of photovoltaic and wind power is preferentially dispatched.

[0152] Specifically, the "Grid Peak-Valley Period Factor" PeakValley is introduced into the dynamic weight factor adjustment mechanism, and the calculation formula is , where the peak period is from 8:00 to 22:00 , the efficiency weight increases, and more attention is paid to the power generation efficiency; the valley period is from 22:00 to 8:00 , Decrease.

[0153] Considering that the fluctuations in new energy output will affect system security, a dynamic adjustment mechanism for safety weights based on new energy output fluctuations is established.

[0154] Specifically, the fluctuations in new energy output are measured by calculating the change rate of the actual output of new energy within a 15-minute statistical period. The formula for calculating the amplitude of new energy output fluctuations is: , where is the output of new energy at the current moment, is the output of new energy 15 minutes ago, is the maximum output of this new energy device.

[0155] Specifically, when the amplitude of new energy output fluctuations > 20%, the safety weight is additionally increased by 0.1 on the original basis to enhance the system's anti-interference ability. The specific adjustment logic includes: if the amplitude of new energy output fluctuations ≤ 10%, maintain the base value ; if 10% < the amplitude of new energy output fluctuations ≤ 20%, ; if the amplitude of new energy output fluctuations > 20%, .

[0156] Specifically, this mechanism works in coordination with the regulation during peak and valley periods of the power grid: during peak periods, if the amplitude of new energy output fluctuations is large, is significantly increased to give priority to ensuring system security; during valley periods, even if the amplitude of new energy output fluctuations is large, is only moderately increased to balance economy and security.

[0157] Specifically, new optimization objectives are added to adapt to the new energy grid connection scenario. On the basis of the original safety risk ratio, efficiency loss ratio, and redundancy retention ratio, the "new energy consumption degree" index is added: , and the constraint condition is ; considering the economy on the power grid side, the "power purchase cost" index is added: , where represents the power purchase power at the i-th time period, is the real-time electricity price at the i-th time period, is the time period duration, and n is the total number of statistical time periods.

[0158] Specifically, based on the characteristics of the peak-valley electricity price difference, during valley periods when the electricity price is low, the gas turbine units are preferentially scheduled to operate at full load to make full use of the low-price power periods, and at the same time, part of the electric energy is stored in the energy storage system; during peak periods when the electricity price is high, gas power generation is reduced, the energy storage is preferentially called to discharge, and the proportion of new energy output is increased, so as to reduce the overall power purchase cost through the peak-valley electricity price difference and improve economic benefits.

[0159] Reference Figure 4 In the asymmetric load transfer strategy, a new new energy priority dispatching mechanism is added: when the temperature of the photovoltaic inverter is < 50 °C and sufficient sunlight is predicted, stable loads are preferentially allocated to the energy storage battery pack to reduce the thermal stress of traditional units; when the power of the wind power converter exceeds 110% of the rated power and is determined to be overloaded, the load is automatically transferred to a gas generator with a high redundancy, and the adjustment rate ≤ 15% of the rated capacity / minute to avoid wind power disconnection from the grid.

[0160] Specifically, the auxiliary system linkage control adds the collaborative control of the energy storage system. Based on the asymmetric load transfer strategy, a three-level control strategy is established, including the peak shaving and valley filling mode, the reserve capacity mode, and the risk response mode. In the peak shaving and valley filling mode, it discharges during peak periods and charges during valley periods, and the charge and discharge power ≤ 50% of the rated capacity; in the reserve capacity mode: 30% of the capacity is reserved as a reserve for sudden risk response; in the risk response mode, when a high-risk area is triggered, the energy storage power is preferentially released to reduce the adjustment range of traditional units.

[0161] Specifically, according to the status of different energy equipment and the results of risk area division, the load of the main equipment is adjusted, and the new energy prediction data is updated every 15 minutes to recalculate the optimal load distribution plan.

[0162] Specifically, when recalculating, the factors to be comprehensively considered include the predicted value of new energy power, the thermal safety threshold of equipment, the system reserve capacity, the status of the energy storage system, the new energy consumption rate, and the power purchase cost. An optimal load distribution plan is obtained by using an optimization algorithm.

[0163] Specifically, the new load distribution plan is sent to the execution mechanisms of each device through control instructions to achieve precise adjustment of the device load.

[0164] Specifically, when the predicted power of the wind power drops suddenly by > 30%, the load increase program of the gas turbine unit is started 30 minutes in advance, and the increase rate ≤ 5% of the rated capacity / minute; when the temperature of the photovoltaic array exceeds 75 °C, the conversion efficiency of the inverter is automatically reduced to 95% to reduce heat generation by reducing power output.

[0165] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, in any aspect, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

Claims

1. A multi-objective constraint solving method for collaborative optimization of fire monitoring and generator sets, characterized in that, The following steps are involved: S1. Collect safety monitoring data and equipment operation data in real time through the sensor network, access the distributed real-time database through the industrial communication network and standardized protocols, and then pre-process the safety monitoring data and equipment operation data to form structured data and store them in the distributed real-time database; S2. Based on structured data, according to the three-dimensional layout of the plant and the distribution of physical equipment, the plant is discretized into grid units, and then the heat source intensity of each grid unit is calculated through the heat source modeling method. The heat source intensity is combined with the boundary conditions to solve the three-dimensional non-steady-state heat conduction equation to generate a temperature field cloud map. Based on the temperature field cloud map combined with the pre-processed safety monitoring data and the equipment thermal tolerance parameters, the risk area is divided, and the risk area division results are stored in real time in a distributed real-time database; S3. Based on the risk area division results and the pre-processed equipment operation data, three types of core constraint parameters are first defined, including safety threshold, minimum efficiency index, and equipment redundancy. Then, a normalized objective function is constructed. The safety risk ratio, efficiency loss ratio, and redundancy retention ratio in the normalized objective function are weighted through a dynamic weight factor adjustment mechanism that is linked in real time with the risk level and scheduling requirements. This achieves dynamic decoupling of multi-objective constraints, generates optimization results including load adjustment instructions and effect prediction data, and finally stores the optimization results in a distributed real-time database. S4. Based on the optimization results and risk area division results, the asymmetric load migration strategy is first implemented. The load adjustment instructions of the main equipment are generated, and the linkage control of the auxiliary system is synchronously triggered. Then, the execution status of the control instructions is collected and fed back in real time, triggering the triple constraint verification. At the same time, it is combined with the quantitative evaluation of the control effect and the cross-step data association storage.

2. The multi-objective constraint solving method for collaborative optimization of fire monitoring and generator sets according to claim 1, wherein The safety monitoring data includes regional temperature values, gas concentration parameters, and image data. The preprocessing of the safety monitoring data includes: using a noise reduction algorithm to process the regional temperature values and gas concentration parameters, interpolating the single-point monitoring data into regional grid temperature distribution and concentration distribution data in combination with the three-dimensional coordinates of the plant area, and establishing a sensor spatial position mapping table to form a continuous spatial thermal safety parameter field. For image data, a convolutional neural network combined with a threshold segmentation and edge detection algorithm is used to identify the flame boundary and calculate the geometric center coordinates; the preprocessing of the equipment operation data includes: for key parameters reflecting the operating status of the equipment, the sensor clock deviation is calibrated through a global clock synchronization protocol, the missing data is supplemented by a linear interpolation method, and invalid data that exceeds the physical limit of the equipment is eliminated. The heterogeneous data is converted into a unified industrial data model, standardized data labels are defined, and a data quality assessment mechanism is established to add "valid, invalid, suspicious" quality stamps to the preprocessed data.

3. The multi-objective constraint solving method for collaborative optimization of fire monitoring and generator sets according to claim 2, characterized in that, The data quality assessment mechanism includes integrity assessment, accuracy assessment, consistency assessment, and timeliness assessment. In the integrity assessment, the system calculates the proportion of missing data points within the data collection cycle and marks the sensor data with a missing rate exceeding the preset threshold as "suspicious"; the accuracy assessment verifies through the physical limit threshold of the device and analyzes the correlation of adjacent sensor data, marking the data outside the reasonable range as "invalid"; the consistency assessment compares the real-time values of the same parameter from different data sources, triggering a manual verification process and marking it as "suspicious" when the deviation exceeds 3%; the timeliness assessment determines whether the data update cycle meets the regulations according to the device type, marking the data with a delay exceeding 2 cycles as "invalid".

4. The multi-objective constraint solving method for collaborative optimization of fire monitoring and generator sets according to claim 1, characterized in that, The discretization of the grid cells adopts a hierarchical adaptive strategy. The area is divided into functional modules according to the plant equipment topology and process flow. A millimeter-level fine grid is used for core heat-generating equipment and easily heat-generating connection components, and a meter-level coarse grid is used for non-critical areas including fluid transmission components and power transmission components. When the equipment load suddenly increases by more than 10% or the safety monitoring data fluctuates abnormally, the grid is automatically locally encrypted to 0.05 m.

5. The multi-objective constraint solving method for collaborative optimization of fire monitoring and generator sets according to claim 1, characterized in that, The heat source modeling includes heat source type classification and heat source intensity calculation. The heat source type classification divides the plant heat sources into three categories: electric conversion heat sources, fluid transmission heat sources, and contact loss heat sources according to the equipment function and heat generation mechanism. The heat source intensity calculation uses the electromagnetic loss formula for electric conversion heat sources for real-time calculation, and integrates copper loss and iron loss for fluid transmission heat sources based on structured data. For contact loss heat sources, the fluid energy equation is used, and the heat dissipation is calculated through the fluid energy equation .

6. The multi-objective constraint solving method for collaborative optimization of fire monitoring and generator sets according to claim 1, characterized in that The boundary condition setting integrates environmental physical properties including environmental temperature, wind speed, and equipment surface material, and equipment heat dissipation parameters including cooling medium flow rate, insulation layer thickness, and thermal conductivity, as follows: Environmental heat dissipation uses Newton's cooling formula Quantify natural convection heat dissipation, where is the equipment surface temperature, is the real-time environmental temperature, and h is preset based on the equipment surface material and environmental wind speed; The cooling system efficiency is calculated by , where V is the volume flow rate of the cooling medium, , are the inlet and outlet temperatures of the cooling medium; The thermal resistance of the insulation structure is determined according to Fourier's law , where is calculated from the insulation layer thickness d and the thermal conductivity .

7. The multi-objective constraint solving method for collaborative optimization of fire monitoring and generator sets according to claim 1, wherein The risk area division first constructs a risk threshold system that includes device-level thermal safety thresholds and area-level thermal safety thresholds. Based on the risk threshold system, combined with the real-time temperature field cloud map and preprocessed safety monitoring data, the plant area is divided into a high-risk area where the temperature exceeds 120% of the device-level threshold and lasts for ≥ 30 s or the detected flame area ≥ 0.5 m², a medium-risk area where the temperature is in the range of 100% - 120% of the device-level threshold or the smoke concentration ≥ 5% LEL, and a safe area where the temperature is below 100% of the device-level threshold and there is no abnormality.

8. The multi-objective constraint solving method for collaborative optimization of fire monitoring and generator sets according to claim 1, characterized in that, The normalized objective function is constructed based on the conversion of safety risk, efficiency loss, and redundancy retention into dimensionless normalized indicators, including: safety risk ratio , which quantifies the risk of thermal runaway of the system; efficiency loss ratio , which measures the impact of load adjustment on power generation efficiency; redundancy retention ratio , which evaluates the emergency response ability of the system, where is the real-time temperature of the device, is the device-level safety threshold, is the ignition point limit of the material, is the importance weight of the device, is the system efficiency before optimization, is the efficiency under the current load distribution, is the minimum efficiency index, is the rated capacity of the device, is the real-time load, is the minimum load rate.

9. The multi-objective constraint solving method for collaborative optimization of fire monitoring and generator sets according to claim 1, characterized in that, The dynamic weight factor adjustment mechanism adopts a weight adjustment algorithm that is linked in real time with the risk level, scheduling requirements, and redundancy status, corresponding to the weight factors of the safety risk ratio, efficiency loss ratio, and redundancy retention ratio respectively , , , satisfying , where the safety weight is positively correlated with the proportion of the area of the high-risk area , the efficiency weight is positively correlated with the emergency degree index of the scheduling instruction , and the redundancy weight , when the system redundancy retention ratio is automatically increased to limit the load adjustment range.

10. The multi-objective constraint solving method for collaborative optimization of fire monitoring and generator sets according to claim 1, characterized in that The asymmetric load migration strategy is implemented according to the risk level and real-time redundancy of the area where the equipment is located. Reduce load, keep ≥15% rated capacity margin, and adjust importance weights first For core equipment in safe areas and medium-risk areas, the real-time redundancy Proportional distribution of incoming load After migration, the load rate shall not exceed 85% of the rated capacity, and priority shall be given to equipment in the high-efficiency area with a load rate of 60%-80%. During the load adjustment process, the temperature field cloud map and constraint conditions are called in real time for triple verification. If the equipment temperature exceeds the safety threshold, the system efficiency is lower than the minimum index, or the load rate of a single device exceeds the limit, an emergency fallback is automatically triggered and the abnormal code is marked. At the same time, the cooling system medium flow is proportionally adjusted according to the load change, and the removable insulation baffle of the high-risk area equipment is automatically closed based on the risk area boundary update result, thereby improving the local insulation performance by more than 30%.

Images