Dynamic generation and execution system for multi-stage linkage fire-fighting strategy of thermal power plant

Abstract

The invention belongs to the technical field of intelligent fire fighting, and particularly relates to a dynamic generation and execution system for a multi-stage linkage fire fighting strategy of a thermal power plant, which comprises the following steps of: dividing a plant into a plurality of discrete space units according to equipment functions and a space adjacency relation, and constructing a dynamic risk scoring model for each space unit, the risk scoring model takes the combustible calorific value density, the deviation amplitude of the equipment surface temperature relative to a safety threshold value, the electrical load instantaneous fluctuation ratio, the electrical / spatial topology distance attenuation factor of the steam turbine or the main control room and the historical fault event frequency as input variables, and a real-time risk score is generated through weighted fusion; and performing fire-fighting response grade division on each space unit based on the risk score, and endowing an initial response priority corresponding to the grade. According to the method, the perspectiveness, the collaboration and the safety of fire disposal can be remarkably improved by constructing an intelligent decision framework based on space unit dynamic risk scoring and a multi-physics field coupling conduction model.

Description

Technical Field

[0001] This invention belongs to the field of intelligent fire protection technology, specifically relating to a dynamic generation and execution system for multi-level linkage fire protection strategies in thermal power plants. Background Technology

[0002] Thermal power plants are characterized by dense equipment, concentrated combustibles, and highly coupled thermal and electrical systems. Once a fire occurs, it can easily trigger cross-regional cascading accidents through thermal radiation, combustible gas diffusion, or electrical cascading. Existing fire protection systems mostly adopt single-point alarm and fixed plan modes, relying on preset thresholds to trigger local sprinklers or power outages. They lack the ability to dynamically assess the transmission of risks between regions. In recent years, although some studies have attempted to introduce multi-source sensing methods such as infrared temperature measurement and gas monitoring, core issues such as static risk assessment, isolated strategy generation, and open-loop execution processes have not yet been resolved, making it difficult to achieve proactive prevention and collaborative handling of high-risk scenarios.

[0003] Problems with existing technology: The current fire protection systems of thermal power plants have the following main defects: risk assessment relies on static thresholds and does not dynamically quantify based on equipment operating status, meteorological conditions, and topological relationships; there is a lack of preventive intervention mechanisms for high-risk neighboring areas that are not on fire, making it impossible to stop the fire spread chain; the generation of fire protection strategies does not consider the multi-objective conflicts of personnel evacuation, equipment preservation, and power grid continuity, which can easily lead to secondary losses; and the system operates in an open-loop mode, making it impossible to dynamically adjust strategies according to the post-disaster status and difficult to cope with the risk of reignition. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic generation and execution system for multi-level linkage fire fighting strategies in thermal power plants. By constructing an intelligent decision-making framework based on spatial unit dynamic risk scoring and multi-physics field coupling transmission model, it can significantly improve the foresight, coordination and safety of fire response.

[0005] The specific technical solution adopted by this invention is as follows: A method for dynamically generating and executing multi-level coordinated fire-fighting strategies in thermal power plants. This method, based on the plant's spatial topology and multi-physics risk coupling mechanism, achieves graded response and preventative coordination through the following collaborative mechanism, specifically including the following steps: The plant area is divided into multiple discrete spatial units according to equipment functions and spatial adjacency relationships. A dynamic risk scoring model is constructed for each spatial unit. The risk scoring model uses combustible calorific value density, deviation of equipment surface temperature from the safety threshold, instantaneous fluctuation rate of electrical load, electrical / spatial topological distance attenuation factor with respect to the turbine or main control room, and frequency of historical fault events as input variables. The real-time risk score is generated by weighted fusion. Based on risk scores, fire response levels are classified for each spatial unit, and initial response priorities are assigned to each level. By integrating the surface temperature field distribution characterized by infrared thermal imaging, the volume fraction of combustible gas output by laser gas detector, the wind speed and direction vectors provided by the plant's micro-meteorological station, the equipment vibration spectrum characteristics, and the video flame recognition results, a multimodal dynamic risk situation map with spatiotemporal consistency is constructed. Based on the inverse square law of thermal radiation, the Gaussian plume diffusion model, and electrical topology interlocking analysis, the risk transmission intensity to other units when a fire occurs in any spatial unit is quantified. The risk transmission intensity characterizes the combined effect of the probability of ignition by heat flux, the risk of exceeding the limit of cumulative concentration of combustible gas in the downwind direction, the possibility of electrical short circuit interlocking tripping, and the rate of spread of high-temperature flue gas along the cable interlayer. In response to a fire alarm confirmation signal, a main response strategy is generated for the source unit; at the same time, for unburned adjacent units where the risk transmission intensity exceeds a preset dynamic threshold, a preventive fire-fighting strategy is automatically generated, which includes at least one of the following: pre-release of inerting medium, emergency interlocking of ventilation system, power outage of non-critical loads, and pre-lowering of fireproof roller shutters. With personnel evacuation time window constraints, fire resistance limit assurance of core thermal equipment, and grid dispatch continuity maintenance as multi-objective optimization dimensions, an interpretable reinforcement learning agent searches for Pareto optimal collaborative solutions of the main response strategy and the preventive strategy under resource constraints, and outputs an executable instruction set with Bayesian confidence intervals. After the strategy is implemented, the risk index of post-disaster reignition is assessed based on the distribution of infrared residual hotspots, structural vibration attenuation characteristics and residual concentration of combustible gases. Based on this, subsequent isolation, cooling or monitoring strategies are dynamically adjusted to achieve closed-loop optimization of the entire cycle from fire suppression to post-disaster safety management.

[0006] According to another aspect of the present invention, in the dynamic risk scoring model, the calorific value density of combustibles is calculated by multiplying the calorific value of fuel per unit volume by the stock volume; the deviation of equipment surface temperature is defined as the absolute value of the difference between the current temperature and the fire resistance limit threshold of the equipment material; the instantaneous fluctuation rate of electrical load is characterized by the sliding window standard deviation of the current derivative; and the topological distance attenuation factor is determined by an exponential decay function. Calculate, where d is the Euclidean or electrical path distance, and λ is the attenuation coefficient.

[0007] According to another aspect of the present invention, the dynamic threshold of the risk transmission intensity is adaptively set based on the current risk score of the target unit: Once a fire alarm is confirmed in the source unit, if the predicted heat flux to the target unit exceeds 80% of the ignition threshold on the surface of the target unit, or if the predicted concentration of combustible gas exceeds 50% of the lower explosive limit within 10 minutes, a preventative strategy is triggered.

[0008] According to another aspect of the present invention, the interpretable reinforcement learning agent embeds a physical constraint rule base during the training process. The rule base includes symbolic strategies based on accident simulation and historical case summarization. The symbolic strategies are automatically extracted from historical accident data through symbolic regression to guide the policy search direction and improve the interpretability of decisions.

[0009] According to another aspect of the present invention, the post-disaster reignition risk index is composed of three weighted components: the proportion of residual hot spot area in infrared thermal imaging, the 72-hour attenuation slope of the laser methane detector reading, and the vibration frequency offset of the supporting structure.

[0010] According to another aspect of the present invention, a dynamic generation and execution system for multi-level linkage fire-fighting strategies in thermal power plants is also provided, comprising: The dynamic risk scoring and zoning management unit is configured to calculate risk scores and classify fire response levels for each spatial unit in the plant area in real time based on combustible calorific value density, equipment temperature deviation, electrical load fluctuation, topological distance attenuation factor and historical fault frequency. The multimodal perception fusion unit is configured to fuse infrared thermal imaging, laser gas detection, micro-meteorological data, vibration spectrum and video flame recognition results to generate a dynamic risk situation map with spatiotemporal consistency. A cross-unit risk transmission modeling engine with built-in physical propagation models for thermal radiation, gas diffusion, and electrical interlocks is used to quantify the risk transmission intensity from a fire alarm unit to other units. The hierarchical strategy generation unit is configured to generate a main response strategy for fire alarm units and a preventive fire-fighting strategy for neighboring areas with high transmission risk. A multi-objective dynamic optimization and execution unit, integrating an interpretable reinforcement learning agent, is used to search for Pareto optimal cooperative solutions under constraints of personnel evacuation, equipment preservation, and power grid continuity, and outputs an executable instruction set with Bayesian confidence intervals; The post-disaster assessment and closed-loop control unit is configured to assess the risk of reignition based on residual hotspots, gas residues, and structural vibration data, and dynamically adjust subsequent strategies.

[0011] According to another aspect of the present invention, the dynamic risk scoring and zoning management unit communicates with the power plant distributed control system in real time to obtain fuel flow, equipment start-up and shutdown status and load curves, and dynamically updates the combustible material inventory and temperature deviation parameters of each unit to achieve rapid updating of risk scores.

[0012] According to another aspect of the present invention, the cross-unit risk transmission modeling engine integrates a CFD flue gas diffusion simulation module and an electrical topology analyzer, and combines real-time wind direction data to dynamically predict the propagation path of high-temperature flue gas in the cable interlayer and the probability of short-circuit cascading trip.

[0013] According to another aspect of the present invention, the hierarchical strategy generation unit is configured with a human-machine collaborative decision-making interface for correcting AI recommendation strategies, recording correction behavior, and dynamically adjusting the weight ratio of AI and human decision-making through a Bayesian update mechanism to ensure the credibility of the strategy.

[0014] According to another aspect of the present invention, the multi-objective dynamic optimization and execution unit is hardwired and interlocked with the SIS safety instrumented system. All fire commands must be verified by the SIS interlocking logic before execution to ensure that no secondary safety accidents are caused, and the execution status is fed back to the risk situation map update module in real time.

[0015] According to another aspect of the present invention, an electronic device is also provided, the electronic device including a memory and a processor; the memory is used to store a program; the processor executes the program to implement the method described in any one of the foregoing.

[0016] According to another aspect of the present invention, a computer-readable storage medium is also provided, the storage medium storing a computer program that, when executed by a processor, implements the method described in any one of the preceding embodiments.

[0017] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the method described in any one of the preceding embodiments.

[0018] The technical effects achieved by this invention are as follows: This invention, by constructing an intelligent decision-making system based on spatial unit dynamic risk scoring and multi-physics field coupled transmission models, realizes the transformation of fire protection strategies in thermal power plants from static threshold response to dynamic risk prediction. The plant area is divided into multiple discrete spatial units, and multi-dimensional parameters such as combustible calorific value density, equipment temperature deviation, electrical load fluctuation, and topological distance attenuation factor are integrated to generate risk scores in real time. Combined with multi-modal sensing data such as infrared, gas, meteorological, and vibration, a spatiotemporally consistent dynamic risk situation map is constructed. On this basis, the physical model of thermal radiation, gas diffusion, and electrical interlocking is used to quantify the cross-regional risk transmission intensity. For high-risk neighboring areas that have not yet caught fire, preventive strategies such as inertization, power outage, or interlocking are automatically generated to effectively block the cascading fire spread chain and significantly improve the foresight and accuracy of fire response.

[0019] This invention uses an interpretable reinforcement learning agent to search for Pareto optimal cooperative solutions among conflicting objectives of personnel evacuation, core equipment preservation, and power grid continuity, and outputs an executable instruction set with Bayesian confidence intervals to ensure policy credibility. In the post-disaster phase, the system comprehensively assesses the risk of reignition by considering residual hotspots, gas decay slope, and structural vibration frequency offset, and dynamically adjusts subsequent control measures to achieve closed-loop optimization throughout the entire cycle from fire suppression to safety confirmation, thereby improving the intrinsic safety level in high-risk industrial scenarios. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0021] To make the objectives and advantages of this invention clearer, the invention will be specifically described below with reference to embodiments. It should be understood that the following text is merely used to describe one or more specific embodiments of the invention and does not strictly limit the scope of protection specifically claimed by the invention.

[0022] It should be noted that the terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] According to an embodiment of the present invention, a method embodiment for dynamically generating and executing a multi-level linkage fire protection strategy for a thermal power plant is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0024] like Figure 1 As shown, the dynamic generation and execution method of multi-level linkage fire protection strategy for thermal power plants is based on the plant's spatial topology and multi-physics risk coupling mechanism. It achieves graded response and preventative linkage through the following collaborative mechanism, specifically including the following steps: S1. Divide the plant area into multiple discrete spatial units according to equipment functions and spatial adjacency relationships, and construct a dynamic risk scoring model for each spatial unit. The risk scoring model uses the calorific value density of combustibles, the deviation of equipment surface temperature from the safety threshold, the instantaneous fluctuation rate of electrical load, the electrical / spatial topological distance attenuation factor with respect to the turbine or main control room, and the frequency of historical fault events as input variables, and generates a real-time risk score through weighted fusion. S2. Based on the risk score, classify the fire response level of each space unit and assign an initial response priority corresponding to the level; S3. By integrating the surface temperature field distribution characterized by infrared thermal imaging, the volume fraction of combustible gas output by the laser gas detector, the wind speed and direction vectors provided by the plant's micro-meteorological station, the equipment vibration spectrum characteristics, and the video flame recognition results, a multimodal dynamic risk situation map with spatiotemporal consistency is constructed. S4. Based on the inverse square law of thermal radiation, Gaussian plume diffusion model and electrical topology interlocking analysis, the risk transmission intensity to other units when a fire occurs in any spatial unit is quantified. The risk transmission intensity characterizes the combined effect of the probability of ignition by heat flux, the risk of exceeding the limit of cumulative concentration of combustible gas in the downwind direction, the possibility of electrical short circuit interlocking tripping and the rate of spread of high temperature flue gas along the cable interlayer. S5. In response to the fire alarm confirmation signal, generate a main response strategy for the source unit; at the same time, for unburned adjacent units where the risk transmission intensity exceeds a preset dynamic threshold, automatically generate a preventive fire-fighting strategy, which includes at least one of the following: pre-release of inerting medium, emergency interlocking of ventilation system, power outage of non-critical loads, and pre-lowering of fireproof roller shutter. S6. Taking personnel evacuation time window constraints, fire resistance limit guarantee of core thermal equipment and grid dispatch continuity maintenance as multi-objective optimization dimensions, an interpretable reinforcement learning agent searches for Pareto optimal cooperative solutions of main response strategy and preventive strategy under resource constraints, and outputs an executable instruction set with Bayesian confidence intervals. Among them, resource constraints include the upper limit of inert medium storage, the maximum flow rate of fire pump and the timing limit of circuit breaker action. S7. After the strategy is implemented, the risk index of post-disaster reignition is assessed based on the distribution of infrared residual hotspots, structural vibration attenuation characteristics and residual concentration of combustible gas, and the subsequent isolation, cooling or monitoring strategies are dynamically adjusted accordingly to achieve closed-loop optimization of the entire cycle from fire suppression to post-disaster safety management.

[0025] Based on the above, in step S1, the equipment layout of thermal power plants is highly non-uniform. Different areas differ in the types of combustibles (such as pulverized coal, lubricating oil, and hydrogen), operating temperatures (boiler furnace and control room), electrical loads (main transformer and lighting circuits), and topological relationships with key facilities (such as steam turbines and main control rooms). If a uniform risk threshold is adopted, it will lead to insufficient response in high-risk areas and frequent false alarms in low-risk areas. Therefore, this method first divides the plant area into multiple discrete spatial units according to the functional attributes of the equipment (such as fuel transportation, heat energy conversion, and power control) and spatial adjacency relationships (such as shared air ducts, cable tray connectivity, and physical distance). Each unit serves as an independent risk assessment node.

[0026] Therefore, based on this, a dynamic risk scoring model is constructed, whose input variables all have physical or engineering significance: in, Calorific value density of combustibles: reflects the potential energy released by combustion per unit volume, and is the decisive factor in the rate of heat release in a fire; Equipment surface temperature deviation: defined as the absolute value of the difference between the current temperature and the material's fire resistance limit threshold, characterizing the risk space for spontaneous combustion or ignition of nearby objects; Instantaneous fluctuation of electrical load: quantified by the standard deviation of the sliding window of the current derivative; high fluctuation indicates the risk of arcing or short circuit. Topological distance decay factor: using an exponential decay function ( ) Quantify the impact intensity of a unit's failure on the core facility; the closer the distance, the higher the weight. Historical failure frequency: As an empirical prior, it reflects the vulnerability of the unit in historical operation.

[0027] The above variables are weighted and fused (the weights can be based on accident statistics or expert knowledge) to generate a real-time updated risk score, realizing the transformation from static thresholds to dynamic profiles.

[0028] In step S2, the risk score itself is a continuous value and needs to be mapped to an operable discrete response level (e.g., Level 1: immediate system-wide linkage; Level 2: local isolation + monitoring; Level 3: early warning standby). Each level corresponds to a set of initial response priorities to guide the subsequent resource scheduling order. For example, when a Level 1 unit is triggered, the system prioritizes its fire extinguishing resources, even if it sacrifices the cooling capacity of the Level 3 unit.

[0029] It should be noted that the response level classification is not fixed, but dynamically adjusted according to the operating conditions: for example, during boiler ignition, the furnace unit automatically upgrades to level one; when the coal conveying system is shut down, the trestle unit downgrades to level three, etc.

[0030] In step S3, the static risk score only reflects potential hazards, while fire is a spatiotemporal evolution process. Therefore, it is necessary to integrate multi-source real-time monitoring data. Infrared thermal imaging: provides full-field surface temperature distribution and identifies abnormal hot spots; Laser gas detection: accurately measures the volume fraction of combustible gases such as CH4 and CO; Micro-weather station: provides wind speed and direction vectors, which determine the direction of smoke diffusion; Vibration spectrum: reflects the mechanical condition of the equipment; abnormal friction may indicate a fire. Video flame recognition: Confirms the presence of open flame through a deep learning model.

[0031] Furthermore, the aforementioned heterogeneous data are integrated into a multimodal dynamic risk situation map with spatiotemporal consistency through a spatiotemporal alignment mechanism (such as unified timestamps and spatial coordinate mapping to the plant BIM model), ensuring that subsequent analysis is based on the same spatiotemporal benchmark.

[0032] In step S4, the fire hazard lies not only in the ignition point, but also in the cross-regional cascading effect, which is quantified based on a recognized physical model: Inverse square law of thermal radiation: Calculate the heat flux from the high-temperature unit to the surface of adjacent equipment to determine whether the ignition threshold has been reached; Gaussian plume diffusion model: Combined with real-time wind direction, it predicts the concentration accumulation of combustible gas in downstream units; Electrical topology interlocking analysis: Identifies the path of protection device malfunction or cascading tripping caused by short circuits.

[0033] Based on the above, the risk transmission intensity is a multi-dimensional comprehensive indicator that characterizes the coupling effect of three types of risks: heat, gas, and electricity.

[0034] In step S5, once a fire alarm is confirmed (e.g., dual verification of video flames and gas concentration), a main response strategy is generated (e.g., activating water cannons or cutting off power). Simultaneously, for unburned adjacent areas where the risk transmission intensity exceeds a dynamic threshold, preventative strategies are automatically generated, for example: If the coal conveyor bridge catches fire and the wind direction is towards the coal bunker, the inerting medium in the coal bunker should be pre-released. If the temperature gradient of the cable interlayer smoke is abnormal, then the ventilation should be locked and non-critical loads disconnected.

[0035] wait By employing the above strategies, the chain of fire spread can be broken through preventative measures against potential fire hazards.

[0036] In step S6, fire safety decisions need to weigh multiple conflicting objectives, such as: Personnel safety: Rapid evacuation is required, but premature power outages will affect emergency lighting; Equipment maintenance: Continuous cooling is required, but spraying may damage electrical equipment; Grid continuity: requires avoiding unplanned outages, but local isolation is unavoidable.

[0037] Using an interpretable reinforcement learning agent, Pareto optimal cooperative solutions are searched under resource constraints (water, inerting gas, electricity), i.e., policy combinations that cannot improve one objective without compromising another. The output instructions are accompanied by Bayesian confidence intervals to characterize the policy reliability.

[0038] Step S7: After the fire is extinguished, the system does not terminate, but continues based on: Infrared residual hotspot distribution: identifying potential reignition points; Structural vibration attenuation characteristics: assessing the integrity of the supporting structure; Residual concentration of combustible gas: used to assess explosion risk.

[0039] Calculate the risk index of post-disaster reignition and dynamically adjust subsequent strategies (such as extending inertia and prohibiting power restoration) to achieve full-cycle monitoring and control from emergency response to post-disaster safety management.

[0040] Based on the above steps, by constructing a system from fire perception to post-fire monitoring, the dynamic risk scoring of spatial units is combined with a multi-physics field coupling transmission model, realizing the transformation of risk assessment from point to network. Through the automatic generation mechanism of preventive strategies, preventive fire suppression can be achieved. Furthermore, by constructing a system from prevention to the prevention of secondary fires after ignition, the system is made to have synergy and adaptability.

[0041] As an optional embodiment, in the dynamic risk scoring model, the calorific value density of combustibles is calculated by multiplying the calorific value of fuel per unit volume by the stock volume; the deviation of equipment surface temperature is defined as the absolute value of the difference between the current temperature and the fire resistance limit threshold of the equipment material; the instantaneous fluctuation rate of electrical load is characterized by the sliding window standard deviation of the current derivative; and the topological distance attenuation factor is determined by an exponential decay function. Calculate, where d is the Euclidean or electrical path distance, and λ is the attenuation coefficient.

[0042] Based on the above, the calorific value density of combustibles is used to represent the potential energy released during combustion per unit space. It is a parameter that determines the heat release rate (HRR) of a fire, and its calculation method is as follows: Calorific value density of combustible material = calorific value of fuel per unit volume × current inventory; The calorific value of fuel per unit volume is determined by the type of fuel (e.g., pulverized coal approximately 24 MJ / m³). 3 Lubricating oil approximately 35 MJ / m 3 The current inventory can be obtained in real time through level gauges, flow integrals, or material balance models in distributed control systems. The higher this index, the more rapidly the fire will spread once it starts, and the stronger the thermal shock to nearby equipment. By using the above variables, the system can distinguish the risk differences between a small amount of high-calorific-value oil and a large amount of low-calorific-value coal powder, thus achieving a refined assessment.

[0043] Furthermore, the deviation of the equipment surface temperature is: |current measured temperature − equipment material fire resistance limit threshold|; The current temperature is collected in real time by infrared thermal imaging or thermocouples; the fire resistance limit threshold is determined based on the equipment material (such as carbon steel, stainless steel, insulation material) and its design specifications (for example, the fire resistance limit of cable insulation is usually 180°C). The larger the value, the closer the equipment is to the spontaneous combustion critical point, or the easier it is to be ignited by external heat sources.

[0044] Furthermore, the instantaneous fluctuation rate of the electrical load is used to quantify the instability of the current. Its calculation is based on the first derivative of the current signal (i.e., di / dt) and statistically analyzed using the sliding window standard deviation. Within the time window [t−Δt,t], calculate The standard deviation of is used as a volatility indicator; High volatility often foreshadows of faults such as arcing, short circuits, or poor contact, and is a significant cause of electrical fires. This variable enables the system to identify hidden risks in advance, compensating for the inadequacy of relying solely on overcurrent protection.

[0045] Furthermore, the topological distance attenuation factor is used to quantify the functional correlation strength between a spatial unit and key facilities (such as the main control room or steam turbine), and it employs an exponential attenuation function: , where d is the Euclidean distance (used for modeling thermal radiation and flue gas diffusion) or electrical path distance (such as the number of tripping stages on a relay protection link, used for cascading fault analysis); λ is the attenuation coefficient, which can be calibrated based on historical accident data (e.g., if the radius of influence of thermal radiation is approximately 30 meters, then λ≈0.1). This technology improves the accuracy of conduction modeling by automatically identifying different risk patterns: those that are physically close but functionally isolated and those that are physically distant but electrically tightly coupled.

[0046] In summary, risks are characterized from four dimensions: energy reserves (calorific value density), thermal state (temperature deviation), electrical stability (load fluctuation), and system coupling (topological distance). The combination of these dimensions is not a simple superposition, but constitutes a multi-dimensional risk state space. These four heterogeneous physical quantities are integrated through a unified mathematical form using four variables, and used to drive the generation of preventive strategies.

[0047] As an optional embodiment, the dynamic threshold of risk transmission intensity is adaptively set according to the current risk score of the target unit: when the source unit fire alarm is confirmed, if its predicted heat flux to the target unit exceeds 80% of the target unit surface ignition threshold, or the predicted flammable gas concentration exceeds 50% of the lower explosive limit within 10 minutes, a preventive strategy is triggered.

[0048] Based on the above, by establishing a proportional relationship between the trigger threshold of the preventive fire-fighting strategy and the material ignition threshold and lower explosive limit of the gas in the target unit (such as 80% heat flux threshold and 50% LEL concentration), and dynamically adjusting the sensitivity in conjunction with risk scoring, the fixed threshold can be converted into an adaptive warning. This not only provides a reference for physical limits, but also improves the reliability of the warning by using time trends (such as concentration increase within 10 minutes).

[0049] Furthermore, heat flux and gas concentration are not based on real-time measurement, but rather on forward prediction based on physical simulations such as the thermal radiation law and Gaussian plume model. This dynamic threshold method deeply integrates the target unit state, material properties, and risk transmission model to achieve timely response.

[0050] As an optional embodiment, the interpretable reinforcement learning agent embeds a physical constraint rule base during training. The rule base contains symbolic strategies based on accident simulation and historical case summarization. The symbolic strategies are automatically extracted from historical accident data through symbolic regression and are in the form of IF condition THEN action expressions, which are used to guide the policy search direction and improve the interpretability of decisions.

[0051] Furthermore, it can be explained that a physical constraint rule base is embedded in the training process of the reinforcement learning agent. The rule base is not a fixed logic preset by humans, but a set of conditional and action-oriented policy rules extracted by symbolic induction based on a large amount of fire accident simulation data and historical handling cases (such as a sudden increase in CO concentration on the coal conveyor bridge and wind direction pointing towards the coal bunker, to the pre-start inertization medium). This serves as a policy search space for reinforcement learning constrained by prior knowledge, avoiding it from exploring invalid or even dangerous actions that violate physical laws or safety procedures.

[0052] Furthermore, it not only improves the security and convergence efficiency of strategy generation, but also makes the AI ​​decision-making process causally explainable. Schedulers can understand why the system recommends this operation, thereby enhancing human-machine trust and deeply integrating domain knowledge with data-driven learning.

[0053] As an optional embodiment, the post-disaster reignition risk index is composed of three weighted components: the proportion of residual hot spot area in infrared thermal imaging, the 72-hour attenuation slope of the laser methane detector reading, and the vibration frequency offset of the supporting structure. When the post-disaster reignition risk index exceeds the preset safety threshold, the system automatically extends the inerting medium supply time and prohibits personnel from entering.

[0054] Based on the above, the post-disaster reignition risk index integrates three types of post-disaster monitoring indicators with clear physical significance: the proportion of residual hot spots in infrared thermal imaging (reflecting combustion sources that have not been completely cooled), the concentration decay slope of laser methane detector readings over 72 hours (reflecting the residual and diffusion trend of combustible gases), and the vibration frequency offset of the supporting structure (reflecting the stiffness degradation of load-bearing components due to high temperatures). This is used to construct a quantitative assessment model. Since it no longer relies on a single parameter to determine the post-disaster safety status, it collaboratively assesses the possibility of reignition and secondary disasters from three aspects: residual heat, gas phase risk, and structural integrity.

[0055] Furthermore, it can continuously conduct risk assessments after a fire is extinguished and dynamically adjust subsequent strategies accordingly (such as extending inerting, prohibiting power restoration, or restricting personnel access), thus achieving control from the completion of fire extinguishing to safety confirmation.

[0056] As an optional embodiment, a dynamic generation and execution system for multi-level linkage fire-fighting strategies in thermal power plants includes: The dynamic risk scoring and zoning management unit is configured to calculate risk scores and classify fire response levels for each spatial unit in the plant area in real time based on combustible calorific value density, equipment temperature deviation, electrical load fluctuation, topological distance attenuation factor and historical fault frequency. The multimodal perception fusion unit is configured to fuse infrared thermal imaging, laser gas detection, micro-meteorological data, vibration spectrum and video flame recognition results to generate a dynamic risk situation map with spatiotemporal consistency. A cross-unit risk transmission modeling engine with built-in physical propagation models for thermal radiation, gas diffusion, and electrical interlocks is used to quantify the risk transmission intensity from a fire alarm unit to other units. The hierarchical strategy generation unit is configured to generate a main response strategy for fire alarm units and a preventive fire-fighting strategy for neighboring areas with high transmission risk. A multi-objective dynamic optimization and execution unit, integrating an interpretable reinforcement learning agent, is used to search for Pareto optimal cooperative solutions under constraints of personnel evacuation, equipment preservation, and power grid continuity, and outputs an executable instruction set with Bayesian confidence intervals; The post-disaster assessment and closed-loop control unit is configured to assess the risk of reignition based on residual hotspots, gas residues, and structural vibration data, and dynamically adjust subsequent strategies.

[0057] Furthermore, the dynamic risk scoring system communicates in real time with the zoned management unit and the power plant's distributed control system to obtain fuel flow, equipment start-up and shutdown status, and load curves, and dynamically updates the combustible material inventory and temperature deviation parameters of each unit, thereby enabling rapid updates to the risk score.

[0058] Furthermore, the cross-unit risk transmission modeling engine integrates a computational fluid dynamics (CFD) flue gas diffusion simulation module with an electrical topology analyzer, and combines real-time wind direction data to dynamically predict the propagation path of high-temperature flue gas in the cable interlayer and the probability of short-circuit cascading trips.

[0059] Furthermore, the hierarchical strategy generation unit is equipped with a human-machine collaborative decision-making interface to correct the AI ​​recommendation strategy, record the correction behavior, and dynamically adjust the weight ratio of AI and human decision-making through a Bayesian update mechanism to ensure the credibility of the strategy.

[0060] Furthermore, the multi-objective dynamic optimization and execution unit is hardwired and interlocked with the SIS safety instrumented system. All fire commands must be verified by the SIS interlocking logic before execution to ensure that no secondary safety accidents are caused, and the execution status is fed back to the risk situation map update module in real time.

[0061] Based on the above, an intelligent fire protection process is achieved through the collaboration of dynamic risk scoring and zoning management units, multimodal perception fusion units, cross-unit risk transmission modeling engines, hierarchical strategy generation units, multi-objective dynamic optimization and execution units, and post-disaster assessment and closed-loop control units. Among them, dynamic risk scoring and zoning management units realize spatial risk profiling, multimodal perception fusion units ensure the spatiotemporal alignment of multi-source heterogeneous data, and cross-unit risk transmission modeling engines quantify the coupling strength of fire spread across regions based on physical models such as thermal radiation, gas diffusion, and electrical topology, providing a basis for preventive intervention.

[0062] Based on this, the hierarchical strategy generation unit distinguishes between active response and preventive actions, while the multi-objective dynamic optimization and execution unit uses interpretable reinforcement learning to find the Pareto optimal solution among personnel safety, equipment preservation and power grid continuity, and attaches a Bayesian confidence interval to improve decision credibility; the post-disaster assessment and closed-loop control unit continuously monitors residual risks and dynamically adjusts subsequent strategies to achieve closed-loop management from fire extinguishing to safety confirmation.

[0063] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0064] According to another aspect of the present invention, an electronic device is also provided, the electronic device including a memory and a processor; the memory is used to store a program; the processor executes the program to implement the method of any of the foregoing.

[0065] According to another aspect of the present invention, a computer-readable storage medium is also provided, the storage medium storing a computer program that, when executed by a processor, implements the method of any of the foregoing.

[0066] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the method described in any of the foregoing.

[0067] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A method for dynamic generation and execution of multi-stage cascading fire fighting strategy in thermal power plants, characterized in that, Based on the spatial topology of the plant site and the coupling mechanism of multiple physical fields, the method comprises the following steps: Divide the plant site into multiple spatial units, and construct a dynamic risk scoring model with the input of combustible heat value density, equipment temperature deviation, electrical load fluctuation, topological distance attenuation factor, and historical failure frequency; Divide the fire response level based on the score; Fusion infrared thermal imaging, gas detection, microclimate, vibration spectrum and video flame data to construct a spatiotemporal consistent multi-modal risk situation map; Based on the thermal radiation, gas diffusion and electrical interlocking model, the risk transmission intensity of the fire alarm unit to other units is quantified; Generate the main response strategy for the fire alarm unit, and automatically generate the preventive fire fighting strategy for the high transmission risk adjacent area; With personnel evacuation, equipment preservation and power grid continuity as multiple objectives, search for the Pareto optimal collaborative solution through interpretable reinforcement learning, and output the instruction set with Bayesian confidence interval; After the disaster, the residual hot spot, gas residue and structure vibration are evaluated to assess the risk of rekindling, and the subsequent strategy is dynamically adjusted to realize the whole cycle closed loop optimization.

2. The method for dynamic generation and execution of multi-stage cascading fire-fighting strategy of thermal power plant according to claim 1, characterized in that: In the dynamic risk score model, the combustible heat value density is calculated by the product of the fuel heat value and the stock in unit volume, the equipment surface temperature deviation amplitude is defined as the absolute value of the difference between the current temperature and the fireproof limit threshold value of the equipment material, the electrical load instantaneous fluctuation rate is represented by the sliding window standard deviation of the current derivative, and the topological distance attenuation factor is an exponential decay function is calculated, where d is the Euclidean or electrical path distance, and λ is the attenuation coefficient.

3. The method for dynamic generation and execution of multi-stage cascading fire-fighting strategy in thermal power plants as claimed in claim 1 wherein, The dynamic threshold of the risk transmission intensity is adaptively set according to the current risk score of the target unit: When the source unit fire alarm is confirmed, if the heat flux prediction value of the source unit to the target unit exceeds 80% of the surface ignition threshold of the target unit, or the combustible gas concentration prediction value exceeds 50% of the lower limit of explosion within 10 minutes, the preventive strategy is triggered.

4. The method for dynamic generation and execution of multi-stage cascading fire-fighting strategy in thermal power plants as claimed in claim 1 wherein: The interpretable reinforcement learning agent embeds a physical constraint rule base in the training process, which contains symbolic strategies based on accident simulation and historical case induction. The symbolic strategies are automatically extracted from historical accident data through symbolic regression, which is used to guide the strategy search direction and improve the decision interpretability.

5. The method for dynamic generation and execution of multi-stage cascading fire fighting strategy in thermal power plant as claimed in claim 1 wherein, After the execution of the strategy, the post-disaster rekindling risk index is evaluated based on the residual hot spot distribution, structure vibration attenuation characteristics and combustible gas residual concentration. The post-disaster rekindling risk index is composed of three parts: the residual hot spot area ratio in infrared thermal imaging, the 72-hour decay slope of laser methane detector reading, and the support structure vibration main frequency offset.

6. A system for dynamic generation and execution of multi-stage cascading fire fighting strategy in thermal power plants using the method as claimed in any one of claims 1 to 5, wherein, It includes: Dynamic risk scoring and partition management unit, configured to calculate the risk score of each spatial unit of the plant site in real time and divide the fire response level based on the combustible heat value density, equipment temperature deviation, electrical load fluctuation, topological distance attenuation factor and historical failure frequency; Multi-modal perception fusion unit, configured to fuse infrared thermal imaging, laser gas detection, microclimate data, vibration spectrum and video flame recognition results to generate a dynamic risk situation map with spatiotemporal consistency; Cross-unit risk transmission modeling engine, which internally builds physical propagation models of thermal radiation, gas diffusion and electrical interlocking, for quantifying the risk transmission intensity of the fire alarm unit to other units; Hierarchical strategy generation unit, configured to generate the main response strategy for the fire alarm unit, and to generate the preventive fire fighting strategy for the high transmission risk adjacent area; Multi-objective dynamic optimization and execution unit, integrated with an interpretable reinforcement learning agent, for searching for the Pareto optimal collaborative solution under the constraints of personnel evacuation, equipment preservation and power grid continuity, and outputting the executable instruction set with Bayesian confidence interval; The post-disaster assessment and closed-loop control unit is configured to assess the risk of rekindling based on residual hot spots, gas residues, and structural vibration data, and dynamically adjust subsequent strategies.

7. The system for dynamic generation and execution of multi-stage cascading fire-fighting strategy in thermal power plants as claimed in claim 6 wherein: The dynamic risk score and partition management unit communicates with the power plant distributed control system in real time to obtain fuel flow, equipment start-stop state, and load curve, dynamically update the combustible material inventory and temperature deviation parameters of each unit, and realize rapid refresh of the risk score.

8. The system for dynamic generation and execution of multi-stage cascading fire fighting strategy in thermal power plant as claimed in claim 6 wherein: The cross-unit risk transmission modeling engine integrates a CFD smoke diffusion simulation module and an electrical topology analyzer, combined with real-time wind direction data, to dynamically predict the spread path of high-temperature flue gas in the cable interlayer and the short-circuit chain trip probability.

9. The system for dynamic generation and execution of multi-stage cascading fire fighting strategy in thermal power plant as claimed in claim 6 wherein: The hierarchical strategy generation unit is configured with a human-machine collaborative decision-making interface for correcting AI recommended strategies, recording correction behavior, and dynamically adjusting the weight ratio of AI and manual decision-making through a Bayesian update mechanism to ensure the credibility of the strategy.

10. The system for dynamic generation and execution of multi-stage cascading fire fighting strategy in thermal power plant as claimed in claim 6 wherein: The multi-target dynamic optimization and execution unit is hard-wired with the SIS safety instrument system interlock, and all fire-fighting instructions need to pass through SIS interlock logic verification before execution to ensure that secondary safety accidents are not caused, and the execution status is fed back to the risk situation map update module in real time.

Images