Basement ventilation and smoke exhaust system and method based on digital management

Abstract

The invention discloses a basement ventilation and smoke exhaust system and method based on digital management, and the method comprises the steps: constructing a digital twin architecture comprising a terminal layer, an edge layer and a cloud layer, and carrying out the real-time mapping and interaction of a physical system for basement ventilation and smoke exhaust and a virtual model; carrying out integrated monitoring on the terminal layer by utilizing multiple sensors, and collecting relevant data parameters of ventilation and smoke exhaust of the basement; and a high-integration-level edge computing gateway is adopted to preprocess the collected data. The regional controller adopts a dynamic Bayesian network to fuse the preprocessed multi-source data, and constructs a local environment state model; the edge layer utilizes a deep reinforcement learning algorithm and a digital twinborn model to simulate multiple scenes to carry out state real-time prediction, and generates an emergency plan library; and the cloud layer central server operates a double-track decision engine and performs adaptive adjustment in combination with a deterministic rule base and an adaptive deep reinforcement learning model, and the technical scheme of the invention realizes efficient and accurate ventilation and smoke exhaust cooperative control.

Description

Technical Field

[0001] The present invention relates to the technical field of ventilation and smoke exhaust, and in particular to a basement ventilation and smoke exhaust system and method based on digital management. Background Art

[0002] In the existing complex basement environment, the ventilation and smoke exhaust system is mainly composed of smoke exhaust terminals, manual control terminals, smoke control terminals and master control terminals, which are respectively set up in multiple different areas. These components are used to work together, and the core components and their corresponding control terminals, such as regional manual terminals and centralized ventilation and smoke exhaust terminals, are usually deployed in different areas that are physically dispersed and logically isolated. Their operating status, fault alarms, start and stop signals and other key data information are stored separately or enclosed in their own independently operated subsystems, such as independent building automation subsystems, fire alarm subsystems, and equipment-specific monitoring platforms.

[0003] Therefore, the data and information generated by the components of a distributed deployment are often scattered across independent systems. The physical dispersion and logical isolation of data make real-time global situational awareness difficult. Management terminals are unable to seamlessly integrate and intuitively monitor all dynamic information, such as valve opening and closing status, fan operating parameters, terminal control instructions, and module feedback signals, on a single interface, making it difficult to quickly identify abnormal correlations across devices and regions. Furthermore, system response is delayed and linkage efficiency is low. Data between different subsystems cannot achieve low-latency, high-fidelity interaction and fusion calculations, resulting in low accuracy in cross-system collaborative analysis. In emergency situations, control instructions must be transmitted layer by layer through multiple systems. The manual intervention commonly used in existing systems mainly relies on decentralized on-site terminals, resulting in the inability to adaptively respond to real-time conditions in actual situations, nor is it possible to dynamically optimize strategies and adjust controls based on real-time, integrated global data. Summary of the Invention

[0004] The object of the present invention is to provide a basement ventilation and smoke exhaust system and method based on digital management to solve the problems raised in the above background technology.

[0005] The specific technical solution provided by the present invention is as follows: A basement ventilation and smoke exhaust system based on digital management, including the following functional units: a global perception fusion unit, an edge intelligent gateway unit, a digital twin drive unit, a deep reinforcement learning decision unit, a dynamic execution verification unit, and an adaptive adjustment control unit. The basement ventilation and smoke exhaust method based on digital management includes the following steps: S1: Build a digital twin architecture including the terminal layer, edge layer, and cloud layer to perform real-time mapping and interaction between the physical system and virtual model of basement ventilation and smoke exhaust.

[0006] Preferably, the real-time mapping and interaction between the physical system and the virtual model for basement ventilation and smoke exhaust specifically include: the terminal layer sensors in the digital twin architecture collect various parameters of the physical system according to the set sampling frequency; the edge layer pre-processes the received data and transmits it to the cloud layer, and at the same time updates the corresponding parameters in the virtual model in real time based on these data; the virtual model uses the updated parameters to perform simulation calculations to simulate the operation of the physical system under different working conditions, and feeds the simulation results back to the edge layer and the cloud layer; the edge layer and the cloud layer then formulate control strategies based on the simulation results of the virtual model and actual needs, and send control instructions to the actuators at the terminal layer. The actuators adjust the physical system according to the instructions to realize real-time interaction between the physical system and the virtual model.

[0007] S2: Use multiple sensors for integrated monitoring at the terminal level to collect data parameters related to basement ventilation and smoke exhaust.

[0008] S3: Use a highly integrated edge computing gateway to pre-process the collected data.

[0009] S4: The regional controller uses a dynamic Bayesian network to fuse pre-processed multi-source data and build a local environmental state model.

[0010] Preferably, constructing the local environment state model specifically includes: S41: Determine the state variables and observation variables of the local environment; S42: Construct a dynamic Bayesian network structure including the initial network structure and the transfer network structure; S43: Determine the parameters of the dynamic Bayesian network to obtain the conditional probability distribution between the variables; S44: Inference of dynamic Bayesian networks using preprocessed multi-source data; S45: Construct the corresponding local environmental state model as the basis for edge layer state prediction and emergency plan generation.

[0011] S5: The edge layer uses deep reinforcement learning algorithms and digital twin models to simulate multiple scenarios for real-time status prediction and generate an emergency plan library.

[0012] Preferably, simulating multiple scenarios to perform real-time status prediction specifically includes: S51: Determine the state space, action space and reward function; S52: Constructing a DQN+LSTM model; the DQN+LSTM model uses an LSTM network and a deep neural network; S53: Calculate the target value using the target network, select actions and update parameters based on the evaluation network; S54: Use the digital twin model to simulate multiple scenarios, use the output of the digital twin model as environmental feedback, and interact with the DQN+LSTM model; S55: Obtain feedback based on the reward function, gradually optimize the action strategy, and based on the optimized strategy, make real-time predictions and outputs of the system status in different scenarios.

[0013] S6: The central server at the cloud layer runs a dual-track decision engine, combining a deterministic rule base with an adaptive deep reinforcement learning model to perform overall adaptive collaborative control.

[0014] Preferably, the operation process of the dual-track decision engine specifically includes: S61: The dual-track decision engine simultaneously uses a deterministic rule base and an adaptive deep reinforcement learning model to analyze the global environment state; S62: The deterministic rule library matches the current state according to the preset rules. If there is a completely matching rule, the corresponding control instruction is directly generated. If there is no completely matching rule or the rule execution effect is poor, the adaptive deep reinforcement learning model is started; S63: The reinforcement learning model calculates the optimal control strategy based on the global state and supplements or adjusts the deterministic rules; S64: Fusing and evaluating the control strategies generated by the two decision paths, and selecting the control strategy with the best comprehensive performance as the final decision; S65: The final control instructions are sent to each edge layer, which then forwards them to the actuators in the terminal layer to achieve coordinated control of the entire system. S66: Synchronously feed back the decision-making process and execution results to the reinforcement learning model for continuous updating and optimization of the model.

[0015] S7: Synchronously retain the hybrid decision-making mechanism, perform adaptive collaborative control of the system under normal conditions, and use manual intervention terminals in emergency conditions to convert manual experience into system knowledge to form a human-machine collaborative decision-making mechanism.

[0016] Compared with existing technologies, this invention effectively balances the needs of real-time performance and complex decision-making through a three-tier architecture. The edge intelligent gateway unit implements local data preprocessing and primary policy caching through distributed edge computing, and adopts distributed ledger technology to promote the sharing of key states among adjacent edge nodes, thus achieving cross-regional policy collaboration. Simultaneously, it utilizes a dual-track decision engine, integrating a deterministic rule base with an adaptive reinforcement learning model. This not only ensures standardized responses in conventional scenarios, but also optimizes dynamic policies through global data. This addresses the limitations of traditional systems that rely on fixed rules and cannot be adaptively adjusted. It comprehensively improves the global perception capability, response speed, decision-making accuracy, and operational reliability of basement ventilation and smoke exhaust systems, providing a digital and intelligent solution for the safety management and control of complex underground spaces. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of the functional units of a basement ventilation and smoke exhaust system based on digital management provided by an embodiment of the present invention; Figure 2 This is a flow chart of an operating method of a basement ventilation and smoke exhaust system based on digital management provided by an embodiment of the present invention; Figure 3 This is a flow chart of an operating method for performing data fusion and model building provided by an embodiment of the present invention; Figure 4 This is a flow chart of an edge layer operation method provided by an embodiment of the present invention; Figure 5 This is a flow chart of the operating method of the dual-track decision engine provided by an embodiment of the present invention; Figure 6 This is a schematic diagram of the logical execution of the cloud layer provided by an embodiment of the present invention. DETAILED DESCRIPTION The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of the present invention.

[0018] Example 1: like Figure 1 As shown, the basement ventilation and smoke exhaust system based on digital management described in this embodiment includes: a global perception fusion unit, an edge intelligent gateway unit, a digital twin drive unit, a deep reinforcement learning decision unit, a dynamic execution verification unit, and an adaptive adjustment control unit.

[0019] In this embodiment, the global perception fusion unit is used to collect device status and environmental data in all dimensions in real time.

[0020] For example, the control terminals involved in the existing basement ventilation and smoke exhaust system are equipped with multiple sensors, including current, vibration, temperature and humidity, and open and closed status. These sensors are uniformly connected to the data of subsystems such as fire protection and building control through industrial protocol converters. Based on the implanted RFID tags, physical devices and digital objects are bound, standardizing the data streams of heterogeneous data sources generated by each terminal. This unit performs spatial topology registration based on RFID device binding. That is, by adding a spatial relationship modeling engine, it automatically constructs a three-dimensional topological map of "device-area-subsystem", such as "B1-Zone 1 Fan-Fire Damper V1", and simultaneously forms virtual sensing of blind spots. The CFD compensation capability of the digital twin drive unit is integrated to dynamically generate key parameters for uncovered areas, achieving compatibility with data from subsystems such as fire protection and building control, and automatically identifying conflicting data sources. For example, a verification is triggered when the temperature difference between fire protection and building control exceeds a threshold. Based on the existing data binding achieved only through devices, this unit adds the dual dimensions of spatial relationships and data credibility to solve the initial data association problem caused by the difficulty of correlation analysis due to physical dispersion, providing a highly correlated data foundation for subsequent digital management.

[0021] In this embodiment, the edge intelligent gateway unit is used to utilize a separate edge computing unit configured in each area to receive perception data, perform multi-source data fusion, align multi-source heterogeneous data timestamps, generate a primary control strategy cache, and realize local preprocessing of sensor data.

[0022] For example, this unit runs a lightweight DRL model and generates a control strategy based on local data when the network is disconnected, such as The concentration gradient autonomously adjusts the air valve, conducts cross-edge consensus coordination, and uses distributed ledger technology to enable adjacent edge nodes to share key states. For example, by sharing the exhaust pressure value of area A, cross-area strategy coordination can be achieved, that is, to avoid overload of the fan in area B. It can not only perform local preprocessing, but also realize autonomous coordination between edge nodes, eliminating cross-regional linkage lag.

[0023] In this embodiment, the digital twin drive unit is used to solve the three-dimensional CFD flow field in real time, preview the impact of control instructions on the flow field, and generate virtual sensor data of uncovered areas for blind spot compensation.

[0024] For example, in this unit, the airflow / smoke diffusion effect is simulated based on control instructions, and the feasibility of the strategy is previewed, such as displaying the backflow area caused by closing a valve, virtual and real interactive verification, comparing virtual sensor data with actual feedback, automatically marking the model error area when the deviation is greater than the set threshold, simulating the fire spread path, and pre-calculating the smoke exhaust efficiency of multiple strategies and ranking them.

[0025] In this embodiment, a deep reinforcement learning decision unit is used for multi-scale anomaly detection based on graph neural networks and real-time state change analysis, and realizes intelligent switching of system modes through a multi-threshold event trigger mechanism.

[0026] For example, this unit uses a multi-scale graph neural network to simultaneously perform abnormal correlation analysis on the device layer, regional layer, and system layer to achieve cross-scale global optimization, and automatically adjust the alarm threshold based on historical event data. It can improve the sensitivity of electrical fire warnings when the humidity rises in summer, and conduct policy conflict arbitration. When the energy-saving mode conflicts with the emergency smoke exhaust command, it automatically makes a decision according to the preset rule matrix.

[0027] In this embodiment, a dynamic execution verification unit is used to convert the strategy into a device control signal, compare the predicted value with the actual sensor feedback and dynamically correct the CFD parameter error, and roll back to the safe strategy when the deviation exceeds the threshold.

[0028] Exemplarily, this unit uses dual channels for execution verification, uses the prediction channel to send control instructions to the three-dimensional situation unit for preview and verification of effectiveness, uses the actual channel to compare device feedback with the expected state in real time, triggers rollback when the deviation exceeds the limit, and automatically starts the parameter calibration algorithm when the CFD model continues to predict inaccurately, such as correcting the boundary turbulence coefficient. At the same time, the safety policy container is used to store the pre-verified backup policy library, which is directly called when an emergency such as the "anti-backflow safety valve opening sequence" occurs. Under the existing verification method that only requires post-correction, this unit establishes a triple guarantee of pre-execution verification-real-time feedback-model self-healing to solve the risk problem of instruction transmission.

[0029] In this embodiment, the adaptive adjustment control unit is used to optimize the sensor network structure in combination with dynamic topology adjustment. A controller based on a deep reinforcement learning algorithm performs operating condition prediction and self-learning, optimizing operating parameters based on historical data.

[0030] For example, this unit automatically updates the system topology based on equipment failures and new additions. For example, after fan 1 shuts down, the associated valves are automatically assigned to the backup fan control. At the same time, the large DRL model trained in the cloud is compressed into a lightweight version that can be deployed on the edge, and regular incremental updates are performed. Finally, full-link traceability is performed to record the data versions of each link of perception, decision-making, and execution to trace the root cause of the fault.

[0031] Example 2: like Figure 2 As shown, the operation method of the basement ventilation and smoke exhaust system based on digital management described in this embodiment includes the following steps: S1: Build a digital twin architecture including the terminal layer, edge layer, and cloud layer to perform real-time mapping and interaction between the physical system and virtual model of basement ventilation and smoke exhaust.

[0032] In this embodiment, the terminal layer serves as the data entry point for the entire system. It consists of hardware devices such as sensors and actuators deployed throughout the basement. These devices are responsible for real-time sensing of various physical environment parameters and executing control commands. These include temperature sensors installed on the ceiling, smoke detectors distributed across different areas, and control terminals for smoke exhaust fans. These devices all directly interact with the physical space. The edge layer, primarily responsible for preliminary data processing and localized decision-making, comprises devices such as edge computing gateways and regional controllers. It enables rapid analysis and response before data is transmitted to the cloud, reducing data volume and latency. For example, if smoke concentrations suddenly rise in a particular area, the edge layer can quickly mobilize smoke exhaust equipment in that area for preliminary processing, preventing the spread of the damage. The cloud layer, the system's "brain," comprises high-performance servers and a cloud computing platform. It is responsible for global management and optimization of the basement's ventilation and smoke exhaust system. It stores massive amounts of historical data and, through big data analysis and artificial intelligence algorithms, provides decision support for the edge layer. It also provides real-time monitoring and remote control of the entire system's operating status.

[0033] For example, the specific implementation process of real-time mapping and interaction between the physical system and the virtual model includes: the terminal layer sensors will collect various parameters of the physical system, such as temperature, humidity, smoke concentration, wind speed, etc., according to the set sampling frequency, and transmit these data through Ethernet, LoRa, NB-IoT, etc. The data is transmitted to the edge layer through wired or wireless communication methods such as wireless communication; after the edge layer pre-processes the received data, it will transmit the processed data to the cloud layer, and at the same time update the corresponding parameters in the virtual model in real time based on this data, so that the virtual model can accurately reflect the current state of the physical system; the virtual model uses the updated parameters to perform simulation calculations to simulate the operation of the physical system under different working conditions, and feeds the simulation results back to the edge layer and cloud layer; the edge layer and cloud layer then formulate control strategies based on the simulation results of the virtual model and actual needs, and send control instructions to the actuators of the terminal layer. The actuators adjust the physical system according to the instructions, thereby realizing real-time interaction between the physical system and the virtual model. When the virtual model simulates that the smoke exhaust effect in a certain area is poor, the cloud layer will analyze the cause and formulate an optimization plan. The edge layer will control the smoke exhaust fan in the area to increase the speed or turn on the standby fan according to the optimization plan. At the same time, the virtual model will update the smoke exhaust effect of the area in real time until the expected goal is achieved.

[0034] For example, for a large underground parking lot that adopts a digital twin architecture, multiple temperature sensors, smoke detectors, and wind speed sensors are deployed at the terminal layer, distributed in various areas of the parking lot; corresponding edge computing gateways are set up at the edge layer, and each gateway is responsible for managing the corresponding sensors and the smoke exhaust equipment in the corresponding area; a cloud computing platform is built at the cloud layer, equipped with corresponding high-performance servers. When a vehicle spontaneously combusts in a certain area of ​​a parking lot, the temperature sensors and smoke detectors in that area quickly collect data indicating elevated temperatures and excessive smoke concentrations, and transmit the data to the corresponding edge computing gateway. After preprocessing the data, the edge gateway transmits the data to the cloud platform and updates the temperature and smoke concentration parameters of that area in the virtual model. The virtual model simulates the fire spread trend and smoke diffusion path based on these parameters, and feeds the simulation results back to the edge gateway and cloud platform. The cloud platform combines historical data and simulation results to develop a global smoke exhaust plan, and the edge gateway controls the area's smoke exhaust fans to turn on according to the plan, adjusts the fan speed to the maximum, and closes the fire curtains connecting the area to other areas. The actuators at the terminal layer perform corresponding operations according to the instructions, and the changes in the physical system are reflected in real time in the virtual model. The virtual model continues to simulate and feedback the results until the fire is under control and the smoke is effectively exhausted.

[0035] S2: Use multiple sensors for integrated monitoring at the terminal level to collect data parameters related to basement ventilation and smoke exhaust.

[0036] In this embodiment, the present invention collects basement smoke exhaust and ventilation related data through a distributed sensor network. The terminal layer deploys three layers of sensor nodes including top-level environmental parameter sensors, middle-level smoke and thermal imaging sensors, and bottom-level airflow dynamic sensors through a layered sensor network. The top-level environmental parameter sensors are used to collect temperature, humidity, 、 , PM2.5, and other environmental parameters. The mid-level smoke and thermal imaging sensors collect smoke data and thermal imaging data from the basement, while the bottom-level airflow dynamics sensors collect airflow dynamics parameters. Each sensor collects target parameters according to the varying characteristics of each parameter and a preset sampling period. The collected raw data undergoes processing within the sensor, including amplification, filtering, and A / D conversion, converting the analog signal into a digital signal. The data transmission module then transmits the digital signal to the sink node at the terminal layer or directly to the edge computing gateway at the edge layer. The sink node aggregates and converts the data received from multiple sensors before transmitting it to the edge computing gateway, forming a high-resolution perception network and improving spatial resolution.

[0037] For example, sensors are connected to the aggregation node in each fire zone via an RS485 bus. The aggregation node then transmits data to the edge computing gateway via Ethernet. During daily operation, temperature sensors monitor temperature changes within the mall in real time, uploading data when the temperature in a particular area exceeds a set threshold. Smoke sensors constantly monitor for smoke and, upon detection, immediately send an alarm signal and smoke concentration data. Wind speed sensors monitor wind speed at the air supply and exhaust vents to ensure proper ventilation. CO concentration sensors monitor airborne CO levels to ensure personnel safety. Current and vibration sensors monitor the operating status of the fans, issuing fault warnings when abnormal current flow or excessive vibration occurs. These sensors work together to enable comprehensive, real-time collection of ventilation and smoke exhaust data parameters within the underground mall.

[0038] S3: Use a highly integrated edge computing gateway to pre-process the collected data.

[0039] In this embodiment, the highly integrated edge computing gateway provides preprocessing operations for the original collected data, and performs preprocessing operations including data cleaning, data conversion, data fusion and data compression on the original data collected by different types of connected sensors. Based on data cleaning, noise, outliers and missing values ​​in the original data are removed, and data in different formats and units are converted into a unified format and unit through data conversion. Data fusion is used to comprehensively process the same type or related types of data from multiple sensors, and a data compression algorithm is used to compress the preprocessed data, thereby improving the efficiency of data transmission and storage while ensuring data accuracy.

[0040] S4: The regional controller uses a dynamic Bayesian network to fuse pre-processed multi-source data and build a local environmental state model.

[0041] In this embodiment, if Figure 3 As shown in the figure, the regional controller uses a dynamic Bayesian network for data fusion and model construction. The specific operation process is as follows: S41: Determine the state variables and observation variables of the local environment; For example, the state variables describe the key parameters of the local environmental state, such as the state of normal, high, or too high temperature; the state of normal, excessive, or seriously excessive smoke concentration; and the state of good, average, or poor ventilation; the observed variables are the preprocessed data from each sensor, such as the measurement value of the temperature sensor, the measurement value of the smoke sensor, the measurement value of the wind speed sensor, etc.

[0042] S42: Construct a dynamic Bayesian network structure including the initial network structure and the transfer network structure; For example, the initial network structure is used to calculate the initial time The probability relationship between state variables and observation variables is described, and the dependency relationship between variables is determined through expert knowledge or data learning. For example, the temperature state depends on the measurement value of the temperature sensor, and the smoke concentration state depends on the measurement value of the smoke sensor. The transfer network structure is used to describe the and The probabilistic relationship between state variables at different times describes the evolution of environmental conditions over time, that is, the temperature state at the current moment will affect the temperature state at the next moment, and the current ventilation state will affect the temperature and smoke concentration state at the next moment, etc.

[0043] S43: Determine the parameters of the dynamic Bayesian network and obtain the conditional probability distribution between the variables.

[0044] For example, the parameters of the initial network are obtained by statistical analysis of historical data or expert experience estimation, such as the probability that the temperature state is normal at the initial moment, the probability that the temperature sensor measurement value is within a certain range when the temperature state is normal, etc.; the parameters of the transfer network are also obtained by learning historical data, such as the probability that the temperature state will become high at the next moment when the current temperature state is normal, etc.

[0045] S44: Use the preprocessed multi-source data to perform inference on the dynamic Bayesian network, that is, estimate the posterior probability distribution of the state variables based on the values ​​of the observed variables.

[0046] For example, the inference process uses a forward-backward algorithm, based on the forward algorithm to calculate at each moment , given before The posterior probability of the state variable at the time of observation is the forward probability; then the backward algorithm is used to calculate the probability at each moment. , given from arrive The posterior probability of the state variable when the observation value is , that is, the backward probability; then combine the forward probability and the backward probability to obtain at each moment The posterior probability distribution of the state variables is used to determine the state of the local environment. Finally, based on the posterior probability distribution of the state variables obtained by reasoning, a local environment state model is constructed. This model can reflect the dynamic changes of the local environment in real time and provide a basis for subsequent state prediction and control decisions.

[0047] S45: Construct the corresponding local environmental state model as the basis for edge layer state prediction and emergency plan generation.

[0048] S5: The edge layer uses deep reinforcement learning algorithms and digital twin models to simulate multiple scenarios for real-time status prediction and generate an emergency plan library.

[0049] In this embodiment, combined with Figure 4 As shown in the figure, the edge layer combines the deep reinforcement learning algorithm of the deep reinforcement learning network and the long short-term memory network to effectively process time series data and optimize decisions. Combined with the simulation capabilities of the digital twin model, it can predict the status of the basement ventilation and smoke exhaust system in real time and simulate multiple scenarios, thereby generating an emergency plan library. The specific process steps include: S51: Determine the state space, action space and reward function.

[0050] For example, the state space includes the state parameters of the current local environment and the operating status of the equipment; the action space includes the control actions that can be taken by the edge layer; the reward function is used to evaluate the effect of the action. For example, when an action is taken, a positive reward will be obtained if the smoke concentration decreases, the temperature drops, the energy consumption of the equipment is reduced, etc., and a negative reward will be obtained otherwise.

[0051] S52: Build a DQN+LSTM model.

[0052] For example, LSTM networks are used to process historical state sequence data, capturing long-term dependencies in time series data. Their output serves as input to deep reinforcement learning algorithms. DQN, on the other hand, uses a deep neural network to fit an action-value function, which is used to evaluate the expected benefit of taking an action in a specific state. During model training, an experience replay mechanism is used to store the experience generated by the agent's interaction with the environment, including state, action, reward, and next state, in an experience pool. This experience is then randomly sampled for training to improve model stability. For example, the target network is used synchronously to calculate the target value, actions are selected and parameters are updated based on the evaluation network, and the parameters of the evaluation network are copied to the target network every certain number of steps to reduce fluctuations during training.

[0053] S53: Use digital twin models to simulate various scenarios.

[0054] Exemplary scenarios include daily operations, equipment failures, and fires. During the simulation process, the output of the digital twin model is used as environmental feedback to interact with the DQN+LSTM model. The agent continuously tries different actions, receives feedback based on the reward function, and gradually optimizes its action strategy. Based on the optimized strategy, real-time predictions of the system state in different scenarios are made. These predictions include future temperature trends, smoke density diffusion paths, wind speed distribution, and more.

[0055] S54: The LSTM network uses historical state data and current state data to output the predicted value of the future state.

[0056] For example, DQN selects the optimal control action based on the predicted state value, ensuring the system is prepared for any response. Finally, a library of emergency response plans is generated based on the simulated scenarios and the corresponding optimal control actions. Each emergency response plan includes a scenario description, triggering conditions, control measures, and expected effects. When a scenario matching a specific plan occurs during actual operation, that plan is quickly invoked for processing.

[0057] S6: The central server at the cloud layer runs a dual-track decision engine, combining a deterministic rule base with an adaptive deep reinforcement learning model to perform overall adaptive collaborative control.

[0058] In this embodiment, a dual-track decision engine at the cloud layer combines the stability of deterministic rules with the adaptability of reinforcement learning to globally optimize and coordinate control of the entire basement ventilation and smoke exhaust system. The deterministic rule base is a set of fixed control rules developed based on fire safety regulations, engineering experience, and expert knowledge. These rules define standardized control measures to be taken under specific conditions, such as "When the smoke concentration in a certain area exceeds 0.1 mg / m³, immediately turn on all smoke exhaust fans in that area and close fire shutters in adjacent areas" and "When the operating current of a smoke exhaust fan exceeds 120% of the rated current, automatically reduce its speed to 80% of the rated speed." These rules are clear, fast, and reliable, and can handle common and regular situations. The adaptive deep reinforcement learning model uses a similar but larger-scale deep reinforcement learning algorithm as the edge layer, including a deep deterministic policy gradient algorithm. This model takes the entire basement's environmental conditions, including temperature, smoke concentration, and equipment operating status, as input and outputs a global control policy. Through interaction with the global environment, the policy is continuously learned and optimized to adapt to complex and changing conditions.

[0059] For example, combined Figure 5 and Figure 6As shown in Figure 1, the implementation steps of the dual-track decision engine are as follows: The central server receives preprocessed data and local environmental state models uploaded by each edge layer. It integrates this information to construct a global environmental state model, reflecting the operating status of the entire basement's ventilation and smoke exhaust system. The dual-track decision engine simultaneously uses a deterministic rule base and an adaptive deep reinforcement learning model to analyze the global environmental state. The deterministic rule base matches the current state against pre-set rules. If a fully matching rule exists, the corresponding control instructions are directly generated. If no fully matching rule exists or the rule execution is ineffective (e.g., smoke concentration continues to rise after the rule is executed), the adaptive deep reinforcement learning model is activated. Based on the global state, the reinforcement learning model calculates the optimal control strategy. This strategy may supplement or adjust the deterministic rules. For example, in addition to the rule-mandated operation of some exhaust fans, it may also activate backup fans to enhance exhaust efficiency. The control strategies generated by the two decision paths are then integrated and evaluated using metrics such as exhaust efficiency, energy consumption, and safety. The control strategy with the best overall performance is selected as the final decision. Finally, the final control instructions are sent to each edge layer, which then forwards them to the actuators at the terminal layer, achieving coordinated control of the entire system. Simultaneously, the decision-making process and execution results are fed back to the reinforcement learning model for continuous updating and optimization, ensuring that the model can continuously adapt to new situations.

[0060] S7: Synchronously retain the hybrid decision-making mechanism, perform adaptive collaborative control of the system under normal conditions, and use manual intervention terminals in emergency conditions to convert manual experience into system knowledge to form a human-machine collaborative decision-making mechanism.

[0061] In this embodiment, a hybrid decision-making mechanism combines the advantages of automatic control and manual decision-making to ensure efficient and reliable operation of the basement ventilation and smoke exhaust system under all circumstances. Under normal conditions, adaptive collaborative control means that the system automatically adjusts ventilation and smoke exhaust equipment based on real-time monitoring data, state prediction results, and optimized control strategies, achieving coordinated operation of equipment in various areas without manual intervention. Specifically, the edge layer and cloud layer continuously update the system's global state model through real-time data interaction and analysis. The cloud layer's dual-track decision engine generates globally optimal control instructions based on the global state model, combining a deterministic rule base and an adaptive deep reinforcement learning model. These control instructions are transmitted via the edge layer to the actuators at the terminal layer, which adjust the equipment's operating state, such as fan speed and valve opening, according to the instructions. Simultaneously, the system monitors and evaluates the equipment's operating performance in real time. If actual performance deviates from expected targets, the control strategy is automatically adjusted, forming a closed-loop control system. This adaptive collaborative control system can adapt to environmental changes in real time, improving system efficiency and energy conservation.

[0062] It should be noted that, in the present invention, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0063] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. The basement ventilation and smoke exhaust system based on digital management is characterized by: It includes the following functional units: global perception fusion unit, edge intelligent gateway unit, digital twin drive unit, deep reinforcement learning decision unit, dynamic execution verification unit, and adaptive adjustment control unit; The global perception fusion unit is used to collect device status and environmental data in real time across all dimensions; The edge intelligent gateway unit; Used to configure a separate edge computing unit in each area to receive perception data and perform multi-source data fusion; The digital twin drive unit is used to solve the three-dimensional CFD flow field in real time, preview the impact of control instructions on the flow field, and generate virtual sensor data for uncovered areas to compensate for blind spots. The deep reinforcement learning decision unit is used to perform real-time state change analysis and realize intelligent switching of system modes through a multi-threshold event trigger mechanism. The dynamic execution verification unit is used to convert the strategy into device control signals, compare the predicted value with the actual sensor feedback, and dynamically correct the CFD parameter error. When the deviation exceeds the threshold, it rolls back to the safe strategy. The adaptive adjustment control unit is used to optimize the sensor network structure in combination with dynamic topology adjustment, and perform working condition prediction and self-learning based on the controller of the deep reinforcement learning algorithm to perform adaptive adjustment control.

2. A basement ventilation and smoke exhaust method based on digital management, characterized by: The method comprises the following steps: S1: Build a digital twin architecture consisting of terminal, edge, and cloud layers to perform real-time mapping and interaction between the physical system and virtual model of basement ventilation and smoke exhaust; S2: Use multiple sensors for integrated monitoring at the terminal level to collect data parameters related to basement ventilation and smoke exhaust; S3: Use a highly integrated edge computing gateway to pre-process the collected data; S4: The regional controller uses a dynamic Bayesian network to fuse pre-processed multi-source data and build a local environment state model; S5: The edge layer uses deep reinforcement learning algorithms and digital twin models to simulate multiple scenarios for real-time status prediction and generate an emergency plan library; S6: The cloud-layer central server runs a dual-track decision engine, combining a deterministic rule base with an adaptive deep reinforcement learning model to perform overall adaptive collaborative control; S7: Synchronously retain the hybrid decision-making mechanism, perform adaptive collaborative control of the system under normal conditions, and use manual intervention terminals in emergency conditions to convert manual experience into system knowledge to form a human-machine collaborative decision-making mechanism.

3. The basement ventilation and smoke exhaust method based on digital management according to claim 2, characterized in that: The real-time mapping and interaction between the physical system and the virtual model for basement ventilation and smoke exhaust specifically includes: The terminal layer sensors collect various parameters of the physical system according to the set sampling frequency; The edge layer pre-processes the received data and transmits it to the cloud layer, while updating the corresponding parameters in the virtual model in real time based on this data; The virtual model uses the updated parameters to perform simulation calculations, simulating the operation of the physical system under different working conditions and feeding the simulation results back to the edge layer and cloud layer; The edge layer and cloud layer then formulate control strategies based on the simulation results of the virtual model and actual needs, and send control instructions to the actuators at the terminal layer. The actuators adjust the physical system according to the instructions, realizing real-time interaction between the physical system and the virtual model.

4. The basement ventilation and smoke exhaust method based on digital management according to claim 3 is characterized by: The terminal layer deploys three layers of sensor nodes including top layer, middle layer and bottom layer through a layered sensor network, and performs real-time parameter collection in layers.

5. The basement ventilation and smoke exhaust method based on digital management according to claim 4 is characterized in that: The S4 specifically includes: S41: Determine the state variables and observation variables of the local environment; The state variables are key parameters used to describe the local environment state, and the observed variables are pre-processed data values ​​from each sensor; S42: Construct a dynamic Bayesian network structure including the initial network structure and the transfer network structure; The initial network structure is used to determine the dependency relationship between variables through expert knowledge or data learning, and obtain the probabilistic relationship between the state variables at the initial moment and the observed variables; the transition network structure is used to describe the probabilistic relationship between the state variables at different moments, and obtain the evolution law of the environmental state over time; S43: Determine the parameters of the dynamic Bayesian network to obtain the conditional probability distribution between the variables; The parameters of the initial network and the transfer network are obtained by statistical analysis of historical data or expert experience estimation; S44: Inference of dynamic Bayesian networks using preprocessed multi-source data; S45: Construct the corresponding local environmental state model as the basis for edge layer state prediction and emergency plan generation.

6. The basement ventilation and smoke exhaust method based on digital management according to claim 5 is characterized by: The reasoning of the dynamic Bayesian network specifically includes: Estimate the posterior probability distribution of the state variable based on the value of the observed variable; Based on the forward algorithm, calculation at each moment , given before The posterior probability of the state variable at the time of observation is used to obtain the forward probability; Calculate at each moment through the backward algorithm , given from To the specified period The posterior probability of the state variable when the observation value is , and the backward probability is obtained; Then combine the forward probability and the backward probability to get The posterior probability distribution of the state variables determines the state of the local environment; Finally, based on the posterior probability distribution of the state variables obtained by reasoning, a local environment state model is constructed to reflect the dynamic changes of the local environment in real time.

7. The basement ventilation and smoke exhaust method based on digital management according to claim 6, characterized in that: The S5 specifically includes: S51: Determine the state space, action space and reward function; S52: Constructing a DQN+LSTM model; the DQN+LSTM model uses an LSTM network and a deep neural network; S53: Calculate the target value using the target network, select actions and update parameters based on the evaluation network; S54: Use the digital twin model to simulate multiple scenarios, use the output of the digital twin model as environmental feedback, and interact with the DQN+LSTM model; S55: Obtain feedback based on the reward function, gradually optimize the action strategy, and based on the optimized strategy, make real-time predictions and outputs of the system status in different scenarios.

8. The basement ventilation and smoke exhaust method based on digital management according to claim 7, characterized in that: During the prediction, the LSTM network uses historical state data and current state data to output the predicted value of the future state. The DQN selects the optimal control action based on the predicted state value. Based on multiple simulated scenarios and corresponding optimal control actions, an emergency plan library is generated. Each emergency plan includes scenario description, trigger conditions, control measures, and expected effects. When a scenario matching a plan occurs in actual operation, the plan is called for processing.

9. The basement ventilation and smoke exhaust method based on digital management according to claim 8, characterized in that: The dual-track decision-making engine includes a deterministic rule base that formulates fixed control rules based on fire regulations, engineering experience, and expert knowledge, and adaptive deep reinforcement learning. The adaptive deep reinforcement learning model uses a deep reinforcement learning algorithm, taking the environmental status data of the entire basement as input, to optimize the output of the global control strategy.

10. The basement ventilation and smoke exhaust method based on digital management according to claim 9, characterized in that: The operation process of the dual-track decision engine includes: S61: The dual-track decision engine simultaneously uses a deterministic rule base and an adaptive deep reinforcement learning model to analyze the global environment state; S62: The deterministic rule library matches the current state according to the preset rules. If there is a completely matching rule, the corresponding control instruction is directly generated. If there is no completely matching rule, the adaptive deep reinforcement learning model is started; S63: The deep reinforcement learning model calculates the optimal control strategy based on the global state, supplementing or adjusting the deterministic rules; S64: Fusing and evaluating the control strategies generated by the two decision paths, and selecting the control strategy with the best comprehensive performance as the final decision; S65: The final control instructions are sent to each edge layer, which then forwards them to the actuators in the terminal layer to achieve coordinated control of the entire system. S66: Synchronously feed back the decision-making process and execution results to the deep reinforcement learning model for continuous updating and optimization of the model.

Images