A high-rise building wind effect active suction and blowing control method based on deep reinforcement learning

Abstract

This invention relates to an active intake and blowing control method for wind effects in high-rise buildings based on deep reinforcement learning. This invention relates to the fields of active flow control and wind-induced vibration control in high-rise buildings. Addressing the shortcomings of existing wind-resistant measures for high-rise buildings in complex wind environments and the insufficient feasibility of active flow control engineering, this invention discloses an active intake and blowing control method for wind effects in high-rise buildings based on deep reinforcement learning. The method includes: constructing an active intake and blowing control device for wind effects in high-rise buildings, constructing a wind tunnel test environment, establishing a deep reinforcement learning model, and realizing interactive training between the model and the wind tunnel test environment. This method can adaptively adjust the intake and blowing flow rates according to real-time wind environment information and structural wind-induced vibration response information, changing the flow field structure around the building, reducing unfavorable flow separation and wake vortex shedding, thereby reducing wind load and wind-induced vibration response in high-rise buildings.

Description

Technical Field

[0001] This invention relates to the field of active flow control and high-rise building wind vibration control technology, and is a high-rise building wind effect active intake and blowing control method based on deep reinforcement learning. Background Technology

[0002] High-rise buildings in my country are characterized by their large number and wide distribution. Especially in coastal areas where strong typhoons are frequent, high-rise buildings are prone to increased wind load on the main structure, intensified wind-induced vibrations, and wind-induced damage to the external envelope under such extreme wind conditions. Therefore, improving the wind resistance of high-rise buildings has become a key technical problem that urgently needs to be solved in the field of high-rise building wind engineering.

[0003] Existing wind-resistant measures for high-rise buildings mainly include structural measures, energy-dissipating and vibration-damping measures, and aerodynamic control measures. Structural measures resist wind loads by increasing structural mass and stiffness, but are easily constrained by structural form and economic costs. Energy-dissipating and vibration-damping measures mainly improve structural wind resistance by installing damping devices, but these devices have high maintenance and replacement costs during the building's service life. In contrast, aerodynamic control measures weaken unfavorable flow effects by adjusting the flow field structure near the building surface, showing greater application potential. In particular, active flow control methods such as intake and blowing can significantly reduce wind loads and wind-induced vibration response by regulating flow separation and wake vortex shedding.

[0004] Furthermore, in the past decade or so, research on using double-skin facades (DSFs) to improve the comfort and structural wind resistance of high-rise buildings has received increasing attention. A double-skin facade can be defined as a building envelope consisting of two layers of curtain walls, with a cavity or channel between them. This type of channel provides feasible space for active flow control, making it possible to install air intake and exhaust devices on the outer curtain wall. By installing air intake and exhaust devices on the outer curtain wall, the flow field near the building surface can be adjusted, thereby reducing wind load and wind-induced vibration response on the building surface. However, existing research on double-skin facades mainly focuses on environmental regulation and aerodynamic control under single conditions. There is a lack of engineering implementation plans that integrate with the building envelope system for wind-induced vibration control under extreme wind conditions in high-rise buildings. At the same time, the adaptability of existing aerodynamic control methods under complex conditions such as multiple wind speeds and directions remains insufficient. Therefore, there is an urgent need to propose an active air intake and exhaust control method for wind effects in high-rise buildings that combines engineering feasibility, closed-loop control capability, and multi-condition adaptability. Summary of the Invention

[0005] To address the shortcomings of existing wind-resistant measures for high-rise buildings in adapting to complex wind environments and the lack of feasibility of active flow control engineering, this invention provides an active air intake and blowing control method for wind effects in high-rise buildings based on deep reinforcement learning, so as to achieve active regulation of wind load and wind vibration response in high-rise buildings.

[0006] This invention provides the following technical solutions: A method for active intake and exhaust control of wind effects in high-rise buildings based on deep reinforcement learning, the method comprising the following steps: Step 1: Construct an active air intake and blowing control device for high-rise building wind effects; Step 2: Design and manufacture a wind tunnel test model based on the shape and structural dynamic characteristics of the high-rise building, and simulate the wind environment of the high-rise building in the wind tunnel; set up test devices corresponding to the externally perforated curtain wall, the air intake and blowing wind-resistant components, the wind environment monitoring sensors and the structural response monitoring sensors on the wind tunnel test model to obtain wind environment information and structural wind vibration response information under different working conditions. Step 3: Construct a deep reinforcement learning model, taking wind environment information and structural wind vibration response information as state information inputs, taking the control parameters of the air intake and blowing wind-resistant components as action signal outputs, and constructing a reward function based on the wind effect control target of high-rise buildings to establish a deep reinforcement learning model; Step 4: Connect the deep reinforcement learning model to the wind tunnel test environment, so that the deep reinforcement learning model outputs control of the intake and blowing air flow based on the real-time collected state information, and iteratively updates the model based on the feedback results of the wind tunnel test to obtain the active intake and blowing air control strategy for high-rise building wind effects.

[0007] Preferably, an externally perforated curtain wall is installed around the high-rise building, and a double-layer curtain wall channel is formed between the externally perforated curtain wall and the exterior facade of the high-rise building through a perimeter connecting plate; multiple holes are provided on the surface of the externally perforated curtain wall, and air-absorbing and air-blowing wind-resistant components are arranged at the holes. The air intake and air blowing wind-resistant assembly includes an intake fan, an air blowing fan, a rotating shaft, fan blades, intake holes, and air blowing holes; Wind environment monitoring sensors and structural response monitoring sensors are installed on the top floor or in the middle of high-rise buildings.

[0008] Preferably, the air intake and air blowing wind-resistant components are arranged in the flow separation area on the side of the building, the wake area on the leeward side, and the area affected by the incoming flow on the windward side. Two or more perimeter-shaped connecting plates are arranged along the height of the high-rise building to form multiple double-layer curtain wall channels between the externally perforated curtain wall and the exterior facade of the high-rise building. In the air intake and air blowing wind resistance assembly, the air intake and air blowing fans are connected to the rotating shaft. The air intake and air blowing fans are used to provide power. The rotating shaft is used to adjust the position of the air intake hole and the air blowing hole. The air intake hole is connected to the air intake fan through a pipe. The air blowing hole is connected to the air blowing fan through a pipe. The air intake and air blowing flow rate is adjusted by adjusting the fan voltage. Wind environment monitoring sensors are used to monitor incoming wind speed and direction, while structural response monitoring sensors are used to monitor the displacement, velocity, and acceleration response of high-rise buildings.

[0009] Preferably, the wind tunnel test model is a scaled-down aeroelastic model corresponding to the high-rise building. The wind tunnel test environment simulates the wind conditions of the high-rise building under different wind environments by adjusting the incoming wind speed and turbulence intensity, pulsating wind speed spectrum and wind direction angle of different wind field types.

[0010] Preferably, the state space of the deep reinforcement learning model includes structural wind vibration response information and wind environment information from multiple monitoring points of the building; the action space of the deep reinforcement learning model includes control signals for each air intake and air blowing wind-resistant component, and the control signals are used to adjust the air intake flow rate and the air blowing flow rate.

[0011] Preferably, the reward function of the deep reinforcement learning model It is based on the structural response and wind pressure configuration within a single action cycle, and is expressed by the following formula:

[0012] ; in, , , , and These are the displacement, velocity, acceleration, extreme wind pressure, and energy consumption performance evaluation indicators for the k-th action cycle. , , , , These are the reference displacement, reference velocity, reference acceleration, reference extreme wind pressure, and reference energy consumption performance evaluation indicators for the k-th action cycle. , , , and These are non-negative weighting coefficients.

[0013] Preferably, the deep reinforcement learning training process is carried out in a closed loop of "state perception - strategy decision - control execution - environmental feedback". That is, the state information is used as the input of the deep reinforcement learning algorithm model, the deep reinforcement learning algorithm model outputs control action signals, the intake or blowing air flow control is implemented by the intake and blowing air wind resistance component, and the wind tunnel test environment provides feedback on the next state and reward value.

[0014] Preferably, the deep reinforcement learning algorithm model is trained under basic operating conditions to obtain a basic policy model, and the basic policy model is transferred to the target operating conditions to directly call or output the inhalation and exhalation control signal after iterative updates.

[0015] A computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement an active intake and exhaust control method for wind effects in high-rise buildings based on deep reinforcement learning.

[0016] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement an active intake and exhaust control method for wind effects in high-rise buildings based on deep reinforcement learning.

[0017] The present invention has the following beneficial effects: This invention combines an externally perforated curtain wall, four-sided connecting plates, and an air-suction and air-blowing wind-resistant component to form an active wind-resistant implementation carrier that can be practically deployed on the periphery of high-rise buildings, thus solving the problem of insufficient engineering feasibility of existing active flow control methods.

[0018] This invention combines deep reinforcement learning algorithm models with interactive training in a wind tunnel test environment, enabling control strategies to be built on experimental feedback. This allows for a more realistic reflection of the wind load and wind vibration response characteristics of high-rise buildings under complex wind conditions, thereby improving the reliability and effectiveness of the control strategies.

[0019] This invention controls the flow field structure around a building by implementing air intake or blowing control over the flow separation area on the side of the building, the wake area on the leeward side, and the incoming flow area on the windward side. This can weaken unfavorable flow separation and wake vortex shedding, thereby reducing the wind load and wind vibration response of high-rise buildings.

[0020] This invention introduces a control strategy for basic operating condition training and target operating condition migration, enabling the trained strategy to be directly invoked or quickly updated under complex wind field conditions such as different wind speeds and directions, reducing retraining costs and improving the generalization ability and application efficiency of the method under multiple operating conditions.

[0021] This invention can be integrated into the design phase of high-rise buildings and can also be used for the renovation of the external envelope system of existing high-rise buildings, thus having good engineering applicability and promotional value. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 Displayed as a schematic diagram of a high-rise building; Figure 2 The diagram shows a four-sided connecting plate. Figure 3 The diagram shows an externally perforated curtain wall. Figure 4 The diagram shows a wind-resistant component that is designed for air intake and air blowing. Figure 5 The diagram shows an active air intake and blowing device for wind effect in high-rise buildings. Figure 6 The image shown is an exploded view of an active air intake and blowing device for wind effects in a high-rise building. Figure 7 The diagram shows a double-layered curtain wall passageway. Figure 8 This diagram illustrates the impact of active air intake and blowing on the wake of a high-rise building.

[0024] Among them, 1-high-rise building, 2-four-sided connecting plate, 3-externally perforated curtain wall, 4-intake fan, 5-blowing fan, 6-rotating shaft, 7-fan blade, 8-intake hole, 9-blowing hole. Detailed Implementation

[0025] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] The present invention will be described in detail below with reference to specific embodiments. Specific Implementation Example 1: according to Figures 1 to 8 As shown, the specific optimized technical solution adopted by the present invention to solve the above-mentioned technical problems is: The present invention relates to an active air intake and blowing control method for wind effects in high-rise buildings based on deep reinforcement learning.

[0028] This invention provides a method for active intake and exhaust control of wind effects in high-rise buildings based on deep reinforcement learning. The method includes the following steps: A method for active intake and exhaust control of wind effects in high-rise buildings based on deep reinforcement learning includes the following steps: S1. Constructing an active air intake and blowing control device for wind effects in high-rise buildings: An externally perforated curtain wall is installed around the high-rise building, and a double-layer curtain wall channel is formed between the externally perforated curtain wall and the exterior facade of the high-rise building through four-sided connecting plates; multiple holes are set on the surface of the externally perforated curtain wall, and air intake and blowing wind-resistant components are arranged at the holes. The air intake and blowing wind-resistant components include an air intake fan, a blowing fan, a rotating shaft, fan blades, air intake holes, and air blowing holes; wind environment monitoring sensors and structural response monitoring sensors are arranged on the top of the high-rise building or on key floors. S2. Constructing a wind tunnel test environment: Design and manufacture a wind tunnel test model based on the shape and structural dynamic characteristics of the high-rise building, and simulate the wind environment of the high-rise building in the wind tunnel; arrange test devices corresponding to the externally perforated curtain wall, the air intake and blowing wind-resistant components, the wind environment monitoring sensors and the structural response monitoring sensors on the wind tunnel test model to obtain wind environment information and structural wind vibration response information under different working conditions. S3. Constructing the DRL algorithm model: The wind environment information and structural wind vibration response information are used as state information inputs, the control parameters of the air intake and blowing wind-resistant components are used as action signals outputs, and a reward function is constructed based on the wind effect control target of high-rise buildings to establish a deep reinforcement learning algorithm model. S4. Implement interactive training between DRL algorithm and wind tunnel test: Connect the deep reinforcement learning algorithm model with the wind tunnel test environment, so that the deep reinforcement learning algorithm model outputs control of intake and blowing flow based on the real-time collected state information, and iteratively updates the model based on the feedback results of the wind tunnel test to obtain the active intake and blowing control strategy for wind effect of high-rise buildings.

[0029] Further, in step S1, the perforations on the surface of the externally perforated curtain wall and the air intake and blowing wind-resistant components are arranged in the flow separation area on the side of the building, the wake area on the leeward side, and the incoming flow area on the windward side. The air intake and blowing fans are connected to a rotating shaft, which provides power. The rotating shaft is used to adjust the positions of the air intake and blowing holes. The air intake holes are connected to the air intake fans via pipes, and the air blowing holes are connected to the air blowing fans via pipes. The air intake and blowing flow rates are adjusted by regulating the fan voltage. Two or more perimeter connecting plates are arranged along the height of the high-rise building to form multiple double-layer curtain wall channels between the externally perforated curtain wall and the exterior facade of the high-rise building. The wind environment monitoring sensor is used to monitor the incoming wind speed and direction, and the structural response monitoring sensor is used to monitor the displacement, velocity, and acceleration response of the high-rise building.

[0030] In step S2, the wind tunnel test model is a scaled-down aeroelastic model corresponding to a high-rise building. The wind tunnel test environment is used to simulate the wind conditions of a high-rise building under different wind environments by adjusting the incoming wind speed and turbulence intensity, pulsating wind speed spectrum and wind direction angle of different wind field types.

[0031] In step S3, the reward function of the deep reinforcement learning algorithm model is... Based on the structural response and wind pressure configuration within one action cycle, its expression is: .

[0032] in, ; in, , , , and These are the displacement, velocity, acceleration, extreme wind pressure, and energy consumption performance evaluation indicators for the k-th action cycle. , , , , These are the reference displacement, reference velocity, reference acceleration, reference extreme wind pressure, and reference energy consumption performance evaluation indicators for the k-th action cycle. , , , and These are non-negative weighting coefficients.

[0033] In step S4, the deep reinforcement learning training process follows a closed-loop approach of "state perception—policy decision-making—control execution—environmental feedback." Specifically, state information is used as input to the deep reinforcement learning algorithm model, which outputs control action signals. The inhalation and blowing wind-resistant components implement inhalation or blowing control at different intensities. The wind tunnel test environment then provides feedback on the next state and reward value. Furthermore, the deep reinforcement learning algorithm model is trained under basic operating conditions to obtain a basic policy model. This basic policy model is then transferred to the target operating conditions, directly invoked, or iteratively updated to output inhalation and blowing control signals.

[0034] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement an active intake and exhaust control method for wind effects in high-rise buildings based on deep reinforcement learning.

[0035] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement an active air intake and exhaust control method for wind effects in high-rise buildings based on deep reinforcement learning. Specific Implementation Example 2: The only difference between Embodiment 2 and Embodiment 1 of this application is that: This invention provides a method for active wind-induced air intake and blowing control of high-rise buildings based on deep reinforcement learning. The method comprises four steps: constructing an active wind-induced air intake and blowing control device for high-rise buildings, building a wind tunnel testing environment, establishing a deep reinforcement learning algorithm model, and interactive training of the wind tunnel test and the deep reinforcement learning algorithm model. The final result is an active wind-induced air intake and blowing control strategy suitable for high-rise buildings. This method can be used for wind-resistant design of new high-rise buildings as well as for functional upgrading and renovation of the external envelope systems of existing high-rise buildings.

[0037] like Figure 1 , Figure 2 , Figure 3 and Figure 5 As shown, an externally perforated curtain wall 3 is installed on the periphery of the high-rise building 1 and connected to the main body of the high-rise building via perimeter connecting plates, forming a double-layer curtain wall passage between the externally perforated curtain wall 3 and the exterior facade of the high-rise building 1. Two or more perimeter connecting plates 2 can be arranged along the building height direction, thereby forming multiple independent or interconnected double-layer curtain wall passages vertically. The surface of the externally perforated curtain wall 3 has multiple holes that can be opened or closed.

[0038] like Figure 4 , Figure 5 and Figure 6 As shown, the air intake and air blowing wind resistance assembly includes an intake fan 4, an air blowing fan 5, a rotating shaft 6, a fan blade 7, an intake hole 8, and an air blowing hole 9. The rotating shaft 6 drives the fan blade 7 to rotate, changing the position of the intake and air blowing holes so that the hole positions can be located in the horizontal direction, vertical direction, or 45° direction, thereby forming a directional airflow at the intake hole or air blowing hole.

[0039] By adjusting the input control signals of the suction fan 4 and the blowing fan 5, the suction and blowing flow rates of the suction hole 8 and the blowing hole 9 can be changed to achieve suction, blowing, or stop control in the corresponding areas. The control signal can be a control voltage. In a preferred embodiment, the control signal is a voltage signal, which is amplified by the drive module and then input to the fan to achieve continuous adjustment of the suction and blowing flow rates.

[0040] like Figure 5 , Figure 7 and Figure 8As shown, when the air intake and blowing wind-resistant components are working, airflow channels are formed within the double-layer curtain wall. The local intake and blowing generated by the intake holes 8 and blowing holes 9 interact with the flow field around the building, thereby changing the flow characteristics around the high-rise building. For the crosswind flow separation area, blowing can delay flow separation or adjust the shear layer trajectory; for the leeward wake area, blowing can change the near-wake vortex structure and suppress periodic vortex shedding; for the windward incoming flow area, intake and / or blowing can reduce local positive pressure peaks and achieve a guiding effect.

[0041] Wind environment monitoring sensors are installed on the top of high-rise buildings or on key floors to monitor incoming wind speed and direction; structural response monitoring sensors are installed on the top of high-rise buildings or on key floors to monitor one or more of displacement, velocity, and acceleration.

[0042] A wind tunnel test model corresponding to the target high-rise building is designed and constructed, and a wind environment matching the actual working conditions is created in the wind tunnel. The wind tunnel test model is preferably a scaled-down aeroelastic model to reflect the geometric shape and structural dynamic characteristics of the high-rise building. The wind tunnel test environment can simulate the wind conditions of the high-rise building under uniform wind fields, atmospheric boundary layer wind fields, deflecting wind fields, or gusts by adjusting the incoming wind speed, wind direction angle, surrounding disturbances, and turbulence characteristics.

[0043] During wind tunnel testing, status information is sent to the deep reinforcement learning algorithm model via a communication interface. This communication interface can employ industrial communication protocols such as serial communication, TCP / IP, or RS485.

[0044] A deep reinforcement learning algorithm model is used to generate active control strategies based on real-time state information. The state space includes structural wind vibration response information, wind environment information, the current operating state of the intake and blowing wind-resistant components, and their state information from the previous moment. The action space includes the intake flow rate and blowing flow rate of each intake and blowing wind-resistant component. In a preferred embodiment, the deep reinforcement learning algorithm model employs one of the following: soft actor-critic algorithm (SAC), PPO, or DDPG, for the continuous action space.

[0045] To quantitatively describe the wind effect control performance of high-rise buildings, performance evaluation indices for high-rise buildings can be constructed based on the structural response at multiple monitoring points, and performance evaluation indices for building envelopes such as glass curtain walls can be constructed based on the wind pressure coefficient at multiple monitoring points. For the k-th action cycle, periodic statistics of displacement, velocity, acceleration, wind pressure, and energy consumption are calculated respectively, and a comprehensive index is obtained through weighted synthesis. The weight of each monitoring point can be determined based on the importance of the measuring point location, the degree of participation of the principal vibration mode, and the area of ​​the external envelope it represents.

[0046] The weights of each monitoring point can also be the same. For example, the displacement performance evaluation index can be expressed as:

[0047] in, Let be the root mean square value of the i-th displacement monitoring point during the k-th action cycle. For the corresponding weight coefficients, and satisfying .

[0048] Wind pressure index of building envelope The maximum absolute wind pressure at each wind pressure measuring point within one operating cycle is used to construct the system.

[0049] Control energy consumption indicators It is characterized by the integral of the input power of each intake fan and blowing fan within one operating cycle.

[0050]

[0051] in, , , , and These are the displacement, velocity, acceleration, extreme wind pressure, and energy consumption performance evaluation indicators for the k-th action cycle. , , , , These are the reference displacement, reference velocity, reference acceleration, reference extreme wind pressure, and reference energy consumption performance evaluation indicators for the k-th action cycle. , , , and The weight coefficient is non-negative; when an indicator does not participate in the control objective, its corresponding weight coefficient can be 0.

[0052] The reward function, constructed based on the aforementioned comprehensive performance index, can be expressed as:

[0053] When the overall performance index decreases, the reward value increases, thereby guiding the deep reinforcement learning model to search for an active air intake and blowing control strategy that can simultaneously reduce the wind vibration response of high-rise buildings and the wind pressure of curtain walls while also controlling energy consumption.

[0054] The interactive training of wind tunnel testing and the deep reinforcement learning algorithm model follows a closed-loop approach of "state perception—policy decision-making—control execution—environmental feedback." Specifically: First, wind environment monitoring sensors and structural response monitoring sensors collect wind environment information and structural wind vibration response information in real time; second, the collected state information is input into the deep reinforcement learning algorithm model, which outputs control action signals for the current action cycle; subsequently, the control action signals are sent to the air intake and air blowing wind-resistant components, enabling the corresponding orifice area to implement air intake or air blowing control of a predetermined intensity; finally, after the action cycle ends, the wind tunnel test environment returns new state information and corresponding reward values ​​as the basis for the next round of policy updates.

[0055] A calibration relationship is pre-established between the control signals and actuator outputs of the air intake and air blowing wind-resistant components. Taking voltage control as an example, a mapping relationship between the control voltage and the average intake flow rate and average blowing flow rate can be established through pre-testing. During training, the deep reinforcement learning algorithm model outputs a control voltage, which is amplified by the drive module and applied to the fan. The fan achieves different power according to different control voltages, thereby forming corresponding intake and blowing flow rates at the intake and blowing orifices.

[0056] The action cycle is a pre-set fixed duration, which can be determined comprehensively based on the structural response frequency, sensor sampling frequency, wind tunnel test duration, and algorithm convergence speed.

[0057] The deep reinforcement learning algorithm model is first trained under basic operating conditions to obtain a basic policy model. These basic operating conditions can be control conditions under typical wind speeds, typical wind direction angles, typical turbulence intensities, or typical wind field types. Subsequently, this basic policy model is transferred to target operating conditions with similar incoming flows, directly calling the model or outputting intake and exhaust control signals after a few iterations. These target operating conditions include, but are not limited to, conditions with different wind speeds, different wind direction angles, different turbulence intensities, and different wind field types. Through transfer learning, the time required for retraining under new operating conditions can be reduced, improving the generalization ability and application efficiency of the control policy under multiple operating conditions.

[0058] The working process of this invention is as follows: First, based on the geometric shape of the high-rise building, the wind resistance target, and the expected control area, the arrangement of the externally perforated curtain wall, the double-layer curtain wall passage, and the air intake and blowing wind resistance components are completed; second, wind environment monitoring sensors and structural response monitoring sensors are arranged on the top of the building or on key floors, and the wind tunnel scale model and corresponding test environment are constructed simultaneously; third, wind environment information and structural wind vibration response information under different control conditions are collected using the wind tunnel test platform, and the state space, action space, and reward function of the deep reinforcement learning algorithm model are established; then, the deep reinforcement learning algorithm model is connected to the wind tunnel test environment in real time, and the active air intake and blowing control strategy is obtained through closed-loop interactive training; finally, the trained control strategy is transferred and deployed to the actual building control system, and the air intake and blowing control signals of each control area are output according to the real-time monitoring status to realize the active regulation of the wind effect of the high-rise building.

[0059] The above description is merely a preferred embodiment of a method for active air intake and blowing control of wind effects in high-rise buildings based on deep reinforcement learning. The scope of protection for this method is not limited to the above embodiments; all technical solutions falling within this framework are within the scope of protection of this invention. It should be noted that for those skilled in the art, any improvements and variations made without departing from the principles of this invention should also be considered within the scope of protection of this invention.

Claims

1. A method for active intake and exhaust control of wind effects in high-rise buildings based on deep reinforcement learning, characterized by: The method includes the following steps: Step 1: Construct an active air intake and blowing control device for high-rise building wind effects; Step 2: Design and manufacture a wind tunnel test model based on the shape and structural dynamic characteristics of the high-rise building, and simulate the wind environment of the high-rise building in the wind tunnel; set up test devices corresponding to the externally perforated curtain wall, the air intake and blowing wind-resistant components, the wind environment monitoring sensors and the structural response monitoring sensors on the wind tunnel test model to obtain wind environment information and structural wind vibration response information under different working conditions. Step 3: Construct a deep reinforcement learning model, taking wind environment information and structural wind vibration response information as state information inputs, taking the control parameters of the air intake and blowing wind-resistant components as action signal outputs, and constructing a reward function based on the wind effect control target of high-rise buildings to establish a deep reinforcement learning model; Step 4: Connect the deep reinforcement learning model to the wind tunnel test environment, so that the deep reinforcement learning model outputs control of the intake and blowing air flow based on the real-time collected state information, and iteratively updates the model based on the feedback results of the wind tunnel test to obtain the active intake and blowing air control strategy for high-rise building wind effects.

2. The method according to claim 1, characterized in that: exist The high-rise building is surrounded by an externally perforated curtain wall, and a double-layer curtain wall channel is formed between the externally perforated curtain wall and the facade of the high-rise building through a four-sided connecting plate; multiple holes are set on the surface of the externally perforated curtain wall, and air-absorbing and air-blowing wind-resistant components are arranged at the holes. The air intake and air blowing wind-resistant assembly includes an intake fan, an air blowing fan, a rotating shaft, fan blades, intake holes, and air blowing holes; Wind environment monitoring sensors and structural response monitoring sensors are installed on the top floor or in the middle of high-rise buildings.

3. The method according to claim 2, characterized in that: The air intake and blowing wind-resistant components are arranged in the flow separation area on the side wind side of the building, the wake area on the leeward side, and the area affected by the incoming flow on the windward side. Two or more perimeter-shaped connecting plates are arranged along the height of the high-rise building to form multiple double-layer curtain wall channels between the externally perforated curtain wall and the exterior facade of the high-rise building. In the air intake and air blowing wind resistance assembly, the air intake and air blowing fans are connected to the rotating shaft. The air intake and air blowing fans are used to provide power. The rotating shaft is used to adjust the position of the air intake hole and the air blowing hole. The air intake hole is connected to the air intake fan through a pipe. The air blowing hole is connected to the air blowing fan through a pipe. The air intake and air blowing flow rate is adjusted by adjusting the fan voltage. Wind environment monitoring sensors are used to monitor incoming wind speed and direction, while structural response monitoring sensors are used to monitor the displacement, velocity, and acceleration response of high-rise buildings.

4. The method according to claim 3, characterized in that: The wind tunnel test model is a scaled-down aeroelastic model corresponding to a high-rise building. The wind tunnel test environment simulates the wind conditions of a high-rise building under different wind environments by adjusting the incoming wind speed and turbulence intensity, pulsating wind speed spectrum and wind direction angle of different wind field types.

5. The method according to claim 4, characterized in that: The state space of the deep reinforcement learning model includes structural wind vibration response information and wind environment information from multiple monitoring points of the building; the action space of the deep reinforcement learning model includes control signals for each intake and blowing wind-resistant component, which are used to adjust the intake flow rate and the blowing flow rate.

6. The method according to claim 5, characterized in that: Reward function of deep reinforcement learning model It is based on the structural response and wind pressure configuration within a single action cycle, and is expressed by the following formula: ； in, , , , and These are the displacement, velocity, acceleration, extreme wind pressure, and energy consumption performance evaluation indicators for the k-th action cycle. , , , , These are the reference displacement, reference velocity, reference acceleration, reference extreme wind pressure, and reference energy consumption performance evaluation indicators for the k-th action cycle. , , , and These are non-negative weighting coefficients.

7. The method according to claim 6, characterized in that: The deep reinforcement learning training process follows a closed-loop approach of "state perception - strategy decision-making - control execution - environmental feedback". That is, the state information is used as the input of the deep reinforcement learning algorithm model, the deep reinforcement learning algorithm model outputs control action signals, the intake or blowing air flow control is implemented by the intake and blowing air wind resistance components, and the wind tunnel test environment provides feedback on the next state and reward value.

8. The method according to claim 7, characterized in that: The deep reinforcement learning algorithm model is trained under basic working conditions to obtain a basic policy model. The basic policy model is then transferred to the target working conditions and the inhalation and exhalation control signals are output directly or after iterative updates.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the method as claimed in any one of claims 1-8.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the method of any one of claims 1-8.

Images