Control law derivation device, structure, and vibration control method

Abstract

The present invention provides a control law derivation device for deriving a control law specific to the unique environment of a structure, a structure equipped with a vibration damping device that performs vibration damping control using the control law, and a vibration damping method using the control law. [Solution] The control law derivation device 30 includes: a simulation model creation unit 31 that creates a simulation model that reproduces or simulates actual phenomena related to the vibration of a structure; a learning processing unit 32 that repeatedly performs a first numerical analysis to calculate the response of the structure by inputting a learning wave representing the external force on the structure into the simulation model, and creates a reinforcement learning model that calculates vibration damping action using the response of the structure obtained by the first numerical analysis; a second numerical analysis unit 35 that performs a second numerical analysis by inputting a verification wave representing the external force on the structure into the simulation model and the reinforcement learning model; and a control law analysis unit 36 ​​that derives a control law according to conditions based on the results of a comparative analysis of the response of the structure obtained by inputting the verification wave into the simulation model and the vibration damping action obtained by inputting that response into the reinforcement learning model.

Description

Technical Field

[0001] The present disclosure relates to a control law derivation device for deriving a control law used for vibration control of a structure, a structure in which a vibration control device for performing vibration control using the control law is installed, and a vibration control method for controlling the structure.

Background Art

[0002] Conventionally, for example, as in Patent Document 1, an active vibration control device in which a weight (mass body) is installed at the top of a structure such as a building and the structure and the weight are connected by an actuator or the like is known. Such a vibration control device is called an active mass damper when the excitation force of the actuator can be controlled, and a semi-active damper when the damping force generated by the damper can be controlled without using an actuator. The control of such a vibration control device is generally performed according to a control law based on control theory. In addition, a control device that calculates a vibration control operation by reinforcement learning using the result of a simulation regarding the vibration of a structure has also been proposed.

Prior Art Documents

Patent Documents

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] : However, the control law based on control theory can only perform control based on a theory conceived by humans. Therefore, it was difficult to say that the vibration control device was being controlled by a control law specialized for the specific environment of the structure. In addition, it was difficult to ensure the explainability of the vibration control operation calculated by reinforcement learning.

Means for Solving the Problems

[0006] In the structure that solves the above problems, a vibration control device is installed. The vibration control device performs vibration control of the structure according to the control law derived by the control law derivation device. The control law derivation device includes a learning processing unit that calculates a vibration control operation by reinforcement learning using a simulation that reproduces or simulates the actual phenomenon related to the vibration of the structure, and a control law analysis unit that derives a control law used for vibration control of the structure by referring to the vibration control operation.

[0007] A vibration control method that solves the above problems is a vibration control method of a structure using a vibration control device. The vibration control device performs vibration control of the structure according to the control law derived by the control law derivation device. The control law derivation device includes a learning processing unit that calculates a vibration control operation by reinforcement learning using a simulation that reproduces or simulates the actual phenomenon related to the vibration of the structure, and a control law analysis unit that derives a control law used for vibration control of the structure by referring to the vibration control operation.

[0008] According to the above configuration, a vibration control operation is calculated by reinforcement learning using a simulation related to the vibration of the structure, and a control law used for vibration control of the structure is derived based on the vibration control operation. As a result, a vibration control law specialized for the specific environment of the structure can be obtained.

[0009] In the above configuration, the calculation of the vibration damping action by the learning processing unit comprises: a simulated model creation step of creating a simulated model which is a model that reproduces or simulates the actual phenomenon related to the vibration of the structure; a reinforcement learning model creation step of repeatedly performing a first numerical analysis in which a learning wave representing the external force on the structure is input to the simulated model and the response of the structure is calculated, and a reinforcement learning model is created that calculates the vibration damping action using the response of the structure obtained by the first numerical analysis; and a second numerical analysis step in which the reinforcement learning model calculates the vibration damping action using the response of the structure obtained by inputting a verification wave representing the external force on the structure to the simulated model.

[0010] According to the above configuration, a reinforcement learning model is created that calculates vibration damping action corresponding to the structural response obtained from the training wave. Then, the reinforcement learning model can obtain vibration damping action corresponding to the structural response corresponding to the verification wave.

[0011] In the above configuration, the reference of the vibration damping action involves comparing and analyzing the response of the structure obtained by inputting the verification wave into the simulation model with the vibration damping action obtained by inputting the response of the structure corresponding to the verification wave into the reinforcement learning model. In this way, the vibration damping action can be referenced by comparing and analyzing the response of the structure and the vibration damping action.

[0012] In the above configuration, the control law analysis unit derives the control law according to the conditions based on the results of the comparative analysis. According to the above configuration, a control law can be derived that simulates vibration damping action corresponding to the response of the structure to the verification wave. [Brief explanation of the drawing]

[0013] [Figure 1] In the first embodiment, Figure 1(a) is a diagram showing an example of a schematic configuration of an active mass damper, which is a vibration damping device installed on a structure, and Figure 1(b) is a diagram showing another example of a schematic configuration of an active mass damper, which is a vibration damping device installed on a structure. [Figure 2] In the first embodiment, Figure 2 is a diagram showing an example of the hardware configuration of the information processing device. [Figure 3] In the first embodiment, Figure 3 is a flowchart showing the procedure for deriving the control law. [Figure 4] In the first embodiment, Figure 4 is a functional block diagram showing a control law derivation device. [Figure 5] In the first embodiment, Figure 5 is a schematic diagram showing a model of a structure equipped with an active mass damper. [Figure 6] In the first embodiment, Figure 6 shows an example of response parameters obtained by the second numerical analysis. [Figure 7] In the first embodiment, Figure 7 is a diagram illustrating the method for deriving the control law. [Figure 8] In the second embodiment, Figure 8 shows an example of a schematic configuration of a semi-active mass damper, which is a vibration damping device installed in a structure. [Figure 9] In the second embodiment, Figure 9 is a schematic diagram showing a model of a structure equipped with a semi-active mass damper. [Figure 10] In the second embodiment, Figure 10 shows an example of the response and damping coefficient of a structure obtained by the second numerical analysis. [Figure 11] In the second embodiment, Figure 11 shows an example of a damping coefficient calculated according to the conditions. [Modes for carrying out the invention]

[0014] (First Embodiment) Referring to Figures 1 to 7, a first embodiment of the control law derivation device, structure, and vibration control method will be described.

[0015] As shown in Figure 1(a), the vibration damping device 20 is an active mass damper installed at the top of a structure 11, for example, an n-story building. The vibration damping device 20 comprises a mass 21, an actuator 22, a first sensor 23, a second sensor 24, and a control device 25. As shown in Figure 1(b), the vibration damping device 20 may also have a restoring force device 26, composed of a spring element or the like, connected in parallel to the actuator 22.

[0016] Mass 21 is a weight (mass body) for damping vibrations of structure 11. Mass 21 is positioned at the top of structure 11. Mass 21 is supported by rolling bearings on structure 11 so as to be able to move relative to it in the horizontal direction.

[0017] The actuator 22 connects the mass 21 and the structure 11. The actuator 22 reduces the horizontal vibration of the structure 11 by exciting the mass 21 in the horizontal direction when the structure 11 vibrates due to an external force.

[0018] The first sensor 23 is a sensor for acquiring the state of the structure 11. The first sensor 23 is, for example, an acceleration sensor. The first sensor 23 measures the first acceleration, which is the acceleration of the structure 11. The first sensor 23 outputs the measured first acceleration to the control device 25.

[0019] The second sensor 24 is a sensor for acquiring the state of the mass 21. The second sensor 24 is, for example, an acceleration sensor. The second sensor 24 measures the second acceleration, which is the acceleration of the mass 21. The second sensor 24 outputs the measured second acceleration to the control device 25.

[0020] In the first embodiment, the control device 25 acquires the displacement and velocity of the top floor of the structure 11 as the state of the structure 11 based on the first acceleration. The control device 25 acquires the relative displacement and relative velocity of the mass 21 with respect to the top floor of the structure 11 as the state of the mass 21 based on the first and second accelerations. The control device 25 controls the actuator 22 with respect to the state of the structure 11 and the state of the mass 21. NThe control device 25 calculates the calculated control force u. N The actuator 22 is controlled so that the mass 21 is excited at (t). The control device 25 controls the control force u according to the control law shown in equation (1). N (t) is calculated. This control law is derived by the control law derivation device 30 and then implemented in the control device 25.

[0021]

number

[0022] (Hardware configuration of the control device and control law derivation device) Figure 2 shows an example of the hardware configuration of an information processing device H10 that functions as a control device 25 and a control law derivation device 30. The information processing device H10 includes a communication device H11, an input device H12, a display device H13, a storage device H14, and a processor H15. Note that this hardware configuration is just one example, and other hardware may be included.

[0023] Communication device H11 is an interface that performs data transmission and reception by establishing a communication path with other devices. Communication device H11 is, for example, a network interface card or a wireless interface.

[0024] Input device H12 is a device that receives input from the user or other users. Input device H12 is, for example, a mouse or keyboard. Display device H13 is a display or touch panel that displays various information.

[0025] A storage device H14 is a storage device that stores data and programs for executing various functions. Examples of storage devices H14 include ROM (Read Only Memory), RAM (Random Access Memory), and hard disks.

[0026] The processor H15 uses programs and data stored in the memory device H14 to control each process in the information processing device H10, which functions as a control device 25 and a control law derivation device 30. Examples of processor H15 include a CPU (Central Processing Unit) and an MPU (Micro Processor Unit). This processor H15 executes various processes corresponding to various operations by loading programs stored in ROM, etc., into RAM. For example, when a predetermined application program is started, the processor H15 operates the processes that execute the operations described later.

[0027] The processor H15 is not limited to performing all of its operations using software. For example, the processor H15 may include dedicated hardware circuits (e.g., application-specific integrated circuits: ASICs) that perform hardware operations for at least some of the operations it performs. In other words, the processor H15 can be configured as follows:

[0028] [1] One or more processors that operate according to a computer program (software) [2] One or more dedicated hardware circuits that perform at least some of the various processes, [3] Circuits that include combinations of these.

[0029] A processor includes the CPU and memory such as RAM and ROM. Memory stores program code or instructions configured to cause the CPU to perform processing. Memory, or non-temporary computer-readable media, includes any available media accessible by a general-purpose or dedicated computer.

[0030] (Method for deriving control laws) Referring to Figures 3 to 7, the method for deriving the control law from the control law derivation device 30 will be explained.

[0031] As shown in Figure 3, the method for deriving the control law comprises creating a simulated model (step S101), creating a reinforcement learning model (step S102), performing a second numerical analysis (step S103), and deriving the control law (step S104).

[0032] As shown in Figure 4, the control law derivation device 30 has a simulated model creation unit 31, a first numerical analysis unit 33 and a reinforcement learning unit 34 which constitute the learning processing unit 32, a second numerical analysis unit 35, and a control law analysis unit 36 ​​as functional units that function through the execution of a program.

[0033] (Creation of a simulated model) In the creation of the simulated model (step S101), the simulated model creation unit 31 creates a simulated model that simulates the structure 11 to be controlled.

[0034] Figure 5 shows a model of the structure 11 of an n-story building and an active mass damper. The control force u(t) of the actuator 22 acts on the top floor of the structure 11 and the mass 21. The simulation model creation unit 31 creates a simulation model by expressing the model shown in Figure 5 using equations of motion. The control law derivation device 30 uses the equation of motion shown in equation (2) to derive the external force f j By solving the problem using numerical analysis with (t) as input, it is possible to simulate the behavior of the n+1 mass system. Note that in equation (2), the external force f acting on the structure 11 is j (t) is defined as the wind load.

[0035]

number

[0036] (Creating reinforcement learning models) In the creation of the reinforcement learning model (step S102), the first numerical analysis unit 33 of the learning processing unit 32 uses the simulated model created in step S101 to analyze the external force f jPerform the first numerical analysis with the learning wave indicating (t) as the input. The learning wave may be an actual wave such as wind or earthquake, or a simulated wave simulating the actual wave. For example, when the purpose is to reduce horizontal vibration due to wind excitation, the learning wave is a composite wave of a sine wave and a DC component (for example, the sine wave has a single amplitude of 20 kN and a period equal to the natural period of the structure 11).

[0037] In the first numerical analysis, the first numerical analysis unit 33 calculates the response of the structure 11 at a certain time t, such as the displacement, velocity, and acceleration of the structure 11. The first numerical analysis unit 33 repeatedly performs numerical analysis with the learning wave as the input.

[0038] Also, the reinforcement learning unit 34 of the learning processing unit 32 executes reinforcement learning based on the response of the structure 11 obtained by the numerical analysis of the first numerical analysis unit 33 that is repeatedly performed. Reinforcement learning learns actions that maximize value through trial and error even when there is no teacher data. As an algorithm for reinforcement learning, for example, Q-Learning, DQN (Deep Q-network) can be used. In reinforcement learning, first, as various parameters, action a t , state s t , and reward r t are set.

[0039] For action a t , the control force output to the actuator 22 at time t is set. More specifically, for the control force of the actuator 22, an arbitrary excitation force between 0 and the maximum excitation force is set. When Q-Learning is used for reinforcement learning, for the control force of the actuator 22, the number of discretizations (for example, 5 divisions, etc.) between the minimum excitation force and the maximum excitation force is set.

[0040] State s t is what is used as the input to determine action a t . State s t [ is set with the states of the structure 11 and the mass 21 at time t. State s tFor example, as shown in equation (3), the displacement and velocity for the uppermost layer of the structure 11 and the mass 21 are set. Note that when Q-Learning is used for reinforcement learning, state s t The aforementioned elements and the number of discretizations for each element (e.g., 3 divisions) are set in this field.

[0041]

number

[0042] In reinforcement learning, state s t Depending on the circumstances, the rewards that can be obtained in the future will be t Action a that maximizes the sum t The system learns to surpass the response of the uppermost layer of the structure 11 under optimal control, and receives a reward r in accordance with the comparison with the result of the optimal control. For example, as shown in equation (4), the goal is to surpass the response of the uppermost layer of the structure 11 under optimal control. t This provides optimal control, including optimal feedback control, LQ (Linear Quadratic) control, LQR (Linear Quadratic Regulator) control, and optimal regulator control.

[0043]

number

[0044] Also, for example, as in equation (5), the reward r corresponds to the square of the acceleration at the top layer of the structure 11. t Give.

[0045]

number

[0046] The reinforcement learning unit 34 uses the analysis results of the first numerical analysis to perform reinforcement learning according to various algorithms as follows. Note that Q(s t ,a t ) is a certain state s t Action a t The reward r obtained when you continue to select this option tThis defines the sum Gt of the expected values.

[0047] If the reinforcement learning algorithm is Q-Learning, then Q(s) is calculated for each step according to equation (6). t ,a t The update of ) is repeated.

[0048]

number

[0049] If the reinforcement learning algorithm is DQN, the neural network parameters θ are updated sequentially so as to minimize the loss function L(θt) defined in equation (7).

[0050]

number

[0051] Through numerical analysis by the first numerical analysis unit 33 and reinforcement learning by the reinforcement learning unit 34, the learning processing unit 32 controls the actuator 22's control force u as a vibration damping action that effectively reduces the horizontal vibration of the structure 11 in a certain state. A Create a reinforcement learning model to calculate (t). The reinforcement learning model is represented as a network in which the input layer, hidden layer, and output layer are connected by multiple layers of neurons.

[0052] (Second numerical analysis) In the second numerical analysis (step S103), the second numerical analysis unit 35 performs a numerical analysis using a verification wave representing an external force as input. The verification wave may be a real wave such as wind or earthquake, or a simulated wave that simulates such a real wave. The verification wave may also be the same as the learning wave or different from the learning wave.

[0053] As shown in Figure 6, in this numerical analysis, the second numerical analysis unit 35 uses the simulated model created in step S101 to calculate the displacement of the top floor of the structure 11 (building displacement), the velocity of the top floor of the structure 11 (building velocity), the relative displacement of the mass 21 with respect to the top floor of the structure 11 (mass relative displacement), and the relative velocity of the mass 21 with respect to the top floor of the structure 11 (mass relative velocity) at a certain time t. The second numerical analysis unit 35 also inputs the various response parameters calculated by the simulated model into a reinforcement learning model to determine the control force u of the actuator 22 at a certain time t. A Calculate (t).

[0054] (Derivation of the control law) In the derivation of the control law (step S105), the control law analysis unit 36 ​​refers to the analysis results from the second numerical analysis unit 35 and derives a control law that simulates the reinforcement learning model. The control law analysis unit 36 ​​uses the response of the structure 11 and the control force u of the actuator 22 to derive a control law. A The control law is derived according to the conditions set based on the comparative analysis results with (t).

[0055] As shown in Figure 7, the response of the structure 11 and the control force u of the actuator 22 A When comparing and analyzing (t), the behavior of various response parameters and the control force u of the actuator 22 can be seen. A It can be confirmed that the behavior is similar to that of (t). In the first embodiment, the control force u is obtained by multiplying the various response parameters calculated by the second numerical analysis by gains G1, G2, G3, and G4 and adding them together. N (t) (see equation (1)) where the control force u A The condition was to simulate (t).

[0056] The control law analysis unit 36 ​​calculates the control force u N (t) and control force u AThe various gains G1, G2, G3, and G4 are calculated to minimize the error with (t). For example, the control law derivation device 30 uses an optimization algorithm to find the gains G1, G2, G3, and G4 that minimize the value of the evaluation function J defined as in equation (8). Examples of optimization algorithms include GA (Genetic Algorithm) and Particale Swarm Optimization (Particle Swarm Optimization).

[0057]

number

[0058] Here, L is the number of data points in the time history data used. For example, if the sampling frequency of the time history data used is 100 Hz and the data spans 200 seconds, then L = 100 × 200 = 20000.

[0059] (Operation of the first embodiment) The control law derived by the control law derivation device 30 in this manner is implemented in the control device 25. The control device 25 acquires the displacement and velocity of the structure 11 based on the first acceleration. The control device 25 acquires the relative displacement and relative velocity of the mass 21 with respect to the structure 11 based on the first and second accelerations. Then, the control device 25 calculates the control force u by substituting the acquired parameters into the control law. N (t) controls actuator 22.

[0060] The effects of the first embodiment will be described. (1-1) Reinforcement learning using simulations of vibrations of structure 11 is used to control the actuator 22's control force u as a vibration damping action. A (t) is calculated. And that control force u A Based on (t), a control law is derived for the vibration control of structure 11.

[0061] With this configuration, the control force u of the actuator 22 NWith respect to (t), a control law that was difficult to realize using optimal control theory can be obtained. As a result, a vibration control law specific to the unique environment of structure 11 can be obtained.

[0062] (1-2) The control law derivation device 30 uses the state of the structure 11 as a parameter to control the control force u N Derive the control law from which (t) is calculated. That is, the control force u N The basis for determining (t) is the state of structure 11. As a result, the basis for determining the control force does not become a black box, as in reinforcement learning models, thus increasing the safety of the vibration damping device 20.

[0063] (1-3) The calculation of vibration damping action by the learning processing unit 32 comprises: a simulated model creation step (step S101) which creates a simulated model that reproduces or simulates the actual phenomena related to the vibration of the structure 11; a reinforcement learning model creation step (step S102) which repeatedly performs a first numerical analysis in which a learning wave representing the external force on the structure 11 is input into the simulated model to calculate the response of the structure 11, and creates a reinforcement learning model that calculates vibration damping action using the response of the structure 11 obtained by the first numerical analysis; and a second numerical analysis step (step S103) in which a verification wave representing the external force on the structure 11 is input into the simulated model to perform a second numerical analysis in which the reinforcement learning model calculates vibration damping action using the response of the structure 11 obtained.

[0064] With this configuration, a reinforcement learning model is created that calculates vibration damping action corresponding to the response of the structure 11 obtained from the training wave. Then, the vibration damping action corresponding to the response of the structure 11 corresponding to the verification wave can be obtained by the reinforcement learning model.

[0065] (1-4) The reference for vibration damping action is obtained by comparing and analyzing the response of structure 11 obtained by inputting a verification wave into a simulation model with the vibration damping action obtained by inputting the response of structure 11 corresponding to the verification wave into a reinforcement learning model. In this way, the vibration damping action can be referenced by comparing and analyzing the response of structure 11 and the vibration damping action.

[0066] (1-5) The control law analysis unit 36 ​​derives a control law according to the conditions based on the results of the comparative analysis. This makes it possible to derive a control law that simulates the vibration damping action corresponding to the response of the structure 11 corresponding to the verification wave.

[0067] (1-6) The control device 25 controls the actuator 22 by the control force u according to the above formula (1). N (t) is controlled. That is, the control device 25 controls the displacement and velocity of the top floor of the structure 11, the relative displacement and relative velocity of the mass 21 with respect to the top floor of the structure 11, and the control force u based on these four parameters. N (t) is controlled. This controls the control force u of the actuator 22 so that high vibration damping performance is achieved according to the state of the structure 11 at that time. N (t) can be controlled.

[0068] (Second Embodiment) Referring to Figures 8 to 11, a second embodiment of the control law derivation device, structure, and vibration damping method will be described. Note that the control law derivation device, structure, and vibration damping method of the second embodiment have the same main configuration as those of the first embodiment. Therefore, in the second embodiment, the parts that differ from the first embodiment will be described in detail, while parts that are the same as those in the first embodiment will be denoted by the same reference numerals, and their detailed descriptions will be omitted. Specifically, the configuration of the vibration damping device, the simulated model, the response parameters of the structure obtained by the first numerical analysis, the behavior in reinforcement learning, and the conditions for deriving the control law differ.

[0069] (Vibration damping device) As shown in Figure 8, the vibration damping device 50 is a semi-active mass damper installed at the top of the structure 11. The vibration damping device 50 comprises a mass 21, a spring element 52, a variable damping damper 53, a first sensor 23, a second sensor 24, and a control device 25.

[0070] The spring element 52 and the variable damping damper 53 connect the mass 21 and the structure 11. The spring element 52 and the variable damping damper 53 are connected in parallel to the mass 21 and the structure 11. The vibration damping device 50 absorbs the vibration energy of the structure 11 by vibrating the mass 21 when the structure 11 vibrates due to an external force, thereby reducing the horizontal vibration of the structure 11.

[0071] The control device 25 controls the damping coefficient c of the variable damping damper 53 according to a predetermined control law, based on the state of the structure 11 and the state of the mass 21. N (t) is calculated. Then, the control device 25 calculates the damping coefficient c N The variable damping damper 53 is controlled so that (t) is expressed. As a result, the vibration damping device 50 exhibits vibration damping performance appropriate to the situation at any given time. The control law is derived by the control law derivation device 30 and then implemented in the control device 25.

[0072] (Creation of a simulated model) In the creation of the simulated model (step S101), the simulated model creation unit 31 creates a simulated model by expressing the model shown in Figure 9 using equations of motion. Figure 9 shows a model of the structure 11 of an n-story building and a semi-active mass damper. The control law derivation device 30 uses the equation of motion shown in equation (9) to express the external force f j By solving the problem using numerical analysis with (t) as input, it is possible to simulate the behavior of the n+1 mass system. Note that in equation (9), the external force f acting on the structure 11 is j (t) is defined as the wind load.

[0073]

number

[0074] (Creating reinforcement learning models) In the creation of the reinforcement learning model (step S102), the first numerical analysis unit 33 of the learning processing unit 32 uses the simulated model to calculate the external force f jNumerical analysis is repeatedly performed using the learning wave (t) as input. The reinforcement learning unit 34 of the learning processing unit 32 performs reinforcement learning based on the response of the structure 11 obtained by the numerical analysis of the first numerical analysis unit 33, which is performed repeatedly. In reinforcement learning, action a t The damping coefficient of the variable damping damper 53 is set as follows. Through this reinforcement learning, the reinforcement learning unit 34 sets the damping coefficient c of the variable damping damper 53 as a vibration damping action that effectively reduces the horizontal vibration of the structure 11 in a certain state. A Create a reinforcement learning model to calculate (t).

[0075] (Second numerical analysis) In the second numerical analysis (step S103), the second numerical analysis unit 35 calculates the external force f j Numerical analysis is performed using the verification wave (t) as input.

[0076] As shown in Figure 10, in this numerical analysis, the second numerical analysis unit 35 uses a simulated model to calculate the response of the structure 11 at a certain time t, for example, the acceleration of the top floor. The second numerical analysis unit 35 also inputs the response parameters calculated by the simulated model into a reinforcement learning model to determine the damping coefficient c of the variable damping damper 53 at a certain time t. A Calculate (t).

[0077] (Derivation of the control law) In the derivation of the control law (step S104), the control law analysis unit 36 ​​analyzes the response of the structure 11 and the damping coefficient c of the variable damping damper 53. A The control law is derived according to the conditions set based on the comparative analysis results with (t).

[0078] For example, the acceleration and damping coefficient c of the structure 11 shown in Figure 10. A A comparative analysis with (t) reveals the following: The damping coefficient is repeatedly switched ON / OFF at half-cycle intervals of acceleration at the top floor of structure 11. • The magnitude of the damping coefficient tends to be proportional to the magnitude of the acceleration amplitude.

[0079] Based on these analysis results, in the second embodiment, the conditions for deriving the control law are set as follows. • If the sign of the derivative of the acceleration of structure 11 at a certain time t(s) is positive, the damping coefficient should be set to 0. • If the sign of the derivative of the acceleration of structure 11 at a certain time t(s) is negative, the damping coefficient shall be a value greater than 0. The value of the damping coefficient in that case shall be x times the amplitude of the acceleration at the time when the sign of the above derivative changed from positive to negative.

[0080] As shown in Figure 11, if the control law analysis unit 36 ​​obtains the results shown in Figure 10 in the second numerical analysis (step S103), it determines the damping coefficient c at time t(s) according to the above conditions. N (t) is calculated. That is, the control law analysis unit 36 ​​calculates the damping coefficient c from time t1, t3, t5, t7 when the sign of the differential value of acceleration changes from positive to negative, to time t2, t4, t6, t8 when the sign of the differential value of acceleration changes from negative to positive, by α times the amplitude of acceleration at each time t1, t3, t5, t7. N We derive the control law (t).

[0081] (Operation of the second embodiment) The control law derived by the control law derivation device 30 in this manner is implemented in the control device 25. The control device 25 obtains the acceleration of the structure 11 based on the first acceleration. The control device 25 then calculates the damping coefficient c based on the sign of the differential value of the acceleration and the amplitude of the acceleration at the time when the sign changes from positive to negative. N (t) controls the variable damping damper 53.

[0082] According to the second embodiment, in addition to the effects similar to those described in (1-1) to (1-5) above, the following effects can be obtained. (2-1) The control device 25 controls the damping coefficient c of the variable damping damper 53 based on the change in acceleration of the structure 11. N (t) is controlled. This allows the damping coefficient c of the variable damping damper 53 to be controlled based on one parameter. N (t) can be controlled.

[0083] Each of the above embodiments can be implemented with the following modifications. Each embodiment and the following modifications can be combined with each other to the extent that they do not contradict each other technically. In the first embodiment, the control law derivation device 30 calculates the control force u based on the displacement and velocity of the structure 11 and the relative displacement and relative velocity of the mass 21. N The control law for calculating (t) was derived. However, the control law derivation device 30 can derive the control law based on the state of the structure 11 and the state of the mass 21. For example, the control law derivation device 30 may derive the control law based on the acceleration of the structure 11 and the acceleration of the mass 21.

[0084] In the second embodiment, the control law derivation device 30 calculates the damping coefficient c based on the acceleration of the structure 11. N The control law for calculating (t) was derived. However, the control law derivation device 30 can derive the control law based on the state of the structure 11 and the state of the mass 21. For example, the control law derivation device 30 may derive the control law based on the displacement and velocity of the structure 11 and the relative displacement and relative velocity of the mass 21. [Explanation of symbols]

[0085] H10... Information processing device, H11... Communication device, H12... Input device, H13... Display device, H14... Memory device, H15... Processor, 11... Structure, 20... Vibration damping device, 21... Mass, 22... Actuator, 23... First sensor, 24... Second sensor, 25... Control device, 26... Attenuator, 30... Control law derivation device, 31... Simulation model creation unit, 32... Learning processing unit, 33... First numerical analysis unit, 34... Reinforcement learning unit, 35... Second numerical analysis unit, 36... Control law analysis unit, 50... Vibration damping device, 52... Spring element, 53... Variable damping damper.

Claims

1. A control law derivation device for deriving control laws used for vibration control of structures using vibration damping devices, A learning processing unit that calculates vibration damping action by reinforcement learning using a simulation that reproduces or simulates actual phenomena related to the vibration of the aforementioned structure, The system includes a control law analysis unit that derives a control law used for vibration control of the structure by referring to the vibration damping operation. Control law derivation device.

2. A structure on which vibration damping devices are installed, The vibration damping device, The vibration control of the structure is performed according to the control law derived by the control law derivation device. The control law derivation device is, A learning processing unit that calculates vibration damping action by reinforcement learning using a simulation that reproduces or simulates actual phenomena related to the vibration of the aforementioned structure, The system includes a control law analysis unit that derives a control law used for vibration control of the structure by referring to the vibration damping operation. structure.

3. A method for controlling vibrations of a structure using a vibration control device, The vibration damping device, The vibration control of the structure is performed according to the control law derived by the control law derivation device. The control law derivation device is, A learning processing unit that calculates vibration damping action by reinforcement learning using a simulation that reproduces or simulates actual phenomena related to the vibration of the aforementioned structure, The system includes a control law analysis unit that derives a control law used for vibration control of the structure by referring to the vibration damping operation. Vibration damping methods.

4. The calculation of the vibration damping operation by the learning processing unit is as follows: A simulation model creation step involves creating a simulation model that reproduces or simulates the actual phenomena related to the vibration of the aforementioned structure, A reinforcement learning model creation step involves repeatedly performing a first numerical analysis in which a learning wave representing the external force on the structure is input to the simulation model to calculate the response of the structure, and creating a reinforcement learning model that calculates the vibration damping action using the response of the structure obtained from the first numerical analysis. The system includes a second numerical analysis step in which a verification wave representing the external force on the structure is input to the simulation model, and the reinforcement learning model performs a second numerical analysis using the response of the structure obtained from this input to calculate the vibration damping action. The vibration damping method according to claim 3.

5. The reference to the vibration damping action involves comparing and analyzing the response of the structure obtained by inputting the verification wave into the simulation model with the vibration damping action calculated by the reinforcement learning model. The vibration damping method according to claim 4.

6. The control law analysis unit derives the control law according to the conditions based on the results of the comparative analysis. The vibration damping method according to claim 5.

Images