Quasi-zero stiffness vibration control-energy harvesting system
Through the control of the dielectric elastomer drive unit and neural network, the load and balance position of the quasi-zero stiffness vibration control-energy acquisition system is realized, which solves the problems of insufficient vibration energy collection and difficulty in high-frequency response in the existing technology, and reduces the problem of low vibration energy collection and difficult high-frequency response. The system energy consumption is increased and the vibration isolation bandwidth is expanded.
Patent Information
- Application Number
- CN202311873030.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-31
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2043-12-31
AI Technical Summary
The existing quasi-zero stiffness vibration isolation device cannot realize the active regulation of load and balanced position, cannot perform fast response control of high-frequency vibration, insufficient vibration collection of the system, and traditional magnetorheological control methods have high power consumption, slow response, and obvious hysteresis.
Design a quasi-zero stiffness vibration control-energy acquisition system including dielectric elastomer driving unit, sensor, voltage amplifier and active control unit, generate active control instructions through neural network framework, adjust the balance position of dielectric elastomer driving unit and Stiffness, achieving wide frequency vibration isolation and energy acquisition.
While achieving high static stiffness and low dynamic stiffness, the balance position and stiffness of the vibration isolation system are actively regulated according to load changes, expand the vibration isolation bandwidth, and effectively collect vibration energy to reduce the demand for external energy.
Smart Images

Figure CN117833589B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a technology in the field of vibration isolation, in particular to a quasi-zero stiffness vibration control-energy collection system. Background Art
[0002] Most of the existing quasi-zero stiffness vibration isolation devices are realized by connecting linear springs and negative stiffness structures in parallel. The fact that all the loads are borne by the linear springs can easily lead to the negative stiffness not being involved in the actual operation process, causing the system to lose its quasi-zero stiffness characteristics. On the other hand, energy is often dissipated in traditional vibration isolation systems and is not fully utilized. If an energy harvesting unit is introduced, the vibration energy can be fully utilized, and the energy consumed by the control function can also be reduced. In addition, existing control methods such as magnetorheological control have high power consumption, slow response, and obvious hysteresis. The introduction of dielectric materials is conducive to reducing system energy consumption, improving response speed and high-frequency control effects. Summary of the invention
[0003] In view of the shortcomings of the prior art that the load and balance position cannot be actively regulated, high-frequency vibration rapid response control and system vibration energy collection cannot be performed, the present invention proposes a quasi-zero stiffness vibration control-energy harvesting system that can be actively regulated. The system can achieve high static stiffness and low dynamic stiffness characteristics, while actively regulating the balance position and stiffness of the system vibration isolation system according to load changes, thereby achieving a larger vibration isolation bandwidth, and can realize the collection of vibration energy, reducing the demand for external energy.
[0004] The present invention is achieved through the following technical solutions:
[0005] The present invention relates to a quasi-zero stiffness vibration control-energy harvesting system, comprising: a quasi-zero stiffness vibration isolation-energy harvesting mechanical module with a dielectric elastomer drive unit and a sensor, a voltage amplifier and an active control unit connected thereto, wherein: the active control unit generates active control instructions through a constructed neural network framework according to the displacement, velocity and acceleration information of the quasi-zero stiffness vibration isolation-energy harvesting mechanical module collected by the sensor, and outputs the active control instructions to the dielectric elastomer drive unit of the quasi-zero stiffness vibration isolation-energy harvesting mechanical module through the voltage amplifier to generate deformation, thereby actively adjusting the equilibrium position and stiffness of the quasi-zero stiffness vibration isolation-energy harvesting mechanical module to achieve broadband vibration isolation and energy harvesting.
[0006] The quasi-zero stiffness active vibration isolation-energy harvesting module comprises: a fixed base, a load platform, a dielectric elastomer drive unit, a repulsive magnet and an induction coil, wherein: the fixed base and the load platform are connected by four connecting rods, the bottom ends of the connecting rods are rotatably arranged on the slider guide rails on the fixed base, the dielectric elastomer drive unit is arranged between adjacent sliders, a pair of repulsive magnets and an induction coil are relatively arranged between the fixed base and the load platform through a direction constraint rod to provide supporting force and constrain the load platform to move in a direction perpendicular to the fixed base, the relative movement between the load platform and the fixed base causes a change in the magnetic field, and then generates an induced current in the induction coil, which is used to provide auxiliary energy support for vibration control, and quasi-zero stiffness vibration isolation is achieved by setting the structural parameters of the dielectric elastomer drive unit and adjusting the stiffness of the dielectric elastomer drive unit.
[0007] The dielectric elastomer driving unit undergoes different deformations and generates corresponding tensile forces and equivalent stiffness according to different voltages output by the active control unit.
[0008] The active control unit includes: a strategy network module, an evaluation network module, a target network module, a prediction network module, a proportional-integral-differential module, a reward function module and a data integration module, wherein: the strategy network module obtains three proportional, integral and differential parameter values through an encoder and a decoder according to the data output by the data integration module, and outputs them to the proportional-integral-differential module; the evaluation network module obtains the goodness and badness evaluation value of the current strategy through an encoder and a decoder based on the output of the data integration module and the reward function value; the target network module obtains the ideal state value or reward signal through an encoder and a decoder according to the output of the data integration module; the prediction network module obtains the goodness and badness evaluation value of the current strategy ... The four voltage signals output by the sub-modules are passed through the encoder and decoder to obtain the predicted displacement, velocity and acceleration response data of the vibration isolation-energy harvesting mechanical module, which are used as part of the training samples for weight adjustment of the strategy network module, the evaluation network module and the target network module; the proportional-integral-differential module outputs four voltage signals according to the proportional, integral and differential parameters; the reward function module calculates the reward value according to the displacement deviation and the voltage value of the dielectric elastomer, and provides it to the strategy network module, the evaluation network module and the target network module; the data integration module integrates the historical status data of the vibration isolation device and the current status data of the vibration isolation device, and provides them to the strategy network module, the evaluation network module and the target network module.
[0009] The reward function module calculates the reward function value according to the platform displacement d, the square of the four voltage signals V1, V2, V3, and V4 output by the proportional-integral-differential module, and the termination function. Where: d0 is the amplitude of the vibration source, g is the termination function, and the four voltage signals correspond to the input voltages of the four dielectric elastomer drive units. When the platform displacement does not exceed the set limit threshold d c Time <d c , the termination function is g=0, otherwise the termination function is g=-200.
[0010] Technical Effects
[0011] The present invention forms a stiffness adjustment module of a hexahedral quasi-zero stiffness vibration isolation platform through four dielectric elastomer drive units; the drive voltage of the dielectric elastomer drive unit is adjusted based on an algorithm that characterizes the mapping relationship between the drive voltage of the dielectric elastomer drive unit and the system response displacement and other information; compared with the prior art, the present invention can realize active regulation of the quasi-zero stiffness vibration isolation equilibrium position and stiffness based on the hexahedral vibration isolation platform. The mutually repelling magnetic poles and electromagnetic coils installed on the inner surface of the upper and lower platforms of the central axis can collect vibration energy while realizing vibration control. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] Figure 1 It is a schematic diagram of the system of the present invention;
[0013] Figure 2 and Figure 3 Schematic diagram of the quasi-zero stiffness vibration isolation platform structure;
[0014] Figure 4 Schematic diagram of the structure of a dielectric elastomer drive unit;
[0015] Figure 5 is a flow chart of an embodiment;
[0016] In the figure: a fixed base 1, a load platform 2, a connecting rod 3, a guide rail 4, a slider 5, a dielectric elastomer drive unit 6, a magnet base 7, a magnet 8, a direction constraint rod 9, a magnet 10, a magnet base 11, a fixed block 12, a sensor 13, an induction coil 14, and a direction constraint rod 15. DETAILED DESCRIPTION
[0017] like Figure 1 As shown, the present embodiment relates to a quasi-zero stiffness vibration control-energy harvesting system, comprising: a quasi-zero stiffness vibration isolation-energy harvesting mechanical module having a dielectric elastomer drive unit 6, a sensor 13 connected thereto, and an active control unit, wherein: the active control unit generates active control instructions through a neural network based on the displacement, velocity and acceleration information of the quasi-zero stiffness vibration isolation-energy harvesting mechanical module collected by the sensor, and outputs the active control instructions to the dielectric elastomer drive unit of the quasi-zero stiffness vibration isolation-energy harvesting mechanical module to generate deformation, thereby actively adjusting the equilibrium position and stiffness of the quasi-zero stiffness vibration isolation-energy harvesting mechanical module to achieve broadband vibration isolation and energy harvesting.
[0018] like Figure 2 As shown, the quasi-zero stiffness vibration isolation-energy harvesting mechanical module includes: a fixed base 1, a load platform 2, a dielectric elastomer drive unit 6, repulsive magnets 8, 10 and an induction coil 14, wherein: the fixed base 1 and the load platform 2 are connected by four connecting rods 3, the bottom ends of the connecting rods 3 are rotatably arranged on the sliders 5 and the guide rails 4 on the fixed base, the dielectric elastomer drive unit 6 is arranged between adjacent sliders 5, a pair of repulsive magnets 8, 10 and the induction coil 14 are relatively arranged between the fixed base and the load platform through a direction constraint rod 15 to provide support force and constrain the load platform to move in a direction perpendicular to the fixed base, the relative movement between the load platform and the fixed base causes a change in the magnetic field, and then generates an induced current in the induction coil, which is used to provide auxiliary energy support for vibration control, and quasi-zero stiffness vibration isolation is achieved by setting the structural parameters of the dielectric elastomer drive unit and adjusting the stiffness of the dielectric elastomer drive unit.
[0019] The connecting rod 3 and the slider 5 are rotatably connected via a hinge.
[0020] like Figure 4 As shown, the dielectric elastomer driving unit 6 includes: a multi-layer dielectric elastomer film sleeve 601 and a driving rod 602, wherein: both ends of the driving rod 602 are respectively connected to adjacent sliders 5, and the multi-layer dielectric elastomer film sleeve 601 stretches after being triggered by voltage, so that the driving rod 602 controls the slider 5 to move on the guide rail 4, thereby controlling the posture adjustment of the connecting rod 3.
[0021] The multilayer dielectric elastomer film sleeve 601 is composed of several layers of concentrically arranged dielectric elastomer films, each layer of the dielectric elastomer film includes a dielectric elastomer located in the middle and conductive graphite layers arranged on its two side surfaces as electrodes, the conductive graphite layers are connected to the active control unit, and the dielectric elastomer generates a corresponding elongation according to the driving voltage to achieve active regulation of the equilibrium position.
[0022] like Figure 2 and Figure 3 As shown, the upper end of the connecting rod 3 is fixed to the load platform 2 by a hinge 12, and the lower end of the connecting rod is fixed to the slider 5 by a hinge. The slider 5 moves along a fixed direction on the guide rail 4, and the guide rail is set on the fixed base 1. When the slider moves toward the center, the bottom of the connecting rod 3 is close to the center of the base, and the height of the load platform 2 increases. When the slider 5 moves in the opposite direction, the bottom of the connecting rod 3 is away from the center of the base, and the height of the load platform is reduced.
[0023] like Figure 2 and Figure 3 As shown, the magnets 8 and 10 are mutually repelling magnets, which provide a supporting force perpendicular to the movement direction of the platform by generating magnetic repulsion to resist the load force on the platform.
[0024] like Figure 5 As shown, the active control unit includes: a strategy network module, an evaluation network module, a target network module, a prediction network module and a proportional-integral-differential module, a reward function module and a data integration module, wherein: the strategy network module using the Transformer architecture obtains three proportional, integral and differential parameter values according to the data output by the data integration module through an encoder and a decoder; the reward function module calculates the reward value according to the displacement deviation and the voltage value of the dielectric elastic body, and provides it to the strategy network module, the evaluation network module and the target network module; the evaluation network module using the Transformer architecture obtains the evaluation value of the current strategy through an encoder and a decoder based on the output of the data integration module and the reward function value output by the reward function module, which is used to guide the parameter update of the strategy network module; the Transformer architecture is used to obtain the evaluation value of the current strategy through an encoder and a decoder based on the output of the data integration module and the reward function value output by the reward function module. The target network module of the former architecture obtains the ideal state value or reward signal through the encoder and decoder according to the output of the data integration module, which is used to guide the learning process of the strategy network module; the four sets of proportional parameter values, integral parameter values and differential parameter values of the proportional-integral-differential module are obtained by the strategy network and output as four voltage signals, which are passed through the voltage amplifier to adjust the four sets of input voltages required by the four sets of dielectric elastomer drivers; the prediction network module using the Transformer architecture obtains the predicted displacement, velocity and acceleration response of the vibration isolation-energy harvesting mechanical module according to the four voltage signals output by the proportional-integral-differential module through the encoder and decoder, accurately predicts the dynamic response characteristics of the vibration isolation system, and is used to supplement the vibration isolation platform response sample data in the experience buffer to accelerate the training speed of the control strategy.
[0025] The data integration module includes new state data sampled from the experience buffer (platform displacement, velocity, acceleration and four-way output voltage of the strategy network module at the current moment), and window data (platform displacement, velocity, acceleration and four-way output voltage of the strategy network module), and the number of sampling points of the window data is not less than 10. In the initial stage, when the number of sampling points of the buffer sample is n (less than 10), the data of window data 1 to (10-n) are filled with 0. The new state data is expanded from 1 state to 10, and the size is kept consistent with the window data dimension. Then the new state data and window data with uniform data size are input into the evaluation network, target network and decision network respectively.
[0026] The current state includes sensor data such as platform displacement, velocity, acceleration, etc.
[0027] Preferably, the strategy network module generates optimal proportional, integral and differential parameters through training to adapt to different working conditions and environmental conditions.
[0028] The input voltage required for adjusting the dielectric elastomer driver is: according to the current position y q With the ideal position y s The deviation value e=(y q -y s ), as well as the current control parameters, the neural network control strategy automatically calculates the output voltage and makes the system tend to a stable state.
[0029] The active control unit is further provided with an experience buffer module for storing the response data collected by the vibration isolation platform and the prediction data generated by the prediction network as training data for sampling and training of the strategy network module.
[0030] like Figure 5 As shown, a quasi-zero stiffness vibration control-energy harvesting method based on the above system includes:
[0031] Step 1, pre-training stage: When the base vibrates, the sensors on the base and the platform sense the displacement, velocity, and acceleration signals and feed them back to the active control unit;
[0032] Step 2: The active control unit inputs the received displacement, velocity, and acceleration data into the data buffer, then samples a portion of the displacement, velocity, and acceleration data from the data buffer, and calculates the reward value based on the sampled data. The data integration module extracts the window data and new state data from the data buffer.
[0033] Step 3: Input the output data and reward value of the data integration module into the strategy network, evaluation network and target network. Based on the gradient descent principle, update the weights of each network structure, and generate proportional, integral and differential parameter values. After proportional, integral and differential transformation, obtain the control voltage.
[0034] Step 4: Under the action of the output voltage, the four sets of dielectric elastomers on the vibration isolation platform generate corresponding strains, change the stretching amount and elastic modulus of the dielectric elastomers, and adjust the vibration isolation balance position and vibration isolation bandwidth. The sensors located on the base and the platform sense the changes in displacement, velocity, and acceleration, and feed back the new displacement, velocity, and acceleration to the active control unit.
[0035] Step 5, repeat steps 1-4, after multiple rounds of iterations and weight updates of each network, until the reward value of the position control effect reaches the set threshold, stop training, and consider that the network model that meets the vibration reduction requirements is obtained at this time, and fix the model weights for actual vibration isolation. During training, every 20 rounds of training, the output voltage and system displacement response data of 20 rounds are used as training data for the prediction module, of which 16 rounds of data are used as sample sets and 4 rounds of data are used as test sets. The input of the prediction module is the voltage value of the dielectric elastomer, and the error is the square root value of the difference between the predicted displacement output by the prediction module and the displacement generated by the actual system. Based on gradient descent, through multiple rounds of iterations, until the error is less than 2%, it is considered that the model accuracy of the prediction module meets the requirements, and then the prediction module replaces the physical platform, and performs rapid simulation together with other modules to generate 100 sets of simulated sample data, which are stored in the experience buffer as a supplement to the entire strategy training data.
[0036] Step 6. In actual vibration isolation, the collected displacement, velocity, and acceleration values of the base and platform are input into the data integration module, and then enter the trained network model to output different proportional, integral, and differential parameter values. After proportional, integral, and differential transformations, the control voltage is obtained to drive the dielectric elastomer to change the vibration isolation equilibrium position and bandwidth, thereby realizing active regulation of the vibration isolation capability.
[0037] Step 7: When the base and the platform move relative to each other, the magnets on the base and the platform also move relative to each other, the magnetic field around the coil changes, an induced current is generated in the coil, and electrical energy is output.
[0038] Compared with the existing technology, this device realizes the active adjustment of the vibration isolation bandwidth of the hexahedral quasi-zero stiffness system by driving the dielectric elastomer drive unit to deform through voltage, and generates induced current in the induction coil through the relative movement of the magnet to realize the collection of vibration energy and reduce the demand for external energy; by establishing an active control unit containing a data integration module and multiple neural networks, the historical data time series characteristics and the new state characteristics of the device are extracted at the same time to achieve better control performance; by feeding back the system displacement response to the reward function of the algorithm framework, the reward value is calculated according to the combination of displacement deviation, the cubic absolute value of the output value of the proportional-integral-differential module and the termination function, and the policy network, evaluation network and target network are updated according to the reward value to optimize the output control strategy. The data in the experience buffer will also be used as the simulated system displacement response feedback.
[0039] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose of the present invention. The protection scope of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. Each implementation scheme within its scope shall be subject to the constraints of the present invention.
Claims
1. A quasi-zero stiffness vibration control-energy harvesting system, characterized in that: include: A quasi-zero stiffness vibration isolation-energy harvesting mechanical module with a dielectric elastomer drive unit and a sensor, a voltage amplifier and an active control unit connected thereto, wherein: the active control unit generates active control instructions through a constructed neural network framework according to the displacement, velocity and acceleration information of the quasi-zero stiffness vibration isolation-energy harvesting mechanical module collected by the sensor, and outputs the active control instructions to the dielectric elastomer drive unit of the quasi-zero stiffness vibration isolation-energy harvesting mechanical module through the voltage amplifier to generate deformation, thereby actively adjusting the equilibrium position and stiffness of the quasi-zero stiffness vibration isolation-energy harvesting mechanical module to achieve broadband vibration isolation and energy harvesting; The quasi-zero stiffness vibration isolation-energy harvesting mechanical module comprises: a fixed base, a load platform, a dielectric elastomer drive unit, a repulsive magnet and an induction coil, wherein: the fixed base and the load platform are connected by four connecting rods, the bottom ends of the connecting rods are rotatably arranged on the slider guide rails on the fixed base, the dielectric elastomer drive unit is arranged between adjacent sliders, a pair of repulsive magnets and an induction coil are relatively arranged between the fixed base and the load platform through a direction constraint rod to provide a supporting force and constrain the load platform to move in a direction perpendicular to the fixed base, the relative movement between the load platform and the fixed base causes a change in the magnetic field, and then generates an induced current in the induction coil, which is used to provide auxiliary energy support for vibration control, and quasi-zero stiffness vibration isolation is achieved by setting the structural parameters of the dielectric elastomer drive unit and adjusting the stiffness of the dielectric elastomer drive unit; The dielectric elastomer drive unit undergoes different deformations and generates corresponding tensile forces and equivalent stiffness according to different voltages output by the active control unit, and specifically comprises: a multi-layer dielectric elastomer film sleeve and a drive rod, wherein: both ends of the drive rod are respectively connected to adjacent sliders, and the multi-layer dielectric elastomer film sleeve is extended after being triggered by the voltage, so that the drive rod controls the slider to move on the guide rail, thereby controlling the position adjustment of the connecting rod; The multilayer dielectric elastomer film sleeve is composed of several layers of concentrically arranged dielectric elastomer films, each layer of the dielectric elastomer film includes a dielectric elastomer in the middle and conductive graphite layers as electrodes arranged on both sides of the dielectric elastomer film, the conductive graphite layers are connected to the active control unit, and the dielectric elastomer generates a corresponding elongation according to the driving voltage to realize active regulation of the equilibrium position; The active control unit includes: a strategy network module, an evaluation network module, a target network module, a prediction network module, a proportional-integral-differential module, a reward function module and a data integration module, wherein: the strategy network module obtains three proportional, integral and differential parameter values through an encoder and a decoder according to the data output by the data integration module, and outputs them to the proportional-integral-differential module; the evaluation network module obtains the evaluation value of the current strategy through an encoder and a decoder based on the output of the data integration module and the reward function value; the target network module obtains the ideal state value or reward signal through an encoder and a decoder according to the output of the data integration module; the prediction network module obtains the reward signal through the proportional-integral-differential module The four-way voltage signal output by the block is used through the encoder and decoder to obtain the predicted displacement, velocity and acceleration response data of the quasi-zero stiffness vibration isolation-energy harvesting mechanical module, which is used as part of the training sample for weight adjustment of the strategy network module, the evaluation network module and the target network module; the proportional-integral-differential module outputs four-way voltage signals according to the proportional, integral and differential parameters; the reward function module calculates the reward value according to the displacement deviation and the voltage value of the dielectric elastic body, and provides it to the strategy network module, the evaluation network module and the target network module; the data integration module integrates the historical state data of the vibration isolation device and the current state data of the vibration isolation device, and provides them to the strategy network module, the evaluation network module and the target network module; The reward function module is based on the platform displacement , four voltage signals output by the proportional-integral-differential module , , , The square of the reward function and the termination function are used to calculate the reward function value. ,in: is the vibration source amplitude, The four voltage signals correspond to the input voltages of the four dielectric elastomer drive units. When the platform displacement does not exceed the set limit threshold, hour , the termination function is , otherwise the abort function is .
2. The quasi-zero stiffness vibration control-energy harvesting system according to claim 1 is characterized in that: The upper end of the connecting rod is fixed to the load platform by a hinge, and the lower end of the connecting rod is fixed to the slider by a hinge. The slider moves along a fixed direction on the guide rail, and the guide rail is set on a fixed base. When the slider moves toward the center, the bottom of the connecting rod is close to the center of the base, and the height of the load platform increases. When the slider moves in the opposite direction, the bottom of the connecting rod is away from the center of the base, and the height of the load platform decreases.
3. A quasi-zero stiffness vibration control-energy harvesting method based on the system of claim 1 or 2, characterized in that: include: Step 1, pre-training stage: When the base vibrates, the sensors on the base and the platform sense the displacement, velocity, and acceleration signals and feed them back to the active control unit; Step 2, the active control unit inputs the received displacement, velocity, and acceleration data into the data buffer, then samples a portion of the displacement, velocity, and acceleration data from the data buffer, and calculates the reward value based on the sampled data, and the data integration module extracts the window data and the new state data from the data buffer; Step 3: Input the output data and reward value of the data integration module into the strategy network, evaluation network and target network. Based on the gradient descent principle, update the weights of each network structure, generate proportional, integral and differential parameter values, and obtain the control voltage after proportional, integral and differential transformation. Step 4: Under the action of the output voltage, the four groups of dielectric elastomers of the vibration isolation platform generate corresponding strains, change the stretching amount and elastic modulus of the dielectric elastomers, and adjust the vibration isolation balance position and the vibration isolation bandwidth; The sensors located on the base and platform sense the changes in displacement, velocity, and acceleration, and feed the new displacement, velocity, and acceleration back to the active control unit; Step 5: Repeat steps 1-4, after multiple rounds of iterations and updating the weights of each network, until the reward value of the position control effect reaches the set threshold, stop training, and consider that a network model that meets the vibration reduction requirements is obtained at this time, and fix the model weights for actual vibration isolation; Step 6: In actual vibration isolation, the displacement, velocity, and acceleration values of the base and platform collected are input into the data integration module, and then enter the trained network model to output different proportional, integral, and differential parameter values. After proportional, integral, and differential transformation, the control voltage is obtained to drive the dielectric elastomer to change the vibration isolation equilibrium position and bandwidth, thereby realizing active regulation of the vibration isolation capability. Step 7: When the base and the platform move relative to each other, the magnets on the base and the platform also move relative to each other, the magnetic field around the coil changes, an induced current is generated in the coil, and electrical energy is output.
4. The quasi-zero stiffness vibration control-energy harvesting method according to claim 3 is characterized in that: During training, every 20 rounds of training, the output voltage and system displacement response data of 20 rounds are used as training data for the prediction module, among which 16 rounds of data are used as sample sets and 4 rounds of data are used as test sets. The input of the prediction module is the voltage value of the dielectric elastomer, and the error is the square root of the difference between the predicted displacement output by the prediction module and the displacement generated by the actual system. Based on gradient descent, multiple rounds of iterations are performed until the error is less than 2%. It is considered that the model accuracy of the prediction module meets the requirements. Then the prediction module replaces the physical platform and performs rapid simulation together with other modules to generate 100 sets of simulated sample data, which are stored in the experience buffer as a supplement to the entire strategy training data.
