A method of controlling a pneumatic flexible drive and related apparatus

Abstract

The application discloses a kind of pneumatically flexible driver control method and related equipment.The method includes: parameter updating is carried out to RMGPI model according to gradient descent method, wherein the RMGPI model is based on the change rate of input quantity and the hysteresis model related to frequency;The model output value and actual output value of the above-mentioned RMGPI model are obtained;According to the model output value and the actual output value, the target control rate is calculated by neural network controller to realize closed-loop control to the pneumatically flexible driver.

Description

Technical Field

[0001] This specification relates to the fields of control science and engineering technology, and more specifically, this application relates to a pneumatic flexible actuator control method and related equipment. Background Technology

[0002] Pneumatic flexible actuators have advantages such as light weight, high force-to-weight ratio, natural compliance, and high safety, and are currently widely used in wearable robots, surgical robots, soft grippers, and other fields. However, due to the hysteresis effect introduced by the pneumatic system and elastomer, high-precision trajectory tracking control still faces many challenges. Summary of the Invention

[0003] The summary section introduces a series of simplified concepts, which will be further explained in detail in the detailed description section. This summary section is not intended to limit the key and essential technical features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0004] In a first aspect, this application proposes a control method for a pneumatic flexible actuator, the method comprising:

[0005] The parameters of the RMGPI model are updated according to the gradient descent method. The RMGPI model is a hysteresis model based on the rate of change of the input and related to the frequency.

[0006] Obtain the model output value and actual output value of the above RMGPI model;

[0007] Based on the above model output value and the above actual output value, the target control rate is calculated by the neural network controller to achieve closed-loop control of the above pneumatic flexible actuator.

[0008] Optionally, the envelope function of the above RMGPI model includes a left envelope function and a right envelope function.

[0009] The above parameter update of the RMGPI model using gradient descent includes:

[0010] Based on the current output value of the operator of the RMGPI model, the output value of the operator of the RMGPI model in the previous cycle, the output value of the RMGPI model, the cost function, the output error value, and the chain rule, the gradient of the cost function with respect to the model parameters in each operator of the envelope function is obtained, so as to update the parameters of the RMGPI model.

[0011] Optionally, the above methods also include:

[0012] The squared loss function is determined as the cost function;

[0013] The difference between the output value of the RMGPI model and the actual output value is taken as the output error value.

[0014] Optionally, the above-mentioned calculation of the target control law using a neural network controller based on the model output value and the actual output value to achieve closed-loop control of the pneumatic flexible actuator includes:

[0015] The first input information and the second input information are obtained based on the above model output value and the above actual output value, wherein the first input information is the output difference between the above model output value and the above actual output value, and the second input information is the first derivative corresponding to the above output difference;

[0016] The first input information and the second input information are input to the neural network controller to calculate the target control rate in order to achieve closed-loop control of the pneumatic flexible actuator.

[0017] Optionally, the aforementioned neural network controller includes an input layer, a fuzzification layer, a hidden layer, and an output layer. The input parameters of the input layer include the input-output error value and the first-order derivative corresponding to the output error value. The fuzzification layer uses a Gaussian membership function. The hidden layer includes two fully connected hidden layers. The output layer determines the output by linear addition. The fuzzification layer and the hidden layer use the ReLU function as the activation function.

[0018] Optionally, the above neural network controller is an FNN controller.

[0019] The above methods also include:

[0020] The parameters of the FNN controller are updated based on the gradient descent method.

[0021] Optionally, the above-mentioned update of the parameters of the FNN controller based on gradient descent includes:

[0022] The squared loss function is selected as the cost function of the above FNN controller;

[0023] The partial differential relationships between the cost function and the output layer, fuzzification layer, and hidden layer are obtained using the chain rule.

[0024] The gradients of the output layer, fuzzy layer, and hidden layer are obtained based on the discrete form of the input equation of the RMGPI model and the partial differential equations mentioned above, in order to update the parameters.

[0025] Secondly, this application also proposes a pneumatic flexible actuator control device, comprising:

[0026] The update unit is used to update the parameters of the RMGPI model according to the gradient descent method. The RMGPI model is a hysteresis model based on the rate of change of the input and related to the frequency.

[0027] The acquisition unit is used to acquire the model output value and the actual output value of the RMGPI model mentioned above.

[0028] The control unit is used to calculate the target control rate through a neural network controller based on the above model output value and the above actual output value in order to achieve closed-loop control of the above pneumatic flexible actuator.

[0029] Thirdly, an electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program stored in the memory to implement the steps of the pneumatic flexible actuator control method as described in any of the first aspects above.

[0030] Fourthly, this application also proposes a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the pneumatic flexible actuator control method of any of the above claims in the first aspect.

[0031] In summary, the pneumatic flexible actuator control method of this application includes: updating the parameters of the RMGPI model according to the gradient descent method, wherein the RMGPI model is a hysteresis model based on the rate of change of the input and related to frequency; obtaining the model output value and the actual output value of the RMGPI model; and calculating the target control law through a neural network controller based on the model output value and the actual output value to achieve closed-loop control of the pneumatic flexible actuator. The flexible actuator control method provided in this application is based on an improved online hysteresis model (RMGPI model) of the Prandtl-Ishlinskii (PI) model, combined with gradient descent and model predictive control methods, to achieve end-to-end model predictive control of the flexible actuator. The proposed online hysteresis model is applicable to symmetric or asymmetric hysteresis systems with arbitrary frequency dependence, and the model can be updated online based on the gradient descent strategy to ensure model accuracy. At the same time, this method avoids the cumbersome inversion step required when using the traditional PI model to solve control problems, reducing the impact of model uncertainty and inaccurate modeling.

[0032] The pneumatic flexible actuator control method proposed in this application, along with other advantages, objectives, and features of this application, will be partly apparent from the following description and partly understood by those skilled in the art through study and practice of this application. Attached Figure Description

[0033] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0034] Figure 1 This is a schematic flowchart of a pneumatic flexible actuator control method provided in an embodiment of this application;

[0035] Figure 2 A cross-sectional structural schematic diagram of a pneumatic flexible actuator provided in an embodiment of this application;

[0036] Figure 3 A schematic diagram illustrating the principle of an RMGPI model operator provided in this application embodiment.

[0037] Figure 4 A schematic diagram of the structure of an RMGPI model provided in this application embodiment;

[0038] Figure 5 A schematic diagram of a controller structure provided in an embodiment of this application;

[0039] Figure 6 A schematic diagram of the structure of a control system provided in an embodiment of this application;

[0040] Figure 7 This is a structural schematic diagram of a pneumatic flexible actuator control device provided in an embodiment of this application;

[0041] Figure 8 This is a schematic diagram of a pneumatic flexible actuator control electronic device provided in an embodiment of this application. Detailed Implementation

[0042] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The technical solutions of the embodiments of this application will now be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0043] like Figure 2 The diagram shown is a cross-sectional schematic of a pneumatic flexible actuator provided in an embodiment of this application. The actuator body is made of vulcanized silicone rubber, which has excellent elongation and tear resistance. The upper side of the actuator employs a fast Pneu-net structure with high bending efficiency to construct rectangular chambers, which are connected at the bottom. The bottom of the flexible actuator contains a binding layer. When gas is introduced, the elongation of the binding layer is smaller than that of the vulcanized silicone rubber body, resulting in inconsistent deformation of the entire flexible actuator: when gas is introduced, the upper rectangular chamber structure expands, while the bottom does not deform significantly due to the presence of the binding layer. Therefore, when gas is introduced, the rectangular chambers and the bottom deform to different degrees, resulting in a downward bending motion.

[0044] Pneumatic flexible actuators have advantages such as light weight, high force-to-weight ratio, natural compliance, and high safety, and are currently widely used in wearable robots, surgical robots, soft grippers, and other fields. However, due to the hysteresis effect introduced by the pneumatic system and elastomer, high-precision trajectory tracking control still faces many challenges.

[0045] Please see Figure 1 This is a schematic flowchart of a pneumatic flexible actuator control method provided in an embodiment of this application, which may specifically include:

[0046] S110. Update the parameters of the RMGPI model according to the gradient descent method. The RMGPI model is a hysteresis model based on the rate of change of the input and related to the frequency.

[0047] For example, such as Figure 4The diagram shown is a structural schematic of an RMGPI model proposed in this application. The RMGPI (Rate-dependent modified generalized Prandtl-Ishlinskii) model introduces the rate of change of quantity, thus enabling the modeling of frequency-dependent hysteresis.

[0048] The RMGPI model incorporates a differential term into the input to account for the rate of change of the input air pressure. The input equation is as follows:

[0049]

[0050] Where u represents the air pressure input into the air chamber. q is the differential coefficient.

[0051] The RMGPI model is obtained by linearly adding N operators, and the graphs of the operators are as follows: Figure 3 As shown. The difference between the RMGPI model and the classic PI model is that the RMGPI model divides the envelope function into a left envelope function β. L and the right envelope function β R Therefore, it can describe asymmetric hysteresis. The output of each operator can be expressed by the following formula:

[0052] F β (t)=f β (u * (t),Fβ(t - (2)

[0053] Among them, F β (t) represents the output of a single operator of the pneumatic flexible actuator, F β (t - ) is the output of the operator at the previous time step, and we have:

[0054]

[0055] in,

[0056]

[0057] k represents "L or R", which is the left or right envelope function of operator j (j = 1, 2, ..., N). A k B k C k Indicate the model parameters in each operator:

[0058]

[0059] Where N represents the number of operators, and different operators have different parameters.

[0060] Ultimately, the model output is represented as a linear sum of individual operators:

[0061]

[0062] Where y m (t) represents the model prediction output of the aerodynamic flexible actuator at time t.

[0063] The RMGPI model significantly improves upon the Prandtl-Ishlinskii model, demonstrating greater generalizability in frequency-related tasks. However, the system is highly complex, and the model itself contains uncertainties. Gradient-based online parameter update methods can achieve high-precision models.

[0064] S120. Obtain the model output value and actual output value of the above RMGPI model;

[0065] For example, such as Figure 6 The diagram shown is a structural schematic of a control system provided in an embodiment of this application. The role of the RMGPI model is to approximate the real system, and the controller performs control based on the real output value and the model output of the RMGPI model.

[0066] S130. Based on the above model output value and the above actual output value, the target control rate is calculated by the neural network controller to achieve closed-loop control of the above pneumatic flexible actuator.

[0067] For example, in this embodiment, the target control law is calculated using a neural network controller based on the model output value and the actual output value to achieve closed-loop control of the pneumatic flexible actuator. Predictive control is performed using the proposed system dynamics model. This explicit dynamics model allows for better system motion control performance, namely shorter settling time and better trajectory tracking performance. Furthermore, the explicit model theoretically optimizes the fitting results for hysteresis effects.

[0068] The neural network model controller proposed in this application can employ neural network controllers such as FCN (Fully Convolutional Network) and FNN (Fuzzy Neural Network). With the rapid development of artificial intelligence and machine learning, FCN, as a powerful model, is widely used in control system design. FCN is a feedforward neural network with multiple layers of neurons, where each neuron is connected to all neurons in the previous layer. It has many advantages, such as strong nonlinear modeling capabilities, high adaptability, and learning and adaptive abilities. This makes FCN an ideal choice for designing high-performance controllers.

[0069] Fuzzy Neural Networks (FNNs) are a hybrid model combining fuzzy logic and artificial neural networks. By integrating the fuzzy set theory of fuzzy logic with the learning and adaptive capabilities of neural networks, they can handle fuzzy information and uncertainty, exhibiting a degree of fault tolerance and robustness. Compared to traditional neural networks, fuzzy neural networks have advantages in handling fuzzy information and uncertainty. They can cope with fuzzy and incomplete data in the real world and make reasonable decisions when faced with ambiguous problems.

[0070] Neural networks possess strong learning capabilities and adaptability, exhibiting high versatility in complex scenarios. The design of neural networks is based on the backpropagation algorithm. The backpropagation algorithm enables very fast online estimation of the neural network weight matrix based on tracking error signals.

[0071] In summary, the flexible actuator control method provided in this application is based on an improved online hysteresis model (RMGPI model) of the Prandtl-Ishlinskii (PI) model, combined with gradient descent and model predictive control methods, to achieve end-to-end model predictive control of the flexible actuator. The proposed online hysteresis model is applicable to symmetric or asymmetric hysteresis systems with arbitrary frequency correlation, and the model can be updated online based on the gradient descent strategy, ensuring model accuracy. Simultaneously, this method avoids the cumbersome inversion steps required when using the traditional PI model to solve control problems, reducing the impact of model uncertainty and inaccurate modeling.

[0072] In some examples, the above method also includes:

[0073] The squared loss function is determined as the cost function;

[0074] The difference between the output value of the RMGPI model and the actual output value is taken as the output error value.

[0075] For example, the online update process for RMGPI model parameters involves backpropagating the error err_m between the actual system output and the model output at each step, and updating the model parameters through gradient descent to achieve better prediction results.

[0076] First, define the cost function as the squared loss function.

[0077]

[0078] The optimization objective is to minimize the loss function J1. The model output y is defined. m The error between the actual system output y and the actual system output y is

[0079] err m =y m -y.(8)

[0080] In some examples, the envelope function of the RMGPI model described above includes a left envelope function and a right envelope function.

[0081] Step S110, the above-mentioned parameter update of the RMGPI model based on gradient descent may specifically include:

[0082] Based on the current output value of the operator of the RMGPI model, the output value of the operator of the RMGPI model in the previous cycle, the output value of the RMGPI model, the cost function, the output error value, and the chain rule, the gradient of the cost function with respect to the model parameters in each operator of the envelope function is obtained, so as to update the parameters of the RMGPI model.

[0083] For example, the current output value of the operator of the RMGPI model, the output value of the operator of the RMGPI model in the previous cycle, the model output value of the RMGPI model, the cost function, and the output error value can be obtained by one or a combination of formulas (1)-(7), and the gradient of the cost function with respect to the model parameters in each operator of the envelope function can be obtained by the chain rule. Specifically, it can include:

[0084] From equations (2)(3)(4)(6), we can obtain

[0085]

[0086] Then, according to the chain rule and equations (7), (8), and (9), calculate the cost function J1 for β. k parameters gradient:

[0087]

[0088] Write the gradients of N operators in vector form:

[0089]

[0090] The parameter matrix of the RMGPI model is defined as follows:

[0091]

[0092] Define the gradient descent learning rate as lr1, W PI i (i = 1, ..., N) represents W PI If the i-th column is given, then the update rule for the RMGPI model is:

[0093]

[0094] In some examples, the above-mentioned calculation of the target control law using a neural network controller based on the model output value and the actual output value to achieve closed-loop control of the pneumatic flexible actuator includes:

[0095] The first input information and the second input information are obtained based on the above model output value and the above actual output value, wherein the first input information is the output difference between the above model output value and the above actual output value, and the second input information is the first derivative corresponding to the above output difference;

[0096] The first input information and the second input information are input to the neural network controller to calculate the target control rate in order to achieve closed-loop control of the pneumatic flexible actuator.

[0097] For example, the specific structure of a neural network controller is as follows: Figure 5 As shown, it consists of an input layer, a fuzzification layer, a hidden layer (fully connected layer), and an output layer. The input signal of the controller is the error err. rm First derivative of the sum of errors

[0098] In some examples, the neural network controller includes an input layer, a fuzzification layer, a hidden layer, and an output layer. The input parameters of the input layer include the input-output error value and the first derivative corresponding to the output error value. The fuzzification layer uses a Gaussian membership function. The hidden layer includes two fully connected hidden layers. The output layer determines the output by linear addition. The fuzzification layer and the hidden layer use the ReLU function as the activation function.

[0099] In some examples, the neural network controller described above is an FNN controller.

[0100] The above methods also include:

[0101] The parameters of the FNN controller are updated based on the gradient descent method.

[0102] For example, a neural network controller includes an input layer, a fuzzing layer, a hidden layer, and an output layer.

[0103] First, define:

[0104] err rm =y m -y r (13)

[0105] W = [W out [W1,W2,W3] T (14)

[0106] Among them, y m y is the predicted output, and y is the actual output. Similarly, yr , representing the reference input signal of the pneumatic flexible actuator.

[0107] The specific structure of the controller is as follows: Figure 5 As shown, it consists of an input layer, a fuzzification layer, a hidden layer (fully connected layer), and an output layer. The input signal of the controller is the error err. rm First derivative of the sum of errors Therefore, the input equation can be written as:

[0108]

[0109] The second part of the controller is the fuzzification layer. The fuzzy neural network combines the nonlinear transformation mechanism of fuzzy logic reasoning with the learning ability of neural networks. Here, the fuzzification layer uses a Gaussian membership function, and each input node corresponds to K nodes in the fuzzification layer:

[0110]

[0111] Where c i and σ i The parameters are Gaussian function parameters. Choosing the ReLU function as the activation function is computationally efficient and avoids the gradient vanishing problem. The output of the blurring layer (i.e., the input of the hidden layer) is:

[0112] h0 = [max(W1F, 0), 1] T (17)

[0113] Where W1 is the bias parameter.

[0114] The third part of the controller is the hidden layer, which uses two fully connected hidden layers, and the activation function is also the ReLU function. The outputs of the two hidden layers are as follows:

[0115] h1 = [max(W2F, 0)] T ,1] T (18)

[0116] h2 = [max(W3F, 0), 1] T (19)

[0117] Similarly, W2 and W3 are bias parameters.

[0118] The final part of the controller is the output layer, which uses linear addition to obtain the final controller output as follows:

[0119] u = W out ×h2 (20)

[0120] Among them, W out These are the bias parameters of the output layer.

[0121] In some examples, the parameters of the FNN controller are updated based on gradient descent, including:

[0122] The squared loss function is selected as the cost function of the above FNN controller;

[0123] The partial differential relationships between the cost function and the output layer, fuzzification layer, and hidden layer are obtained using the chain rule.

[0124] The gradients of the output layer, fuzzy layer, and hidden layer are obtained based on the discrete form of the input equation of the RMGPI model and the partial differential equations mentioned above, in order to update the parameters.

[0125] For example, this application first obtains RMGPI model parameters that approximate the real model through genetic algorithm parameter identification, and then updates them online to make the predicted output y m The algorithm continuously approximates the true output y. Since the RMGPI model is an explicit mathematical model and its gradient is known, the gradients of the controller parameters based on the neural network can be derived. Similarly, using err... rm By calculating the gradient of the controller parameters and backpropagating, the controller parameters can be updated online to make y m Gradually approaching y r W represents the bias term in the controller. The controller employs a gradient descent-based control method; the backpropagation of the controller will be calculated below.

[0126] Since the controller is a neural network, gradient descent is used to update the parameters. First, the squared loss function is chosen as the cost function of the controller, denoted as...

[0127]

[0128] The optimization objective is to minimize the cost function J2, which means that the controller rate solution is transformed into an optimization problem.

[0129] According to the chain rule, we can obtain the following formula:

[0130]

[0131] Based on equation (1), u* can be written in discrete form:

[0132]

[0133] Where t s Given the sampling time, we can obtain the partial derivative of u* with respect to u:

[0134]

[0135] Substituting (24) into (22) yields the cost function J2 for the output layer parameter W. out gradient:

[0136]

[0137] Similarly, the gradients of the cost function J2 with respect to the parameters W3 and W2 of the two hidden layers and the parameter W1 of the fuzzy layer can be obtained:

[0138]

[0139]

[0140]

[0141] Let the partial differential matrix be:

[0142]

[0143] If the learning rate for the controller parameters is defined as lr2, then the update rule for the controller parameters is:

[0144] W = W - lr² × Δ. (30)

[0145] Choose the Lyapunov equation:

[0146]

[0147] Where N is the prediction step size, we can obtain:

[0148]

[0149] From formula (30), we can obtain:

[0150]

[0151] Substituting the above equation into equation (32), we get:

[0152]

[0153] At the same time:

[0154]

[0155] remember:

[0156]

[0157] z(t)=Q(t) T *lr2*Q(t) (37)

[0158] Substituting equations (36) and (37) into equation (35), we get:

[0159]

[0160] Substituting equation (38) into equation (34), we get:

[0161] e m (t+N+1)=(1-z(t))*e m (t+N)+O(t+N) (39)

[0162] Substituting equation (39) into equation (32) and scaling it up, we get:

[0163] ΔV(t)<[(1-z(t))*e(t+N)+O(t+N)] 2 +[z 2 (t)-1]*e 2 (t+N) (40)

[0164] Assuming O(t+N)→0, to prove ΔV(t)<0, we only need to have:

[0165]

[0166] The above formula is equivalent to:

[0167]

[0168] According to equation (37), z is a quadratic form and is always greater than or equal to zero, therefore the above equation always holds. Thus, ΔV(t) < 0 always holds, and the convergence is proven.

[0169] This concludes the convergence proof process.

[0170] In summary, the pneumatic flexible actuator control method proposed in this application avoids solving the inverse PI model while realizing model predictive control based on a fuzzy neural network using an improved PI model that is updated online. The method proposed in this embodiment is characterized by high computational efficiency and ease of implementation, and can be applied to any system with asymmetric and frequency-dependent hysteresis characteristics. Furthermore, based on Lyapunov stability theory, a stability proof is provided, guaranteeing the convergence of the closed-loop system.

[0171] Please see Figure 7 One embodiment of the pneumatic flexible actuator control device in this application may include:

[0172] Update unit 21 is used to update the parameters of the RMGPI model according to the gradient descent method, wherein the RMGPI model is a hysteresis model based on the rate of change of the input and related to the frequency.

[0173] The acquisition unit 22 is used to acquire the model output value and the actual output value of the above RMGPI model;

[0174] The control unit 23 is used to calculate the target control rate through the neural network controller based on the above model output value and the above actual output value to achieve closed-loop control of the above pneumatic flexible actuator.

[0175] The pneumatic flexible actuator control unit can also operate in the following ways:

[0176] The envelope function of the RMGPI model described above includes a left envelope function and a right envelope function.

[0177] The above parameter update of the RMGPI model using gradient descent includes:

[0178] Based on the current output value of the operator of the RMGPI model, the output value of the operator of the RMGPI model in the previous cycle, the output value of the RMGPI model, the cost function, the output error value, and the chain rule, the gradient of the cost function with respect to the model parameters in each operator of the envelope function is obtained, so as to update the parameters of the RMGPI model.

[0179] In one feasible implementation, the above method further includes:

[0180] The squared loss function is determined as the cost function;

[0181] The difference between the output value of the RMGPI model and the actual output value is taken as the output error value.

[0182] In one feasible implementation, the above-mentioned calculation of the target control law by a neural network controller based on the above-mentioned model output value and the above-mentioned actual output value to achieve closed-loop control of the above-mentioned pneumatic flexible actuator includes:

[0183] The first input information and the second input information are obtained based on the above model output value and the above actual output value, wherein the first input information is the output difference between the above model output value and the above actual output value, and the second input information is the first derivative corresponding to the above output difference;

[0184] The first input information and the second input information are input to the neural network controller to calculate the target control rate in order to achieve closed-loop control of the pneumatic flexible actuator.

[0185] In one feasible implementation, the neural network controller includes an input layer, a fuzzification layer, a hidden layer, and an output layer. The input parameters of the input layer include the input-output error value and the first derivative corresponding to the output error value. The fuzzification layer uses a Gaussian membership function. The hidden layer includes two fully connected hidden layers. The output layer determines the output by linear addition. The fuzzification layer and the hidden layer use the ReLU function as the activation function.

[0186] In one feasible implementation, the aforementioned neural network controller is an FNN controller.

[0187] The above methods also include:

[0188] The parameters of the FNN controller are updated based on the gradient descent method.

[0189] In one feasible implementation, updating the parameters of the FNN controller based on gradient descent includes:

[0190] The squared loss function is selected as the cost function of the above FNN controller;

[0191] The partial differential relationships between the cost function and the output layer, fuzzification layer, and hidden layer are obtained using the chain rule.

[0192] The gradients of the output layer, fuzzy layer, and hidden layer are obtained based on the discrete form of the input equation of the RMGPI model and the partial differential equations mentioned above, in order to update the parameters.

[0193] like Figure 8 As shown, this application embodiment also provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, it implements the steps of any of the above-described methods for controlling the pneumatic flexible actuator, specifically including:

[0194] The parameters of the RMGPI model are updated according to the gradient descent method. The RMGPI model is a hysteresis model based on the rate of change of the input and related to the frequency.

[0195] Obtain the model output value and actual output value of the above RMGPI model;

[0196] Based on the above model output value and the above actual output value, the target control rate is calculated by the neural network controller to achieve closed-loop control of the above pneumatic flexible actuator.

[0197] In one feasible implementation, the envelope function of the aforementioned RMGPI model includes a left envelope function and a right envelope function.

[0198] The above parameter update of the RMGPI model using gradient descent includes:

[0199] Based on the current output value of the operator of the RMGPI model, the output value of the operator of the RMGPI model in the previous cycle, the output value of the RMGPI model, the cost function, the output error value, and the chain rule, the gradient of the cost function with respect to the model parameters in each operator of the envelope function is obtained, so as to update the parameters of the RMGPI model.

[0200] In one feasible implementation, the above method further includes:

[0201] The squared loss function is determined as the cost function;

[0202] The difference between the output value of the RMGPI model and the actual output value is taken as the output error value.

[0203] In one feasible implementation, the above-mentioned calculation of the target control law by a neural network controller based on the above-mentioned model output value and the above-mentioned actual output value to achieve closed-loop control of the above-mentioned pneumatic flexible actuator includes:

[0204] The first input information and the second input information are obtained based on the above model output value and the above actual output value, wherein the first input information is the output difference between the above model output value and the above actual output value, and the second input information is the first derivative corresponding to the above output difference;

[0205] The first input information and the second input information are input to the neural network controller to calculate the target control rate in order to achieve closed-loop control of the pneumatic flexible actuator.

[0206] In one feasible implementation, the neural network controller includes an input layer, a fuzzification layer, a hidden layer, and an output layer. The input parameters of the input layer include the input-output error value and the first derivative corresponding to the output error value. The fuzzification layer uses a Gaussian membership function. The hidden layer includes two fully connected hidden layers. The output layer determines the output by linear addition. The fuzzification layer and the hidden layer use the ReLU function as the activation function.

[0207] In one feasible implementation, the aforementioned neural network controller is an FNN controller.

[0208] The above methods also include:

[0209] The parameters of the FNN controller are updated based on the gradient descent method.

[0210] In one feasible implementation, updating the parameters of the FNN controller based on gradient descent includes:

[0211] The squared loss function is selected as the cost function of the above FNN controller;

[0212] The partial differential relationships between the cost function and the output layer, fuzzification layer, and hidden layer are obtained using the chain rule.

[0213] The gradients of the output layer, fuzzy layer, and hidden layer are obtained based on the discrete form of the input equation of the RMGPI model and the partial differential equations mentioned above, in order to update the parameters.

[0214] Since the electronic device described in this embodiment is the device used to implement a pneumatic flexible actuator control device in the embodiments of this application, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the method described in the embodiments of this application. Therefore, how the electronic device implements the method in the embodiments of this application will not be described in detail here. Any device used by those skilled in the art to implement the method in the embodiments of this application is within the scope of protection of this application.

[0215] In practical implementation, when the computer program 311 is executed by the processor, it can achieve the following: Figure 1 Any of the corresponding implementation methods in the embodiments.

[0216] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0217] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0218] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0219] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0220] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0221] This application also provides a computer program product, which includes computer software instructions. When the computer software instructions are executed on a processing device, the processing device performs the pneumatic flexible actuator control process described in the corresponding embodiment.

[0222] A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0223] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0224] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0225] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0226] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0227] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0228] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A control method for a pneumatic flexible actuator, characterized in that, include: The parameters of the RMGPI model are updated according to the gradient descent method, wherein the RMGPI model is a hysteresis model based on the rate of change of the input and related to frequency. Obtain the model output value and actual output value of the RMGPI model; wherein, the model output value is used to characterize the model predicted output of the aerodynamic flexible actuator; and the actual output value is used to characterize the actual output of the aerodynamic flexible actuator. Based on the model output value and the actual output value, the target control rate is calculated by the neural network controller to achieve closed-loop control of the pneumatic flexible actuator; The envelope function of the RMGPI model includes a left envelope function and a right envelope function. The parameter update of the RMGPI model based on gradient descent includes: The gradients of the cost function with respect to the model parameters in each operator of the envelope function are obtained based on the current output value of the operator of the RMGPI model, the output value of the operator of the RMGPI model in the previous cycle, the model output value of the RMGPI model, the cost function, the output error value, and the chain rule, so as to update the parameters of the RMGPI model. The step of calculating the target control rate using a neural network controller based on the model output value and the actual output value to achieve closed-loop control of the pneumatic flexible actuator includes: First input information and second input information are obtained based on the model output value and the actual output value, wherein the first input information is the output difference between the model output value and the actual output value, and the second input information is the first derivative corresponding to the output difference; The first input information and the second input information are input to the neural network controller to calculate the target control rate in order to achieve closed-loop control of the pneumatic flexible actuator.

2. The control method according to claim 1, characterized in that, Also includes: The squared loss function is determined as the cost function; The difference between the output value of the RMGPI model and the actual output value is used as the output error value.

3. The control method according to claim 1, characterized in that, The neural network controller includes an input layer, a fuzzification layer, a hidden layer, and an output layer. The input parameters of the input layer include the input-output error value and the first derivative corresponding to the output error value. The fuzzification layer uses a Gaussian membership function. The hidden layer includes two fully connected hidden layers. The output layer determines the output by linear addition. The fuzzification layer and the hidden layer use the ReLU function as the activation function.

4. The method according to claim 3, characterized in that, The neural network controller is an FNN controller. The method further includes: The parameters of the FNN controller are updated based on the gradient descent method.

5. The method according to claim 4, characterized in that, The step of updating the parameters of the FNN controller based on gradient descent includes: The squared loss function is selected as the cost function of the FNN controller; The partial differential relationships between the cost function and the output layer, fuzzification layer, and hidden layer are obtained using the chain rule. The gradients of the output layer, fuzzy layer, and hidden layer are obtained based on the discrete form of the input equation of the RMGPI model and the partial differential relation, respectively, to update the parameters.

6. A pneumatic flexible actuator control device, characterized in that, include: An update unit is used to update the parameters of the RMGPI model according to the gradient descent method, wherein the RMGPI model is a hysteresis model based on the rate of change of the input and related to frequency. The acquisition unit is used to acquire the model output value and the actual output value of the RMGPI model; wherein, the model output value is used to characterize the model prediction output of the aerodynamic flexible actuator; and the actual output value is used to characterize the actual output of the aerodynamic flexible actuator. The control unit is used to calculate the target control rate through a neural network controller based on the model output value and the actual output value in order to achieve closed-loop control of the pneumatic flexible actuator. The envelope function of the RMGPI model includes a left envelope function and a right envelope function. The parameter update of the RMGPI model based on gradient descent includes: The gradients of the cost function with respect to the model parameters in each operator of the envelope function are obtained based on the current output value of the operator of the RMGPI model, the output value of the operator of the RMGPI model in the previous cycle, the model output value of the RMGPI model, the cost function, the output error value, and the chain rule, so as to update the parameters of the RMGPI model. The step of calculating the target control rate using a neural network controller based on the model output value and the actual output value to achieve closed-loop control of the pneumatic flexible actuator includes: First input information and second input information are obtained based on the model output value and the actual output value, wherein the first input information is the output difference between the model output value and the actual output value, and the second input information is the first derivative corresponding to the output difference; The first input information and the second input information are input to the neural network controller to calculate the target control rate in order to achieve closed-loop control of the pneumatic flexible actuator.

7. An electronic device, comprising: The memory and processor are characterized in that the processor is used to execute a computer program stored in the memory to implement the steps of the control method of the pneumatic flexible actuator as described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the control method of the pneumatic flexible actuator as described in any one of claims 1-5.

Images