A method, system, device and medium for controlling flapping wing energy harvesting performance

Abstract

The present invention discloses a control method, system, device and medium for flapping wing energy harvesting performance, which relates to the field of flow control technology, including: real-time measurement of flow characteristic parameters of incoming flow and motion parameters of flapping wings; converting flow characteristic parameters and motion parameters into numerical vectors and inputting them into a neural network model to generate a predicted value of energy harvesting performance; adjusting the control parameters of flapping wings according to the predicted value of energy harvesting performance and the actual value of energy harvesting performance; inputting the adjusted control parameters into an actuator, controlling the movement of flapping wings through the actuator, and optimizing the energy harvesting performance. The present invention can monitor and adjust the movement and energy harvesting performance of flapping wings in real time without artificially setting control parameters, thereby improving the efficiency and stability of the control system.

Description

Technical Field

[0001] The present invention relates to the field of flow control technology, and in particular to a method, system, device and medium for controlling flapping wing energy harvesting performance. Background Art

[0002] Flapping wing energy harvesting is a method of harvesting energy from incoming flow using the aerodynamic force generated by flapping wing motion. The optimization of flapping wing motion or control parameters needs to consider the changing conditions of incoming flow, such as speed, direction, turbulence, etc., as well as the characteristic parameters of flapping wing, such as airfoil, size, flexibility, motion parameters, etc. These factors will affect the aerodynamic force and energy harvesting efficiency of flapping wing. The key to the control system is how to optimize the flapping wing motion or control parameters to maximize the energy harvesting efficiency. Traditional flapping wing energy harvesting research is based on numerical simulation or experimental testing methods, which can accurately reflect the flow characteristics and energy harvesting performance of flapping wing. However, the disadvantage is that the calculation or testing cost is high, it is difficult to consider the complex nonlinear and unsteady flow characteristics of flapping wing, and it is impossible to meet the rapid analysis of flapping wing performance under time-varying incoming flow conditions, so it is impossible to achieve real-time control. In addition, most of the design parameters of flapping wing are continuous variables, and the traditional method only studies discrete points, which cannot accurately obtain the optimal working conditions. It has strong limitations and is difficult to adapt to the rapid optimization control of flapping wing energy harvesting performance under non-uniform time-varying incoming flow conditions.

[0003] At present, there are some methods based on feedback control to improve the energy harvesting performance of flapping-wing energy harvesting devices, such as methods based on PID controllers and methods based on fuzzy controllers. The common feature of these methods is that they all need to establish a mathematical model of the flapping-wing energy harvesting device, and then calculate the error and control parameters based on the output and target value of the model to adjust the movement of the flapping wing. The main problems are as follows: 1. The mathematical model of the flapping-wing energy harvesting device is often complex, difficult to establish and solve, and there may be uncertainty and nonlinearity, resulting in low accuracy and adaptability of the model. 2. PID controllers, fuzzy controllers, etc. all need to manually set some parameters, such as proportional coefficients, integral coefficients, differential coefficients, fuzzy rules, adaptive gains, etc. The setting of these parameters often depends on experience and experiments, and may fail with changes in the flow field, resulting in low performance and stability of the control system. Summary of the invention

[0004] The present invention provides a control method, system, device and medium for flapping wing energy harvesting performance, which solves the problems that the existing mathematical model may have uncertainty and nonlinearity, resulting in low accuracy and adaptability of the model, the controller needs to manually set some parameters, and cannot achieve fast and accurate control.

[0005] The present invention provides a method for controlling flapping wing energy harvesting performance, comprising:

[0006] Real-time measurement of the flow characteristic parameters of the incoming flow and the motion parameters of the flapping wings;

[0007] The flow characteristic parameters and motion parameters are converted into numerical vectors and input into the neural network model to generate the predicted value of energy harvesting performance;

[0008] Adjusting the control parameters of the flapping wings according to the predicted value of the energy harvesting performance and the actual value of the energy harvesting performance;

[0009] The adjusted control parameters are input into the actuator, and the movement of the flapping wing is controlled by the actuator to optimize the flapping wing energy harvesting performance;

[0010] The deep neural network model includes:

[0011] The input layer is used to receive numerical vectors;

[0012] Convolutional layer, used to extract local features from numerical vectors;

[0013] The conversion layer is used to rearrange the output of the convolutional layer according to the time dimension and feature dimension;

[0014] LSTM layer, used to extract time series features from the output of the conversion layer;

[0015] The fully connected layer is used to extract high-level features from the output of the LSTM layer;

[0016] The output layer is used to classify or regress the output of the fully connected layer and output the predicted value of the energy harvesting performance.

[0017] Preferably, the flow characteristic parameters of the incoming flow are measured by the first sensor component, the motion parameters of the flapping wings are measured by the second sensor component, and the real value of the energy harvesting performance is output by the flapping-wing energy harvesting device.

[0018] Preferably, the flow characteristic parameters include time-varying speed, direction and turbulence, and the motion parameters include flapping wing position, speed, acceleration, angle, angular velocity and angular acceleration.

[0019] Preferably, the input layer is composed of a plurality of neurons, the number of neurons is the same as the dimension of the numerical vector, and the output is a (0) = x, where a (0) is an N×1 vector;

[0020] The convolution layer includes 64 neurons, and its activation function is a nonlinear function, which is convolved by the following formula:

[0021] a (1) =g (1) (W(1) *a (0) +b (1) )

[0022] In the formula, a (1) is a 64×1 vector, b (1) is a 64×1 bias vector, W (1) is the convolution kernel, * indicates the convolution operation, g (1) (·) is the activation function;

[0023] The output of the conversion layer is as follows:

[0024] a (1.5) = reshape(a (1) ,T,F)

[0025] In the formula, a (1.5) It is a T×F matrix, where T is the number of time steps and F is the feature dimension;

[0026] The LSTM layer includes 64 neurons, which are composed of multiple LSTM units. Each LSTM unit has a cell state and a hidden state for storing and transmitting long-term and short-term information. Each LSTM unit has three gates, namely a forget gate, an input gate, and an output gate, which are used to control the update and output of the cell state. The output of the LSTM layer is as follows:

[0027] a (2) =g (2) (W (2) a (1) +b (2) )

[0028] a (2) is a 64×1 vector, W (2) is a 64×64 weight matrix, b (2) is a 64×1 bias vector, g (2) (·) is the activation function;

[0029] The fully connected layer includes 64 neurons, and its output is as follows:

[0030] a (3) =g (3) (W (3) *a (2) +b (3) )

[0031] In the formula, a (3) is a 64×1 vector, W (3) is a 64×64 group convolution kernel, b (3) is a 64×1 bias vector, * indicates group convolution operation, g(3) (·) is the activation function;

[0032] The output layer includes 4 neurons, and its output is as follows:

[0033] a (4) =g (4) (W (4) a (3) +b (4) )

[0034] In the formula, a (4) is a 4×1 vector, W (4) is a 4×64 weight matrix, b (4) is a 4×1 bias vector, g (4) (·) is the activation function.

[0035] Preferably, before converting the flow characteristic parameters and motion parameters into numerical vectors and inputting them into the neural network model, the neural network model needs to be trained, including the following steps:

[0036] Initialize the parameters of the neural network model;

[0037] Get a data set, including multiple input features and corresponding multiple output labels;

[0038] Input the input features to the initialized neural network model for forward propagation to obtain the predicted value;

[0039] Calculate the loss function of the neural network model based on the predicted value and the corresponding output label;

[0040] Input the input features to the initialized neural network model for back propagation, and calculate the gradient and error of each layer;

[0041] Update the weight matrix and bias vector of each layer according to the gradient descent algorithm;

[0042] Repeat the above forward propagation and back propagation process until the loss function converges or the maximum number of iterations is reached, completing the training of the neural network model.

[0043] Preferably, adjusting the control parameters of the flapping wings according to the predicted value of the energy harvesting performance and the actual value of the energy harvesting performance comprises the following steps:

[0044] calculating an error and a gradient between a predicted value of the energy harvesting performance and an actual value of the energy harvesting performance;

[0045] The control parameters of flapping wings are adjusted by gradient descent algorithm.

[0046] Preferably, the actuator includes a motor, a servo motor and a solenoid valve; the motor is used to control the wingspan and wing sweep of the flapping wing, the servo motor is used to control the phase and angle of attack of the flapping wing, and the solenoid valve is used to control the curvature and thickness of the flapping wing.

[0047] A control system for flapping wing energy harvesting performance, comprising the following steps:

[0048] An acquisition module is used to measure the flow characteristic parameters of the incoming flow and the motion parameters of the flapping wings in real time;

[0049] An input module, used for converting flow characteristic parameters and motion parameters into numerical vectors and inputting the numerical vectors into a neural network model to generate a predicted value of energy harvesting performance;

[0050] An adjustment module, used for adjusting the control parameters of the flapping wings according to the predicted value of the energy harvesting performance and the actual value of the energy harvesting performance;

[0051] A control module, used for inputting the adjusted control parameters into the actuator, controlling the movement of the flapping wing through the actuator, and optimizing the flapping wing energy harvesting performance;

[0052] The deep neural network model includes:

[0053] The input layer is used to receive numerical vectors;

[0054] Convolutional layer, used to extract local features from numerical vectors;

[0055] The conversion layer is used to rearrange the output of the convolutional layer according to the time dimension and feature dimension;

[0056] LSTM layer, used to extract time series features from the output of the conversion layer;

[0057] The fully connected layer is used to extract high-level features from the output of the LSTM layer;

[0058] The output layer is used to classify or regress the output of the fully connected layer and output the predicted value of the energy harvesting performance.

[0059] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned control method for flapping-wing energy harvesting performance when executing the program.

[0060] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the control method of flapping-wing energy harvesting performance is implemented.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] The present invention first measures the flow characteristic parameters of the incoming flow and the motion parameters of the flapping wings in real time, and inputs these parameters into the deep neural network model, overcoming the limitations of the traditional method, without the need to establish a complex mathematical model, and can predict and control the energy collection performance of the flapping wings in real time according to the changes in the flow field and the movement of the flapping wings, thereby improving the energy conversion rate and efficiency of the flapping wings. The present invention can monitor and adjust the movement and energy collection performance of the flapping wings in real time without artificially setting the control parameters, thereby improving the efficiency and stability of the control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0064] Figure 1 A control system diagram of a flapping wing energy harvesting performance control method of the present invention;

[0065] Figure 2 This is a framework diagram of the deep neural network model of the present invention. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] A control method for flapping wing energy harvesting performance, referring to Figure 1 , specifically a real-time control method for flapping wing energy harvesting performance under time-varying incoming flow conditions, comprising the following steps:

[0068] Step 1: Real-time measurement of the flow characteristic parameters of the incoming flow and the motion parameters of the flapping wings.

[0069] The flow characteristic parameters of the time-varying incoming flow conditions and the motion parameters of the flapping wings are measured by multiple sensor components. The multiple sensor components include at least two sets of sensor components, wherein the first sensor is used to measure the flow characteristics of the time-varying incoming flow conditions, including the speed, direction, turbulence, etc. of the incoming flow, and send these data to the controller in the form of a digital signal. The second sensor is used to measure the motion parameters of the flapping wings, such as the position, speed, acceleration, angle, angular velocity, angular acceleration, etc. of the flapping wings.

[0070] Step 2: Convert the flow characteristic parameters and motion parameters into numerical vectors and input them into the neural network model to generate predicted values ​​of energy harvesting performance.

[0071] The controller converts the digital signal of the sensor into a numerical vector as the input of the deep neural network model. The deep neural network model is used for data analysis and prediction. Specifically, it extracts features from the sensor data, then inputs them into the neural network, performs a series of nonlinear transformations, extracts the features and rules of the data, and finally outputs the predicted value of the flapping wing's energy harvesting performance and sends it to the controller.

[0072] like Figure 2 As shown, the deep neural network model of the present invention includes an input layer, a convolutional layer, a conversion layer, an LSTM layer, a fully connected layer and an output layer.

[0073] The input layer is used to receive the numerical vector from the data conversion module as the input of the deep neural network model. It consists of N neurons. The number of neurons N in the input layer is the same as the dimension of the sensor data, and each neuron corresponds to a data feature. The input layer receives external data x and passes it to the next layer. The output of the input layer can be expressed as a (0) = x, where a (0) is an N×1 vector.

[0074] The convolutional layer has 64 neurons, corresponding to the number of local features. The convolutional layer uses the convolution kernel W (1) Extract local features of the input data, such as edges, textures, etc. The activation function of the convolutional layer can be a ReLU function or other nonlinear function, expressed as The output of the convolutional layer is represented as where a (1) is a 64×1 vector, b (1) is a 64×1 bias vector, and * represents the convolution operation.

[0075] The conversion layer is used to rearrange the data in terms of time and feature dimensions, so that the output of the convolutional layer can be considered as a sequence, each element of which is a feature vector, and then input into the LSTM layer. The output of the conversion layer can be represented as a (1.5) = reshape(a (1) ,T,F), where a (1.5) It is a T×F matrix, where T is the number of time steps and F is the feature dimension. The number of time steps and feature dimension are selected based on the actual type and quantity of the data.

[0076] The LSTM layer has 64 neurons, corresponding to the number of time series features. The LSTM layer uses the Long Short-Term Memory (LSTM) unit to extract time series features from the output of the convolutional layer, such as trends and cycles. It is a special recurrent neural network (RNN) that uses time series to analyze inputs and solve long-term dependency problems and gradient vanishing problems. The activation function of the LSTM layer can be a tanh function or other nonlinear function, expressed as g (2) (·). The output of the LSTM layer can be expressed as a (2) =g (2) (W (2) a (1) +b (2) ), where a (2) is a 64×1 vector, W (2) is a 64×64 weight matrix, b (2) is a 64×1 bias vector. The LSTM layer consists of multiple LSTM units, each of which has a cell state and a hidden state, which are used to store and transmit long-term and short-term information. The LSTM unit has three gates, namely the forget gate, input gate, and output gate, which are used to control the update and output of the cell state. This is taken as an example of two layers of stacked LSTM units. The specific calculation formula is as follows:

[0077] The calculation formula of the LSTM unit in the first layer is:

[0078] f 1t =σ(W f1 x t +U f1 h 1t-1 +b f1 )

[0079] i 1t =σ(W i1 x t +U i1 h 1t-1 +b i1 )

[0080] o 1t =σ(W o1 x t +U o1 h 1t-1 +b o1 )

[0081]

[0082]

[0083] h 1t =o 1t⊙tanh(c 1t )

[0084] Among them, f 1t : The forget gate of the t-th time step of the first layer, i 1t : The input gate of the tth time step of the first layer, o 1t : The output gate of the first layer at the tth time step, by passing the input data x t , the hidden state h and bias vector b of the previous time step are linearly combined through the weight matrices W and U, and then calculated by nonlinear transformation through the sigmoid function. The candidate cell state at the t-th time step of the first layer is calculated by nonlinear transformation using the tanh function. 1t : The cell state of the tth time step of the first layer, calculated by adding the cell state c of the previous time step 1t-1 and the candidate cell states at the current time step Through the forget gate 1t and input gate i 1t Perform weighted combination. 1t : The hidden state of the first layer at the tth time step, which is used to output to the next layer or subsequent layers. The calculation method is to convert the cell state c at the current time step 1t Through the output gate o 1t It is obtained by weighted transformation with tanh function.

[0085] The calculation formula of the LSTM unit in the second layer is:

[0086] f 2t =σ(W f2 h 1t +U f2 h 2t-1 +b f2 )

[0087] i 2t =σ(W i2 h 1t +U i2 h 2t-1 +b i2 )

[0088] o 2t =σ(W o2 h 1t +U o2 h 2t-1 +b o2 )

[0089]

[0090]

[0091] h 2t =o 2t ⊙tanh(c 2t )

[0092] Among them, f 2t : The forget gate of the t-th time step of the second layer, i 2t : The input gate of the t-th time step of the second layer, o 2t : The output gate of the second layer at the tth time step, by 1t , the hidden state h and bias vector b of the previous time step are linearly combined through the weight matrices W and U, and then calculated by nonlinear transformation through the sigmoid function. The candidate cell state at the t-th time step of the second layer is calculated by nonlinear transformation using the tanh function. 2t : The cell state of the tth time step of the second layer, calculated by adding the cell state c of the previous time step 2t-1 and the candidate cell states at the current time step Through the forget gate 2t and input gate i 2t Perform weighted combination. 2t : The hidden state of the t-th time step of the second layer, which is used to output to the subsequent layers. The calculation method is to convert the cell state c of the current time step 2t Through the output gate o 2t The output of the entire LSTM layer can be expressed as a (2) =h 2t =g (2) (W (2) a (1) +b (2) ).

[0093] The fully connected layer has 64 neurons, corresponding to the number of high-level features. The fully connected layer uses grouped convolution to extract high-level features from the output of the LSTM layer. The activation function of the fully connected layer can be a sigmoid function or other nonlinear function g (3) (·). The output of the fully connected layer can be expressed as a (3) =g (3) (W (3) *a (2) +b (3) ), where a (3) is a 64×1 vector, W (3) is a 64×64 group convolution kernel, b (3) is a 64×1 bias vector, and * indicates grouped convolution operation.

[0094] The output layer consists of 4 neurons, which are the output of the deep neural network model, corresponding to the number of categories of the task, that is, the predicted value of the energy harvesting performance of the flapping wing. The output layer uses the output of the fully connected layer for final classification or regression, such as prediction, decision-making, etc. The activation function of the output layer can be a softmax function or other nonlinear function g (4) (·). The output of the output layer can be expressed as a (4) =g (4) (W (4) a (3) +b (4) ), where a (4) is a 4×1 vector, W (4) is a 4×64 weight matrix, b (4) is a 4×1 bias vector.

[0095] The training process of a neural network refers to using samples from a data set, inputting them into the neural network, performing a forward propagation and a back propagation, updating the parameters of the neural network, and repeating this process until the loss function of the neural network converges or the maximum number of iterations is reached. The training process of a neural network can be specifically divided into the following steps:

[0096] (1) Initialize the parameters of the neural network, such as the weight matrix W (l) and the bias vector b (l) , where l represents the lth layer, which is initialized using random numbers or other methods.

[0097] (2) Select a batch of samples, such as x (i) and (i) , where i represents the i-th sample, x (i) represents the input feature, y (i) Represents the output labels, selected using random sampling or other methods.

[0098] (3) Perform forward propagation, that is, from the input layer to the output layer, calculate the output of each layer in turn, such as a (l) =g (l) (W (l) a (l-1) +b (l) ), where g (l) (·) represents the activation function of the lth layer, a (l) represents the output of layer l, a (0) =x (i) , Where L represents the last layer, Represents the predicted value.

[0099] (4) Calculate the loss function, that is, the difference between the predicted value and the true value. The loss function is defined as: Where K represents the number of categories, and They respectively represent the true value and predicted value of the kth category of the ith sample. The cross entropy loss function is used here, and other loss functions can also be used.

[0100] (5) Perform back propagation, that is, from the output layer to the input layer, calculate the gradient of each layer in turn, which is defined as follows: where z (l) =W (l) a (l-1) +b (l) , represents the error of the lth layer, represents the derivative of the activation function of the lth layer,

[0101] (6) Update parameters, that is, update the weight matrix and bias vector of each layer according to the gradient descent algorithm, such as Where α represents the learning rate, and other optimization algorithms can also be used, such as the Adam optimizer.

[0102] (7) Repeat steps (2) to (6) until the loss function converges or the maximum number of iterations is reached, completing the training process of the neural network.

[0103] Step 3: Adjust the control parameters of the flapping wings according to the predicted value of the energy harvesting performance and the actual value of the energy harvesting performance.

[0104] The flapping-wing energy harvesting device is used to utilize the movement of flapping wings to collect energy from the flow field and convert the energy into electrical energy or other forms of energy.

[0105] The controller adjusts the control parameters of the flapping wings according to the predicted value of the energy harvesting performance and the actual value of the energy harvesting performance.

[0106] The present invention can monitor and adjust the motion and energy collection performance of the flapping wing in real time without manually setting the control parameters, thereby improving the efficiency and stability of the control. When the flow field conditions change, the sensor can sense the changes in the motion parameters of the flapping wing and the characteristics of the flow field, and transmit the data to the controller. The controller calculates the error and gradient based on the predicted value and target value of the deep neural network model, and uses the gradient descent algorithm to adjust the control parameters of the flapping wing. The adjusted control parameters are transmitted to the flapping wing energy collection device, and the effective airfoil, phase, angle of attack and motion of the flapping wing are changed, so that the energy collection performance of the flapping wing adapts to the new flow field conditions, completing an adaptive control cycle.

[0107] Step 4: Input the adjusted control parameters into the actuator, control the movement of the flapping wings through the actuator, and optimize the energy harvesting performance.

[0108] The actuator includes a motor, a servo motor and a solenoid valve, which changes the shape and movement of the flapping wing according to the output signal of the controller.

[0109] The motor uses a brushless DC motor, and controls the wingspan and wing sweep of the flapping wing, that is, the horizontal and vertical movement of the flapping wing, by adjusting the speed and direction of the motor. The input signal of the motor is the first two components of the output signal of the controller, which respectively represent the expected values ​​of the wingspan and wing sweep of the flapping wing. The output signal of the motor is the actual value of the wingspan and wing sweep of the flapping wing, which is measured by the encoder and gyroscope respectively. The working principle of the motor is that when the difference between the input signal and the output signal is greater than a threshold, the motor starts and adjusts the speed and direction so that the wingspan and wing sweep of the flapping wing gradually approach the expected value. When the difference between the input signal and the output signal is less than a threshold, the motor stops or keeps running at a constant speed so that the wingspan and wing sweep of the flapping wing remain near the expected value. The performance indicators of the motor are response time of 0.1 second, accuracy of 0.01 degree, and stability of 0.001 degree / second.

[0110] The servo motor uses a digital servo motor to control the phase and angle of attack of the flapping wing, that is, the angle and position of the flapping wing relative to the incoming flow, by adjusting the angle and speed of the servo motor. The input signal of the servo motor is the last two components of the output signal of the controller, which respectively represent the expected values ​​of the phase and angle of attack of the flapping wing. The output signal of the servo motor is the actual value of the phase and angle of attack of the flapping wing, which is measured by a rotary encoder respectively. The working principle of the servo motor is that when the difference between the input signal and the output signal is greater than a threshold, the servo motor starts and adjusts the angle and speed so that the phase and angle of attack of the flapping wing gradually approach the expected value. When the difference between the input signal and the output signal is less than a threshold, the servo motor stops or maintains constant speed operation so that the phase and angle of attack of the flapping wing remain near the expected value. The performance indicators of the servo motor are response time of 0.05 seconds, accuracy of 0.005 degrees, and stability of 0.0005 degrees / second.

[0111] The solenoid valve controls the effective airfoil of the flapping wing, that is, the curvature and thickness of the flapping wing, by adjusting the switch and voltage. The input signal of the solenoid valve is the last two components of the output signal of the controller, which respectively represent the expected values ​​of the curvature and thickness of the flapping wing. The output signal of the solenoid valve is the actual value of the curvature and thickness of the flapping wing, which is measured by the pressure sensor and the distance sensor respectively. The working principle of the solenoid valve is that when the difference between the input signal and the output signal is greater than a threshold, the solenoid valve opens and adjusts the voltage so that the curvature and thickness of the flapping wing gradually approach the expected value. When the difference between the input signal and the output signal is less than a threshold, the solenoid valve closes or maintains constant pressure operation so that the curvature and thickness of the flapping wing remain near the expected value. The performance indicators of the solenoid valve are response time of 0.01 seconds, accuracy of 0.001 mm, and stability of 0.0001 mm / s.

[0112] The present invention has the feature of self-learning, and can make an online estimate of one or some unknown parameters of the deep neural network model, and then substitute the latest estimated values ​​of these parameters into and update the gain of the designed controller, so that the output of the model is closer to the true value, and the accuracy and adaptability of the model are improved. Specifically, the logic of the self-learning control implemented by the present invention is: when the energy collection performance of the flapping wing reaches the optimal or near-optimal, the sensor can transmit the motion parameters of the flapping wing and the characteristics of the flow field, as well as the optimal value of the energy collection performance of the flapping wing, as the training data of the deep neural network model to the controller, and the controller calculates the error and gradient according to the predicted value and the true value of the deep neural network model, and uses the back propagation algorithm to update and optimize the parameters of the deep neural network model, so as to improve the accuracy and adaptability of the model and make the output of the model closer to the true value. The controller transmits a control signal to the flapping wing energy collection device based on the updated model parameters, improves the energy collection performance of the flapping wing, and completes a self-learning control cycle.

[0113] Based on the same concept, the present invention also provides a control system for flapping-wing energy collection performance, including a collection module, an input module, an adjustment module and a control module.

[0114] The acquisition module is used to measure the flow characteristic parameters of the incoming flow and the motion parameters of the flapping wings in real time.

[0115] The input module is used to convert flow characteristic parameters and motion parameters into numerical vectors and input them into the neural network model to generate predicted values ​​of energy harvesting performance.

[0116] The adjustment module is used to adjust the control parameters of the flapping wings according to the predicted value of the energy harvesting performance and the actual value of the energy harvesting performance.

[0117] The control module is used to input the adjusted control parameters into the actuator, control the movement of the flapping wings through the actuator, and optimize the flapping wing energy harvesting performance.

[0118] Furthermore, the deep neural network model includes an input layer, a convolutional layer, a transformation layer, an LSTM layer, a fully connected layer, and an output layer.

[0119] The input layer is used to receive numeric vectors.

[0120] The convolutional layer is used to extract local features from numerical vectors.

[0121] The conversion layer is used to rearrange the output of the convolutional layer in terms of time and feature dimensions.

[0122] The LSTM layer is used to extract time series features from the output of the transformation layer.

[0123] The fully connected layer is used to extract high-level features from the output of the LSTM layer.

[0124] The output layer is used to classify or regress the output of the fully connected layer and output the predicted value of the energy harvesting performance.

[0125] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned control method for flapping-wing energy harvesting performance when executing the program.

[0126] The present invention also provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the control method for flapping-wing energy harvesting performance is implemented.

[0127] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0128] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for controlling flapping wing energy harvesting performance, characterized in that: The following steps are involved: Real-time measurement of the flow characteristic parameters of the incoming flow and the motion parameters of the flapping wing; the flow characteristic parameters include the time-varying speed, direction and turbulence, and the motion parameters include the position, speed, acceleration, angle, angular velocity and angular acceleration of the flapping wing; The flow characteristic parameters and motion parameters are converted into numerical vectors and input into the neural network model to generate the predicted value of energy harvesting performance; Adjusting the control parameters of the flapping wings according to the predicted value of the energy harvesting performance and the actual value of the energy harvesting performance; The adjusted control parameters are input into the actuator, and the movement of the flapping wing is controlled by the actuator to optimize the flapping wing energy harvesting performance; The deep neural network model includes: The input layer is used to receive numerical vectors; Convolutional layer, used to extract local features from numerical vectors; The conversion layer is used to rearrange the output of the convolutional layer according to the time dimension and feature dimension; LSTM layer, used to extract time series features from the output of the conversion layer; The fully connected layer is used to extract high-level features from the output of the LSTM layer; The output layer is used to classify or regress the output of the fully connected layer and output the predicted value of energy harvesting performance; The actuator comprises a motor, a servo motor and a solenoid valve; the motor is used to control the wingspan and wing sweep of the flapping wing, the servo motor is used to control the phase and angle of attack of the flapping wing, and the solenoid valve is used to control the curvature and thickness of the flapping wing.

2. A method for controlling flapping wing energy harvesting performance as claimed in claim 1, characterized in that: The flow characteristic parameters of the incoming flow are measured by the first sensor component, the motion parameters of the flapping wings are measured by the second sensor component, and the real value of the energy collection performance is output by the flapping wing energy collection device.

3. A method for controlling flapping wing energy harvesting performance as claimed in claim 1, characterized in that: The input layer consists of multiple neurons, the number of neurons is the same as the dimension of the numerical vector, and the output is a (0) = x, where a (0) is an N×1 vector; The convolution layer includes 64 neurons, and its activation function is a nonlinear function, which is convolved by the following formula: a (1) =g (1) (W (1) *a (0) +b (1) ) In the formula, a (1) is a 64×1 vector, b (1) is a 64×1 bias vector, W (1) is the convolution kernel, * indicates the convolution operation, g (1) (·) is the activation function; The output of the conversion layer is as follows: a (1.5) =reshape(a (1) ,T,F) In the formula, a (1.5) It is a T×F matrix, where T is the number of time steps and F is the feature dimension; The LSTM layer includes 64 neurons, which are composed of multiple LSTM units. Each LSTM unit has a cell state and a hidden state for storing and transmitting long-term and short-term information. Each LSTM unit has three gates, namely a forget gate, an input gate, and an output gate, which are used to control the update and output of the cell state. The output of the LSTM layer is as follows: a (2) =g (2) (W (2) a (1) +b (2) ) a (2) is a 64×1 vector, W (2) is a 64×64 weight matrix, b (2) is a 64×1 bias vector, g (2) (·) is the activation function; The fully connected layer includes 64 neurons, and its output is as follows: a (3) =g (3) (W (3) *a (2) +b (3) ) In the formula, a (3) is a 64×1 vector, W (3) is a 64×64 group convolution kernel, b (3) is a 64×1 bias vector, * indicates group convolution operation, g (3) (·) is the activation function; The output layer includes 4 neurons, and its output is as follows: a (4) =g (4) (W (4) a (3) +b (4) ) In the formula, a (4) is a 4×1 vector, W (4) is a 4×64 weight matrix, b (4) is a 4×1 bias vector, g (4) (·) is the activation function.

4. A method for controlling flapping wing energy harvesting performance as claimed in claim 1, characterized in that: Before converting the flow characteristic parameters and motion parameters into numerical vectors and inputting them into the neural network model, the neural network model needs to be trained, including the following steps: Initialize the parameters of the neural network model; Get a data set, including multiple input features and corresponding multiple output labels; Input the input features to the initialized neural network model for forward propagation to obtain the predicted value; Calculate the loss function of the neural network model based on the predicted value and the corresponding output label; Input the input features to the initialized neural network model for back propagation, and calculate the gradient and error of each layer; Update the weight matrix and bias vector of each layer according to the gradient descent algorithm; Repeat the above forward propagation and back propagation process until the loss function converges or the maximum number of iterations is reached, completing the training of the neural network model.

5. A method for controlling flapping wing energy harvesting performance as claimed in claim 1, characterized in that: According to the predicted value of the energy harvesting performance and the actual value of the energy harvesting performance, the control parameters of the flapping wing are adjusted, including the following steps: calculating an error and a gradient between a predicted value of the energy harvesting performance and an actual value of the energy harvesting performance; The control parameters of flapping wings are adjusted by gradient descent algorithm.

6. A control system for flapping wing energy harvesting performance, characterized in that: The following steps are involved: An acquisition module is used to measure the flow characteristic parameters of the incoming flow and the motion parameters of the flapping wings in real time; the flow characteristic parameters include time-varying speed, direction and turbulence, and the motion parameters include the position, speed, acceleration, angle, angular velocity and angular acceleration of the flapping wings; An input module, used for converting flow characteristic parameters and motion parameters into numerical vectors and inputting the numerical vectors into a neural network model to generate a predicted value of energy harvesting performance; An adjustment module, used for adjusting the control parameters of the flapping wings according to the predicted value of the energy harvesting performance and the actual value of the energy harvesting performance; A control module, used for inputting the adjusted control parameters into the actuator, controlling the movement of the flapping wing through the actuator, and optimizing the flapping wing energy harvesting performance; The deep neural network model includes: The input layer is used to receive numerical vectors; Convolutional layer, used to extract local features from numerical vectors; The conversion layer is used to rearrange the output of the convolutional layer according to the time dimension and feature dimension; LSTM layer, used to extract time series features from the output of the conversion layer; The fully connected layer is used to extract high-level features from the output of the LSTM layer; The output layer is used to classify or regress the output of the fully connected layer and output the predicted value of energy harvesting performance; The actuator comprises a motor, a servo motor and a solenoid valve; the motor is used to control the wingspan and wing sweep of the flapping wing, the servo motor is used to control the phase and angle of attack of the flapping wing, and the solenoid valve is used to control the curvature and thickness of the flapping wing.

7. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for controlling flapping-wing energy harvesting performance as described in any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the control method for flapping wing energy harvesting performance described in any one of claims 1 to 5 is implemented.

Images