Piezoelectric actuator temperature rise prediction method based on Bayesian optimization CNN-LSTM

Abstract

According to the Bayesian optimization CNN-LSTM-based piezoelectric actuator temperature rise prediction method provided by the invention, the prediction precision is improved by combining a convolutional neural network and a long-short-term memory network and through a prediction result of a Bayesian optimization model; according to the method, the convolutional neural network part can effectively extract local features of piezoelectric actuator data, the long-term and short-term memory network part is good at capturing the long-term dependency relationship of time sequence data, the design of the LSTM network structure enhances the anti-noise capability of the model, the parameters of the CNN-LSTM model are automatically adjusted and optimized through Bayesian optimization, and the accuracy of the model is improved. In the Bayesian optimization, an objective function is approached by constructing an agent model, and a search process is guided by using an acquisition function, so that the evaluation frequency of the objective function is reduced, the calculation efficiency is improved, and the prediction effect of the temperature rise phenomenon of the piezoelectric actuator is further improved. The method comprises the steps of temperature rise data collection, training data construction, CNN data input, LSTM data input and Bayesian optimization, and finally a final prediction result is obtained.

Description

Technical Field

[0001] The invention relates to a piezoelectric actuator temperature rise prediction method based on Bayesian optimization CNN-LSTM, and specifically belongs to the technical field of electronic appliances. Background Art

[0002] With the rapid development of industries and manufacturing, optical instruments, medical industry, aerospace, military field, consumer electronics and other fields, piezoelectric actuators, devices that can convert electrical energy into mechanical energy, have emerged. Due to the advantages of high precision, flexible travel, no external electromagnetic interference and scalable structure, actuators using piezoelectric drive technology can accurately convert input electrical energy into displacement, force or motion output through electromechanical coupling;

[0003] Existing piezoelectric actuators have good linearity when working at the same temperature. When working in an environment with changing temperature, due to changes in material properties, there will be a phenomenon of sensitivity temperature rise, and its performance parameters (such as displacement, force, voltage, etc.) will change, which will affect the detection accuracy. In order to improve the output performance of piezoelectric actuators and enhance the electromechanical conversion capability of piezoelectric actuators, a piezoelectric actuator temperature rise prediction method based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM) is proposed for piezoelectric electromechanical conversion systems, and the prediction results are optimized by Bayesian, in the hope that this prediction method can improve system stability. Summary of the invention

[0004] The purpose of the present invention is to provide a piezoelectric actuator temperature rise prediction method based on Bayesian optimization CNN-LSTM to solve the adverse effects of temperature rise in the piezoelectric actuator system on system accuracy and stability. A piezoelectric actuator temperature rise prediction model based on CNN-LSTM is designed to predict the temperature rise of the piezoelectric actuator online and optimize the prediction results, so that the piezoelectric actuator can obtain higher stability by predicting the temperature rise.

[0005] In order to solve the above technical problems, the technical solution steps adopted by the present invention specifically include:

[0006] Step 1, collect temperature rise data;

[0007] Step 2: Build training data;

[0008] Step 3: CNN input data;

[0009] Step 4: LSTM input data;

[0010] Step 5: Bayesian optimization;

[0011] Step 6: Get the final prediction result;

[0012] Furthermore, the parameters of the CNN-LSTM model are automatically tuned through Bayesian optimization. Bayesian optimization approximates the objective function by constructing a proxy model and uses the acquisition function to guide the search process, reducing the number of evaluations of the objective function, thereby improving computational efficiency and further improving the prediction effect of the temperature rise phenomenon of the piezoelectric actuator.

[0013] The specific operations of collecting temperature rise data in step 1 include: selecting a thermocouple as a temperature sensor according to the operating temperature range and accuracy requirements of the piezoelectric actuator; amplifying and filtering the electrical signal through a transmitter; and transmitting the data to a PLC or monitoring system wirelessly for further processing, recording and display.

[0014] Furthermore, thermocouples can detect temperature changes and convert them into electrical signals. Since the thermocouple signal is weak, it needs to be amplified and filtered by a transmitter to eliminate noise and ensure the accuracy of the signal. The thermoelectric potential current signal generated by the thermocouple is then converted into a standard voltage signal by the transmitter.

[0015] The temperature rise drift data processed in step 1 are merged into a data matrix in chronological order, with the first 67% of the data as the training set and the last 33% of the data as the test set.

[0016] Furthermore, the training set in step 2 is used as CNN input data, and the CNN method is used to extract and fuse the data features of the temperature rise data. The CNN process is as follows:

[0017] Y(i,j)=δ([w]*[X i,j ]+b) (1),

[0018] In formula (1), Y(i, j) is the feature data output by CNN, δ is the tanh activation function, * is the convolution symbol, [w] is the two-dimensional convolution kernel, [Xi, j] is the deep displacement of the landslide and the groundwater level in the convolution area, m is the number of rows where the data is located, n is the number of columns where the data is located, l is the convolution kernel moving step, and b is the bias parameter.

[0019] Furthermore, the feature data output by CNN in step 3 is used as LSTM input data, and the output data is the prediction result of the training set. The loss function is calculated based on the prediction result, and the parameters of CNN and LSTM are optimized. The optimal parameters of the CNN-LSTM method are determined through an iterative cycle. The LSTM process is as follows:

[0020] f t =σ(W ih ·h t-1 +W fx ·x t +b f ) (2),

[0021] i t =σ(W ih ·h t-1 +W ix ·x t +b i ) (3),

[0022] O t =σ(W ih ·h t-1 +W Ox ·x t +b O ) (4),

[0023] h t =O t tanh(C t ) (5),

[0024] In formulas (2) to (5), f is the forget gate, i is the input gate, O is the output gate, σ is the Sigmoid function, and h is (t-1) is the output at time t-1, x t is the deep displacement characteristic data of the landslide input at time t, W is the weight of each data, b is each bias term, C t is the cell state at time t, C (t-1) is the cell state at time t-1, h t is the output at time t.

[0025] Furthermore, in step 5, the Bayesian theorem is used to guide the search to find the minimum or maximum value of the objective function, that is, the previous search point is used to determine the next search point, and the Gaussian process is used to establish the prior and posterior models of the objective function; the Bayesian optimization process is as follows:

[0026] Step 501: define an objective function, where the objective function F(x) is a function to be maximized or minimized, where x is a parameter vector;

[0027] Step 502: Select a proxy model and use Gaussian Process (GP) as the proxy model to simulate the objective function. The Gaussian process is defined as:

[0028] F(x)~GP(m(x),k(x,x'))(6),

[0029] In formula (6), m(x) is the mean function (usually assumed to be 0), and k(x, x') is the kernel function;

[0030] Step 503: Select an acquisition function;

[0031] Step 504: Initialization, select the initial point X initEvaluate and use these points to train a Gaussian process model;

[0032] Step 505, iterative optimization, in each iteration, use the acquisition function to select the next evaluation point x next

[0033] x next = argmax x α(x) (7),

[0034] Step 506: Evaluate the objective function f(x next ) at this point, update the Gaussian process model, including the mean and variance:

[0035] μ(x new )=k(x new ,X)K -1 yσ 2 (x new )=k(x new ,x new )-k(x new ,X)K -1 k(X,x new ) (8),

[0037] In formula (8), K is the kernel matrix, k(x new ,X) is the kernel function value vector between the new point and all training points;

[0038] Step 507, termination condition, repeating the iteration steps until the termination condition is met, such as reaching the maximum number of iterations or the improvement is less than a certain threshold;

[0039] Step 508: Return the best parameters. Return the most optimized parameters x* according to the proxy model.

[0040] x * = arg max x∈X μ(x) (9),

[0041] In formulas (7) and (9), argmax x It is a function that finds parameters for a function.

[0042] The beneficial effects of the present invention are:

[0043] 1. By combining the convolutional neural network (CNN) and the long short-term memory network (LSTM), and then using the prediction results of the Bayesian optimization model, the prediction accuracy is improved; the convolutional neural network (CNN) part can effectively extract the local features of the piezoelectric actuator data, while the long short-term memory network (LSTM) part is good at capturing the long-term dependencies of time series data. The design of the LSTM network structure enhances the model's anti-noise ability. The parameters of the CNN-LSTM model are automatically tuned through Bayesian optimization. Bayesian optimization approximates the objective function by constructing a proxy model and uses the acquisition function to guide the search process, reducing the number of evaluations of the objective function, thereby improving the calculation efficiency and further improving the prediction effect of the temperature rise phenomenon of the piezoelectric actuator. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the Bayesian optimization process of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the attached Figure 1 , clearly and completely describe the technical solutions in the embodiments. Figure 1 As shown in the figure, firstly, a temperature rise prediction model of piezoelectric actuator based on CNN-LSTM is designed. The temperature rise prediction model of piezoelectric actuator needs to collect temperature rise data first. According to the working temperature range and accuracy requirements of piezoelectric actuator, thermocouple is selected as temperature sensor. Thermocouple can detect temperature change and convert it into electrical signal. Since the thermocouple signal is weak, it needs to be amplified and filtered by transmitter to eliminate noise and ensure the accuracy of signal. The transmitter converts the thermoelectric potential current signal generated by thermocouple into standard voltage signal, and transmits the data to PLC or monitoring system by wireless means (WiFi, Bluetooth, etc.) for further processing, recording and display, so as to obtain randomly generated initialization points.

[0046] Then the recorded and displayed data are processed to obtain temperature rise drift data, the processed temperature rise drift data are merged into a data matrix in time sequence, the first 67% of the data are used as a training set, and the last 33% of the data are used as a test set to construct training data, thereby constructing training data;

[0047] The constructed training data is used as CNN input data, and the CNN method is used to extract and fuse the data features of the temperature rise data. The CNN process is as follows:

[0048] Y(i,j)=δ([w]*[X i,j ]+b) (1),

[0049] In formula (1), Y(i, j) is the feature data output by CNN, δ is the tanh activation function, * is the convolution symbol, [w] is the two-dimensional convolution kernel, [Xi, j] is the deep displacement of the landslide and the groundwater level in the convolution area, m is the number of rows where the data is located, n is the number of columns where the data is located, l is the convolution kernel moving step, and b is the bias parameter;

[0050] The feature data output by CNN is then used as the input data of LSTM. The output data is the prediction result of the training set. The loss function is calculated based on the prediction result. The parameters of CNN and LSTM are optimized. The optimal parameters of the CNN-LSTM method are determined through an iterative cycle. The LSTM process is as follows:

[0051] f t =σ(W ih ·h t-1 +W fx ·x t +b f ) (2),

[0052] i t =σ(W ih ·h t-1 +W ix ·x t +b i ) (3),

[0053] O t =σ(W ih ·h t-1 +W Ox ·x t +b O ) (4),

[0054] h t =O t tanh(C t ) (5),

[0055] In formulas (2) to (5), f is the forget gate, i is the input gate, O is the output gate, σ is the Sigmoid function, and h is (t-1) is the output at time t-1, x t is the deep displacement characteristic data of the landslide input at time t, W is the weight of each data, b is each bias term, C t is the cell state at time t, C (t-1) is the cell state at time t-1, h t is the output at time t;

[0056] Finally, Bayes’ theorem is used to guide the search to find the minimum or maximum value of the objective function, that is, the previous search point is used to determine the next search point, and the Gaussian process is used to establish the prior and posterior models of the objective function;

[0057] The Bayesian optimization process is as follows: define the objective function, where the objective function F(x) is the function to be maximized or minimized, and x is a parameter vector; select a proxy model, and the present invention uses a Gaussian process (GP) as a proxy model to simulate the objective function. The Gaussian process is defined as:

[0058] F(x)~GP(m(x),k(x,x')) (6),

[0059] In formula (6), m(x) is the mean function (usually assumed to be 0), k(x, x') is the kernel function, which is used to define the similarity in the input space; select the acquisition function, which is used to balance exploration and exploitation. Common acquisition functions include expected improvement (EI), probability of improvement (PI) and upper confidence bound (UCB); initialization, select the initial point X init Evaluate and use these points to train the Gaussian process model; iterate and optimize, and in each iteration, use the acquisition function to select the next evaluation point x next

[0060] x next = arg max x α(x) (7),

[0061] Evaluate the objective function f(x next ) at this point, update the Gaussian process model, including the mean and variance:

[0062] μ(x new )=k(x new ,X)K -1 yσ 2 (x new )=k(x new ,x new )-k(x new ,X)K -1 k(X,x new )

[0063] (8)

[0064] In formula (8), K is the kernel matrix, k(xnext , X) is the kernel function value vector between the new point and all training points; termination condition, repeat the iteration steps until the termination condition is met, such as reaching the maximum number of iterations or the improvement is less than a certain threshold; return the best parameter, return the optimized parameter x* according to the proxy model

[0065] x * = arg max x∈X μ(x) (9),

[0066] In formulas (7) and (9), argmax x It is a function that finds parameters for a function. The best parameters of Bayesian optimization are used as the final parameters of the model, and the test set is input into the CNN-LSTM method. The obtained prediction result is the final prediction result of CNN-LSTM.

[0067] The above is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent replacement and improvement made to the above embodiments without departing from the content of the technical solution of the present invention, based on the technical essence of the present invention, within the spirit and principles of the present invention, still fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for predicting temperature rise of piezoelectric actuator based on Bayesian optimization CNN-LSTM, characterized in that: Specifically include: Step 1, collect temperature rise data; Step 2: Build training data; Step 3: CNN input data; Step 4: LSTM input data; Step 5: Bayesian optimization; Step 6: Get the final prediction result.

2. The method for predicting temperature rise of a piezoelectric actuator based on Bayesian optimization CNN-LSTM according to claim 1, characterized in that: The specific operations for collecting temperature rise data in step 1 include: Step 101, according to the operating temperature range and accuracy requirements of the piezoelectric actuator; Step 102, selecting a thermocouple as a temperature sensor; Step 103, amplifying and filtering the electrical signal through the transmitter; Step 104: wirelessly transmit the data to a PLC or monitoring system for further processing, recording and display.

3. The method for predicting temperature rise of a piezoelectric actuator based on Bayesian optimization CNN-LSTM according to claim 1, characterized in that: The temperature rise drift data processed in step 1 are merged into a data matrix in chronological order, with the first 67% of the data as the training set and the last 33% of the data as the test set.

4. The method for predicting temperature rise of a piezoelectric actuator based on Bayesian optimization CNN-LSTM according to claim 1, characterized in that: The training set in step 2 is used as CNN input data. The CNN method is used to extract and fuse the data features of the temperature rise data. The CNN process is as follows: Yi,j=δ(w*X i,j +b)(1), In formula (1), Y(i, j) is the feature data output by CNN, δ is the tanh activation function, * is the convolution symbol, [w] is the two-dimensional convolution kernel, [Xi, j] is the deep displacement of the landslide and the groundwater level in the convolution area, m is the number of rows where the data is located, n is the number of columns where the data is located, l is the convolution kernel moving step, and b is the bias parameter.

5. The method for predicting temperature rise of a piezoelectric actuator based on Bayesian optimization CNN-LSTM according to claim 1, characterized in that: In step 3, the feature data output by CNN is used as the LSTM input data. The output data is the prediction result of the training set. The loss function is calculated based on the prediction result. The parameters of CNN and LSTM are optimized. The optimal parameters of the CNN-LSTM method are determined through an iterative cycle. The LSTM process is as follows: f t =σ(Q ih ·h t-1 +W fx ·x t +b f ) (2), i t =σ(Q ih ·h t-1 +W ix ·x t +b i ) (3), O t =σ(Q ih ·h t-1 +W Ox ·x t +b O ) (4), h t =O t ·tanh(C t ) (5), In formulas (2) to (5), f is the forget gate, i is the input gate, 0 is the output gate, σ is the Sigmoid function, and h is (t-1) is the output at time t-1, x t is the deep displacement characteristic data of the landslide input at time t, W is the weight of each data, b is each bias term, C t is the cell state at time t, C (t-1) is the cell state at time t-1, h t is the output at time t.

6. The method for predicting temperature rise of a piezoelectric actuator based on Bayesian optimization CNN-LSTM according to claim 1, characterized in that: In step 5, the Bayesian theorem is used to guide the search to find the minimum or maximum value of the objective function, that is, the previous search point is used to determine the next search point, and the Gaussian process is used to establish the prior and posterior models of the objective function.

7. The method for predicting temperature rise of a piezoelectric actuator based on Bayesian optimization CNN-LSTM according to claim 1, characterized in that: The Bayesian optimization process in step 5 is as follows: Step 501: define an objective function, where the objective function F(x) is a function to be maximized or minimized, where x is a parameter vector; Step 502: Select a proxy model and use Gaussian Process (GP) as the proxy model to simulate the objective function. The Gaussian process is defined as: F(x)~GP(m(x),k(x,x′))(6), In formula (6), m(x) is the mean function (usually assumed to be 0), and k(x, x') is the kernel function; Step 503: Select an acquisition function; Step 504: Initialization, select the initial point X init Evaluate and use these points to train a Gaussian process model; Step 505, iterative optimization, in each iteration, use the acquisition function to select the next evaluation point x next x next =argmax x α(x)(7), Step 506: Evaluate the objective function f(x next ) at this point, update the Gaussian process model, including the mean and variance: μx new =kx new ,XK -1 yσ 2 x new =kx new ,x new -k(x new ,X)K -1 k(X,x new ) (8), In formula (8), K is the kernel matrix, kx new ,X is the kernel function value vector between the new point and all training points; Step 507, termination condition, repeating the iteration steps until the termination condition is met, such as reaching the maximum number of iterations or the improvement is less than a certain threshold; Step 508: Return the best parameters. Return the most optimized parameters x* according to the proxy model. x * =arg max x∈X μ(x) (9), In formulas (7) and (9), arg max x It is a function that finds parameters for a function.

8. The method for predicting temperature rise of a piezoelectric actuator based on Bayesian optimization CNN-LSTM according to claim 1, characterized in that: The best parameters of Bayesian optimization in step 5 are used as the final parameters of the model, and the test set is input into the CNN-LSTM method. The obtained prediction results are the final prediction results of CNN-LSTM.

Images