A gyroscope temperature drift compensation method based on hybrid kernel support vector machine optimized by BAS-GSA
Through the BAS-GSA-optimized hybrid core support vector machine model, combined with linear core and radial basis function core, the problem of unpredictable nonlinear features in fiber gyroscope temperature drift compensation is solved, and high-precision temperature drift compensation and generalization are achieved Improvement of capabilities.
Patent Information
- Application Number
- CN202111330599.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-11
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2041-11-11
AI Technical Summary
The prior art has problems in fiber gyroscope temperature drift compensation that are difficult to accurately predict and compensate for nonlinear features. Hardware compensation increases cost and complexity, and the model accuracy and generalization capabilities of software compensation are insufficient, especially under small sample conditions.
Using a hybrid core support vector machine model based on BAS-GSA optimization, combining linear core and radial basis function core, the hybrid core supports vector machine model is established to improve the prediction accuracy and generalization ability of temperature drift. .
It realizes high-precision prediction and compensation of fiber gyroscope temperature drift, improves the learning ability and generalization ability of support vector machines, avoids the limitations of a single optimization algorithm, and can accurately predict and compensate at different temperature changes rates.
Smart Images

Figure CN114036843B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of inertial navigation devices, and specifically relates to a hybrid kernel support vector machine gyroscope temperature drift compensation method based on BAS-GSA (beard whisker search algorithm-gravitational search algorithm) optimization, which realizes high-precision temperature drift compensation of gyroscopes in view of the characteristics of large temperature drift and strong nonlinearity of optical fiber and micromechanical gyroscopes. Background Art
[0002] The strapdown inertial navigation system is an autonomous and comprehensive navigation system, and the gyro is a key component for achieving high-precision navigation and positioning. However, the measurement error caused by temperature drift severely limits the practical application of gyro in navigation and positioning. Since the characteristics of some components are greatly affected by the ambient temperature, the resulting gyro measurement error will vary with temperature, which has attracted a lot of attention from researchers on the precise compensation of temperature drift. Due to its good simplicity and adaptability, the compensation strategy based on the software model with temperature dependent variable as input and fiber optic gyro temperature drift as output has been widely used in practice. However, since temperature drift has strong nonlinear characteristics, it is necessary to first study how to accurately establish the prediction and compensation model.
[0003] Temperature drift is an important factor affecting the measurement accuracy of fiber optic gyroscopes. The research on compensation technology for temperature drift at home and abroad is mainly divided into two categories: hardware compensation and software compensation. Hardware compensation is mainly divided into improving the temperature characteristics of device materials and establishing a temperature control system that can keep the device environment temperature constant. Hardware compensation technology focuses on fundamentally solving the temperature drift problem, but it will lead to a series of problems such as increased design cost, increased device volume, and complex design and processing. Software compensation technology is to establish a predictive compensation model with temperature-related quantities as input and temperature drift as output. Common modeling methods include least squares, neural networks, support vector machines, etc. The least squares method has too high an order in the nonlinear temperature drift compensation model, and the calculation burden is heavy. The accuracy of neural network training and learning depends on the number of samples and the selection of initial structural parameters, and there are risks of overfitting and underfitting. Support vector machine (SVM) is a classifier based on structural risk, which still has strong robustness and learning accuracy under small sample conditions. Support vector machines have been widely used in classification and regression operations since the 1990s, but their computing performance depends on the selection of kernel functions and initial structural parameters. At present, there is no method to determine the kernel function of support vector machine. The kernel function of support vector machine is usually divided into two categories: local kernel and global kernel. The former has strong local search and learning ability, while the latter has good extrapolation and generalization ability. In order to combine the advantages of these two kernel functions, some studies have proposed support vector machine (HKSVM) based on hybrid kernel function to solve complex nonlinear regression and prediction problems. However, the hybrid kernel makes the structure of support vector machine more complex, and more hyperparameters need to be adjusted and selected. In recent years, with the rapid rise of heuristic algorithms such as particle swarm optimization (PSO), genetic algorithm (GA), gravitational algorithm (GSA), cuckoo algorithm (CSA) and beetle beard algorithm (BAS), the initial structural parameters of intelligent algorithms such as neural network and support vector machine have developed from manual trial and error to automatic optimization, and the accuracy and efficiency of parameter search have been greatly improved. However, heuristic algorithms have their own advantages and disadvantages, and there is a lack of methods with optimal search capabilities that can combine the advantages of multiple algorithms.
[0004] Improving the accuracy of gyro temperature drift compensation can also be achieved through hardware improvements, such as adding an auxiliary gyro temperature control device to keep the gyro temperature constant when the device is in use. This solution can fundamentally solve the temperature drift problem of gyro devices, but it makes the device structure extremely complex, while increasing the device size and cost, which is not conducive to promotion and application in practice. Summary of the invention
[0005] To solve the above problems, the present invention discloses a gyroscope temperature drift compensation method based on a hybrid kernel support vector machine optimized by BAS-GSA, and conducts temperature drift experimental verification on a gyroscope with certain parameters, and the obtained effect can meet the requirements of high-precision temperature drift compensation.
[0006] To achieve the above object, the technical solution of the present invention is as follows:
[0007] A hybrid kernel support vector machine gyroscope temperature drift compensation method based on BAS-GSA optimization includes the following steps:
[0008] Step 1), establish the mathematical model of gyro temperature drift:
[0009] Step 1.1) Establish the mathematical model of fiber optic gyroscope temperature drift:
[0010]
[0011] Where E is the output temperature drift, T is the measured temperature, △T is the temperature change, is the temperature gradient, S i ,S v ,S d is the correlation coefficient of the corresponding variables.
[0012] Step 1.2) Establish the gyro temperature drift transfer function as follows:
[0013]
[0014] Where y is the output temperature drift, and f() is the transfer function of the temperature drift model to be solved.
[0015] Step 2), establish the evaluation function based on mean square error as follows:
[0016]
[0017] Among them, fit represents the objective function value, y i and are the measured and predicted temperature drift values, respectively. n represents the number of samples.
[0018] Step 3), establish a support vector machine hybrid kernel model based on linear kernel and radial basis function kernel as follows:
[0019] κ H (x,x i )=(1-w)κ P (x,x i )+wκ R (x,x i ) (4)
[0020] κ P (x,x i )=(x T x i +1) d (5)
[0021] κR (x,x i )=exp[-||xx i || 2 / (2δ 2 )] (6)
[0022] Among them, κ H (x,x i ), κ P (x,x i ) and κ R (x,x i ) are the hybrid kernel, linear kernel and radial basis function kernel respectively. d and δ are the dimension and width of the linear kernel and radial basis function kernel respectively. w is the relationship coefficient between the linear kernel and the radial basis function kernel.
[0023] Step 4), the steps to establish the hybrid kernel support vector machine model are as follows:
[0024] Step 4.1), based on step 3), establish a 4-dimensional vector composed of parameters to be optimized based on the hybrid kernel support vector machine as follows:
[0025] x(t)=[C,d,δ,w] T (7)
[0026] Among them, C is the support vector machine penalty factor.
[0027] Step 4.2), establish the support vector machine model as follows:
[0028]
[0029] Among them, ω and b are the model parameters of the support vector machine to be trained and solved, ε is the insensitive coefficient, ξ and is the slack variable.
[0030] Step 4.3) Use the Lagrangian method to solve equation (8) and obtain the hybrid kernel support vector machine model as follows:
[0031]
[0032] in, and α i is the Lagrange multiplier, and b is the cutoff coefficient.
[0033] Step 5), establish the BAS optimization steps as follows:
[0034] Step 5.1) Initialize the 4-dimensional positions of N beetles in the search space and calculate their respective evaluation function values fit(x i ):
[0035] x i (t) = [Ci ,d i ,δ i ,w i ] T ,i=1,2,…,N (10)
[0036] Step 5.2) Calculate the positions of the beetle's left and right antennae in space as follows:
[0037]
[0038] Where i = 1, 2, ..., N, x il (t) and x ir (t) The positions of the left and right antennae of the beetle, n i is a randomly generated 4-dimensional unit vector, and d(t) is the length of the tentacle.
[0039] Step 5.3) Calculate the new position of the beetle:
[0040]
[0041] Where, i = 1, 2, ..., N, fit(x il ) and fit(x ir ) are the evaluation function values calculated by the positions of the left and right antennae of the beetle, μ(t) represents the moving step, and sgn is the sign function. At the same time, the evaluation function value corresponding to the new position of the beetle is calculated
[0042] Step 5.4) Update the global position of the beetle in the next step as follows:
[0043]
[0044] Where i=1,2,…,N.
[0045] Step 6), establish the GSA optimization steps as follows:
[0046] Step 6.1) Randomly initialize the 4-dimensional positions of N beetles in the search space as shown in equation (10).
[0047] Step 6.2) Calculate the evaluation function value and quality of N beetles in the search space:
[0048] m i (t) = [fit i (t)-fit worst (t)] / [fit best (t)-fit worst (t)] (14)
[0049]
[0050] Where i = 1, 2, ..., N, fit i (t) and M i (t) are the evaluation function value and quality of the ith beetle, fit best (t) and fit worst (t) are the best and worst evaluation function values, respectively, m i (t) is an intermediate variable.
[0051] Step 6.3) Calculate the component of the total gravitational force on the i-th beetle in the k-th dimension:
[0052]
[0053] Among them, i, j=1,2,…,N, k=1,2,…,4. λ j is a random coefficient between 0 and 1. represents the component of the gravitational force of beetle j on beetle i in the k-th dimension, which is calculated as follows:
[0054]
[0055] in, is a minimum constant value, R i,j (t) represents the Euclidean distance between beetle j and beetle i, and G(t) represents the gravitational constant.
[0056] Step 6.4) Calculate the beetle's next moving speed and position:
[0057]
[0058]
[0059]
[0060] Among them, i,j=1,2,…,N, k=1,2,…,4. is a random coefficient between 0 and 1. represents the acceleration of beetle i, and Indicates the beetle's next moving speed and position, and Indicates the beetle's current movement speed and position.
[0061] Step 7), the steps to establish the hybrid BAS-GSA optimization are as follows
[0062] Step 7.1) Set the threshold of the evaluation function value fit min and the maximum number of iterations T max . Set the starting position x of N beetlesi (t) and its corresponding evaluation function value fit(x i ), i = 1, 2, ..., N. Calculate the optimal evaluation function value fit best as follows:
[0063] fit best =min{fit(x1),fit(x2),…,fit(x N )} (twenty one)
[0064] Step 7.2) Use formula (3) to calculate the evaluation function value fit(x i ), i=1,2,…,N.
[0065] Step 7.3) Use formula (11) to calculate the positions of the left and right antennae of N beetles: li (t) and x ri (t); use formula (12) to calculate the next position of the beetle
[0066] Step 7.4) Use formula (15) to calculate the mass M of N beetles i (t); Use formula (16) to calculate the gravitational force on N particles Use formula (18) to calculate the next moving speed of N particles
[0067] Step 7.5) Use formula (22) to calculate the new positions of the N beetles.
[0068]
[0069] Among them, v i (t+1) is the vector composed of the next moving speed calculated by formula (18), is the next step position vector calculated by formula (12), and α is a coefficient between 0 and 1.
[0070] Step 7.6) Use formula (3) to calculate the evaluation function values corresponding to the new positions of the N beetles
[0071] Step 7.7) Use formula (13) to calculate the global new position x of the N beetles i (t+1).
[0072] Step 7.8) t←t+1. If t≤T max And fit best ≥f min , then return to step 7.1 to continue execution, otherwise end the loop and output the optimal solution x best =[C,d,δ,w] T.
[0073] Step 8), use the optimal solution x output from step 7) best =[C,d,δ,w] T Train and test a support vector machine model.
[0074] The beneficial effects of the present invention are:
[0075] 1. In view of the shortcomings that a single kernel function of a support vector machine cannot have strong extrapolation and interpolation capabilities at the same time, the hybrid kernel function based on a linear kernel and a radial basis function kernel described in the present invention can have strong extrapolation and interpolation capabilities at the same time, which can effectively improve the learning ability and generalization ability of the support vector machine.
[0076] 2. The conventional hyperparameter optimization algorithm GSA is prone to fall into the local minimum, and the local search ability of BAS is weak. The hybrid BAS-GSA search algorithm described in the present invention can avoid the shortcomings of a single optimization algorithm, and at the same time has the local search ability of GSA and the global search ability of BAS. It can realize the hyperparameter optimization of the hybrid kernel support vector machine and has a higher solution accuracy.
[0077] 3. Conventional support vector machine models have weak solution accuracy and generalization ability in solving complex and strongly nonlinear problems such as temperature compensation. The hybrid kernel support vector machine model based on BAS-GSA shown in the present invention has strong prediction accuracy and generalization ability, and can achieve high-precision prediction and compensation of fiber optic gyroscope temperature drift under different temperature change rates. BRIEF DESCRIPTION OF THE DRAWINGS
[0078] Figure 1 It is a principle flow chart of the method of the present invention.
[0079] Figure 2 It is a composition diagram of the experimental equipment of the method of the present invention.
[0080] Figure 3 It is a temperature drift experimental measurement curve (10°C / min) used to train the method of the present invention.
[0081] Figure 4 It is a temperature drift experimental measurement curve (5°C / min) used to test the method of the present invention.
[0082] Figure 5 It is a training error curve using the method of the present invention.
[0083] Figure 6 It is a test error curve using the method of the present invention. DETAILED DESCRIPTION
[0084] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.
[0085] like Figure 1 As shown, the present invention discloses a gyro temperature drift compensation method based on a hybrid kernel support vector machine optimized by BAS-GSA, comprising the following steps:
[0086] Step 1), the experimental composition is as follows Figure 2 As shown, it is mainly composed of a gyroscope, a thermistor, a temperature control box and a monitoring computer. Figure 3 It is the temperature drift curve with a temperature change rate of 10℃ / min, which is used to train the hybrid kernel support vector machine model; Figure 4 It is a temperature drift curve with a temperature change rate of 5℃ / min, which is used to test the trained hybrid kernel support vector machine model. First, the gyro temperature drift mathematical model is established:
[0087]
[0088] Where y is the output temperature drift, T is the measured temperature, △T is the temperature change, is the temperature gradient, and f() is the temperature drift model transfer function.
[0089] Step 2), establish the evaluation function based on mean square error as follows:
[0090]
[0091] Among them, fit represents the objective function value, y i and are the measured and predicted gyro temperature drift values, respectively. n represents the number of samples, and its value is 240.
[0092] Step 3), establish a support vector machine hybrid kernel model based on linear kernel and radial basis function kernel as follows:
[0093] κ H (x,x i )=(1-w)κ P (x,x i )+wκ R (x,x i ) (3)
[0094] κ P (x,x i )=(x T x i +1) d (4)
[0095] κ R (x,xi )=exp[-||xx i || 2 / (2δ 2 )] (5)
[0096] Among them, κ H (x,x i ), κ P (x,x i ) and κ R (x,x i ) are the hybrid kernel, linear kernel and radial basis function kernel respectively. d and δ are the dimension and width of the linear kernel and radial basis function kernel respectively. w is the relationship coefficient between the linear kernel and the radial basis function kernel, and its value is set to 0.5.
[0097] Step 4), based on step 3), a 4-dimensional vector composed of parameters to be optimized based on the hybrid kernel support vector machine is established as follows:
[0098] x(t)=[C,d,δ,w] T (6)
[0099] Among them, C is the support vector machine penalty factor.
[0100] Step 5), establish the BAS optimization steps as follows:
[0101] Step 5.1) Initialize the 4-dimensional positions of 20 beetles in the search space and calculate their respective evaluation function values fit(x i ):
[0102] x i (t) = [C i ,d i ,δ i ,w i ] T ,i=1,2,…,20 (7)
[0103] Step 5.2) Calculate the positions of the beetle's left and right antennae in space as follows:
[0104]
[0105] Where i = 1, 2, ..., 20, x il (t) and x ir (t) The positions of the left and right antennae of the beetle, n i is a randomly generated 4-dimensional vector. d(t) is the length of the tentacle, where d(t) = η·d(t-1)+ρ, ρ = 0.01, η = 0.95.
[0106] Step 5.3) Calculate the new position of the beetle:
[0107]
[0108] Where, i = 1, 2, ..., 20, fit(x il ) and fit(x ir ) are the evaluation function values calculated from the positions of the left and right antennae of the beetle. μ(t) represents the moving step, where μ(t) = 2*0.95 t sgn is the sign function. At the same time, the evaluation function value corresponding to the new position of the beetle is calculated
[0109] Step 5.4) Update the global position of the beetle in the next step as follows:
[0110]
[0111] Among them, i=1,2,…,20.
[0112] Step 6), the steps to establish GSA optimization are as follows:
[0113] Step 6.1) Randomly initialize the 4-dimensional positions of 20 beetles in the search space as shown in equation (7).
[0114] Step 6.2) Calculate the evaluation function values and their quality of the 20 beetles in the search space:
[0115] m i (t) = [fit i (t)-fit worst (t)] / [fit best (t)-fit worst (t)] (11)
[0116]
[0117] Where i = 1, 2, ..., 20, fit i (t) and M i (t) are the evaluation function value and quality of the ith beetle, fit best (t) and fit worst (t) are the best and worst evaluation function values, respectively, m i (t) is an intermediate variable.
[0118] Step 6.3) Calculate the component of the total gravitational force on the i-th beetle in the k-th dimension:
[0119]
[0120] Where, i,j=1,2,…,20, k=1,2,…,4. j is a random coefficient between 0 and 1. represents the component of the gravitational force of beetle j on beetle i in the k-th dimension, which is calculated as follows:
[0121]
[0122] in, is a minimum constant value, R i,j (t) represents the Euclidean distance between beetle j and beetle i, G(t) represents the gravitational constant, where G(t) = 10*exp(-t / 15).
[0123] Step 6.4) Calculate the beetle's next moving speed and position:
[0124]
[0125]
[0126]
[0127] Among them, i,j=1,2,…,20, k=1,2,…,4. is a random coefficient between 0 and 1. represents the acceleration of beetle i, and Indicates the beetle's next moving speed and position, and Indicates the beetle's current movement speed and position.
[0128] Step 7), the steps to establish the hybrid BAS-GSA optimization are as follows
[0129] Step 7.1) Set the threshold of the evaluation function value fit min = 0.0001 and the maximum number of iterations T max = 100. Set the starting position x of 20 beetles i (t) and its corresponding evaluation function value fit(x i ), i = 1, 2, ..., 20. Calculate the optimal evaluation function value fit best as follows:
[0130] fit best =min{fit(x1),fit(x2),…,fit(x N )} (18)
[0131] Step 7.2) Use formula (2) to calculate the evaluation function value fit(x i ), i=1,2,…,20.
[0132] Step 7.3) Use formula (8) to calculate the left and right antenna positions of 20 beetles: li (t) and x ri (t); use formula (9) to calculate the next position of the beetle
[0133] Step 7.4) Use formula (12) to calculate the mass M of the 20 beetles. i (t); Use formula (13) to calculate the gravitational force on the 20 particles Use formula (15) to calculate the next moving speed of 20 particles
[0134] Step 7.5) Use formula (11) to calculate the new positions of the 20 beetles.
[0135]
[0136] Among them, v i (t+1) is the vector composed of the next moving speed calculated by formula (15), is the next step position vector calculated by formula (9), α=0.5.
[0137] Step 7.6) Use formula (2) to calculate the evaluation function values corresponding to the new positions of the 20 beetles
[0138] Step 7.7) Use formula (10) to calculate the global new position x of the 20 beetles i (t+1).
[0139] Step 7.8) t←t+1, if t≤T max And fit best ≥f min , then return to step 7.1 to continue execution, otherwise end the loop and output the optimal solution x best =[C,d,δ,w] T .
[0140] Step 8), use the optimal solution x output from step 7) best =[C,d,δ,w] T Train and test a support vector machine model.
[0141] The parameters used for simulation verification in the present invention are as follows:
[0142] In this system, the output frequency of the single-axis gyro is 100Hz, and the constant drift is 0.1° / h. For the convenience of analysis, the original data of the gyro is subtracted from the constant drift and the earth's rotation rate, and the temperature drift is obtained after using a 10s average filter. The temperature drift output curve of the gyro is shown in the figure below: Figure 3 and Figure 4 As shown, the temperature rates are approximately 10°C / min and 5°C / min, which are used for training and testing the hybrid kernel support vector machine model, respectively. Figure 5 It represents the prediction curve after the hybrid kernel support vector machine is trained with the training data (temperature rate is about 10°C / min), and its mean square error value is 0.00054. Figure 6 It represents the prediction curve after testing the trained hybrid kernel support vector machine with test data (temperature rate is about 5℃ / min), and its mean square error value is 0.00086. The result shows that the prediction model proposed in this paper can achieve high prediction accuracy and realize high-precision compensation of gyroscope temperature drift.
[0143] It should be noted that the above content only illustrates the technical idea of the present invention and cannot be used to limit the protection scope of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications all fall within the protection scope of the claims of the present invention.
Claims
1. A gyro temperature drift compensation method based on hybrid kernel support vector machine optimized by BAS-GSA, Features: The following steps are involved: Step 1), establish the mathematical model of gyro temperature drift: Where y is the output temperature drift, T is the measured temperature, △T is the temperature change, is the temperature gradient, f() is the temperature drift model transfer function, Step 2), establish the evaluation function based on mean square error as follows: Among them, fit represents the objective function value, y i and are the measured and predicted temperature drift values, respectively, n represents the number of samples, Step 3), establish a support vector machine hybrid kernel model based on linear kernel and radial basis function kernel as follows: k H (x,x i )=(1-w)κ P (x,x i )+wκ R (x,x i ) (3) κ P (x,x i )=(x T x i +1) d (4) κ R (x,x i )=exp[-||x-x i || 2 / (2δ 2 )] (5) Among them, κ H (x,x i ), κ P (x,x i ) and κ R (x,x i ) are the hybrid kernel, linear kernel and radial basis function kernel respectively, d and δ are the dimension and width of the linear kernel and radial basis function kernel respectively, w is the relationship coefficient between the linear kernel and the radial basis function kernel, Step 4), based on step 3), a 4-dimensional vector composed of parameters to be optimized based on the hybrid kernel support vector machine is established as follows: x(t)=[C,d,δ,w] T (6) Among them, C is the support vector machine penalty factor, Step 5), establish the BAS optimization steps as follows: Step 5.1) Initialize the 4-dimensional positions of N beetles in the search space and calculate their respective evaluation function values fit(x i ): x i (t)=[C i ,d i ,δ i ,w i ] T ,i=1,2,…,N (7) Step 5.2) Calculate the positions of the beetle's left and right antennae in space as follows: Where i = 1, 2, ..., N, x il (t) and x ir (t) The positions of the left and right antennae of the beetle, n i is a randomly generated 4-dimensional unit vector, d(t) is the length of the tentacle, Step 5.3) Calculate the new position of the beetle: Where, i = 1, 2, ..., N, fit(x il ) and fit(x ir ) are the evaluation function values calculated by the positions of the left and right antennae of the beetle, μ(t) represents the moving step, sgn is the sign function, and the evaluation function value corresponding to the new position of the beetle is calculated at the same time Step 5.4) Update the beetle's global position for the next step as follows: Where i = 1, 2, ..., N, Step 6), the steps to establish GSA optimization are as follows: Step 6.1) Randomly initialize the 4-dimensional positions of N beetles in the search space as shown in equation (7), Step 6.2) Calculate the evaluation function value and quality of N beetles in the search space: m i (t)=[fit i (t)-fit worst (t)] / [fit best (t)-fit worst (t)] (11) Where i = 1, 2, ..., N, fit i (t) and M i (t) are the evaluation function value and quality of the ith beetle, fit best (t) and fit worst (t) are the best and worst evaluation function values, respectively, m i (t) is the intermediate variable, Step 6.3) Calculate the component of the total gravitational force on the i-th beetle in the k-th dimension: Among them, i,j=1,2,…,N,k=1,2,…,4,λ j is a random coefficient between 0 and 1, represents the component of the gravitational force of beetle j on beetle i in the k-th dimension, which is calculated as follows: in, is a minimum constant value, R i,j (t) represents the Euclidean distance between beetle j and beetle i, G(t) represents the gravitational constant, Step 6.4) Calculate the beetle's next moving speed and position: Among them, i,j=1,2,…,N,k=1,2,…,4,θ i is a random coefficient between 0 and 1, represents the acceleration of beetle i, and Indicates the beetle's next moving speed and position, and Indicates the current moving speed and position of the beetle. Step 7), establish a hybrid BAS-GSA optimization algorithm and calculate the optimal hyperparameters. Step 8), use the optimal solution x output from step 7) best =[C,d,δ,w] T Train and test a support vector machine model.
2. According to the hybrid kernel support vector machine gyro temperature drift compensation method based on BAS-GSA optimization according to claim 1, It is characterized in that The detailed steps of step 7) are as follows: Step 7.1) Set the threshold of the evaluation function value fit min and the maximum number of iterations T max , set the starting position x of N beetles i (t) and its corresponding evaluation function value fit(x i ), i = 1, 2, ..., N, calculate the optimal evaluation function value fit best as follows: fit best =min{fit(x 1 ),fit(x 2 ),…,fit(x N )} (18) Step 7.2) Use formula (2) to calculate the evaluation function value fit(x i ), i=1,2,…,N, Step 7.3) Use formula (8) to calculate the positions of the left and right antennae of N beetles: li (t) and x ri (t); use formula (9) to calculate the next position of the beetle Step 7.4) Use formula (12) to calculate the mass M of N beetles i (t); Use formula (13) to calculate the gravitational force F on N particles i k (t), use formula (15) to calculate the next moving speed of N particles Step 7.5) Use formula (11) to calculate the new positions of the N beetles, Among them, v i (t+1) is the vector composed of the next moving speed calculated by formula (15), is the next step position vector calculated by formula (9), α is a coefficient between 0 and 1, Step 7.6) Use formula (2) to calculate the evaluation function values corresponding to the new positions of the N beetles Step 7.7) Use formula (10) to calculate the global new position x of the N beetles i (t+1), Step 7.8) t←t+1, if t≤T max And fit best ≥f min , then return to step 7.1 to continue execution, otherwise end the loop and output the optimal solution x best =[C,d,δ,w] T .
