A scale factor error compensation method for fiber optic gyroscope during startup phase based on CWOA-MLP neural network

Abstract

The present invention discloses a method for compensating the scale factor error of a fiber optic gyroscope startup process based on a CWOA-MLP neural network, and belongs to the technical field of fiber optic gyroscopes. The present invention designs a scale factor calibration experiment for a fiber optic gyroscope startup phase under a continuously changing temperature, obtains the scale factor error of the fiber optic gyroscope at different temperature points, and uses a CWOA-MLP neural network to compensate for the scale factor error of the fiber optic gyroscope. Experiments show that compared with traditional discrete calibration experiments, the present invention greatly shortens the calibration time while improving the calibration accuracy, thereby improving the performance of the fiber optic gyroscope in the startup phase.

Description

Technical Field

[0001] The present invention relates to the technical field of fiber optic gyroscopes, and mainly to a method for compensating a scale factor error of a fiber optic gyroscope during its startup phase based on a CWOA-MLP neural network. Background Art

[0002] As a new generation of inertial sensors, fiber optic gyroscopes have the advantages of being light, small, low power, long life, high reliability, and mass production. Therefore, they are widely used in a large number of precision instruments. However, during the power-on startup of the fiber optic gyroscope, the operation of each optoelectronic device will cause the internal temperature of the fiber optic gyroscope to rise, which will lead to scale factor errors and reduce the accuracy of the fiber optic gyroscope. Therefore, it is necessary to model and compensate for the scale factor error of the fiber optic gyroscope during the startup process.

[0003] In order to model the scale factor error, the scale factor sensitivity of the fiber optic gyroscope should first be measured within the temperature range required for the application. The commonly used measurement method is to evenly select several experimental temperature points within the required temperature range, and then measure the scale factor of the fiber optic gyroscope after it reaches the specified temperature in the incubator. During this measurement process, the gyroscope needs to be kept warm many times, which is time-consuming and cumbersome. In addition, the selection of experimental temperature points makes the obtained temperature and scale factor values ​​discrete values, which makes it difficult to cover the entire temperature range, affecting the accuracy of later modeling and compensation.

[0004] At present, the compensation methods for the scale factor error during the startup process are mostly focused on controlling or compensating the central wavelength of the light source, such as adding a temperature control system to the light source or establishing a wavelength startup model. Although these methods suppress the scale factor error to a certain extent, it is difficult to achieve an ideal compensation effect because the scale factor error of the fiber gyroscope is also related to factors such as the fiber ring change caused by the temperature increase and the change in the modulation coefficient of the feedback channel.

[0005] Existing modeling mostly uses simple and feasible linear or piecewise linear model structures, but the accuracy is limited. The temperature characteristics of fiber optic gyroscopes are affected by many factors and have very complex nonlinear characteristics. Neural networks have a good ability to approximate complex nonlinear functions, which simplifies the temperature compensation steps and improves the compensation accuracy.

[0006] At present, the main method for training neural networks is the back propagation algorithm, namely the BP algorithm. However, the BP algorithm has a slow calculation speed, is greatly affected by the initial value, and is prone to fall into a local optimal solution. Summary of the invention

[0007] The purpose of the invention in this paper is to provide a method for compensating the scale factor error in the startup phase of a fiber optic gyroscope based on a CWOA-MLP neural network, so as to achieve real-time compensation of the scale factor error during the startup process.

[0008] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0009] A method for compensating scale factor error of a fiber optic gyroscope during startup phase based on a CWOA-MLP neural network comprises the following steps:

[0010] 1) Calibrate the scale factor of the fiber optic gyroscope during the startup phase under continuous temperature;

[0011] 2) Construct an MLP neural network based on input and output;

[0012] 3) Use CWOA algorithm to optimize the parameters of MLP neural network;

[0013] 4) Obtain the scale factor error model of the fiber optic gyroscope during the startup phase based on the CWOA-MLP neural network, and compensate the scale factor error of the fiber optic gyroscope in real time.

[0014] Furthermore, the step 1) is specifically as follows:

[0015] 1.1) Place the fiber optic gyroscope on the rate turntable of the temperature-controlled rate stage, set the rotation axis of the rate stage vertically, and make the input reference axis of the fiber optic gyroscope parallel to the rotation axis of the rate stage;

[0016] 1.2) Set the initial temperature of the incubator to T0, so that the fiber optic gyroscope reaches thermal equilibrium at the initial temperature of the incubator;

[0017] 1.3) The temperature rise rate of the temperature box is set to ΔT℃ / min, and a calibration experiment of the fiber optic gyroscope is performed within the working temperature range [T0, T1] of the fiber optic gyroscope. The calibration experiment process includes:

[0018] 1.3.1) Set the turntable speed to ω, power on the fiber optic gyroscope, and collect the internal temperature T and the digital value G output by the fiber optic gyroscope during the startup process at the corresponding test temperature point;

[0019] 1.3.2) After each sampling time t1, the fiber optic gyroscope is powered off and the turntable is controlled to decelerate uniformly to 0; after the turntable is stationary for t2 time, step 1.3.1) is repeated again until the temperature of the incubator rises to T1.

[0020] Furthermore, the step 2) is specifically as follows:

[0021] The MLP neural network is set as a three-layer neural network with dual input and single output. The mathematical model is expressed as:

[0022]

[0023]

[0024]

[0025]

[0026] Among them, h is the number of hidden layers, W 1j is the weight from the first input to the jth hidden layer, W 2j is the weight from the second input to the jth hidden layer, s j is the weighted sum of the inputs to the jth hidden layer, θ j is the threshold of the jth hidden layer; T is the internal temperature of the fiber optic gyroscope during startup, sigmoid(.) is the activation function, S j is the output of the jth hidden layer; θ' is the threshold of the output layer, w j is the weight from the jth hidden layer to the output layer, is the weighted sum of the outputs of the hidden layer, is the final estimated value of the scaling factor. The above weights and thresholds need to be further trained and optimized to obtain the optimal neural network.

[0027] Furthermore, the step 3) is specifically as follows:

[0028] The weight and threshold of the MLP neural network are used as the position vector of the CWOA algorithm, the error of the MLP neural network is used as the fitness function of the CWOA algorithm, and the population size N and the number of iterations t are set. max , construct the CWOA optimization model, and its mathematical model is as follows:

[0029]

[0030]

[0031] in, is the position vector of each individual in CWOA, J is the loss function of the MLP algorithm, which is also the fitness function of the CWOA algorithm. is the weight set from the input layer to the hidden layer in the MLP neural network, is the weight set from the hidden layer to the output layer in the MLP neural network, is the threshold set of the hidden layer of the MLP neural network, K is the calculated value of the scale factor of the fiber optic gyroscope;

[0032] Define the parameter p in the CWOA algorithm, The values ​​are as follows:

[0033]

[0034]

[0035]

[0036]

[0037] Among them, p, is a parameter in the CWOA algorithm. The initial value of p is a random number in the range [0,1]. t is the random number generated by the Tent chaotic map at the tth iteration, is a random number uniformly distributed in the range [0,1]. is the convergence factor, t is the number of iterations, t max is the maximum number of iterations;

[0038] When p < 0.5 and When , the shrinking and encircling mechanism is selected, and the formula is as follows:

[0039]

[0040]

[0041] When p < 0.5 and When random search is selected, the formula is as follows:

[0042]

[0043]

[0044] When p≥0.5, the spiral update position is selected, and the formula is as follows:

[0045]

[0046] in, is the current global optimal position vector corresponding to the tth iteration, is a randomly generated position vector, is the vector from the individual to the current global optimal position, is the vector from an individual to a random position, l is a random number between [-1,1], and b is a constant that limits the shape of the logarithmic spiral.

[0047] Furthermore, the compensation method of step 4) is specifically as follows:

[0048] (1) The internal temperature T and the rate of change of the internal temperature of the fiber optic gyroscope are Input into the CWOA-MLP neural network model to calculate the predicted value of the scaling factor

[0049] (2) Correct the digital quantity G output by the fiber optic gyroscope:

[0050]

[0051] Wherein, K0 is the preset standard value of the scale factor, and G′ is the digital value output by the corrected fiber optic gyroscope.

[0052] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0053] The present invention provides a method for compensating the scale factor of a fiber optic gyroscope based on a CWOA-MLP neural network, studies the influence of the temperature on the scale factor during the startup phase of the fiber optic gyroscope, and designs a calibration experiment at continuous temperature points, so that the output data of the fiber optic gyroscope startup phase at continuous temperature points can be obtained without multiple insulation. In addition, the CWOA optimization algorithm has a simple principle, few parameters, and strong global optimization capability, which is superior to the common BP training method and avoids the problem of easily falling into a local optimal solution. The experimental results show that compared with the traditional discrete calibration scheme, the present invention greatly shortens the calibration time while improving the calibration accuracy, and improves the performance of the fiber optic gyroscope during the startup phase at different temperatures. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is the flow chart of scale factor error compensation in this example;

[0055] Figure 2 It is a schematic diagram of the relationship between the scale factor of the fiber optic gyroscope and the temperature and the temperature change rate before compensation;

[0056] Figure 3 This is a schematic diagram of the relationship between the scale factor of the compensated fiber optic gyroscope and the temperature and the temperature change rate. DETAILED DESCRIPTION

[0057] The following further describes the implementation of the present invention, and examples of the implementation are shown in the accompanying drawings. The embodiments described below are exemplary and intended to be used to explain the present invention, but should not be understood as limiting the present invention.

[0058] like Figure 1 As shown, the present invention designs a calibration experiment of the scale factor of the fiber optic gyroscope startup phase at continuous temperature points. Only one set of experiments is needed to obtain the output data of the fiber optic gyroscope startup phase at continuous temperature points, which specifically includes the following steps:

[0059] Step 1. Place the fiber optic gyroscope on the rate turntable of the temperature-controlled rate table. Set the rotation axis of the rate table to be vertical, with the error within the specified value; set the input reference axis IRA of the fiber optic gyroscope parallel to the rotation axis, with the error within the specified value. Connect the fiber optic gyroscope to the output test equipment, so that the test equipment records the elapsed time and the output of the fiber optic gyroscope.

[0060] Step 2: Set the initial temperature of the incubator to -40°C. After reaching the specified temperature, keep it warm for half an hour to allow the fiber optic gyroscope to reach thermal equilibrium.

[0061] Step 3: Raise the temperature of the temperature box at a rate of 1°C / min to calibrate the scale factor of the start-up process of the fiber optic gyroscope within the operating temperature range of the fiber optic gyroscope [-40°C, 60°C].

[0062] In step 3, the specific implementation steps of starting the process scale factor calibration are as follows:

[0063] (3.1) According to the experimental requirements, the angular velocity of the turntable is set to ω = 60° / s, and the turntable is started to rotate at 10° / s. 2 The angular acceleration changes to 60° / s.

[0064] (3.2) After the angular velocity stabilizes, the FOG power supply is turned on to collect the internal temperature T of the FOG during the startup phase and the digital value G output by the FOG.

[0065] (3.3) After sampling for 180 seconds, the fiber optic gyroscope is powered off and the turntable is controlled to decelerate uniformly to 0.

[0066] (3.4) After the turntable has been stationary for 180 seconds, repeat steps (3.1)-(3.3) until the incubator reaches the target temperature of 60°C.

[0067] Step 4: Calculate the scale factor of the fiber optic gyroscope:

[0068] K=G / ω

[0069] The scale factor values ​​of the fiber optic gyroscope at the startup stage under different input angular rates and temperatures are as follows: Figure 2 As shown, it can be clearly seen that the fiber optic gyroscope scale factor shows a nonlinear trend with temperature and increases with increasing temperature.

[0070] Step 5: The internal temperature T and the temperature change rate of the fiber optic gyroscope during startup As the input of the CWOA-MLP neural network, the scaling factor value K is used as the output of the CWOA-MLP neural network. The parameters of the CWOA-MLP neural network are trained to obtain a set of trained neural networks.

[0071] In step 5, the specific steps of establishing and training CWOA-MLP are as follows:

[0072] (5.1) Set the MLP neural network as a three-layer neural network with dual input and single output, set the hidden layer to be, and the mathematical model is expressed as:

[0073]

[0074]

[0075]

[0076]

[0077] Wherein, h is the number of hidden layers. In this embodiment, h=20; W 1j is the weight from the first input to the jth hidden layer, W 2j is the weight from the second input to the jth hidden layer, s j The weighted sum of the inputs to the jth hidden layer, θ j is the threshold of the jth hidden layer; T is the internal temperature of the fiber optic gyroscope during startup, sigmoid(.) is the activation function, S j is the output of the jth hidden layer; θ' is the threshold of the output layer, w j is the weight from the jth hidden layer to the output layer, is the weighted sum of the outputs of the hidden layer, is the final estimate of the scaling factor.

[0078] (5.2) The weight threshold of MLP is used as the position vector of CWOA, the error of MLP is used as the fitness function of CWOA, the population size N and the number of iterations t are set. max , construct the CWOA optimization algorithm, and its mathematical model is as follows:

[0079]

[0080]

[0081] in, is the position vector of each individual in CWOA, J is the loss function of the MLP algorithm, which is also the fitness function of CWOA. is the weight set from the input layer to the hidden layer in the MLP neural network, is the weight set from the hidden layer to the output layer in the MLP neural network, is the threshold set of the hidden layer of the MLP neural network, K is the calculated value of the scale factor of the fiber optic gyroscope;

[0082] Define the parameter p in the CWOA algorithm, The values ​​are as follows:

[0083]

[0084]

[0085]

[0086]

[0087] Among them, p, is a parameter in the CWOA algorithm. The initial value of p is a random number in the range [0,1], and the chaotic sequence is generated using Tent chaotic mapping. t is the random number corresponding to the tth iteration, is a random number uniformly distributed in the range [0,1]. is the convergence factor that decreases linearly from 2 to 0, t is the number of iterations, t max is the maximum number of iterations.

[0088] When p < 0.5 and When , the shrinking and encircling mechanism is selected, and the formula is as follows:

[0089]

[0090]

[0091] When p < 0.5 and When random search is selected, the formula is as follows:

[0092]

[0093]

[0094] When p≥0.5, the spiral update position is selected, and the formula is as follows:

[0095]

[0096] in, is the current global optimal position vector corresponding to the tth iteration, is a randomly generated position vector, is the vector from the individual to the current global optimal position, is the vector from an individual to a random position, l is a random number between [-1,1], and b is a constant that limits the shape of the logarithmic spiral.

[0097] (5.3) Follow step (5.2) to Perform global optimization and finally obtain the best position vector These are the parameters of the MLP neural network after training.

[0098] Step 6: Perform temperature compensation on the resonant gyroscope based on the trained CWOA-MLP neural network model.

[0099] In step 6, the specific steps of temperature compensation are as follows:

[0100] (6.1) The internal temperature T and the rate of change of the internal temperature of the fiber optic gyroscope are Input into the CWOA-MLP neural network model to calculate the predicted value of the scaling factor

[0101] (6.2) Correct the digital value G output by the fiber optic gyroscope:

[0102]

[0103] Wherein, K0 is the preset standard value of the scale factor, and G′ is the digital value output by the corrected fiber optic gyroscope.

[0104] Compensation effect such as Figure 3 As shown, during the start-up process of the fiber optic gyroscope at full temperature, the change of the scale factor is less than 50ppm, and it can be seen that the compensation effect is obvious.

[0105] Although the present invention has been disclosed as above in the form of preferred embodiments, it is obvious that the present invention is not limited to the above embodiments. Any technical personnel in this field can make possible changes and modifications to the technical solutions of the present invention by using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above examples based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall fall within the protection scope of the technical solutions of the present invention.

Claims

1. A method for compensating scale factor error of a fiber optic gyroscope during startup based on a CWOA-MLP neural network, comprising the following steps: 1) Calibrate the scale factor of the fiber optic gyroscope during the startup phase under continuous temperature; 2) Construct an MLP neural network based on input and output; 3) Use CWOA algorithm to optimize the parameters of MLP neural network; 4) Obtain the scale factor error model of the fiber optic gyroscope during the startup phase based on the CWOA-MLP neural network, and compensate the scale factor error of the fiber optic gyroscope in real time; The step 1) is specifically as follows: 1.1) Place the fiber optic gyroscope on the rate turntable of the temperature-controlled rate stage, set the rotation axis of the rate stage vertically, and make the input reference axis of the fiber optic gyroscope parallel to the rotation axis of the rate stage; 1.2) Set the initial temperature of the incubator to T0, so that the fiber optic gyroscope reaches thermal equilibrium at the initial temperature of the incubator; 1.3) The temperature rise rate of the temperature box is set to ΔT℃ / min, and a calibration experiment of the fiber optic gyroscope is performed within the working temperature range [T0, T1] of the fiber optic gyroscope. The calibration experiment process includes: 1.3.1) Set the turntable speed to ω, power on the fiber optic gyroscope, and collect the internal temperature T and the digital value G output by the fiber optic gyroscope during the startup process at the corresponding test temperature point; 1.3.2) After each sampling time t1, the fiber optic gyroscope is powered off and the turntable is controlled to decelerate uniformly to 0; after the turntable is stationary for t2 time, step 1.3.1) is repeated again until the temperature of the incubator rises to T1; The step 2) is specifically as follows: The MLP neural network is set as a three-layer neural network with dual input and single output. The mathematical model is expressed as: Among them, h is the number of hidden layers, W 1j is the weight from the first input to the jth hidden layer, W 2j is the weight from the second input to the jth hidden layer, s j is the weighted sum of the inputs to the jth hidden layer, θ j is the threshold of the jth hidden layer; T is the internal temperature of the fiber optic gyroscope during startup, sigmoid(.) is the activation function, S j is the output of the jth hidden layer; θ' is the threshold of the output layer, w j is the weight from the jth hidden layer to the output layer, is the weighted sum of the outputs of the hidden layer, is the final estimated value of the scaling factor; the above weights and thresholds need to be further trained and optimized to obtain the optimal neural network; The step 3) is specifically as follows: The weight and threshold of the MLP neural network are used as the position vector of the CWOA algorithm, the error of the MLP neural network is used as the fitness function of the CWOA algorithm, and the population size N and the number of iterations t are set. max , construct the CWOA optimization model, and its mathematical model is as follows: in, is the position vector of each individual in CWOA, J is the loss function of the MLP algorithm, which is also the fitness function of the CWOA algorithm. is the weight set from the input layer to the hidden layer in the MLP neural network, is the weight set from the hidden layer to the output layer in the MLP neural network, is the threshold set of the hidden layer of the MLP neural network, K is the calculated value of the scale factor of the fiber optic gyroscope; Define the parameter p in the CWOA algorithm, The values ​​are as follows: Among them, p, is a parameter in the CWOA algorithm. The initial value of p is a random number in the range [0,1]. t is the random number generated by the Tent chaotic map at the tth iteration, is a random number uniformly distributed in the range [0,1]. is the convergence factor, t is the number of iterations, t max is the maximum number of iterations; When p < 0.5 and When , the shrinking and encircling mechanism is selected, and the formula is as follows: When p < 0.5 and When random search is selected, the formula is as follows: When p≥0.5, the spiral update position is selected, and the formula is as follows: in, is the current global optimal position vector corresponding to the tth iteration, is a randomly generated position vector, is the vector from the individual to the current global optimal position, is the vector from the individual to the random position, l is a random number between [-1,1], and b is a constant that limits the shape of the logarithmic spiral; The compensation method of step 4) is specifically as follows: (1) The internal temperature T and the rate of change of the internal temperature of the fiber optic gyroscope are Input into the CWOA-MLP neural network model to calculate the predicted value of the scaling factor (2) Correct the digital quantity G output by the fiber optic gyroscope: Wherein, K0 is the preset standard value of the scale factor, and G′ is the digital value output by the corrected fiber optic gyroscope.

Images