Inertial device random characteristic simulation method based on Allan variance matching

Abstract

The invention relates to an inertial device random characteristic simulation method based on Allan variance matching, and belongs to the technical field of inertial device random characteristic simulation, and the inertial device random characteristic simulation method comprises the following steps: S1, constructing a fusion random model of inertial device random errors based on quantization noise, white noise, zero offset instability noise, random walk noise and rate slope noise; s2, performing a static experiment on the inertial device to be modeled, and obtaining an Allan variance actual curve of the inertial device based on data of the static experiment; and S3, based on the Allan variance actual curve of the inertial device obtained in the step S2, performing parameter optimization determination on the fusion random model constructed in the step S1 by using an improved genetic algorithm. According to the method, the global matching property of the Allan variance curve of the random error simulation data of the inertial device and the actual data of the inertial device is improved, and the problem that the random characteristic of the inertial device is difficult to simulate more accurately in the prior art is solved.

Description

Technical Field

[0001] The invention belongs to the technical field of inertial device characteristic simulation, and in particular relates to a method for simulating random characteristics of an inertial device based on Allan variance matching. Background Art

[0002] Most of the deterministic errors of the rotating inertial navigation system are suppressed, and the random error of the integral characteristic of the inertial navigation calculation becomes the main error source. In order to realize the selection of rotating inertial navigation system components and the simulation and deduction of system error characteristics, it is necessary to carry out high-fidelity random characteristic simulation of inertial devices.

[0003] At present, the simulation of random characteristics of inertial devices often uses Allan variance, ARMA, EMD and other methods for modeling and analysis. However, the ARMA and EMD methods have limitations in modeling and analysis due to the device's successive power-on error and sampling time, and the model generalization ability is limited. Using Allan variance to characterize the random characteristics of inertial devices has relatively good stability and generalization. For example, the invention patent with publication number CN108225374A proposes an Allan variance analysis method that integrates genetic algorithms, and the invention patent with publication number CN103345569A discloses a noise coefficient calculation method for Allan variance analysis technology.

[0004] However, the existing Allan variance method for simulating the random characteristics of inertial devices has problems such as inaccurate traditional Allan variance parameter fitting modeling because the Allan variance calculation is usually not an ideal smooth curve, and the random modeling of inertial devices often uses white noise plus a first-order Markov process, which makes it difficult to accurately simulate the random characteristics of inertial devices. Summary of the invention

[0005] In view of the deficiencies in the prior art, an object of the present invention is to provide a method for simulating random characteristics of an inertial device based on Allan variance matching, so as to solve or improve the defects in the prior art.

[0006] To achieve the above object, the present invention adopts the following technical solution: a method for simulating random characteristics of an inertial device based on Allan variance matching, comprising the following steps: S1. Based on quantization noise, white noise, zero bias instability noise, random walk noise and rate ramp noise, a fusion random model of random errors of inertial devices is constructed; S2. Conducting a static experiment on the inertial device to be modeled, and obtaining the actual Allan variance curve of the inertial device based on the data of the static experiment; S3. Based on the actual Allan variance curve of the inertial device obtained in step S2, the improved genetic algorithm is used to optimize and determine the parameters of the fusion random model constructed in step S1.

[0007] Preferably, in step S1, the fusion random model of the random error of the inertial device constructed is: δω=δω Q +δω N +δω B +δω K +δω R ; Where δω is the random error of the inertial device, δω Q is the quantization noise in the random error of the inertial device, δω N is the white noise in the random error of the inertial device, δω B is the zero bias instability noise in the random error of the inertial device, δω K is the random walk noise in the random error of the inertial device, δω R is the rate ramp noise in the random error of the inertial device; Specifically, δω Q =Qu1(t),δω N = Nu2(t), δω K =Ku4(t), Where Q is the quantization noise intensity, N is the white noise intensity, B is the zero-bias instability noise, β is the inverse of the correlation time, K is the random walk noise intensity, R is the rate ramp noise intensity, u1(t), u2(t), u3(t), u4(t), u5(t) are mutually uncorrelated unit Gaussian white noises, ω0 is the natural frequency parameter, t is time, and s is the complex frequency; Therefore, the fusion random model of the random error of the inertial device is constructed as follows: Among them, quantization noise intensity Q, white noise intensity N, zero bias instability noise B, correlation time inverse β, random walk noise intensity K, rate ramp noise intensity R, natural frequency parameter ω0, and complex frequency s are model parameters to be optimized.

[0008] Preferably, in step S2, a static experiment is performed on the inertial device to be modeled, and a specific method for obtaining an actual Allan variance curve of the inertial device based on the data of the static experiment includes: S21, the inertial device to be modeled is placed statically on the platform, and is turned on for 2-3 hours to preheat; after preheating, the output data of the inertial device is collected, and the collection time is 6-8 hours to collect static experimental data; S22. Based on the collected static experimental data, an actual Allan variance curve of the inertial device is obtained.

[0009] Preferably, in step S22, the specific method for obtaining the actual Allan variance curve of the inertial device based on the collected static experimental data is as follows: For a data set {y1, y2, …, y N} with N sampling points and a sampling time interval of τ0, the Allan variance analysis modeling process is described as follows: (a) Cluster the entire data set with correlation times τ = τ0, 2τ0, ..., kτ0 (k < N / 2), where each cluster contains k sampling points and the corresponding correlation times vary from small to large; (b) Calculate the mean of each cluster and the difference in means between adjacent clusters; (c) Calculate the Allan variance σ according to the formula 2 ; In the formula, σ 2 (τ) is the Allan variance corresponding to the correlation time τ, N C is the total number of clusters corresponding to the correlation time τ, is the mean of the i-th cluster at this time, is the mean of the (i + 1)-th cluster at this time.

[0010] Preferably, in step S3, based on the actual Allan variance curve of the inertial device obtained in step S2, the specific method for optimizing the parameters of the fusion random model constructed in step S1 using the improved genetic algorithm includes: S31. Encoding and initial population generation: Use binary encoding combined with multi-parameter cascading for encoding. The specific operation is as follows: Let the target function parameter quantization noise intensity Q, white noise intensity N, bias instability noise B, reciprocal of the correlation time β, random walk noise intensity K, rate ramp noise intensity R, natural frequency parameter ω0, and complex frequency s to be optimized and determined correspond to x1, x2, x3, x4, x5, x6, x7, x8, and their value ranges are [x1 min , x1 max , [x2 min , x2 max , [x3 min , x3 max , [x4 min , x4 max , [x5 min , x5 max , [x6 min , x6 max , [x7 min , x7 max , [x8 min , x8 max], whose target parameter accuracy or optimization step size are Δx1, Δx2, Δx3, Δx4, Δx5, Δx6, Δx7, Δx8 respectively; The corresponding encoding lengths of x1, x2, x3, x4, x5, x6, x7, and x8 are: In the formula, bit x Encode length for each variable, bit sum Encode the length for each individual; x max Indicates the corresponding maximum values ​​of x1, x2, x3, x4, x5, x6, x7, and x8; x min Indicates the minimum value of the corresponding values ​​of x1, x2, x3, x4, x5, x6, x7, and x8; Δx represents the average value of Δx1, Δx2, Δx3, Δx4, Δx5, Δx6, Δx7, and Δx8; The initial population is generated randomly, the population size is popsize, and the encoding length of each individual is bit sum , the initial population generation method is expressed as: population initial =round(rand(popsize,bit sum )); Where population initial represents the initial population, rand(popsize,bit sum ) indicates generating popsize row bits sum Column random numbers, round means rounding the above random numbers, that is, generating a binary-coded initial population of the size of popsize population; S32, fitness evaluation: Combined with the objective function to be optimized, the inverse proportional transformation is used for constraint: fitness=1 / f(x); In the formula, f(x) represents the objective function to be optimized; The specific calculation formula of the objective function to be optimized is as follows: Wherein, RMSE is the root mean square error of the Allan variance curve between the simulated inertial device random data and the actual inertial device data, A is the Allan variance curve data of the actual inertial device data, F is the Allan variance curve data of the simulated inertial device data, and W is the weighted scheme value; According to the characteristics of the Allan variance curve, the Allan variance curve is fitted. The segmented fitting method is as follows: In the formula, p, q, a, b, f are fitting coefficients, x c is the curve segmentation point; Based on the above formula, the inverse proportional function is calculated to obtain the variable weight. The specific variable weight calculation formula is as follows: w = 1 / y; In theory, the Allan variance curve under the weight is: Where T(x) is the Allan variance curve under the weight effect, is the Allan variance curve of the generated data; S33, select: Assuming the average fitness of the group is avgf, directly retain the fitness greater than avgf*P better Individual, P better Take 0.5-1; population retain =population last fitness>avgf*P better ; Where population retain Indicates the retained individuals, population last For individuals of the previous generation; At this time, the number of individuals in the new population is less than that of the previous generation, and the individuals in the population are sorted according to fitness as follows: Where population fill Represents the sorted filled individuals, population better represents the optimal individual, population last Indicates the previous generation of individuals, population worse represents the individual with the worst fitness, rand∈(0,1) is a uniformly distributed random number, P s ∈(0,1) is a parameter used to control the ratio; S34, Crossover: Perform one-point crossover operation on different parameter coding segments respectively to ensure that each parameter coding segment is crossovered in each crossover operation; S35, mutation: The random position single-point mutation method is used to mutate the individual codes in the population, that is, 0 is mutated to 1 or 1 is mutated to 0; S36. According to steps S31-S35, based on the fusion random model constructed in step S1 and the actual Allan variance curve of the inertial device obtained in step S2, the optimal solution of parameter quantization noise intensity Q, white noise intensity N, zero bias instability noise B, correlation time inverse β, random walk noise intensity K, rate ramp noise intensity R, natural frequency parameter ω0, and complex frequency s is obtained.

[0011] Compared with the prior art, the present invention has the following beneficial effects: an inertial device random characteristic simulation method based on Allan variance matching of the present invention constructs a fusion random model of random errors based on quantization noise, white noise, zero bias instability noise, random walk noise, and rate ramp noise, and uses an improved genetic algorithm to optimize and determine the parameters of the fusion random model. In the parameter optimization and determination process, considering that the Allan variance result output variation range of the inertial device is several orders of magnitude, the weightless Allan variance evaluation function will lead to general Allan variance global matching performance, and a variable weight target evaluation function is proposed, which improves the global matching of the Allan variance curve between the inertial device random error simulation data and the actual inertial device data, and the simulation data and the real data have high global matching of the Allan variance, simple operation, and good practicality, and solves the existing Allan variance method for simulating the random characteristics of inertial devices, which solves the problems that the traditional Allan variance parameter fitting modeling is inaccurate because the Allan variance calculation is usually not an ideal smooth curve, and the inertial device random modeling often uses white noise plus a first-order Markov process, which makes it difficult to accurately simulate the random characteristics of the inertial device. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on the drawings in the following description without creative work.

[0013] Figure 1 The figure is a flow chart of a method for simulating random characteristics of an inertial device based on Allan variance matching according to an embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments in 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. In order to make the above-mentioned features and advantages of the present invention more obvious and easy to understand, the following embodiments are specifically cited and described in detail with the drawings.

[0015] like Figure 1 As shown, an embodiment of the present invention provides a method for simulating random characteristics of an inertial device based on Allan variance matching, comprising the following steps: S1. Based on quantization noise, white noise, zero bias instability noise, random walk noise and rate ramp noise, a fusion random model of random errors of inertial devices is constructed; S2. Conducting a static experiment on the inertial device to be modeled, and obtaining the actual Allan variance curve of the inertial device based on the data of the static experiment; S3. Based on the actual Allan variance curve of the inertial device obtained in step S2, the improved genetic algorithm is used to optimize and determine the parameters of the fusion random model constructed in step S1.

[0016] In this embodiment, in step S1, the fusion random model of the random error of the inertial device constructed is: δω=δω Q +δω N +δω B +δω K +δω R ; Where δω is the random error of the inertial device, δω Q is the quantization noise in the random error of the inertial device, δω N is the white noise in the random error of the inertial device, δω B is the zero bias instability noise in the random error of the inertial device, δω K is the random walk noise in the random error of the inertial device, δω R is the rate ramp noise in the random error of the inertial device; Specifically, δω Q =Qu1(t),δω N = Nu2(t), δω K =Ku4(t), Wherein, Q is the quantization noise intensity, N is the white noise intensity, B is the zero-bias instability noise, β is the reciprocal of the correlation time, K is the random walk noise intensity, R is the rate ramp noise intensity, u1(t), u2(t), u3(t), u4(t), u5(t) are uncorrelated unit Gaussian white noises, ω0 is the natural frequency parameter, t is the time, and s is the complex frequency; Therefore, the specific fusion random model of the random error of the inertial device is as follows: Wherein, the quantization noise intensity Q, the white noise intensity N, the zero-bias instability noise B, the reciprocal of the correlation time β, the random walk noise intensity K, the rate ramp noise intensity R, the natural frequency parameter ω0, and the complex frequency s are model parameters to be optimized and determined.

[0017] In this embodiment, in step S2, for the inertial device to be modeled, a static experiment is carried out, and the specific method for obtaining the actual Allan variance curve of the inertial device based on the data of the static experiment includes: S21. Place the inertial device to be modeled statically on the platform and warm it up for 2 - 3 hours; after warming up, start collecting the output data of the inertial device, and the collection time is 6 - 8 hours to obtain the static experiment data; S22. Based on the collected static experiment data, obtain the actual Allan variance curve of the inertial device.

[0018] In this embodiment, in step S22, the specific method for obtaining the actual Allan variance curve of the inertial device based on the collected static experiment data is as follows: For a data set {y1, y2,..., y N} with N sampling points and a sampling time interval of τ0, the Allan variance analysis modeling process is described as follows: (a) Cluster the entire data set with correlation times τ = τ0, 2τ0,..., kτ0 (k < N / 2), each cluster contains k sampling points, and the corresponding correlation times change from small to large; (b) Calculate the mean of each cluster and calculate the difference between the means of adjacent clusters; (c) Calculate the Allan variance σ according to the formula 2 ; In the formula, σ 2 (τ) is the Allan variance corresponding to the correlation time τ, N C is the total number of clusters corresponding to the correlation time τ, is the mean of the i-th cluster at this time, is the mean of the (i + 1)-th cluster at this time.

[0019] In this embodiment, in step S3, based on the actual Allan variance curve of the inertial device obtained in step S2, the specific method of optimizing and determining the parameters of the fusion random model constructed in step S1 using the improved genetic algorithm includes: S31, encoding and initial population generation: Coding is an important part of genetic algorithms. Different coding methods need to be used for different problems. The coding operation affects genetic operations such as selection, crossover and mutation. Currently, the commonly used coding methods include binary coding, Gray code coding, floating point coding and multi-parameter cascade coding. The specific object optimized in the present invention is a multi-parameter objective function. Therefore, binary coding combined with multi-parameter cascade coding is used for coding. The specific operation is: Assume that the objective function parameters to be optimized are quantization noise intensity Q, white noise intensity N, zero bias instability noise B, correlation time inverse β, random walk noise intensity K, rate ramp noise intensity R, natural frequency parameter ω0, and complex frequency s, corresponding to x1, x2, x3, x4, x5, x6, x7, x8, and their value range is [x1 min ,x1 max ]、[x2 min ,x2 max ]、[x3 min ,x3 max ]、[x4 min ,x4 max ]、[x5 min ,x5 max ]、[x6 min ,x6 max ]、[x7 min ,x7 max ]、[x8 min ,x8 max ], whose target parameter accuracy or optimization step size are Δx1, Δx2, Δx3, Δx4, Δx5, Δx6, Δx7, Δx8 respectively; The corresponding encoding lengths of x1, x2, x3, x4, x5, x6, x7, and x8 are: In the formula, bit x Encode length for each variable, bit sum Encode the length for each individual; x max Indicates the corresponding maximum values ​​of x1, x2, x3, x4, x5, x6, x7, and x8; x min Indicates the minimum value of the corresponding values ​​of x1, x2, x3, x4, x5, x6, x7, and x8; Δx represents the average value of Δx1, Δx2, Δx3, Δx4, Δx5, Δx6, Δx7, and Δx8; In order to meet the requirements of initial population diversity and randomness, the initial population is generated randomly. The population size is set to popsize and the encoding length of each individual is bit. sum , the initial population generation method is expressed as: population initial =round(rand(popsize,bit sum )); Where population initial represents the initial population, rand(popsize,bit sum ) indicates generating popsize row bits sum Column random numbers, round means rounding the above random numbers, that is, generating a binary-coded initial population of the size of popsize population; S32, fitness evaluation: In the evolutionary search, the genetic algorithm uses fitness as the standard to distinguish the quality of the population. The fitness function is generated by the conversion of the objective function. The quality of the fitness function selection directly affects the quality of the algorithm. The introduction of fitness value adjustment and resource sharing strategy can accelerate the convergence speed and jump out of the local optimal point. Adjusting the fitness value is to change the proportional relationship of the original fitness value by transformation. Commonly used proportional relationships include linear transformation, power function transformation and exponential transformation. The individual fitness evaluation function is usually a variant of the objective function. The objective function is usually a minimization or maximization problem, so the objective function is reasonably transformed according to the needs. In the early stage of the genetic algorithm, some highly competitive individuals may affect the selection and global optimization performance. Therefore, the present invention combines the objective function to be optimized and determines, and uses inverse proportional transformation for constraints: fitness=1 / f(x); In the formula, fitness represents the inverse proportional function after constraint, and f(x) represents the objective function to be optimized; The fitness of most populations in the present invention will not be too good, and combined with the processing of the selection process, the genetic diversity of the population is guaranteed; In the Allan variance matching process, the objective function selected is the root mean square error (RMSE) of the Allan variance curve between the simulated random data of the inertial device and the actual inertial device data. Theoretically, the smaller the RMSE, the better. Therefore, the population fitness is set as an inverse proportional function. On the one hand, it can ensure that individuals with a slight advantage in the early stage of optimization will not affect the global optimization. In the later stage, when the population fitness is higher, it can ensure that the smaller advantage is perceived, realizing the global search in the early stage and the local optimization in the later stage. The specific calculation formula of the objective function to be optimized is as follows: Wherein, RMSE is the root mean square error of the Allan variance curve between the simulated inertial device random data and the actual inertial device data, A is the Allan variance curve data of the actual inertial device data, F is the Allan variance curve data of the simulated inertial device data, and W is the weighted scheme value. If there is no weight, W = 1; However, due to the large variation range of the non-uniform output of the Allan variance curve of the inertial device, using the unweighted root mean square error of the two Allan variance curves as the evaluation index will result in a better matching of the partial curves in the short correlation time and a poor matching of the Allan variance curve in the long correlation time; therefore, the fitness evaluation function based on weight optimization can improve the global matching of the Allan variance curves of the inertial device random error simulation data and the actual inertial device data; According to the characteristics of the Allan variance curve, the Allan variance curve is fitted. The segmented fitting method is as follows: In the formula, p, q, a, b, f are fitting coefficients, x c is the curve segmentation point; The above formula can basically ensure the high-precision fitting of the Allan variance curve. Based on the above formula, the inverse proportional function is calculated to obtain the variable weight. The specific variable weight calculation formula is as follows: w = 1 / y; In theory, the Allan variance curve under the weight is: Where T(x) is the Allan variance curve under the weight effect, is the Allan variance curve of the generated data, which is theoretically close to or consistent with y. Approaching y, the final calculated target value should be near 1, that is, the error weight of each evaluation point in the evaluation function accounts for a uniform proportion; S33, select: Selection is the process of selecting superior individuals from a group to generate a new generation of individuals. Common selection methods include fitness ratio method, best individual preservation method, league selection method and exclusion method, etc. The present invention combines the above methods with the survival of the fittest in nature to design a selection method; Assuming the average fitness of the group is avgf, directly retain the fitness greater than avgf*P better Individual, P better Take 0.5-1; population retain=population last fitness>avgf*P better ; Where population retain Indicates the retained individuals, population last For individuals of the previous generation; At this time, the number of individuals in the new population is less than that of the previous generation. Considering the survival of the fittest and population richness, the individuals in the population are sorted by fitness as follows: Where population fill Represents the sorted filled individuals, population better represents the optimal individual, population last Indicates the previous generation of individuals, population worse represents the individual with the worst fitness, rand∈(0,1) is a uniformly distributed random number, P s ∈(0,1) is a parameter used to control the ratio; That is, when the uniformly distributed random number is greater than P s , from the optimal individual population better Select individuals to fill in order; when the uniformly distributed random number is less than P s , from the individual population with the worst fitness worse Select individuals in order to fill in the vacant individual position or directly select individuals from the previous generation population according to the vacant individual position number. last Filling is performed to ensure population richness and global optimization capabilities; S34, Crossover: The core of biological evolution in nature is the recombination and mutation of biological genes. Genetic algorithms use the crossover method to simulate this process to generate new individuals and improve the search ability of genetic algorithms. Common crossover methods include one-point crossover, two-point crossover and consistent crossover (using a mask word to determine which individual's corresponding gene the new individual inherits). Since multi-parameter binary cascade coding is used in the encoding, one-point crossover operations are performed for different parameter coding segments to ensure that each parameter coding segment is crossed in each crossover operation, thereby ensuring the rapidity of the search. S35, mutation: Mutation is the change of certain genes of individuals in a group. For binary coding, it is mainly the process of mutating 0 to 1 or 1 to 0. Common mutation methods include basic mutation, two-mutation point reversal mutation and adaptive mutation. The present invention adopts a random position single-point mutation method to mutate the individual coding in the group, that is, mutate 0 to 1 or mutate 1 to 0. S36. According to steps S31-S35, based on the fusion random model constructed in step S1 and the actual Allan variance curve of the inertial device obtained in step S2, the optimal solution of parameter quantization noise intensity Q, white noise intensity N, zero bias instability noise B, correlation time inverse β, random walk noise intensity K, rate ramp noise intensity R, natural frequency parameter ω0, and complex frequency s is obtained.

[0020] The method of the present invention constructs a fusion random model of random errors based on quantization noise, white noise, zero bias instability noise, random walk noise and rate ramp noise, and uses an improved genetic algorithm to optimize and determine parameters of the fusion random model. In the process of parameter optimization and determination, considering that the Allan variance result output variation range of the inertial device is in several orders of magnitude, the weightless Allan variance evaluation function will lead to general Allan variance global matching performance, and a variable weight target evaluation function is proposed, so that the global matching of the Allan variance curve between the inertial device random error simulation data and the actual inertial device data is improved, the simulation data and the real data Allan variance global matching is high, the operation is simple, and the practicability is good, and the existing Allan variance method for simulating the random characteristics of inertial devices solves the problems that the traditional Allan variance parameter fitting modeling is inaccurate because the Allan variance calculation is usually not an ideal smooth curve, and the inertial device random modeling often adopts white noise plus a first-order Markov process, which makes it difficult to accurately simulate the random characteristics of the inertial device.

[0021] Parts of the present invention that are not disclosed in detail belong to the common knowledge in the art.

[0022] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for simulating random characteristics of inertial devices based on Allan variance matching, characterized in that: It includes the following steps: S1. Based on quantization noise, white noise, bias instability noise, random walk noise, and rate ramp noise, construct a fusion random model of the random error of the inertial device; S2. Conduct a static experiment on the inertial device to be modeled, and obtain the actual Allan variance curve of the inertial device based on the data of the static experiment; S3. Based on the actual Allan variance curve of the inertial device obtained in step S2, use the improved genetic algorithm to optimize and determine the parameters of the fusion random model constructed in step S1.

2. The method for simulating random characteristics of an inertial device based on Allan variance matching according to claim 1, characterized in that: In step S1, the constructed fusion random model of the random error of the inertial device is: here=here Q +see N +see B +see K +see R ; Where δω is the random error of the inertial device, δω Q is the quantization noise in the random error of the inertial device, δω N is the white noise in the random error of the inertial device, δω B is the zero bias instability noise in the random error of the inertial device, δω K is the random walk noise in the random error of the inertial device, δω R is the rate ramp noise in the random error of the inertial device; Specifically, δω Q =Qu1(t),δω N = Nu2(t), δω K =Ku4(t), Where Q is the quantization noise intensity, N is the white noise intensity, B is the zero-bias instability noise, β is the inverse of the correlation time, K is the random walk noise intensity, R is the rate ramp noise intensity, u1(t), u2(t), u3(t), u4(t), u5(t) are mutually uncorrelated unit Gaussian white noises, ω0 is the natural frequency parameter, t is time, and s is the complex frequency; Therefore, the specifically constructed fusion random model of the random error of the inertial device is: Among them, the quantization noise intensity Q, white noise intensity N, bias instability noise B, reciprocal of the correlation time β, random walk noise intensity K, rate ramp noise intensity R, natural frequency parameter ω0, and complex frequency s are the model parameters to be optimized and determined.

3. The method for simulating random characteristics of an inertial device based on Allan variance matching according to claim 1, characterized in that: In step S2, when conducting a static experiment on the inertial device to be modeled, the specific method for obtaining the actual Allan variance curve of the inertial device based on the data of the static experiment includes: S21. Place the inertial device to be modeled statically on the platform and power it on for 2 - 3 hours for preheating; after preheating, start collecting the output data of the inertial device, and the collection time is 6 - 8 hours to obtain the static experiment data; S22. Based on the collected static experiment data, obtain the actual Allan variance curve of the inertial device.

4. The method for simulating random characteristics of an inertial device based on Allan variance matching according to claim 3, characterized in that: In step S22, the specific method for obtaining the actual Allan variance curve of the inertial device based on the collected static experiment data is: For a data set {y1,y2,…,y N }, and its Allan variance analysis modeling process is described as follows: (a) Cluster the entire data set with the correlation time τ = τ0, 2τ0,..., kτ0 (k < N / 2), each cluster contains k sampling points, and the corresponding correlation time changes from small to large; (b) Calculate the mean of each cluster and calculate the difference in the means between adjacent clusters. (c) According to the formula Calculate Allan variance σ 2 ; In the formula, σ 2 (τ) is the Allan variance corresponding to the correlation time τ, N C is the total number of clusters corresponding to correlation time τ, is the mean of the ith cluster at this time, is the mean of the i+1th cluster at this time.

5. The method for simulating random characteristics of an inertial device based on Allan variance matching according to claim 1, characterized in that: In step S3, based on the actual Allan variance curve of the inertial device obtained in step S2, the specific method for using the improved genetic algorithm to optimize and determine the parameters of the fusion random model constructed in step S1 includes: S31. Encoding and initial population generation: Use binary encoding combined with multi-parameter cascading for encoding. The specific operation is: Assume that the objective function parameters to be optimized are quantization noise intensity Q, white noise intensity N, zero bias instability noise B, correlation time inverse β, random walk noise intensity K, rate ramp noise intensity R, natural frequency parameter ω0, and complex frequency s, corresponding to x1, x2, x3, x4, x5, x6, x7, x8, and their value range is [x1 min ,x1 max ]、[x2 min ,x2 max ]、 [x3 min ,x3 max ]、[x4 min ,x4 max ]、[x5 min ,x5 max ]、[x6 min ,x6 max ]、[x7 min ,x7 max ]、[x8 min ,x8 max ], whose target parameter accuracy or optimization step size are Δx1, Δx2, Δx3, Δx4, Δx5, Δx6, Δx7, Δx8 respectively; Then the encoding lengths corresponding to x1, x2, x3, x4, x5, x6, x7, x8 are: In the formula, bit x Encode length for each variable, bit sum Encode the length for each individual; x max Indicates the corresponding maximum values ​​of x1, x2, x3, x4, x5, x6, x7, and x8; x min Indicates the minimum value of the corresponding values ​​of x1, x2, x3, x4, x5, x6, x7, and x8; Δx represents the average value of Δx1, Δx2, Δx3, Δx4, Δx5, Δx6, Δx7, Δx8; The initial population is generated randomly, the population size is popsize, and the encoding length of each individual is bit sum , the initial population generation method is expressed as: population initial =round(rand(popsize,bit sum )); Where population initial Indicates the initial population, rand(popsize,bit sum ) indicates generating popsize row bits sum Column random numbers, round means rounding the above random numbers, that is, generating a binary-coded initial population of popsize population size; S32, fitness evaluation: Combined with the objective function to be optimized and determined, use inverse proportional transformation for constraint: fitness = 1 / f(x); In the formula, f(x) represents the objective function to be optimized and determined; The specific calculation formula of the objective function to be optimized and determined is as follows: In the formula, RMSE is the root mean square error of the Allan variance curves of the simulated random data of the inertial device and the actual inertial device data, A is the data of the Allan variance curve of the actual inertial device data, F is the data of the Allan variance curve of the inertial device simulation data, and W is the weighted scheme value; According to the characteristics of the Allan variance curve, fit the Allan variance curve, and the piecewise fitting method is as follows: In the formula, p, q, a, b, f are fitting coefficients, x c is the curve segmentation point; Based on the above formula, the inverse proportional function is calculated to obtain the variable weight. The specific variable weight calculation formula is as follows: w = 1 / y; In theory, the Allan variance curve under the weight is: Where T(x) is the Allan variance curve under the weight effect, is the Allan variance curve of the generated data; S33, select: Assuming the average fitness of the group is avgf, directly retain the fitness greater than avgf*P better Individual, P better Take 0.5-1; population retain =population last fitness>avgf*P better ; Where population retain Indicates the retained individuals, population last For individuals of the previous generation; At this time, the number of individuals in the new population is less than that of the previous generation, and the individuals in the population are sorted according to fitness as follows: Where population fill Represents the sorted filled individuals, population better represents the optimal individual, population last Indicates the previous generation of individuals, population worse represents the individual with the worst fitness, rand∈(0,1) is a uniformly distributed random number, P s ∈(0,1) is a parameter used to control the ratio; S34, Crossover: Perform one-point crossover operation on different parameter coding segments respectively to ensure that each parameter coding segment is crossovered in each crossover operation; S35, mutation: The random position single-point mutation method is used to mutate the individual codes in the population, that is, 0 is mutated to 1 or 1 is mutated to 0; S36. According to steps S31-S35, based on the fusion random model constructed in step S1 and the actual Allan variance curve of the inertial device obtained in step S2, the optimal solution of parameter quantization noise intensity Q, white noise intensity N, zero bias instability noise B, correlation time inverse β, random walk noise intensity K, rate ramp noise intensity R, natural frequency parameter ω0, and complex frequency s is obtained.

Images