Sensor error extraction method based on traditional inertial navigation system

Abstract

The invention belongs to the technical field of electronic digital data processing, and particularly relates to a sensor error extraction method based on a traditional inertial navigation system, which comprises the following steps: acquiring measurement data and an error sequence of a sensor; dividing the error sequence of the sensor based on the numerical range of the error sequence of the sensor to obtain a plurality of segments of each sensor; based on the error data of each segment of each sensor, establishing an error model of each segment of each sensor; according to the fitting parameter difference of the error models of the continuous segments of the sensors and the difference between the error data and the error models of the corresponding segments, nonlinear indexes of all the sensors at all the moments are obtained; unit weights of the sensors and time weights of all moments are obtained; and based on the unit weight of the sensor and the time weight of each moment, establishing an error model of each type of sensor index, and extracting an error. According to the invention, the sensor error extraction precision and accuracy of the traditional inertial navigation system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing. More specifically, the present invention relates to a method for extracting sensor errors based on a traditional inertial navigation system. Background Art

[0002] An inertial navigation system (INS) is an autonomous navigation system that does not rely on external information and can operate in environments such as air, ground, and underwater. The INS uses inertial sensors such as accelerometers and gyroscopes to measure the acceleration and angular velocity of a carrier relative to an inertial reference frame, integrates over time, and converts to a navigation coordinate system to obtain motion parameters such as the velocity, position, and attitude of the carrier in the navigation coordinate system. The INS has strong anti-interference ability and is not affected by meteorological conditions, so it is applied to navigation fields such as land, sea, and air.

[0003] Taking a microelectromechanical system inertial measurement unit as an example, it has attracted attention and is used due to advantages such as low cost, small size, and low power consumption. However, this inertial measurement unit has large errors and complex error sources, such as zero bias error and installation error. The low accuracy caused by large errors limits the use of this inertial measurement unit in various fields such as unmanned aerial vehicles. To improve the measurement accuracy of the inertial measurement unit, on the one hand, the design method and manufacturing process are improved, but this method has high R & D costs and long R & D cycles; on the other hand, the errors of the inertial measurement unit are calibrated, an error model of the inertial device and the navigation system is established, the model parameters are determined through an estimation algorithm, and finally error compensation is achieved.

[0004] Currently, the error models of inertial measurement units are mostly in polynomial form. The precise angular position and angular rate provided by a two-axis or three-axis turntable are used, and the least squares method or Kalman filtering technology is used to estimate each error parameter of the inertial measurement unit. However, such error models ignore the time-varying characteristics of the error parameters and do not consider the non-linear characteristics of the errors, resulting in the errors gradually increasing with time accumulation when obtaining information such as the carrier velocity during the carrier movement process. Summary of the Invention

[0005] To solve the technical problem of insufficient accuracy of the error model of the above-mentioned inertial measurement unit, the present invention provides a method for extracting sensor errors based on a traditional inertial navigation system, including: Obtain measurement data of several sensors of several types, as well as reference data of sensor indicators for each type. The sensor indicators corresponding to sensors of the same type are the same; obtain the error sequence of each sensor; based on the numerical range of the error sequence of the sensor, divide the error sequence of the sensor to obtain several segments of each sensor; based on the error data of each segment of each sensor, establish an error model for each segment of each sensor, where the error model includes several fitting parameters and corresponding exponential terms; according to the difference in fitting parameters of the error models of consecutive segments of the sensor and the difference between the error data and the error model of the corresponding segment, obtain the non-linear index of each sensor at each moment; based on the non-linear index of each sensor at each moment, obtain the unit weight of the sensor; according to the difference in the non-linear index of sensors of the same type at each moment, obtain the time weight of each sensor at each moment; based on the unit weight of the sensor and the time weight at each moment, establish an error model for sensor indicators of each type; based on the error model of sensor indicators of each type, extract the error of sensor indicators of each type.

[0006] By analyzing the accuracy of all sensors in the redundant inertial navigation system, the present invention weights to obtain the error of sensor indicators, further improving the measurement accuracy of the redundant inertial navigation system. The present invention performs segmented analysis on each sensor to obtain the error model of each sensor at different times, avoiding the insufficient accuracy of sensor error detection caused by the time-varying and non-linear characteristics of the errors in the inertial navigation system.

[0007] Preferably, the step of dividing the error sequence of the sensor based on the numerical range of the error sequence of the sensor to obtain several segments of each sensor includes: denoting the error sequence of any sensor as the target error sequence, obtaining B error subsequences of the target error sequence, and obtaining the fitting polynomial of each error subsequence; based on the fitting polynomials of each error subsequence, calculate the similarity between any two adjacent error subsequences; for the b-th error subsequence , denote the similarity with the left adjacent and right adjacent error subsequences as , ; if , then incorporate the b-th error subsequence into the segment where the (b - 1)-th error subsequence is located; if , then take the b-th error subsequence as a new segment to obtain several segments of the target error sequence.

[0008] The present invention segments the error sequence of the sensor according to the range difference of the error data of the sensor at different time periods, improving the accuracy of error detection of sensor indicators.

[0009] Preferably, the calculation of the similarity between any two adjacent error subsequences includes: The fitting polynomial includes several fitting parameters; The similarity of adjacent error subsequences satisfies the expression: ; In the formula, represents the similarity between the b-th error subsequence and the (b + 1)-th error subsequence of the target error sequence; represents the number of fitting parameters; , represent the a-th fitting parameter of the fitting polynomials of the b-th and (b + 1)-th error subsequences; , represent the maximum and minimum values of the a-th fitting parameter of the fitting polynomial of the target error sequence; represents the exponential function with the natural constant as the base.

[0010] Preferably, establishing the error models for each segment of each sensor includes: performing polynomial function fitting on the error data of each segment of each sensor using the least squares method, and all fitting parameters and corresponding exponential terms constitute the error model for the corresponding segment of the corresponding sensor.

[0011] Preferably, obtaining the non-linear index of each sensor at each moment includes: denoting any moment of the error sequence of any sensor as the target moment, and denoting C segments in the left neighborhood and C segments in the right neighborhood of the target moment as the neighborhood of the target moment; obtaining the error stability of the target moment according to the difference in fitting parameters of the error models in the neighborhood of the target moment; obtaining the accuracy of the neighborhood error model of the target moment according to the difference between the error data in the neighborhood of the target moment and the error model of the corresponding segment; multiplying the error stability of the target moment by the accuracy of the neighborhood error model of the target moment, and performing negative correlation normalization to obtain the non-linear index of the target moment.

[0012] The present invention obtains the non-linear index of each sensor at each moment according to the error stability and the accuracy of the error model of the sensor, provides a basis for analyzing the non-linear change characteristics of the sensor error, and provides a foundation for further improving the error measurement accuracy of the inertial navigation system.

[0013] Preferably, the error stability of the target moment satisfies the expression: ; In the formula, represents the error stability of the target moment, represents the moment at which the target moment is located; represents the number of fitting parameters; represents the number of neighborhood segments of the target moment; , denote the a-th fitting parameter of the fitting polynomials of the (c + 1)-th and c-th neighborhood segments at the target time; denote the time length of the neighborhood at the target time; denote the exponential function with the natural constant as the base.

[0014] Preferably, the accuracy of the neighborhood error model at the target time satisfies the expression: ; In the formula, denote the accuracy of the neighborhood error model at the target time, denote the time at which the target time is located; denote the number of neighborhood segments at the target time; denote the time length of the c-th neighborhood segment at the target time; denote the shortest distance between the error value at the h-th time of the c-th neighborhood segment at the target time and the corresponding error model.

[0015] Preferably, obtaining the unit weight of the sensor based on the non-linear exponents of each sensor at each time includes: designating any sensor of any type of sensor as the target sensor; obtaining the product of the absolute value of the difference between the non-linear exponents of all adjacent times of the target sensor and the mean value of the non-linear exponents of all times of the target sensor, and performing negative correlation normalization to obtain the unit weight of the target sensor.

[0016] The present invention obtains the unit weight of the sensor, making the reference weight of the sensor with higher accuracy higher, and improving the error measurement accuracy of the redundant inertial navigation system.

[0017] Preferably, obtaining the time weight of each sensor at each time according to the difference in the non-linear exponents of the same type of sensors at each time includes: taking the ratio of the non-linear exponent of the u-th sensor of any type at the i-th time to the sum of the non-linear exponents of all sensors of this type at the i-th time as the time weight of the u-th sensor of this type at the i-th time.

[0018] The present invention attaches different time weights to different sensors according to the different measurement times of the sensors, making the error measurement of the sensors more flexible and more accurate.

[0019] Preferably, establishing the error model of each type of sensor index includes: at the i-th moment of the u-th sensor of any type of sensor index, inputting the measurement data into the corresponding segmented error model to obtain the error factor of the u-th sensor at the i-th moment; multiplying the unit weight of the u-th sensor by the time weight at the i-th moment as the comprehensive weight of the u-th sensor at the i-th moment; taking the ratio of the comprehensive weight of each sensor at the i-th moment to the sum of the comprehensive weights of all sensors at the i-th moment as the final weight of each sensor at the i-th moment; taking the final weight of each sensor at the i-th moment as the weight value, and performing weighted summation on the error factors of all sensors at the i-th moment as the error of the type of sensor index at the i-th moment; the errors at all moments together constitute the error model of the type of sensor index.

[0020] The beneficial effects of the present invention are as follows: (1) The present invention obtains sensor indexes based on multiple inertial measurement units, avoiding the problem of insufficient accuracy in obtaining sensor errors caused by using a single low-precision inertial measurement unit; (2) The present invention determines the errors of sensor indexes according to different times and different sensors, improving the application range of the sensor error model of the present invention and avoiding the problem that a single error model cannot be applied to all measurement stages; (3) The error model of the sensor index of the present invention avoids the problems of traditional error models ignoring the time-varying characteristics and non-linear characteristics of error parameters, improving the accuracy and precision of error measurement of sensor indexes. Description of the Drawings

[0021] Figure 1 is a flowchart schematically showing a method for extracting sensor errors based on a traditional inertial navigation system in the present invention; Figure 2 is a schematic diagram showing a change curve of a fitting parameter of a fitting polynomial. Detailed Embodiments

[0022] An embodiment of the present invention discloses a method for extracting sensor errors based on a traditional inertial navigation system, referring to Figure 1 , including steps S1 - S4: S1: Obtain the measurement data of several sensors of several types, as well as the reference data of each type of sensor index; obtain the error sequence of each sensor.

[0023] It should be noted that due to the limitations of sensor accuracy, low-cost microelectromechanical inertial measurement units cannot be widely applied. Therefore, redundancy techniques are often used to improve accuracy. The redundant inertial navigation system can improve the accuracy of sensor indicators by referring to the sensing data of multiple microelectromechanical inertial elements. Therefore, the present invention sets up an inertial measurement unit array, and by analyzing the sensor data of multiple inertial measurement units, an error model for each type of sensor indicator is established.

[0024] Specifically, an inertial measurement unit array composed of M rows and N columns of inertial measurement units is set up. The inertial measurement unit includes several types of sensors, such as accelerometers and gyroscopes. It should be noted that the sensors included in the inertial measurement units at different positions are the same, and each type of sensor measures one sensor indicator. The sensor indicators include acceleration, angular acceleration, etc. M and N are set by the implementer according to the actual implementation situation. For example, M and N can be set to 16.

[0025] The inertial measurement unit array and the high-precision navigation system are installed on the carrier to obtain the measurement data of each sensor during the movement of the carrier. The reference data of each type of sensor indicator is obtained through the high-precision navigation system. The measurement data of each sensor is subtracted from the reference data of the corresponding type of sensor indicator to obtain the error sequence of each sensor.

[0026] It should be noted that the high-precision navigation system is a positioning technology that can provide higher precision, such as a high-precision turntable, a high-precision inertial measurement unit, etc. The error sequence is obtained by subtracting the measurement data at the same movement moment from the reference data of the corresponding sensor indicator. The sampling rates and sampling times of the measurement data and the reference data are the same, and the sampling rate is set by the implementer according to the actual implementation situation. For example, the sampling rates of the accelerometer and the gyroscope can be set to 200 Hz.

[0027] So far, the error sequences of each sensor have been obtained.

[0028] S2: Based on the numerical range of the error sequence of the sensor, the error sequence of the sensor is divided to obtain several segments of each sensor; based on the error data of each segment of each sensor, an error model for each segment of each sensor is established. The error model includes several fitting parameters and corresponding exponential terms.

[0029] It should be noted that if the error of the sensor indicator conforms to the polynomial error model, using a polynomial function to fit the error sequence of the sensor can obtain an error model with better fitting effect. This error model is composed of several fitting parameters and corresponding exponential terms. However, considering the time-varying characteristics of the error, there may be several segments with different fitting effects in the error sequence. Therefore, the present invention first segments the error sequence and establishes an error model for each segment, so that the error model better conforms to the time-varying characteristics of the sensor data error.

[0030] It should be further noted that the movement of the carrier is continuous, and the change of the error over time is also continuous. Therefore, a window can be set to analyze the sliding process of the window on the error sequence. The fitting effect of the error data within the window to the polynomial and the change of the fitting coefficients are used to complete the segmentation.

[0031] Specifically, a sliding window with a time length of T1 is set, and the sliding window is placed at the starting point of the error sequence of each sensor and slides with a step size of T2 until the end point of the error sequence. The error data covered within the window during the sliding process of the sliding window is recorded as an error subsequence. Then, the error sequence of each sensor contains several error subsequences, and the number of error subsequences is denoted as B. It should be noted that T1 and T2 are set by the implementer according to the actual implementation situation. For example, T1 can be set to 1 minute and T2 can be set to 1 second.

[0032] Preferably, based on the numerical range of the error sequence of the sensor, the error sequence of the sensor is divided to obtain several segments of each sensor: The error sequence of any sensor is denoted as the target error sequence. The least squares method is used to fit the polynomial function to each error subsequence of the target error sequence to obtain the fitting polynomial of each error subsequence. The fitting polynomial contains several fitting parameters and corresponding exponential terms. It should be noted that the number of terms of the polynomial function is set by the implementer according to the actual implementation situation. For example, if the number of terms of the polynomial function is set to 4, the polynomial function includes a cubic term, a quadratic term, a linear term, and a constant term.

[0033] Figure 2 is a schematic diagram of the change curve of a fitting parameter of the fitting polynomial. It should be noted that when the fitting parameters of the fitting polynomials of adjacent error subsequences are similar, the adjacent error subsequences can be merged.

[0034] The similarity of adjacent error subsequences satisfies the expression: ; In the formula, represents the similarity between the b-th error subsequence and the (b + 1)-th error subsequence of the target error sequence; represents the number of fitting parameters; , represent the a-th fitting parameter of the fitting polynomials of the b-th and (b + 1)-th error subsequences; , represent the maximum and minimum values of the a-th fitting parameter of the fitting polynomial of the target error sequence; denotes the exponential function with the natural constant as the base. It should be noted that to ensure the uniformity of the number of fitting parameters and facilitate the corresponding fitting parameters for different error subsequences, the fitting parameter of 0 should also participate in the statistics and calculations.

[0035] In the formula, denotes the difference between the a-th fitting parameter of the fitting polynomials of the b-th and the (b + 1)-th error subsequences of the target error sequence. The larger this value is, the lower the similarity of the a-th fitting parameter of the fitting polynomials of the b-th and the (b + 1)-th error subsequences. denotes the normalization of the difference of the a-th fitting parameter, to avoid the influence of a single exponential term on the similarity of adjacent error subsequences being too large due to the large range of fitting parameters. denotes the sum of the differences of all fitting parameters of the fitting polynomials of adjacent error subsequences, representing the total difference of the fitting polynomials of adjacent error subsequences. The larger this value is, the lower the similarity of adjacent error subsequences.

[0036] Merge the error subsequences of the target error sequence in the order of the error subsequences: for the b-th error subsequence (1 < b < B), if its similarity with the left adjacent error subsequence is greater than or equal to its similarity with the right adjacent error subsequence, then incorporate the b-th error subsequence into the (b - 1)-th error subsequence to form a segment. If its similarity with the left adjacent error subsequence is less than its similarity with the right adjacent error subsequence, then take the b-th error subsequence as a new segment, obtaining several segments of the target error sequence.

[0037] So far, the merging of the error subsequences of the error sequences of all sensors has been completed, and several segments of each sensor have been obtained.

[0038] Preferably, based on the error data of each segment of each sensor, establish the error model of each segment of each sensor: For the error data of each segment of each sensor, use the least squares method to perform polynomial function fitting. All fitting parameters and the corresponding exponential terms constitute the error model of the corresponding segment of the corresponding sensor.

[0039] So far, the error models of each segment of each sensor have been obtained.

[0040] S3: Based on the differences in the fitting parameters of the continuous segmented error models of the sensors, and the differences between the error data and the error models of the corresponding segments, obtain the non - linear indices of each sensor at each moment; based on the non - linear indices of each sensor at each moment, obtain the unit weights of the sensors; according to the differences in the non - linear indices of the same - type sensors at each moment, obtain the time weights of each sensor at each moment; based on the unit weights of the sensors and the time weights at each moment, establish the error models of the indicators of each type of sensor.

[0041] It should be noted that in S2, the error sequence of the sensor is segmented. Although the phased error model can improve the accuracy of the error model, due to the influence of temperature changes and random noise of the sensor during the movement of the carrier, the error has the characteristic of non - linear change. This leads to the change of fitting parameters in some segments due to the non - linear change of the error. In order to more accurately construct the error model of the sensor, the non - linear change part needs to be extracted. When there are large changes in the continuous segmented error models of the sensors and the effect of the error model is not good, it indicates that the error of the corresponding segment has strong non - linear characteristics. Although the same - type sensors measure the same sensor indicators, due to different manufacturing processes and different precisions, for the non - linear characteristics of different segments, the traditional redundant inertial navigation system averages the sensor data of all inertial measurement units as the value of the corresponding sensor indicator, while the present invention assigns higher weights to the sensors with weaker non - linear characteristics, which can improve the accuracy of the sensors.

[0042] Specifically, denote any moment of the error sequence of any sensor as the target moment, and denote the C segments in the left neighborhood and the C segments in the right neighborhood of the target moment as the neighborhood of the target moment. It should be noted that the number of segmented segments C in the neighborhood is set by the implementer according to the actual implementation situation. For example, C is preset to 3.

[0043] It should be noted that the longer the time length of the neighborhood of the target moment and the smaller the difference in the fitting parameters of the error models of the segments within the neighborhood, the more stable the error model of the sensor in the corresponding time period.

[0044] Preferably, the error stability of the target moment: ; In the formula, represents the error stability of the target moment, represents the moment at which the target moment is located; represents the number of fitting parameters; represents the number of segmented segments in the neighborhood of the target moment; 、 represent the a - th fitting parameter of the fitting polynomials of the (c + 1) - th and c - th neighborhood segments of the target moment; represents the time length of the neighborhood of the target moment; represents the exponential function with the natural constant as the base.

[0045] In the formula, represents the difference between the a-th fitting parameter of the fitting polynomials of the (c + 1)-th and c-th neighborhood segments at the target time, represents the sum of the differences between all fitting parameters of the fitting polynomials of all adjacent neighborhood segments at the target time, reflecting the difference of the fitting polynomials of the neighborhood segments at the target time. The larger this value is, the lower the error stability at the target time; at the same time, the time length of the neighborhood at the target time The longer it is, the more stable the error of the neighborhood at the target time is.

[0046] It should be noted that the greater the distance between the error data in the neighborhood at the target time and the error model, the worse the fitting effect of the error model, and the lower the accuracy of the neighborhood error model at the target time.

[0047] Preferably, the accuracy of the neighborhood error model at the target time satisfies the expression: ; In the formula, represents the accuracy of the neighborhood error model at the target time, represents the time at which the target time is located; represents the number of neighborhood segments at the target time; represents the time length of the c-th neighborhood segment at the target time; represents the shortest distance between the error value at the h-th moment of the c-th neighborhood segment at the target time and the corresponding error model; represents the exponential function with the natural constant as the base.

[0048] In the formula, represents the average of the shortest distances from the error values at all moments of the c-th neighborhood segment at the target time to the error model, representing the fitting loss of the error model. The larger this value is, the greater the fitting loss of the error model and the worse the fitting effect; represents the sum of the fitting losses of the error models of all neighborhood segments at the target time. The larger this value is, the greater the total fitting loss, and further represents the lower accuracy of the neighborhood error model at the target time.

[0049] It should be noted that the lower the error stability at the target time and the lower the accuracy of the neighborhood error model, the less the error at the target time conforms to the linear characteristic, and thus the higher the non-linear index. Therefore, based on the error stability at the target time and the accuracy of the neighborhood error model at the target time, the non-linear index at the target time is obtained.

[0050] Preferably, the non-linear index at the target time satisfies the expression: ; In the formula, represents the non - linear exponent at the target time, represents the time when the target time is located; represents the error stability at the target time, represents the accuracy of the neighborhood error model at the target time.

[0051] Thus, the non - linear exponents of each sensor at each time are obtained.

[0052] It should be noted that if the non - linear exponents of each sensor at each time are relatively low and stable, the accuracy of the sensor is higher, so the referenceability of the sensor data is higher, and thus the unit weight of the sensor is greater.

[0053] Preferably, the unit weight of the sensor satisfies the expression: ; In the formula, represents the unit weight of the u - th sensor of any type; represents the number of sensors of the said type; represents the mean value of the non - linear exponents of all times of the u - th sensor of the said type; represents the time length of the error sequence; , represents the non - linear exponents of the (i + 1) - th and i - th times of the u - th sensor of the said type.

[0054] In the formula, represents the difference of the non - linear exponents of all adjacent times of the u - th sensor of the said type. The larger this value is, the more unstable the non - linear exponent of the sensor is, so the unit weight of the corresponding sensor is smaller; the larger the mean value of the non - linear exponents of all times of the u - th sensor of the said type is, the stronger the overall non - linear exponent performance of the sensor is, and the worse the overall accuracy is, so the unit weight of the sensor is lower.

[0055] Thus, the unit weights of each sensor of each type are obtained.

[0056] It should be noted that a sensor with a high unit weight does not mean high accuracy at all times. By comparing the non - linear exponents of different sensors of the same type at each time, it can be more accurate and more applicable to the inertial measurement unit. Therefore, according to the differences of the non - linear exponents of the same - type sensors at each time, the time weights of each sensor at each time are obtained in the present invention.

[0057] Preferably, the time weights of any time of each sensor of each type satisfy the expression: ; Wherein, represents the time weight of the u-th sensor of any type at the i-th moment; represents the non-linear exponent of the u-th sensor of the said type at the i-th moment; represents the number of sensors of the said type.

[0058] Thus, the time weights of each sensor of each type at each moment are obtained.

[0059] It should be noted that for any sensor index at any moment, an error factor can be obtained from the error models of all corresponding type sensors. By weighting all the error factors according to the unit weight and time weight of the sensors, the error of the said sensor index can be formed. The errors of the said sensor index at all moments together constitute the error model of the sensor index.

[0060] Preferably, based on the unit weight of the sensor and the time weights at each moment, an error model of each type of sensor index is established: For any moment of all sensors of any type of sensor index, the measurement data is input into the corresponding segmented error model to obtain the error factors of all sensors at the said moment; The errors of each type of sensor index at each moment satisfy the expression: ; Wherein, represents the error of the i-th moment of any type of sensor index; represents the number of sensors of the said type of sensor index; represents the unit weight of the u-th sensor of the said type of sensor index; represents the time weight of the u-th sensor of the said type of sensor index at the i-th moment; represents the error factor of the u-th sensor of the said type of sensor index at the i-th moment.

[0061] Wherein, represents the comprehensive weight of the u-th sensor at the i-th moment; represents normalizing the comprehensive weight of the u-th sensor at the i-th moment to ensure that the sum of the comprehensive weights of all sensors is equal to 1; represents weighted summation of the error factors of all sensors to form the error of the said type of sensor index at the i-th moment.

[0062] The errors of all moments of any type of sensor index together constitute the error model of the said type of sensor index.

[0063] Thus, the error models of each type of sensor index are obtained.

[0064] S4: Extract the errors of each type of sensor index based on the error models of each type of sensor index.

[0065] During the real-time movement of the carrier, obtain the real-time measurement data of each sensor of each type through the inertial measurement unit array, input it into the error model of the corresponding sensor index, obtain the errors of each type of sensor index, and complete the error extraction of the sensor.

[0066] So far, the error extraction of the sensors in the traditional inertial navigation system has been completed.

[0067] Although this specification has shown and described multiple embodiments of the present invention, it is obvious to those skilled in the art that such embodiments are provided only by way of example. Those skilled in the art will think of many changes, alterations, and alternative ways without departing from the spirit and concept of the present invention.

[0068] The above are all preferred embodiments of the present invention. Without limiting the protection scope of the present invention accordingly, therefore: All equivalent changes made according to the structure, shape, and principle of the present invention shall be covered within the protection scope of the present invention.

Claims

1. A sensor error extraction method based on a traditional inertial navigation system, characterized in that: include: Obtain measurement data of several sensors of several types, as well as reference data of indicators of each type of sensor, where the sensor indicators corresponding to sensors of the same type are the same; obtain the error sequence of each sensor; Based on the numerical range of the error sequence of the sensor, the error sequence of the sensor is divided to obtain several segments of each sensor; based on the error data of each segment of each sensor, an error model of each segment of each sensor is established, wherein the error model includes several fitting parameters and corresponding exponential terms; According to the difference in fitting parameters of the error model of the sensor's continuous segments, the difference between the error data and the error model of the corresponding segments, the nonlinear index of each sensor at each moment is obtained; Based on the nonlinear index of the sensor at each moment, the unit weight of the sensor is obtained; based on the difference in the nonlinear index of the same type of sensors at each moment, the time weight of each sensor at each moment is obtained; based on the unit weight of the sensor and the time weight at each moment, the error model of each type of sensor indicator is established; Based on the error models of various types of sensor indicators, the errors of various types of sensor indicators are extracted.

2. The sensor error extraction method based on the traditional inertial navigation system according to claim 1 is characterized in that: The error sequence of the sensor is divided based on the numerical range of the error sequence of the sensor to obtain a plurality of segments of each sensor, including: The error sequence of any sensor is recorded as the target error sequence, B error subsequences of the target error sequence are obtained, and the fitting polynomial of each error subsequence is obtained; based on the fitting polynomial of each error subsequence, the similarity of any two adjacent error subsequences is calculated; For Error subsequence , and the similarity with the left adjacent and right adjacent error subsequences is recorded as , ;like , then the b-th error subsequence is merged into the segment where the b-1-th error subsequence is located; if , then the bth error subsequence is taken as a new segment to obtain several segments of the target error sequence.

3. The sensor error extraction method based on the traditional inertial navigation system according to claim 2 is characterized in that: The calculating the similarity of any two adjacent error subsequences comprises: The fitting polynomial contains several fitting parameters; The similarity of adjacent error subsequences satisfies the expression: ； In the formula, Represents the similarity between the bth error subsequence and the b+1th error subsequence of the target error sequence; represents the number of fitting parameters; , represents the ath fitting parameter of the fitting polynomial of the bth and b+1th error subsequences; , The maximum and minimum values ​​of the ath fitting parameter of the fitting polynomial representing the target error sequence; Represents an exponential function with a natural constant as its base.

4. The sensor error extraction method based on the traditional inertial navigation system according to claim 1 is characterized in that: The step of establishing the error model of each segment of each sensor includes: The error data of each segment of each sensor is fitted with a polynomial function using the least squares method, and all fitting parameters and corresponding exponential terms constitute the error model of the corresponding segment of the corresponding sensor.

5. The sensor error extraction method based on the traditional inertial navigation system according to claim 1 is characterized in that: The obtaining of the nonlinear index of each sensor at each moment includes: Any moment of the error sequence of any sensor is recorded as the target moment, and the C segments of the left neighborhood and the C segments of the right neighborhood of the target moment are recorded as the neighborhood of the target moment; According to the difference in fitting parameters of the error model in the neighborhood of the target time, the error stability at the target time is obtained; according to the difference between the error data in the neighborhood of the target time and the error model of the corresponding segment, the accuracy of the neighborhood error model at the target time is obtained; The error stability at the target time is multiplied by the accuracy of the neighborhood error model at the target time and normalized by negative correlation to obtain the nonlinear index at the target time.

6. The sensor error extraction method based on the traditional inertial navigation system according to claim 5 is characterized in that: The error stability at the target time satisfies the expression: ； In the formula, represents the error stability at the target time, Indicates the time at which the target moment is located; represents the number of fitting parameters; Indicates the number of neighborhood segments at the target time; , represents the ath fitting parameter of the fitting polynomial of the c+1th and cth neighborhood segments at the target time; represents the time length of the neighborhood of the target moment; Represents an exponential function with a natural constant as its base.

7. The sensor error extraction method based on the traditional inertial navigation system according to claim 5 is characterized in that: The accuracy of the neighborhood error model at the target time satisfies the expression: ； In the formula, represents the accuracy of the neighborhood error model at the target time, Indicates the time at which the target moment is located; Indicates the number of neighborhood segments at the target time; Represents the time length of the cth neighborhood segment at the target time; Indicates the shortest distance between the error value of the cth neighborhood segment at the target time at the hth moment and the corresponding error model.

8. The sensor error extraction method based on the traditional inertial navigation system according to claim 1 is characterized in that: The step of obtaining the unit weight of the sensor based on the nonlinear index of the sensor at each moment includes: Any sensor of any type of sensor is recorded as a target sensor; the absolute value of the difference of the nonlinear index of the target sensor at all adjacent moments is obtained, multiplied by the mean of the nonlinear index of the target sensor at all moments, and negative correlation normalization is performed to obtain the unit weight of the target sensor.

9. The sensor error extraction method based on the traditional inertial navigation system according to claim 1 is characterized in that: The step of obtaining the time weight of each sensor at each moment according to the difference of nonlinear indexes of sensors of the same type at each moment includes: The nonlinear index of the u-th sensor of any type at the i-th moment is compared with the sum of the nonlinear indexes of all sensors of the type at the i-th moment, as the time weight of the u-th sensor of the type at the i-th moment.

10. The sensor error extraction method based on the traditional inertial navigation system according to claim 1, characterized in that: The error model of each type of sensor indicator is established, including: For the i-th moment of the u-th sensor of any type of sensor indicator, the measurement data is input into the error model of the corresponding segment to obtain the error factor of the i-th moment of the u-th sensor; Multiply the unit weight of the u-th sensor by the time weight at the i-th moment to obtain the comprehensive weight of the u-th sensor at the i-th moment; compare the comprehensive weight of each sensor at the i-th moment with the sum of the comprehensive weights of all sensors at the i-th moment to obtain the final weight of each sensor at the i-th moment; Taking the final weight of each sensor at the i-th moment as the weight, the error factors of all sensors at the i-th moment are weighted summed as the error of the type sensor indicator at the i-th moment; the errors at all moments together constitute the error model of the type sensor indicator.

Images