Temperature compensation model training method, application method, electronic equipment and storage medium
The temperature compensation model established by the adaptive neural fuzzy inference system (ANFIS) solves the problem of poor zero deviation compensation effect of MEMS inertial measurement unit, significantly improving navigation accuracy.
Patent Information
- Application Number
- CN202410927974.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-11
- Publication Date
- 2024-11-19
AI Technical Summary
In the prior art, the temperature compensation model of the MEMS inertial measurement unit has poor compensation effect on zero deviation error, resulting in a decrease in navigation accuracy.
The temperature compensation model is established by using the adaptive neural fuzzy inference system (ANFIS). The temperature data and temperature gradient are deeply learned through the fuzzification layer, excitation layer, normalization layer and output layer, and the model is trained to improve the compensation effect of zero-bias error .
Through deep learning of the ANFIS model, the temperature compensation effect of the MEMS inertial measurement unit is significantly improved and the navigation accuracy is improved.
Smart Images

Figure CN118982048A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of navigation technology, and in particular to a temperature compensation model training method, application method, electronic device and storage medium. Background Art
[0002] With the rapid development of Micro Electromechanical System (MEMS) technology, MEMS Inertial Measurement Unit (IMU) has been widely used in navigation fields such as vehicles, drones, and robots due to its advantages of small size, light weight, and low cost. However, MEMS IMU is temperature sensitive. When the temperature changes, its zero bias error parameters will change significantly, resulting in a sharp decline in the performance of the combined navigation system of the Global Navigation Satellite System (GNSS) and Inertial Navigation System (INS) using MEMS IMU.
[0003] In the prior art, in order to improve the problem that the MEMSIMU is sensitive to temperature and thus causes a decrease in navigation accuracy, the MEMSIMU is temperature calibrated before leaving the factory to obtain temperature compensation parameters. A temperature compensation model is established based on the obtained temperature compensation parameters using a polynomial fitting method, and the temperature compensation model is used to compensate for the zero bias error of the MEMSIMU.
[0004] However, the existing temperature compensation model does not have a good effect on compensating zero bias error. Summary of the invention
[0005] The present application provides a temperature compensation model training method, an application method, an electronic device and a storage medium to solve the problem of inaccuracy in establishing a temperature compensation model for an IMU using a traditional polynomial fitting method, thereby improving the IMU temperature compensation effect.
[0006] In a first aspect, the present application provides a temperature compensation model training method, comprising: obtaining a plurality of first sample data and labels of the plurality of first sample data, each of the first sample data comprising: first temperature data, a first temperature gradient corresponding to the first temperature data; the label of the first sample data is a zero bias error of an inertial measurement unit IMU determined by a combined navigation system under the first temperature data; establishing an initial adaptive neuro-fuzzy inference system (Adaptive Neuro-Fuzzy Inference The invention discloses an ANFIS zero-bias temperature compensation model; the initial ANFIS zero-bias temperature compensation model comprises: a fuzzification layer, an excitation layer, a normalization layer, a conclusion layer and an output layer connected in sequence; the fuzzification layer comprises a plurality of fuzzification processing nodes, each of which is used to perform fuzzification processing on the plurality of first sample data inputted according to at least one membership function set in advance to obtain membership results corresponding to the plurality of first sample data respectively; the membership results comprise at least one membership, the at least one membership corresponds to the at least one membership function in a one-to-one manner, and the membership is used to indicate the probability that the first sample data satisfies each fuzzy set; the excitation layer is used to perform a T-normal form operator on the membership data corresponding to the plurality of first sample data outputted by the fuzzification layer respectively, to obtain the excitation intensity corresponding to at least one rule respectively. degree; the rule is used to indicate the zero bias temperature compensation strategy; the normalization layer is used to normalize the excitation intensity corresponding to the at least one rule, and obtain the normalized excitation intensity corresponding to the at least one rule; the conclusion layer is used to process the normalized excitation intensity corresponding to the at least one rule, and obtain the final zero bias error result corresponding to the at least one rule; the output layer is used to fuse the multiple intermediate zero bias error results output by the conclusion layer, and obtain the estimated final zero bias error result; the multiple first sample data are input into the initial ANFIS zero bias temperature compensation model, and the initial ANFIS zero bias temperature compensation model is trained until the model convergence condition is met to obtain the target ANFIS zero bias temperature compensation model; the target ANFIS zero bias temperature compensation model is used to perform zero bias compensation on the output parameters of the IMU.
[0007] In some embodiments, the multiple first sample data are input into an initial ANFIS zero-bias temperature compensation model, and the initial ANFIS zero-bias temperature compensation model is trained based on the labels corresponding to the multiple first sample data, including: inputting the multiple first sample data into the initial ANFIS zero-bias temperature compensation model to obtain the first output data of the initial ANFIS model; the first output data includes the output data of the normalization layer of the ANFIS and the conclusion layer of the ANFIS; using the fading adaptive Kalman filtering algorithm and the first output data, the first conclusion parameter based on which the conclusion layer is constructed is updated to the second conclusion parameter; based on the second conclusion parameter, the initial ANFIS zero-bias temperature compensation model is updated to obtain the target ANFIS zero-bias temperature compensation model.
[0008] In some embodiments, the updating of the initial ANFIS zero bias temperature compensation model based on the second conclusion parameter to obtain a target ANFIS zero bias temperature compensation model includes: performing a first update on the initial ANFIS zero bias temperature compensation model based on the second conclusion parameter to obtain a first ANFIS zero bias temperature compensation model; determining whether a training error of the first ANFIS zero bias temperature compensation model satisfies a model convergence condition; if the model convergence condition is satisfied, using the first ANFIS zero bias temperature compensation model as a target ANFIS zero bias temperature compensation model; if the model convergence condition is not satisfied, using a gradient descent algorithm and the second conclusion parameter to update the first premise parameter based on which the membership function is constructed to a second premise parameter; the first premise parameter is a training error of the first ANFIS zero bias temperature compensation model using fuzzy C-means clustering (Fuzzy The method is to train the second sample data using the ANFIS zero bias temperature compensation model by using the ANFIS C-Means Algorithm (FCM) algorithm; wherein, the first sample data and the second sample data are obtained at different times; the second conclusion parameter is updated to a third conclusion parameter using the fading adaptive Kalman filtering algorithm and the first output data; based on the second premise parameter and the third conclusion parameter, the first ANFIS zero bias temperature compensation model is updated at least once until the training error of the updated ANFIS zero bias temperature compensation model meets the model convergence condition, thereby obtaining the target ANFIS zero bias temperature compensation model.
[0009] In some embodiments, the use of the gradient descent algorithm and the second conclusion parameter to update the first premise parameter based on which the membership function is constructed to the second premise parameter includes: based on the training error of the initial ANFIS zero-bias temperature compensation model, the first premise parameter, the output data of each fuzzification processing node of the fuzzification layer, and the formula:
[0010] Obtaining a first output error rate corresponding to the fuzzification layer;
[0011] in, is the first output error rate; E k is the training error, v j,i and is the first premise parameter; 1,2(j-1)+i is the output data of the 2*(j-1)+ith fuzzy processing node of the fuzzy layer; wherein j is greater than 0 and less than or equal to the number of the first sample data; i is greater than 0 and less than the number of the rules;
[0012] Based on the training error of the initial ANFIS zero-bias temperature compensation model, the first premise parameter, the output data and formula of each fuzzy processing node of the fuzzy layer:
[0013]
[0014] Obtaining a second output error rate corresponding to the fuzzification layer; is the second output error rate; E k is the training error, v j,i and is the first premise parameter; 1,2(j-1)+i It is the output data of the 2*(j-1)+ith fuzzy processing node of the fuzzy layer; wherein, j is greater than 0 and less than or equal to the number of first sample data; i is greater than 0 and less than the number of rules; based on the first output error rate, the second output error rate and the preset learning rate, the first premise parameter is updated to the second premise parameter.
[0015] In some embodiments, based on the first output error rate, the second output error rate and a preset learning rate, updating the first premise parameter to the second premise parameter includes: according to the first output error rate and the preset learning rate, using the formula: Update the first parameter in the first premise parameter; where v on the right side of the equal sign j,i is the first parameter before updating; the v on the left side of the equal sign j,i is the updated first parameter; η is the preset learning rate; according to the second output error rate and the preset learning rate, using the formula: Update the second parameter in the first premise parameter; where σ on the right side of the equal sign j,i is the first parameter before updating; σ on the left side of the equal sign j,i is the updated first parameter; η is the preset learning rate; according to the updated first parameter and the second parameter, the second premise parameter is obtained.
[0016] In some embodiments, obtaining multiple first sample data includes: performing temperature measurement operations and determining zero bias errors when the combined navigation system performs a navigation task; storing the measured first temperature data, and the first temperature gradient determined based on the first temperature data and the zero bias error under the first temperature data in a storage subunit corresponding to a temperature node to which the first temperature data belongs; wherein the IMU includes a first storage unit, and the first storage unit includes multiple storage subunits; the first temperature data belongs to a preset temperature range, and the preset temperature range includes multiple temperature nodes, and the multiple temperature nodes correspond one-to-one to the multiple storage subunits in the first storage unit; when the number of temperature nodes to which all the collected first temperature data belongs is greater than a temperature node number threshold, obtaining multiple first temperature data, corresponding first temperature gradients and corresponding zero bias errors from the first storage unit; wherein the multiple first sample data include the multiple first temperature data and the corresponding first temperature gradients.
[0017] In a second aspect, the present application provides a temperature compensation model application method, comprising: when the integrated navigation system performs a navigation task, obtaining third temperature data currently measured; inputting the third temperature data and a third temperature gradient determined based on the third temperature data as third sample data into the target ANFIS zero bias temperature compensation model described in the first aspect, and obtaining a target zero bias error output by the target ANFIS zero bias temperature compensation model; based on the target zero bias error, performing zero bias compensation on the output parameters of the inertial measurement unit IMU of the integrated navigation system.
[0018] In a third aspect, the present application provides a temperature compensation processing device, comprising a module for executing the method described in the first aspect or the second aspect.
[0019] In a fourth aspect, the present application provides a chip comprising an interface circuit and a logic circuit, wherein the interface circuit is used to receive signals from other chips outside the chip and transmit them to the logic circuit, or to send signals from the logic circuit to other chips outside the chip, and the logic circuit is used to implement the method described in the first aspect or the second aspect.
[0020] In a fifth aspect, the present application provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect or the second aspect.
[0021] In a sixth aspect, the present application provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and wherein the processor implements the method described in the first aspect or the second aspect when executing the program.
[0022] In the present application, in the scenario of performing zero bias compensation on the output parameters of the IMU in the integrated navigation system, an initial ANFIS zero bias temperature compensation model is first established, and the initial ANFIS zero bias temperature compensation model is trained based on multiple first temperature data, corresponding first temperature gradients and zero bias errors until the model convergence conditions are met, and a target ANFIS zero bias temperature compensation model is obtained, and zero bias compensation is performed based on the target ANFIS zero bias temperature compensation model; wherein, the ANFIS zero bias temperature compensation model based on which the zero bias compensation of the output parameters of the IMU is performed integrates a multi-layer neural network and a fuzzy reasoning system, and can perform deep learning on the relationship between the first temperature data, the first temperature gradient and the corresponding zero bias error, thereby improving the compensation effect. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1 A schematic diagram of the implementation flow of the temperature compensation model training method provided in the embodiment of the present application;
[0024] Figure 2 A schematic diagram of the structure of the initial ANFIS zero-bias temperature compensation model provided in an embodiment of the present application;
[0025] Figure 3 A schematic diagram of the implementation flow of the temperature compensation model training method provided in the embodiment of the present application;
[0026] Figure 4 A schematic diagram of the implementation flow of the temperature compensation model training method provided in the embodiment of the present application;
[0027] Figure 5 A schematic diagram of the implementation flow of the temperature compensation model training method provided in the embodiment of the present application;
[0028] Figure 6 A schematic diagram of the implementation flow of the temperature compensation method provided in the embodiment of the present application;
[0029] Figure 7 A schematic diagram of the structure of a temperature compensation model training device provided in an embodiment of the present application;
[0030] Figure 8 A schematic diagram of the structure of a temperature compensation processing device provided in an embodiment of the present application;
[0031] Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0032] In this application, "at least one" means one or more, and "plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c alone can represent: a, b, c, a and b, a and c, or a, b, and c, where a, b, and c can be single or multiple. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.
[0033] The directions or positional relationships indicated by terms such as "center", "longitudinal", "lateral", "up", "down", "left", "right", "front", and "back" are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present application and simplifying the description. They do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as a limitation on the present application.
[0034] The terms "connected" and "connected" should be understood in a broad sense. For example, the "connected" or "connected" of a circuit structure can refer to not only physical connection, but also electrical connection or signal connection. For example, it can be directly connected, that is, physically connected, or indirectly connected through at least one intermediate element, as long as the circuit is connected, or it can be the internal connection of two elements; signal connection can refer to signal connection through a circuit or through a media medium, such as radio waves. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.
[0035] The following introduces some concepts involved in the embodiments of the present application:
[0036] IMU: Inertial Measurement Unit, which consists of a gyroscope and an accelerometer and can measure the angular velocity and acceleration of a moving object.
[0037] Inertial Navigation System (INS) is a system that uses IMU measurement data for navigation. It can calculate the speed and position of an object in the navigation coordinate system based on the IMU module.
[0038] The Global Navigation Satellite System (GNSS) uses measured satellite data for navigation and positioning.
[0039] GNSS / INS integrated navigation system: integrates INS and GNSS, and comprehensively utilizes the advantages of both to obtain accurate results of position, speed and attitude.
[0040] In the prior art, the temperature of the MEMS IMU needs to be calibrated before leaving the factory to obtain the temperature compensation parameters corresponding to the zero bias error. The conventional temperature calibration method is carried out in a temperature chamber, collecting MEMS IMU data at a series of temperature nodes at a certain temperature interval, and then obtaining the error parameters at each temperature value through interpolation or fitting. In order to obtain more accurate temperature error parameters, the selected temperature nodes need to be as dense as possible, at least one temperature collection point needs to be selected every five degrees, and each temperature node needs to be kept at a constant temperature for a long time, which requires a lot of manpower and material resources, and greatly increases the cost of MEMS IMU, thereby limiting the application of MEMS IMU to a certain extent.
[0041] Conventional MEMSIMU temperature compensation models mostly use polynomial fitting methods, such as third-order polynomials, fourth-order polynomials, etc. However, the polynomial fitting method cannot perform deep learning based on the input sample data and the corresponding zero bias error. The zero bias error obtained by fitting is not accurate enough, resulting in poor temperature compensation effect.
[0042] In order to solve the problem that the zero bias error obtained by establishing the MEMSIMU temperature compensation model using the traditional polynomial fitting method is not accurate enough, the present application provides a temperature compensation model training method, establishes an ANFIS zero bias temperature compensation model, uses temperature data and temperature gradient, and the MEMSIMU zero bias error estimated in real time by the navigation system to train the ANFIS zero bias temperature compensation model, and compensates the output parameters of the IMU based on the trained ANFIS zero bias temperature compensation model. Since ANFIS combines the advantages of neural networks and fuzzy reasoning systems, it can deeply learn the relationship between the first temperature data, the first temperature gradient and the corresponding zero bias error, thereby improving the temperature compensation effect.
[0043] The temperature compensation model establishment method provided in the embodiment of the present application is suitable for electronic devices with positioning and navigation functions, and the electronic devices are positioned based on a GNSS / INS combined navigation system.
[0044] Figure 1 A schematic diagram of the implementation flow of the temperature compensation model establishment method provided in the embodiment of the present application is shown in FIG. Figure 1 As shown, the temperature compensation model establishment method includes:
[0045] S101. Acquire a plurality of first sample data and a plurality of labels of the first sample data, each of the first sample data comprising: first temperature data and a first temperature gradient corresponding to the first temperature data.
[0046] The label of the first sample data is the zero bias error of the IMU determined by the integrated navigation system under the first temperature data.
[0047] In the embodiment of the present application, the first sample data is obtained in real time when the electronic device performs a navigation task.
[0048] Specifically, the electronic device collects first temperature data in real time when performing a navigation task, and determines a first temperature gradient and a zero bias error of the IMU based on the first temperature data.
[0049] In the first sample data, the temperature range of the first temperature data is from a to b, wherein a is greater than or equal to the first threshold value, and b is less than or equal to the second threshold value; illustratively, the first threshold value is -10°C, and the second threshold value is 70°C. In other words, the range of the first temperature data is from -10°C to 70°C.
[0050] In some embodiments, the electronic device can set multiple temperature nodes in the temperature range, and the temperature interval between each two temperature nodes is ΔT. ΔT can be set according to actual needs. For example, for application scenarios with high positioning accuracy requirements, ΔT is 0.5°C, and for application scenarios with low positioning accuracy requirements, ΔT is 1°C.
[0051] Exemplarily, ΔT is 1°C, and the temperature range is -10°C to 70°C, then the number of temperature nodes is (70+10) / 1+1=81, namely -10°C, -9°C, ..., 69°C, and 70°C.
[0052] The electronic device comprises a first storage unit, which comprises a plurality of storage subunits. The plurality of temperature nodes correspond to the plurality of storage subunits one by one, and the storage subunits are used to store data related to the corresponding temperature nodes.
[0053] Among them, a small temperature threshold can be set on the left and right sides of each temperature node. If the absolute value of the difference between the measured first temperature data and the temperature node is less than the temperature threshold, the first temperature data, the temperature gradient corresponding to the first temperature data, and the zero bias error corresponding to the first temperature data can be stored in the storage subunit corresponding to the temperature node.
[0054] Exemplarily, an example is given to illustrate how to store the first sample data. Assuming that the temperature threshold is 0.1°C, the first temperature data collected is 69.95°C, and the absolute value of the difference between 69.95 and 70 is less than 0.1, the first temperature data, the first temperature gradient corresponding to the first temperature data, and the zero bias error corresponding to the first temperature data are stored in the storage subunit corresponding to the temperature node of 70°C.
[0055] In some embodiments, if there is no temperature node corresponding to the collected first temperature data in the electronic device, the electronic device does not store the first temperature data.
[0056] Exemplarily, when the first temperature data collected by the electronic device is 69.8° C., and there is no temperature node in the preset range such that the absolute value of the difference between the temperature node and the first temperature data is less than or equal to 0.1, the electronic device does not store the first temperature data.
[0057] It can be understood that the first temperature data in the multiple first sample data are collected at different times. After the electronic device collects the first temperature data, it calculates the first temperature gradient and the zero bias error based on the first temperature data and the time when the first temperature data was collected.
[0058] Specifically, the first temperature gradient corresponding to the first temperature data is calculated as shown in the following formula 1:
[0059] K k =(T k -T k-1 ) / (t k -t k-1 )Formula 1;
[0060] In formula 1, T k , t k The first temperature data and time data are collected at the current k moment respectively; T k-1 , t k-1 They are the first temperature data and time data of the previous temperature node that have been collected,
[0061] The label of the first sample data, that is, the zero bias error, can be obtained by the Kalman filter algorithm as follows:
[0062] After the INS has completed the initial alignment, the Kalman filter algorithm is used to estimate the above-mentioned errors of the INS, including the error of the zero bias error, using the error parameters of the INS position error, velocity error, attitude error and IMU zero bias error as the state of the Kalman filter, and the position and velocity information of the GNSS as the observation quantity. Then, the estimated error quantity is used to correct the state of the INS, so as to obtain more accurate error quantities such as the INS position error, velocity error, attitude error and zero bias error.
[0063] Kalman filtering includes state equations and observation equations. The state equation is shown in the following formula 2:
[0064] x k =φx k-1 +w k Formula 2:
[0065] In formula 2, x k is the state quantity to be estimated at time k. In the GNSS / INS integrated navigation algorithm, the error parameters such as INS position error, velocity error, attitude error, zero bias error, etc. are taken as the state quantity. k-1 is the error of INS at the last moment, is the state transfer matrix of Kalman filter, which can be discretized by INS error propagation equation. k is the Kalman system noise matrix, which can be obtained based on the error characteristics of the IMU device.
[0066] The observation equation of Kalman filtering is shown in the following formula 3:
[0067] z k =Hx k +V k Formula 3;
[0068] In formula 3, z k is the observation at time k, which may include the difference between the position obtained based on GNSS and the position and velocity obtained based on INS. It is a matrix, and H is the observation noise matrix, which can be obtained according to the position and velocity accuracy of GNSS.
[0069] After obtaining the state equation and observation equation of the Kalman filter, according to the calculation process of the Kalman filter, not only the zero bias error at each moment can be estimated in real time, but also the INS position error, velocity error, and attitude error can be estimated in real time.
[0070] It can be seen that when the GNSS / INS integrated navigation system is working in real time, the zero bias error of the IMU at each moment will be estimated in real time based on the first temperature data collected at each moment. When the temperature changes, the IMU zero bias will also change with the change in temperature.
[0071] It should be noted that the bias error in this application includes the gyroscope bias error and accelerometer error of the IMU, both of which will change with temperature.
[0072] The gyroscope bias error includes the bias errors in different x, y, and z directions measured by the gyroscope.
[0073] It should be noted that in the embodiments of the present application, for the sake of simplicity, the ANFIS zero-bias temperature compensation model can also be described as a temperature compensation model, a zero-bias temperature compensation model, an ANFIS model, a model, etc., and the embodiments of the present application do not limit this.
[0074] S102, establishing an initial ANFIS zero bias temperature compensation model.
[0075] The initial ANFIS zero-bias temperature compensation model comprises: a fuzzification layer, an excitation layer, a normalization layer, a conclusion layer and an output layer connected in sequence; the fuzzification layer comprises a plurality of fuzzy processing nodes, each of which is used to perform fuzzification processing on a plurality of first sample data inputted according to at least one membership function set in advance to obtain membership results corresponding to the plurality of first sample data respectively; the membership results comprise at least one membership, at least one membership corresponds to at least one membership function one-to-one, and the membership is used to indicate the probability that the first sample data satisfies the fuzzy set; the excitation layer is used to perform fuzzification processing on the plurality of first sample data outputted by the fuzzification layer According to the respectively corresponding membership results, the T normal form operator is executed to obtain the excitation intensity corresponding to at least one rule; the rule is used to indicate the zero-bias temperature compensation strategy; the normalized excitation intensity layer is used to normalize the excitation intensity corresponding to at least one rule to obtain the normalized excitation intensity corresponding to at least one rule; the conclusion layer is used to process the normalized excitation intensity corresponding to at least one rule according to a preset first conclusion parameter to obtain the intermediate zero-bias error result corresponding to at least one rule; the output layer is used to fuse multiple intermediate zero-bias error results output by the conclusion layer to obtain the estimated final zero-bias error result.
[0076] The structure of Adaptive-Network-based Fuzzy Inference Systems (ANFIS) is as follows: Figure 2 The network shown is a multi-input and single-output network, which consists of five layers of sub-networks, integrating the advantages of neural networks and fuzzy systems. ANFIS includes 5 sub-network layers. The functions of each layer of ANFIS are introduced in detail below:
[0077] (1) The first layer of ANFIS is the fuzzification layer. The input of the fuzzification layer is a plurality of first sample data. Before the first sample data is input into the fuzzification layer, the first sample data is first processed to obtain a first vector corresponding to the first sample data. Then, the plurality of first vectors are input into the fuzzification layer.
[0078] After the fuzzification layer fuzzifies the multiple first vectors using at least one membership function, it outputs the membership degree of the multiple first vectors for each membership function. The membership degree is calculated as shown in the following formula 4:
[0079]
[0080] In formula 4, μ i,j (x j ) is the jth first vector x j The membership degree corresponding to the i-th membership function. is a parameter set, called the first premise parameter. j,i Indicates that the jth first vector corresponds to the center of the ith membership function, It represents the variance of the j-th first vector corresponding to the i-th membership function.
[0081] (2) The second layer of ANFIS is the excitation layer. The excitation layer performs the T-normal operator of “AND” on multiple first membership results, and the excitation intensity corresponding to each rule is obtained as shown in the following formula 5:
[0082]
[0083] In Formula 5, n is the number of first sample data, m is the number of membership functions corresponding to each first sample data, and w is t (t=1,…,m n ) corresponds to the incentive intensity of the tth rule.
[0084] Among them, the incentive intensity can be understood as the satisfaction degree or weight of each rule.
[0085] (3) The third layer of ANFIS is the normalization layer, which can also be described as the normalized excitation intensity layer.
[0086] The normalized excitation intensity layer is used to normalize the excitation intensity of each rule output by the second layer according to Formula 6:
[0087]
[0088] In Formula 6, M is the total number of rules, and M corresponds to the number of nodes in the third layer. is the normalized incentive intensity of the ith rule.
[0089] (4) The fourth layer of ANFIS is the conclusion layer: the number of nodes in the fourth layer is the same as that in the third layer. Each node in the fourth layer corresponds to an output, where the output of the i-th node in the fourth layer is Formula 7:
[0090]
[0091] In formula seven, is the intermediate zero bias error of the ith node output, (a i,j , b i ) is called the first conclusion parameter.
[0092] (5) The fifth layer of ANFIS is the output layer. The fifth layer fuses each intermediate zero bias error of the fourth layer to obtain the estimated final zero bias error result.
[0093] The fusion of each intermediate zero bias error may be performed by weighting and summing each intermediate zero bias error.
[0094] The final zero bias error result is represented by a matrix or a vector, and each element in the matrix or the vector corresponds one-to-one to the final zero bias error corresponding to each first sample data.
[0095] The execution order of S102 and S101 in this embodiment is only an example. In some embodiments, S102 may be executed first and then S101.
[0096] S103, inputting a plurality of first sample data into an initial ANFIS zero bias temperature compensation model, training the initial ANFIS zero bias temperature compensation model based on labels corresponding to the plurality of first sample data, until a model convergence condition is met, and obtaining a target ANFIS zero bias temperature compensation model.
[0097] The label of the first sample data is the expected output of inputting the first sample data into the initial ANFIS zero-bias temperature compensation model.
[0098] Among them, the target ANFIS zero bias temperature compensation model is used to perform zero bias compensation on the output parameters of the IMU at different temperatures.
[0099] Exemplarily, the output parameters of the IMU include an angular velocity output by a gyroscope and / or an acceleration output by an accelerometer;
[0100] The angular velocity bias error and / or accelerometer bias error at different temperatures can be determined based on the target ANFIS bias temperature compensation model, and then the angular velocity bias error and / or accelerometer bias error determined based on the target ANFIS bias temperature compensation model can be used to compensate for the angular velocity and acceleration output by the IMU.
[0101] Training the initial ANFIS zero-bias temperature compensation model is essentially training the first conclusion parameter in Formula 7 until the model convergence condition is met. Before training the initial ANFIS zero-bias temperature compensation model, the first premise parameter in Formula 4 can be obtained in advance and used as the initial premise parameter in the initial ANFIS zero-bias temperature compensation model.
[0102] The initial prerequisite parameters can be obtained before the electronic device performs the positioning and navigation task.
[0103] Exemplarily, the electronic device is obtained based on historical positioning and navigation data, and the historical positioning and navigation data includes: second temperature data, a second temperature gradient, and a zero bias error corresponding to the second temperature data.
[0104] The second temperature data, the second temperature gradient corresponding to the second temperature data, and the zero bias error corresponding to the second temperature data are acquired in the same manner as the first sample data, the first temperature gradient corresponding to the first sample data, and the zero bias error corresponding to the first sample data, and are not elaborated here.
[0105] The second temperature data and the second temperature gradient are described as second sample data. Assume that there are n second sample data, and illustratively, n=2. The number of membership functions corresponding to one second sample data is m, and illustratively, m is 4. Then the number of rules M is 4.
[0106] Initial premise parameter v j,i , The determination of adopts the fuzzy C-means clustering method, and its specific implementation process is shown in the following steps a1 to a2 (not shown in the figure):
[0107] Step a1: The second vector X corresponding to n sets of second sample data k =[x 1,k ,…,x n,k ] T (k=1,2,…N) is divided into m fuzzy sets (fuzzy groups), and the cluster center of each group is calculated to minimize the value function of the dissimilarity index (where n is the number of second sample data at a time, N is the number of times the second sample data is collected; X k Represents a set of second sample data at a time, x nk Indicates the nth sample in a set of sample data). Specifically, FCM uses fuzzy partitioning, so that each given data uses a membership value between 0 and 1 to determine the degree to which it belongs to each group, and the membership of each input constitutes the membership matrix U. The membership matrix U is initialized with a random number between 0 and 1, and it satisfies the constraints of formula 8:
[0108]
[0109] In formula 8, u j,i,k ,i=1,…,m;j=1,…,n;k=1,…,N is an element of the membership matrix U.
[0110] After determining the membership matrix U, the following formula 9 and formula 10 are used to calculate the m groups of cluster centers (i.e., the center v of the Gaussian function as the membership function): j,i ) and the variance of the Gaussian function It is the first premise parameter.
[0111]
[0112] c∈[1,∞] is a weighted exponent, generally taken as 2.
[0113] Step a2: Calculate the value function based on Formula 11:
[0114]
[0115] Among them, J j is the value function value, D j,i,k =||x j,k -v j,i || represents x j,k and v j,i The Euclidean distance between .
[0116] If J j is less than a certain threshold, or the J calculated this time j Compared with the J calculated last time j If the difference is less than the threshold, the premise parameters calculated by the current formula 9 and formula 10 are used as the initial premise parameters. j is greater than a certain threshold, or the J calculated this time j Compared with the J calculated last time j The difference between the distances is greater than or equal to the threshold. Then a new U matrix is calculated based on Formula 12, and then the calculation of x based on the new U matrix is returned. j,k and v j,i and value functions until J i is less than a certain threshold, or, J j Compared with the J calculated last time j The difference is less than the threshold. The premise parameters calculated by the current round of formula nine and formula ten are used as the initial premise parameters.
[0117] In an embodiment of the present application, when performing zero bias compensation on the output parameters of the IMU in the combined navigation system, an initial ANFIS zero bias temperature compensation model is first established, and the initial ANFIS zero bias temperature compensation model is trained based on multiple first temperature data, corresponding first temperature gradients and zero bias errors until the model convergence conditions are met, and a target ANFIS zero bias temperature compensation model is obtained, and zero bias compensation is performed based on the target ANFIS zero bias temperature compensation model; wherein the ANFIS model based on which the zero bias compensation of the output parameters of the IMU is performed integrates a multi-layer neural network and a fuzzy inference system, and can perform deep learning on the relationship between the first temperature data, the first temperature gradient and the corresponding zero bias error, and uses the learned target ANFIS model for temperature compensation, which can improve the compensation effect.
[0118] Next, through Figure 3 The embodiment specifically describes the training process of the initial ANFIS zero-bias temperature compensation model.
[0119] Figure 3 The following is a schematic diagram of the implementation flow of the temperature compensation model training method provided in the embodiment of the present application. Figure 3 As shown, the training method of S103 for the initial ANFIS zero bias temperature compensation model includes:
[0120] Step b1, inputting a plurality of first sample data into an initial ANFIS temperature compensation model to obtain first output data of the initial ANFIS model; the first output data includes output data of some layers of the ANFIS.
[0121] The first sample data includes: the normalized excitation intensity of the third layer output and the zero bias error of the last layer output.
[0122] The zero bias error of the last layer output of the initial ANFIS is shown in Formula 12:
[0123]
[0124] Among them, y k is the zero bias error output by the ANFIS conclusion layer.
[0125] is the normalized excitation intensity and the input vector X k The (n+1)*M-dimensional row vector formed; is the normalized excitation intensity corresponding to each rule at time k; θ k is the first conclusion parameter a corresponding to all rules at time k i,j,k , b i,k (i=1,…,M,j=1,…,n) composed of (n+1)*M-dimensional column vectors, a i,k =[a i,1,k ,…,a i,n,k ] T (i=1,…,M) is a column vector consisting of some first conclusion parameters at time k, M is the total number of rules, and n is the number of input first sample data.
[0126] Step b2: using the fading adaptive Kalman filter algorithm and the first output data, updating the first conclusion parameter to the second conclusion parameter.
[0127] Specifically, the electronic device uses the fading adaptive Kalman filter algorithm to process the first output data, the input vector corresponding to the first sample data, the number n of the first sample data, the number M of the rules, the first conclusion parameter a i,j,kand b i,k and observation noise are calculated to obtain the second conclusion parameter.
[0128] Among them, the state equation of the fading adaptive Kalman filter algorithm is: θ k =θ k-1 ; The observation equation of the fading adaptive Kalman filter algorithm is: y k =A k ·θ k +V k , where θ k is the state quantity, y k is the observed quantity, V k is the observation noise.
[0129] The first conclusion parameter is updated according to the following formula 11 to obtain the second conclusion parameter:
[0130]
[0131] In Formula 11, θ k Used to characterize the first conclusion parameter before updating, then θ k+1 It is used to represent the updated second conclusion parameter. λ is a fading factor, which controls the attenuation of the first conclusion parameter that has not been updated, and enhances the trust in the updated conclusion parameter. In actual use, λ can be greater than or equal to 0.9 and less than or equal to 1. S k =(n+1)M×(n+1)M. In the initial state, θ0=0, S0=ηI. η is any positive number, I is the unit matrix. η is generally taken as a larger value.
[0132] Step b3: updating the initial temperature compensation model based on the second conclusion parameter to obtain an updated temperature compensation model.
[0133] Updating the initial temperature compensation model is an iterative process, which can be done by Figure 4 The embodiments are described in detail and will not be elaborated here.
[0134] In this embodiment, when training the initial temperature compensation model, a fading Kalman filter algorithm is used to update the initial conclusion parameters of the conclusion layer, thereby obtaining a target temperature compensation model.
[0135] As mentioned above, updating the initial temperature compensation model is an iterative process. Next, Figure 4 The embodiment specifically illustrates how to update the temperature compensation model to obtain the target temperature compensation model.
[0136] Figure 4 The temperature compensation model training method provided in the embodiment of the present application is as follows: Figure 4As shown, step b3 may include:
[0137] Step b31: Update the temperature compensation model based on the input conclusion parameters and premise parameters to obtain an updated temperature compensation model.
[0138] In the first cycle, the input conclusion parameter is the second conclusion parameter, the premise parameter is the first premise parameter in the initial temperature compensation model, and the initial temperature compensation model is updated for the first time based on the second conclusion parameter to obtain the first temperature compensation model. After the first cycle, the second conclusion parameter is updated to the third conclusion parameter, and the first premise parameter is updated to the second premise parameter.
[0139] In subsequent cycles, the input conclusion parameters are the conclusion parameters of the previous round of update, and the premise parameters are the premise parameters after the previous round of update.
[0140] Step b32: determine whether the training error of the updated temperature compensation model meets the model convergence condition; if so, execute b33; otherwise, execute b34.
[0141] Among them, the training error at time k is y k is the output result of the model at time k, y d is the expected output at time k, i.e., y d is the label of the second sample data.
[0142] If E k If the training error is less than the preset threshold, or the difference between the training error obtained in this round of update and the training error obtained in the previous round of update is less than the preset threshold, it is determined that the training error meets the model convergence condition.
[0143] Step b33, using the updated temperature compensation model as the target temperature compensation model.
[0144] Exemplarily, in the first cycle, the premise parameters of the target temperature compensation model are initial premise parameters, and the conclusion parameters are first conclusion parameters.
[0145] Step b34: Use the gradient descent algorithm and the input conclusion parameters to update the input premise parameters.
[0146] Exemplarily, in the first cycle, the first premise parameter based on which the membership function is constructed is updated to the second premise parameter.
[0147] As shown in S103 above, the first premise parameter is obtained by training the second sample data using the FCM algorithm. And as mentioned above, the first sample data and the second sample data are obtained at different times.
[0148] The first sample data is obtained by the electronic device during the current positioning and navigation task, the second sample data is obtained by the electronic device during the historical positioning and navigation task, or the second sample data is collected by the electronic device before the current positioning and navigation task is performed.
[0149] Assume that the training error at time k is Among them, y k is the output result of the model at time k, y d is the expected output at time k, i.e., y d is the label of the second sample data.
[0150] When the first premise parameter is updated using the gradient descent algorithm, the error rates of the nodes from the fifth layer to the first layer of ANFIS can be calculated in sequence according to the training error, and the first premise parameter can be updated based on the calculated error rates of the nodes in the first layer.
[0151] Among them, the error rate of the fifth layer output node of ANFIS is:
[0152] The error rate of the output node of the fourth layer of ANFIS is shown in Formula 12:
[0153]
[0154] In Formula 12, O 4,t It represents the output of the tth node in the fourth layer of ANFIS, and M is the number of rules.
[0155] The error rate of each node in the third layer of ANFIS is shown in Formula 13:
[0156]
[0157] In formula 13, O 3,s represents the output of the sth node in the third layer of ANFIS, b s,k is the conclusion parameter corresponding to the sth rule at time k, X k The first vector of input at time k.
[0158] The error rate of each node in the second layer of ANFIS is shown in Formula 14:
[0159] Among them, O 2,p represents the output of the pth node in the second layer of ANFIS, w i ,i=1,…,M is the output of the i-th node in the second layer of ANFIS.
[0160] The error rate of each node in the first layer of ANFIS is shown in Formula 15:
[0161]
[0162] Among them, O 1,q It represents the output of the qth node in the first layer of ANFIS, n is the number of input variables, and m is the number of membership functions corresponding to each input variable.
[0163] Taking the case where the number of nodes in the first layer is 9 as an example, the specific calculation method of the error rate of each node in the first layer is explained:
[0164] When q = 1 or 2,
[0165] When q=3 or q=4,
[0166] When q=5 or q=6,
[0167]
[0168] When q=7 or q=8,
[0169] In some embodiments, the electronic device can obtain the first output error rate corresponding to the fuzzification layer based on the training error of the initial ANFIS temperature compensation model, the first premise parameter, the output data of each fuzzification processing node of the fuzzification layer and Formula 16.
[0170] Based on the training error of the initial ANFIS temperature compensation model, the first premise parameter, the output data of each fuzzy processing node of the fuzzy layer and Formula 16, the second output error rate corresponding to the fuzzy layer is obtained.
[0171] Among them, Formula 16 and Formula 17 are as follows:
[0172]
[0173] In Formula 16 and Formula 17, is the first output error rate, is the second output error rate. E k is the training error, v j,i and is the first premise parameter; 1,2(j-1)+i It is the output data of the 2*(j-1)+ith fuzzy processing node of the fuzzy layer; wherein j is greater than 0 and less than or equal to the number of the first sample data; and i is greater than 0 and less than the number of rules.
[0174] After the first output error rate is obtained, the first parameter in the first premise parameter is updated using Formula 18 based on the first output error rate, the second output error rate and the preset learning rate.
[0175]
[0176] In formula 18, v on the right side of the equal sign j,i is the first parameter before updating; the v on the left side of the equal sign j,i is the first parameter after update; η is the preset learning rate.
[0177] After the second output error rate is obtained, the second parameter in the first premise parameter is updated based on the second output error rate and the preset learning rate using Formula 19.
[0178]
[0179] Among them, σ on the left side of the equal sign j,i is the first parameter after update; η is the preset learning rate.
[0180] The second premise parameters include the updated first parameters and the updated second parameters.
[0181] Step b35: using the fading adaptive Kalman filter algorithm and the first output data, update the input conclusion parameters.
[0182] Exemplarily, in the first cycle, the specific implementation method of updating the second conclusion parameter to the third conclusion parameter using the fading adaptive Kalman filter algorithm and the first output data is the same as the method of updating the first conclusion parameter to the second conclusion parameter using the fading adaptive Kalman filter algorithm and the first output data in step b2. No further details will be given here.
[0183] After executing step b35, based on the second premise parameter and the third conclusion parameter currently calculated, return to execute b31 to update the initial temperature compensation model, and repeat the process until the updated temperature compensation model meets the model convergence condition to obtain the target temperature compensation model.
[0184] As mentioned above, the initial ANFIS zero-bias temperature model is trained based on the first sample data. Figure 5 The embodiment exemplarily illustrates how to obtain the first sample data.
[0185] It should be noted that each first sample data carries a label, and obtaining the first sample data also includes determining the label of the first sample data.
[0186] Figure 5 The following is a schematic diagram of the implementation flow of the temperature compensation model training method provided in the embodiment of the present application. Figure 5 As shown, before S101, it also includes:
[0187] S101a, performing temperature measurement operations and zero bias error determination operations when the integrated navigation system performs a navigation task.
[0188] In some embodiments, the IMU is equipped with a temperature sensor, and the temperature measurement operation can be performed by the IMU; alternatively, the temperature measurement operation can be performed by a temperature sensor module independently provided in the integrated navigation system.
[0189] S101b, storing the measured first temperature data, the first temperature gradient corresponding to the first temperature data, and the zero bias error under the first temperature data in a storage subunit corresponding to the temperature node to which the first temperature data belongs.
[0190] As described in the above embodiment, the IMU includes a first storage unit, which includes multiple storage sub-units; the temperature data belongs to a preset temperature range, which includes multiple temperature nodes, and the multiple temperature nodes correspond one-to-one to the multiple storage sub-units in the first storage unit.
[0191] In some embodiments, after acquiring the first temperature data, as described above, based on the Kalman filtering algorithm, the zero bias error at the moment of acquiring the first temperature data is determined, and the zero bias error and the zero bias error corresponding to the first temperature data or the first sample data are also stored in the storage subunit corresponding to the temperature node to which the first temperature data belongs.
[0192] In some embodiments, when the electronic device stores the first temperature data and the corresponding first temperature gradient, it is also necessary to determine whether there is a temperature node in the electronic device where the absolute value of the difference between the first temperature data is less than a temperature threshold.
[0193] If there is a temperature node in the electronic device where the absolute value of the difference between the measured / collected first temperature data is less than the temperature threshold, the electronic device will store the measured first temperature data and the first temperature gradient determined based on the first temperature data in the corresponding storage subunit corresponding to the temperature node. Also, the estimated zero bias error corresponding to the first temperature data is also stored in the storage subunit corresponding to the temperature node.
[0194] In some embodiments, before writing the first temperature data into the temperature node, it is first determined whether other temperature data and the corresponding temperature gradient and zero bias error are stored in the storage subunit corresponding to the temperature node. If so, the other temperature data, the corresponding temperature gradient and the zero bias error are deleted; then, the first temperature data, the first temperature gradient and the corresponding zero bias error (label) are stored in the storage subunit.
[0195] S101c. When the number of temperature nodes to which all the collected first temperature data belongs is greater than a temperature node number threshold, all the first temperature data, the corresponding first temperature gradient, and the corresponding zero bias error stored in the first storage unit are acquired from the first storage unit.
[0196] The plurality of first sample data includes a plurality of first temperature data and corresponding first temperature gradients.
[0197] The label of the first sample data is the zero bias error under the first temperature gradient.
[0198] The temperature node number threshold can be preset according to actual needs, and the temperature node number threshold is less than or equal to the number of all temperature nodes.
[0199] The number of temperature nodes to which all the collected first temperature data belongs, that is, the number of first temperature data stored in the first storage unit. If this number is greater than the temperature node number threshold, it proves that the number of first sample data is large, and using more sample data to train the temperature compensation model can improve the accuracy of the model.
[0200] In this embodiment, the first sample data is acquired in real time during the process of the electronic device performing a navigation task, so that the initial temperature compensation model is quickly trained based on the first sample data when performing the navigation task, so as to perform zero bias compensation on the parameters output by the IMU based on the trained target temperature compensation model. In this way, real-time training of the initial temperature compensation model can solve the problem of time-consuming and high-cost temperature calibration methods in conventional technical means.
[0201] Above Figure 1 , Figure 3 , Figure 5 The embodiment specifically illustrates the training process of the ANFIS zero-bias temperature compensation model. Next, Figure 6 The embodiment illustrates the use process of the ANFIS zero-bias temperature compensation model.
[0202] Figure 6 A schematic diagram of a temperature compensation method according to an embodiment of the present invention is shown in FIG. Figure 5 As shown, the method includes:
[0203] S601. When the integrated navigation system performs a navigation task, obtain third temperature data currently measured.
[0204] In some embodiments, the operation of acquiring the third temperature data may be performed after determining the target ANFIS zero bias temperature compensation model.
[0205] S602: Input the third temperature data and the third temperature gradient corresponding to the third temperature data as third sample data into a target temperature compensation model to obtain a target zero bias error output by the target temperature compensation model.
[0206] The method for determining the target temperature compensation model is as described in the above embodiment and will not be elaborated here.
[0207] S603: Based on the target zero bias error, perform zero bias compensation on the inertial measurement unit IMU output parameters of the integrated navigation system.
[0208] Taking the angular velocity information output by the x-axis gyroscope of the IMU as an example, the zero bias compensation is explained as follows:
[0209] Assume that the x-axis output of the gyroscope of the MEMSIMU at time k is w x Based on the target ANFIS bias temperature compensation model, the x-axis gyro bias error is y x , using y x to w x After compensation, the x-axis gyro value is w x '=w x -y x . The compensation w x 'Used for strapdown solution, since it has compensated for the gyro's zero bias error, a more accurate attitude result can be obtained after strapdown solution, so that the GNSS / INS integrated navigation system can obtain more accurate position, speed and attitude results.
[0210] In this embodiment, after determining the target zero bias temperature compensation model, the zero bias error estimation can be performed on the temperature data collected in real time based on the target zero bias temperature compensation model to perform real-time compensation on the output parameters of the IMU at different times, so that the GNSS / INS combined navigation system can obtain more accurate position, velocity, and attitude results.
[0211] It should be noted that although the steps of the method in the present application are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.; or, steps in different embodiments may be combined into a new technical solution.
[0212] Based on the foregoing embodiments, the embodiments of the present application provide a training device for a temperature compensation model, which includes the modules included and the units included in the modules, and can be implemented by a processor; of course, it can also be implemented by a specific logic circuit; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP) or a field programmable gate array (FPGA), etc.
[0213] Figure 7 A schematic diagram of the structure of the temperature compensation processing device provided in the embodiment of the present application is shown in FIG. Figure 7 As shown, the temperature compensation processing device 70 includes a first acquisition module 71, a building module 72, and a training module 73, wherein:
[0214] The first acquisition module 71 is used to acquire multiple first sample data, each of which includes: first temperature data and a first temperature gradient corresponding to the first temperature data; the label of the first sample data is the zero bias error of the inertial measurement unit IMU determined by the integrated navigation system under the first temperature data.
[0215] Establishing module 72, for establishing an initial ANFIS zero-bias temperature compensation model; the initial ANFIS zero-bias temperature compensation model comprises: a fuzzification layer, an excitation layer, a normalization layer, a conclusion layer and an output layer connected in sequence; the fuzzification layer comprises a plurality of fuzzification processing nodes, each of which is used to perform fuzzification processing on the plurality of first sample data inputted according to at least one membership function set in advance to obtain membership results corresponding to the plurality of first sample data respectively; the membership results comprise at least one membership, the at least one membership corresponds to the at least one membership function one-to-one, and the membership is used to indicate the probability that the first sample data satisfies each fuzzy set; The excitation layer is used to execute the T-normal form operator on the membership data corresponding to the multiple first sample data output by the fuzzification layer, so as to obtain the excitation intensity corresponding to at least one rule; the rule is used to indicate the zero-bias temperature compensation strategy; the normalization layer is used to normalize the excitation intensity corresponding to the at least one rule, so as to obtain the normalized excitation intensity corresponding to the at least one rule; the conclusion layer is used to process the normalized excitation intensity corresponding to the at least one rule, so as to obtain the intermediate zero-bias error result corresponding to the at least one rule; the output layer is used to fuse the multiple intermediate zero-bias error results output by the conclusion layer, so as to obtain the estimated final zero-bias error result.
[0216] The training module 73 is used to input the multiple first sample data into the initial ANFIS zero bias temperature compensation model, train the initial ANFIS zero bias temperature compensation model based on the labels corresponding to the multiple first sample data, until the model convergence condition is met, and obtain the target ANFIS zero bias temperature compensation model; the target ANFIS zero bias temperature compensation model is used to perform zero bias compensation on the output parameters of the IMU.
[0217] In some embodiments, the training module 73 is used to input the multiple first sample data into the initial ANFIS zero bias temperature compensation model to obtain the first output data of the initial ANFIS model; the first output data includes the output data of the normalization layer of the ANFIS and the conclusion layer of the ANFIS; using the fading adaptive Kalman filtering algorithm and the first output data, the first conclusion parameter based on which the conclusion layer is constructed is updated to the second conclusion parameter; based on the second conclusion parameter, the initial ANFIS zero bias temperature compensation model is updated to obtain the target ANFIS zero bias temperature compensation model.
[0218] In some embodiments, the training module 73 is used to perform a first update on the initial ANFIS zero-bias temperature compensation model based on the second conclusion parameter to obtain a first ANFIS zero-bias temperature compensation model; determine whether the training error of the first ANFIS zero-bias temperature compensation model meets the model convergence condition; if the model convergence condition is met, use the first ANFIS zero-bias temperature compensation model as the target ANFIS zero-bias temperature compensation model; if the model convergence condition is not met, use the gradient descent algorithm and the second conclusion parameter to update the first premise parameter based on which the membership function is constructed to the second premise parameter; the first premise parameter is obtained by training the second sample data using the FCM algorithm; wherein the first sample data and the second sample data are obtained at different times; use the fading adaptive Kalman filtering algorithm and the first output data to update the second conclusion parameter to the third conclusion parameter; based on the second premise parameter and the third conclusion parameter, update the first ANFIS zero-bias temperature compensation model at least once until the training error of the updated ANFIS zero-bias temperature compensation model meets the model convergence condition to obtain the target ANFIS zero-bias temperature compensation model.
[0219] In some embodiments, the training module 73 is used to train the initial ANFIS zero-bias temperature compensation model based on the training error, the first premise parameter, the output data and formula of each fuzzification processing node of the fuzzification layer: The first output error rate corresponding to the fuzzification layer is obtained; wherein, is the first output error rate; E kis the training error, v j,i and is the first premise parameter; 1,2(j-1)+i is the output data of the 2*(j-1)+ith fuzzy processing node of the fuzzy layer; wherein j is greater than 0 and less than or equal to the number of the first sample data; i is greater than 0 and less than the number of the rules; based on the training error of the initial ANFIS zero-bias temperature compensation model, the first premise parameter, the output data and formula of each fuzzy processing node of the fuzzy layer: Obtaining a second output error rate corresponding to the fuzzification layer; is the second output error rate; E k is the training error, v j,i and is the first premise parameter; 1,2(j-1)+i It is the output data of the 2*(j-1)+ith fuzzy processing node of the fuzzy layer; wherein, j is greater than 0 and less than or equal to the number of first sample data; i is greater than 0 and less than the number of rules; based on the first output error rate, the second output error rate and the preset learning rate, the first premise parameter is updated to the second premise parameter.
[0220] In some embodiments, the training module 73 is used to use the formula: Update the first parameter in the first premise parameter; where v on the right side of the equal sign j,i is the first parameter before updating; the v on the left side of the equal sign j,i is the updated first parameter; η is the preset learning rate; according to the second output error rate and the preset learning rate, using the formula: Update the second parameter in the first premise parameter; where σ on the right side of the equal sign j,i is the first parameter before updating; σ on the left side of the equal sign j,i is the updated first parameter; η is the preset learning rate; according to the updated first parameter and the second parameter, the second premise parameter is obtained.
[0221] In some embodiments, a first acquisition module 71 is used to perform temperature measurement operations and zero bias error determination operations when the combined navigation system performs a navigation task; the measured first temperature data, the first temperature gradient determined based on the first temperature data, and the zero bias error under the first temperature data are stored in a storage subunit corresponding to the temperature node to which the first temperature data belongs; wherein the IMU includes a first storage unit, and the first storage unit includes multiple storage subunits; the temperature data belongs to a preset temperature range, and the preset temperature range includes multiple temperature nodes, and the multiple temperature nodes correspond one-to-one to the multiple storage subunits in the first storage unit; when the number of temperature nodes to which all the collected first temperature data belong is greater than a temperature node number threshold, multiple first temperature data, corresponding first temperature gradients and corresponding zero bias errors are obtained from the first storage unit; wherein the multiple first sample data include the multiple first temperature data and the corresponding first temperature gradients.
[0222] Based on the foregoing embodiments, an embodiment of the present application provides a temperature compensation processing device, which includes the modules included and the units included in the modules, which can be implemented by a processor; of course, it can also be implemented by a specific logic circuit; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP) or a field programmable gate array (FPGA), etc.
[0223] Figure 8 A schematic diagram of the structure of the temperature compensation processing device provided in the embodiment of the present application is shown in FIG. Figure 7 As shown, the temperature compensation processing device 80 includes a second acquisition module 81, a zero bias error estimation module 82, and a zero bias error compensation module 83, wherein:
[0224] The second acquisition module 81 is used to acquire the third temperature data currently measured when the integrated navigation system performs a navigation task.
[0225] The zero bias error estimation module 82 is used to input the third temperature data and the third temperature gradient determined based on the third temperature data as third sample data into the target ANFIS zero bias temperature compensation model to obtain the target zero bias error output by the target ANFIS zero bias temperature compensation model.
[0226] The bias error compensation module 83 is used to perform bias compensation on output parameters of an inertial measurement unit IMU of the integrated navigation system based on the target bias error.
[0227] The description of the above device embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the device embodiment of the present application, please refer to the description of the method embodiment of the present application for understanding.
[0228] It should be noted that in the embodiments of this application Figure 7 and Figure 8 The division of modules in the device shown is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional unit in each embodiment of the present application may be integrated into a processing unit, or may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit. It may also be implemented in the form of a combination of software and hardware.
[0229] It should be noted that in the embodiments of the present application, if the above-mentioned method for transmitting network slicing information is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of the present application can be essentially or partly reflected in the form of a software product, which is stored in a storage medium and includes several instructions for enabling an electronic device to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk. In this way, the embodiments of the present application are not limited to any specific combination of hardware and software.
[0230] An embodiment of the present application provides a chip, including an interface circuit and a logic circuit, wherein the interface circuit is used to receive signals from other chips outside the chip and transmit them to the logic circuit, or to send signals from the logic circuit to other chips outside the chip, and the logic circuit is used to implement the temperature compensation training method or temperature compensation method described in the embodiment of the present application.
[0231] An embodiment of the present application provides an electronic device, which may be a home base station gateway in the present application. Fig. 9 is a schematic diagram of the structure of an electronic device according to an embodiment of the present application, such as Fig. 9 As shown, the electronic device 90 includes a memory 901 and a processor 902, wherein the memory 901 stores a computer program that can be run on the processor 902, and the processor 902 implements the steps in the method provided in the above embodiment when executing the program.
[0232] It should be noted that the memory 901 is configured to store instructions and applications executable by the processor 902, and can also cache data to be processed or already processed by the processor 902 and various modules in the electronic device 90 (for example, image data, audio data, voice communication data, and video communication data), which can be implemented through flash memory (FLASH) or random access memory 901 (Random Access Memory, RAM).
[0233] An embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps in the method for transmitting network slice information provided in the above embodiment are implemented.
[0234] An embodiment of the present application provides a computer program product containing instructions, which, when executed on a computer, enables the computer to execute the steps in the method for transmitting network slice information provided in the above method embodiment.
[0235] It should be noted here that the description of the above storage medium and device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects as the method embodiments. For technical details not disclosed in the storage medium, storage medium and device embodiments of this application, please refer to the description of the method embodiments of this application for understanding.
[0236] It should be understood that "one embodiment" or "an embodiment" or "some embodiments" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in one embodiment" or "in some embodiments" appearing throughout the specification may not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present application, the size of the sequence number of the above-mentioned processes does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application. The above-mentioned sequence numbers of the embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments. The above description of each embodiment tends to emphasize the differences between the various embodiments, and the same or similar aspects can be referenced to each other. For the sake of brevity, this article will not repeat them.
[0237] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as: multiple modules or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or modules can be electrical, mechanical or other forms.
[0238] The modules described above as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules; they may be located in one place or distributed on multiple system units; some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment.
[0239] In addition, all functional modules in the embodiments of the present application may be integrated into one processing unit, or each module may be a separate unit, or two or more modules may be integrated into one unit; the above-mentioned integrated modules may be implemented in the form of hardware or in the form of hardware plus software functional units.
[0240] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, etc., various media that can store program codes.
[0241] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application can essentially or in other words, the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling an electronic device to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.
[0242] The methods disclosed in several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.
[0243] The features disclosed in several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.
[0244] The features disclosed in several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.
[0245] The above is only an implementation method of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A temperature compensation model training method, characterized in that: include: Acquire a plurality of first sample data and labels of the plurality of first sample data, each of the first sample data comprising: first temperature data and a first temperature gradient corresponding to the first temperature data; the label of the first sample data is a zero bias error of an inertial measurement unit IMU determined by the integrated navigation system under the first temperature data; An initial adaptive neural fuzzy inference system ANFIS zero-bias temperature compensation model is established; the initial ANFIS zero-bias temperature compensation model comprises: a fuzzification layer, an excitation layer, a normalization layer, a conclusion layer and an output layer connected in sequence; the fuzzification layer comprises a plurality of fuzzification processing nodes, each of which is used to perform fuzzification processing on the input plurality of first sample data respectively according to at least one membership function set in advance to obtain membership results corresponding to the plurality of first sample data respectively; the membership results comprise at least one membership, the at least one membership corresponds to the at least one membership function one-to-one, and the membership is used to indicate whether the first sample data satisfies each fuzzy set The excitation layer is used to perform the T-normal operator on the membership data corresponding to the multiple first sample data output by the fuzzification layer, and obtain the excitation intensity corresponding to at least one rule; the rule is used to indicate the zero-bias temperature compensation strategy; the normalization layer is used to normalize the excitation intensity corresponding to the at least one rule, and obtain the normalized excitation intensity corresponding to the at least one rule; the conclusion layer is used to process the normalized excitation intensity corresponding to the at least one rule, and obtain the intermediate zero-bias error result corresponding to the at least one rule; the output layer is used to fuse the multiple results output by the conclusion layer, and obtain the estimated final zero-bias error result; The multiple first sample data are input into an initial ANFIS zero bias temperature compensation model, and the initial ANFIS zero bias temperature compensation model is trained based on the labels corresponding to the multiple first sample data until the model convergence condition is met, so as to obtain a target ANFIS zero bias temperature compensation model; the target ANFIS zero bias temperature compensation model is used to perform zero bias compensation on the output parameters of the IMU.
2. The method according to claim 1, characterized in that The step of inputting the plurality of first sample data into an initial ANFIS zero-bias temperature compensation model and training the initial ANFIS zero-bias temperature compensation model comprises: Inputting the plurality of first sample data into an initial ANFIS zero-bias temperature compensation model to obtain first output data of the initial ANFIS model; the first output data includes output data of a normalization layer of the ANFIS and a conclusion layer of the ANFIS; Using a fading adaptive Kalman filter algorithm and the first output data, updating the first conclusion parameter based on which the conclusion layer is constructed to a second conclusion parameter; The initial ANFIS zero bias temperature compensation model is updated based on the second conclusion parameter to obtain a target ANFIS zero bias temperature compensation model.
3. The method according to claim 2, characterized in that The updating of the initial ANFIS zero bias temperature compensation model based on the second conclusion parameter to obtain a target ANFIS zero bias temperature compensation model includes: Based on the second conclusion parameter, the initial ANFIS zero bias temperature compensation model is updated for the first time to obtain a first ANFIS zero bias temperature compensation model; Determining whether the training error of the first ANFIS zero-bias temperature compensation model meets a model convergence condition; If the model convergence condition is met, taking the first ANFIS zero bias temperature compensation model as the target ANFIS zero bias temperature compensation model; If the model convergence condition is not met, the first premise parameter based on which the membership function is constructed is updated to the second premise parameter by using the gradient descent algorithm and the second conclusion parameter; the first premise parameter is obtained by training the second sample data by using the FCM algorithm; wherein the first sample data and the second sample data are obtained at different times; Using a fading adaptive Kalman filter algorithm and the first output data, updating the second conclusion parameter to a third conclusion parameter; Based on the second premise parameter and the third conclusion parameter, the first ANFIS zero bias temperature compensation model is updated at least once until the training error of the updated ANFIS zero bias temperature compensation model meets the model convergence condition, thereby obtaining a target ANFIS zero bias temperature compensation model.
4. The method according to claim 3, characterized in that The method of updating the first premise parameter based on which the membership function is constructed to the second premise parameter by using the gradient descent algorithm and the second conclusion parameter includes: Based on the training error of the initial ANFIS zero-bias temperature compensation model, the first premise parameter, the output data and formula of each fuzzy processing node of the fuzzy layer: Obtaining a first output error rate corresponding to the fuzzification layer; in, is the first output error rate; E k is the training error, v j,i and is the first premise parameter; 1,2(j-1)+i is the output data of the 2*(j-1)+ith fuzzy processing node of the fuzzy layer; wherein j is greater than 0 and less than or equal to the number of the first sample data; i is greater than 0 and less than the number of the rules; Based on the training error of the initial ANFIS zero-bias temperature compensation model, the first premise parameter, the output data and formula of each fuzzy processing node of the fuzzy layer: Obtaining a second output error rate corresponding to the fuzzification layer; in, is the second output error rate; E k is the training error, v j,i and is the first premise parameter; 1,2(j-1)+i is the output data of the 2*(j-1)+ith fuzzy processing node of the fuzzy layer; wherein j is greater than 0 and less than or equal to the number of the first sample data; i is greater than 0 and less than the number of rules; Based on the first output error rate, the second output error rate, and a preset learning rate, the first premise parameter is updated to the second premise parameter.
5. The method according to claim 4, characterized in that The updating the first premise parameter to the second premise parameter based on the first output error rate, the second output error rate and a preset learning rate includes: According to the first output error rate and the preset learning rate, using the formula: Update the first parameter in the first premise parameter; where v on the right side of the equal sign j,i is the first parameter before updating; the v on the left side of the equal sign j,i is the updated first parameter; η is the preset learning rate; According to the second output error rate and the preset learning rate, using the formula: Update the second parameter in the first premise parameter; where σ on the right side of the equal sign j,i is the first parameter before updating; σ on the left side of the equal sign j,i is the updated first parameter; η is the preset learning rate; The second premise parameter is obtained according to the updated first parameter and the second parameter.
6. The method according to claim 1, characterized in that The obtaining of a plurality of first sample data and labels of the plurality of first sample data comprises: performing temperature measurement operations and zero bias error determination operations when the integrated navigation system performs a navigation task; The measured first temperature data, the first temperature gradient determined based on the first temperature data, and the zero bias error under the first temperature data are stored in a storage subunit corresponding to a temperature node to which the first temperature data belongs; wherein the IMU includes a first storage unit, and the first storage unit includes a plurality of storage subunits; the first temperature data belongs to a preset temperature range, and the preset temperature range includes a plurality of temperature nodes, and the plurality of temperature nodes correspond one-to-one to the plurality of storage subunits in the first storage unit; When the number of temperature nodes to which all the collected first temperature data belong is greater than a temperature node number threshold, multiple first temperature data, corresponding first temperature gradients and corresponding zero bias errors are obtained from the first storage unit; wherein the multiple first sample data include the multiple first temperature data and the corresponding first temperature gradients.
7. A temperature compensation model application method, characterized in that: The method comprises: When the integrated navigation system performs a navigation task, obtaining third temperature data currently measured; Inputting the third temperature data and a third temperature gradient determined based on the third temperature data as third sample data into the target ANFIS zero bias temperature compensation model according to any one of claims 1 to 6, to obtain a target zero bias error output by the target ANFIS zero bias temperature compensation model; Based on the target zero bias error, zero bias compensation is performed on output parameters of an inertial measurement unit (IMU) of the integrated navigation system.
8. A temperature compensation processing device, characterized in that: include: A module for executing the method according to any one of claims 1 to 6, or a module for executing the method according to claim 7.
9. An electronic device, characterized in that: include: Memory and processor; The memory is configured to store computer program instructions; The processor is configured to execute the computer program instructions so that the electronic device implements the method according to any one of claims 1 to 6, or implements the method according to claim 7.
10. A computer-readable storage medium, characterized in that: include: Computer program instructions; The electronic device runs the computer program instructions so that the electronic device implements the method according to any one of claims 1 to 6, or implements the method according to claim 7.
Citation Information
Cited By
Temperature compensation method and system based on inertial device
CN119223271A
