Method and device for predicting change trends of MEMS inertial sensors under high microgravity in space

Abstract

The present invention provides a method and device for predicting the changing trend of MEMS inertial sensors under high microgravity in space, and relates to the technical field of sensor calibration methods. The method comprises: first, collecting data including time, temperature, acceleration and angular velocity in a high microgravity environment in space, and processing the data in blocks. Then, embedding the time and temperature series and performing feature fusion with the acceleration and angular velocity time series segments respectively. Next, a gated attention mechanism is used to assign weights to the angular velocity and acceleration features. Finally, a CNN-GRU multi-task prediction model is constructed to predict the changing trends of angular velocity and acceleration. The present invention takes into account the interactivity and specificity between the acceleration and angular velocity changing trend prediction subtasks, realizes the effective prediction of the zero bias of the inertial sensor under high microgravity in space, and provides technical support for the calibration of inertial sensors in high microgravity environments in space.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of sensor calibration methods, and specifically to a method and device for predicting the change trend of a MEMS inertial sensor under high microgravity in space. Background Art

[0002] The space station's high-microgravity levitation experiment system is a complex levitation motion control system involving multiple aspects, including optics, mechanics, electricity, heat, magnetism, and control. It utilizes a number of key technologies, including visual measurement cameras, optical detection sensors, pressure sensors, temperature sensors, absolute optical encoders, MEMS inertial measurement technology, MEMS micropropulsion technology, and electromagnetic actuators. Inertial sensors are a key component of inertial measurement units (IMUs) that measure the angular rate of moving objects, manufactured using microelectromechanical systems (MEMS) technology. MEMS sensors in high-microgravity inertial measurement units (IMUs) include accelerometers and gyroscopes. However, MEMS inertial sensors are susceptible to factors such as their own materials, manufacturing quality, and operating environment, resulting in widespread measurement errors. Generally speaking, MEMS inertial sensor errors are primarily categorized as deterministic and random errors. The former primarily refers to errors caused by perturbations (parameter changes in sensitive physical models) and environmental sensitivity (interference from sensitive environments), while the latter primarily refers to random drift caused by uncertain factors. Deterministic errors are the most significant source of error. Therefore, prior to use, it is crucial to determine the variation trend of MEMS inertial sensors so that they can be compensated for.

[0003] Currently, signal processing-based methods are the most widely used in MEMS inertial sensor trend prediction. These methods involve three steps: signal acquisition and processing, feature extraction, and trend prediction. Currently, the signals collected for trend prediction of MEMS three-axis inertial sensors are broadly classified into: time, temperature, x / y / z-axis acceleration, and x / y / z-axis angular velocity. Currently, using a single signal source to calibrate the trend of MEMS three-axis inertial sensors has limitations, such as low trend prediction accuracy and reliability. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to address the deficiencies of the existing technology and provide a method and device for predicting the change trend of MEMS inertial sensors under high microgravity in space.

[0005] On the one hand, the present invention solves the above-mentioned technical problem with the following technical solution: a method for predicting the change trend of a MEMS inertial sensor under high microgravity in space, comprising:

[0006] S1. Collecting raw MEMS time series data and performing data preprocessing on the raw MEMS time series data to obtain multiple initial time series segments, wherein the raw MEMS time series data includes time, temperature, acceleration, and angular velocity obtained under high microgravity in space;

[0007] S2. Blocking the multiple initial time series segments, embedding the block sequence segments containing time data and temperature data, and fusing the embedded time series segments and temperature series segments with the block acceleration time series segments and angular velocity time series segments through a fully connected neural network to obtain fused features;

[0008] S3. Building a gated attention model based on the fusion features, and assigning feature weights of acceleration and angular velocity through the gated attention model;

[0009] S4. Construct a CNN-GRU multi-task prediction model, use the acceleration containing feature weights and the angular velocity containing feature weights as inputs of the CNN-GRU prediction model, perform result mapping restoration on the output features of the CNN-GRU prediction model, and update the network parameters of the CNN-GRU prediction model according to the restoration results, thereby completing the training of the CNN-GRU multi-task prediction model;

[0010] S5. Use the trained CNN-GRU multi-task prediction model to predict the change trend of the acceleration to be processed and the angular velocity to be processed of the MEMS three-axis inertial sensor, and obtain the change trend prediction value corresponding to the acceleration to be processed and the change trend prediction value corresponding to the angular velocity to be processed.

[0011] The beneficial effects of the present invention are as follows: a CNN-GRU multi-task prediction model is constructed, and the fusion features obtained by fusing the sequence segments containing time data and temperature data after segmentation with the acceleration time series segments and angular velocity time series segments after segmentation are used to train the CNN-GRU multi-task prediction model, taking into account the effective interaction between the acceleration and angular velocity change trend prediction subtasks, the specific modeling between the subtasks and the model reasoning accuracy, thereby achieving accurate and rapid change trend prediction.

[0012] On the other hand, another technical solution of the present invention to solve the above technical problem is as follows: a device for predicting the change trend of a MEMS inertial sensor under high microgravity in space, comprising:

[0013] a preprocessing unit, configured to collect raw MEMS time series data and perform data preprocessing on the raw MEMS time series data to obtain a plurality of initial time series segments, wherein the raw MEMS time series data includes time, temperature, acceleration, and angular velocity obtained under high microgravity in space;

[0014] a feature processing unit configured to perform block processing on the plurality of initial time series segments, embed the block sequence segments containing time data and temperature data, and fuse the embedded time series segments and temperature series segments with the block acceleration time series segments and angular velocity time series segments through a fully connected neural network to obtain fused features;

[0015] a weight allocation unit, configured to construct a gated attention model based on the fused features, and allocate feature weights of acceleration and angular velocity using the gated attention model;

[0016] A model training unit is used to construct a CNN-GRU multi-task prediction model, use the acceleration containing feature weights and the angular velocity containing feature weights as inputs of the CNN-GRU prediction model, perform result mapping restoration on the output features of the CNN-GRU prediction model, and update the network parameters of the CNN-GRU prediction model according to the restoration results, thereby completing the training of the CNN-GRU multi-task prediction model;

[0017] The change trend prediction unit is used to predict the change trend of the to-be-processed acceleration and to-be-processed angular velocity of the MEMS three-axis inertial sensor through the trained CNN-GRU multi-task prediction model, and obtain the change trend prediction value corresponding to the to-be-processed acceleration and the change trend prediction value corresponding to the to-be-processed angular velocity.

[0018] On the other hand, another technical solution of the present invention to solve the above-mentioned technical problem is as follows: an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are configured to execute the above-mentioned method for predicting the change trend of a MEMS inertial sensor under high microgravity in space.

[0019] On the other hand, another technical solution of the present invention to solve the above-mentioned technical problem is as follows: a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned method for predicting the change trend of MEMS inertial sensors under high microgravity in space. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of a method for predicting a change trend of a MEMS inertial sensor under high microgravity in space provided by an embodiment of the present invention;

[0021] Figure 2 A flowchart of a method for predicting the change trend of a MEMS inertial sensor under high microgravity in space provided by an embodiment of the present invention;

[0022] Figure 3 A network block diagram of the CNN-GRU multi-task prediction model provided by an embodiment of the present invention;

[0023] Figure 4 This is a module block diagram of the MEMS three-axis inertial sensor change trend prediction device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0025] The following describes the technical solutions in the embodiments of the present application in conjunction with the accompanying drawings. In the description of the present application, unless otherwise specified, " / " indicates that the objects associated with each other are in an "or" relationship. For example, A / B can represent A or B. "And / or" in the present application is only a description of the association relationship between the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural.

[0026] Furthermore, in the description of this application, unless otherwise specified, "plurality" means two or more than two. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0027] In addition, to facilitate the clear description of the technical solutions of the embodiments of the present application, in the embodiments of the present application, the words "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or execution order, and the words "first" and "second" do not necessarily mean different.

[0028] At the same time, in the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner to facilitate understanding.

[0029] like Figure 1 、 Figure 2As shown, the present invention provides a method for predicting the change trend of a MEMS inertial sensor under high microgravity in space, comprising:

[0030] S1. Collecting raw MEMS time series data and performing data preprocessing on the raw MEMS time series data to obtain multiple initial time series segments. The raw MEMS time series data includes time, temperature, acceleration (i.e., x / y / z-axis acceleration), and angular velocity (i.e., x / y / z-axis acceleration) obtained under high microgravity in space.

[0031] S2. Blocking the multiple initial time series segments, embedding the block sequence segments containing time data and temperature data, and fusing the embedded time series segments and temperature time series segments with the block acceleration time series segments and angular velocity time series segments through a fully connected neural network to obtain fused features;

[0032] S3. Building a gated attention model based on the fusion features, and assigning feature weights of acceleration and angular velocity through the gated attention model;

[0033] S4. Construct a CNN-GRU multi-task prediction model, use the acceleration containing feature weights and the angular velocity containing feature weights as inputs of the CNN-GRU prediction model, perform result mapping restoration on the output features of the CNN-GRU prediction model, and update the network parameters of the CNN-GRU prediction model according to the restoration results, thereby completing the training of the CNN-GRU multi-task prediction model;

[0034] S5. Use the trained CNN-GRU multi-task prediction model to predict the change trend of the acceleration to be processed and the angular velocity to be processed of the MEMS three-axis inertial sensor, and obtain the change trend prediction value corresponding to the acceleration to be processed and the change trend prediction value corresponding to the angular velocity to be processed.

[0035] The advantage of the above embodiment is that a CNN-GRU multi-task prediction model is constructed, and the fusion features obtained by fusing the time series segments and the temperature series segments with the blocked acceleration time series segments and the angular velocity time series segments are used to train the CNN-GRU multi-task prediction model, taking into account the effective interaction between the acceleration and angular velocity change trend prediction subtasks, the specific modeling between the subtasks and the model reasoning accuracy, thereby achieving accurate and rapid change trend prediction.

[0036] like Figure 2 As shown, preferably, data preprocessing is performed on the original MEMS time series data to obtain multiple initial time series segments, specifically:

[0037] The time, temperature, acceleration and angular velocity in the original MEMS timing data are normalized by a normalization formula, and the normalization formula is:

[0038]

[0039] Among them, xmean is the mean of different parameters in the original MEMS time series data, xvar is the variance of different parameters, and x i is the i-th value in the input window length L;

[0040] The normalized raw MEMS time series data is decomposed into multiple time series samples according to a set fixed length. The multiple time series samples are clustered and analyzed using a clustering algorithm to obtain the cluster distance of each time series sample.

[0041] Finally, outlier screening is performed. Specifically, time series samples whose clustering distance is greater than the set clustering distance threshold are removed to obtain multiple initial time series segments.

[0042] Specifically, the normalization process is a z-score normalization method.

[0043] In the above embodiment, the consistency and quality of the data are improved by normalization and removal of outliers, which provides a high-quality data foundation for subsequent steps, thereby helping to improve the generalization ability and processing efficiency of the model.

[0044] Preferably, the multiple initial time sequence segments are processed in blocks, specifically:

[0045] The multiple initial time sequence segments are divided into blocks, and each block-processed time sequence segment is expressed as The time series segments include sequence segments containing time data, temperature time series segments containing temperature data, acceleration time series segments containing acceleration data, and angular velocity time series segments containing angular velocity data. Based on the expression of each block, the initial time series segments are expressed as:

[0046]

[0047] Where X is the time series segment after block division, R is the real number field, P×N is a matrix with width P and length N, N is the number of blocks, L is the input window length, P is the length of the time series segment, and S is the block sliding distance.

[0048] Preferably, before the multiple initial time sequence segments are divided into blocks, the method further includes the following steps:

[0049] The normalization process is performed on the plurality of initial time sequence segments using normalization formula (2), wherein the normalization formula (2) is:

[0050]

[0051] Among them, x′mean is the mean of different parameters in the initial time series segment, x′var is the variance of different parameters, and x′ i is the i-th value in the input window of length L.

[0052] In the above embodiment, not only the complete multi-dimensional characteristics of the MEMS inertial data are retained, but also the requirements of different prediction durations can be adapted by adjusting the block parameters, thereby improving the scalability of the model.

[0053] Preferably, the time and temperature time series segments are feature fused with the acceleration and angular velocity.

[0054] The segmented sequence segments containing time data and temperature data are embedded and represented. The embedded time series segments and temperature series segments are fused with the segmented acceleration time series segments and angular velocity time series segments through a fully connected neural network to obtain fused features, specifically:

[0055] Based on the expression of the initial time series segment, the time series segment is mapped into a temperature feature matrix Wtemp∈R D×P , and map the temperature time series segments into a time feature matrix Wtime∈R D×P ;

[0056] Construct the learning matrix W acc ∈R D×P and W ang ∈R D×P , by learning the matrix W acc ∈R D×P and W ang ∈R D×P The acceleration time series segment and the angular velocity time series segment are respectively mapped into acceleration feature matrices X acc ∈R D×N And the angular velocity characteristic matrix X ang ∈R D×N , where D×P is a matrix with width D and length P;

[0057] The temperature feature matrix and the time feature matrix are fused with the acceleration feature matrix and the angular velocity feature matrix respectively through a fully connected neural network to obtain the fused features corresponding to the acceleration and the fused features corresponding to the angular velocity:

[0058]

[0059] in, is the i-th variable in the initial acceleration time series segment, is the i-th variable in the initial angular velocity time series segment.

[0060] In the above embodiment, time series segments and temperature series segments are mapped into feature matrices through embedding representation, and are effectively fused with acceleration and angular velocity feature matrices respectively, thereby generating fused features that contain both original information and are rich in deep relationships. This not only significantly enhances the expressive power of features, but also improves the accuracy of subsequent analysis tasks and the flexibility of the model, providing strong support for processing and analyzing complex MEMS timing data.

[0061] Preferably, a gated attention model is constructed based on the fusion features, and feature weights of acceleration and angular velocity are allocated by the gated attention model, specifically:

[0062] Constructing the mapping matrix And through the mapping matrix and And the attention score calculation formula for the fused acceleration and angular velocity information Calculate and obtain the attention score. The calculation formula of the attention score is as follows:

[0063]

[0064] in, By the mapping matrix get, By the mapping matrix get, By the mapping matrix get;

[0065] Construct the mapping matrix W G ∈R D×D , through the mapping matrix W G For the initial acceleration time series segment X acc and the initial angular velocity time series segment X ang Perform mapping and splicing to obtain the gating matrix G=(x d ) T *W G , the weights of acceleration and angular velocity are matched through the gating matrix G to obtain the characteristic weight of acceleration and the characteristic weight of angular velocity.

[0066] In the above embodiment, by introducing a gated attention model and dynamically assigning weights to acceleration and angular velocity features, not only does the model's adaptability and accuracy in complex data environments improve, but it also improves the interpretability of the importance of acceleration and angular velocity features. Furthermore, the flexible weight assignment strategy further enhances the model's performance, including accuracy, robustness, and generalization, providing a more efficient and accurate solution for MEMS homodyne data prediction.

[0067] like Figure 3 As shown, preferably, a CNN-GRU multi-task prediction model is constructed, the acceleration containing feature weights and the angular velocity containing feature weights are used as inputs of the CNN-GRU prediction model, and the output features of the CNN-GRU prediction model are mapped and restored, and the network parameters of the CNN-GRU prediction model are updated according to the restoration results, thereby completing the training of the CNN-GRU multi-task prediction model, specifically:

[0068] Build feature extraction module, feature splicing module, feature fusion module and prediction module to obtain a preliminary CNN-GRU multi-task prediction model;

[0069] Inputting the acceleration containing the feature weight and the angular velocity containing the feature weight into the feature extraction module to perform feature extraction to obtain a plurality of feature maps;

[0070] Inputting the multiple feature maps into the feature splicing module for flattening to obtain a one-dimensional feature vector; specifically, the feature splicing module is composed of a feature map connection layer and a flattening layer; the feature splicing module is used to connect the feature maps extracted in the previous step together, and then flatten the multi-channel feature map into a one-dimensional feature vector;

[0071] Inputting the one-dimensional feature vector into the feature fusion module for fusion processing to obtain a fusion vector;

[0072] Inputting the fusion vector into the prediction module for prediction to obtain a prediction result, performing mapping restoration on the prediction result, and updating the network parameters of the CNN-GRU prediction model according to the restoration result;

[0073] Through the loss function The parameter gradient of the CNN-GRU model is calculated, and the parameters of the CNN-GRU prediction model are updated in reverse, thereby completing the training of the CNN-GRU multi-task prediction model.

[0074] In the above example, by automatically learning and extracting key features from the data through a CNN, combined with the advantages of a GRU for processing time series data, the model is able to capture complex relationships and temporal dependencies between data. Furthermore, the use of a multi-task prediction framework and a result mapping-reduction mechanism not only improves prediction accuracy but also enhances the model's flexibility and practicality.

[0075] like Figure 3 As shown, preferably, the feature extraction module includes a two-dimensional convolutional neural network, and the feature extraction expression of the two-dimensional convolutional neural network is:

[0076]

[0077] in, represents the j-th input signal feature (i.e., feature map) of the l-th layer, represents the convolution kernel weight between the i-th output feature signal of the l-1th layer and the j-th input feature signal of the lth layer, k represents the width of the lth layer, represents the i-th output feature signal of the l-1th layer, It represents the bias of the j-th signal feature in the l-th layer, and * represents the convolution operation.

[0078] Specifically, if Figure 3 As shown in the figure, the feature extraction module belongs to the CNN network, which is composed of a two-dimensional convolutional neural network, including three convolution layers and three pooling layers. The first convolution layer is a wide convolution kernel used to capture more feature information, and the remaining convolution layers use small convolution kernels.

[0079] In the above embodiment, feature extraction of MEMS time series data is performed through a CNN network, which not only improves the feature extraction capability of the model, but also enhances the computational efficiency, robustness and scalability of the model, providing a more accurate and comprehensive feature representation for subsequent prediction tasks.

[0080] like Figure 3 As shown, preferably, the one-dimensional feature vector is input into the feature fusion module for fusion processing to obtain a fusion vector, specifically:

[0081] The feature fusion module is composed of a GRU-based feature fusion layer;

[0082] The output of the feature fusion module is set using the dropout function, the dropout rate is set to 0.5, and the one-dimensional feature vector is input into the feature fusion module for fusion processing. The expression of the fusion processing is:

[0083]

[0084] R i =σ(U LR L i +W HR H i-1 +b R ),

[0085] S i =Mish(U LS L i +W HS (R i *H i-1 )+bS ),

[0086] H i =(1-Z i )*H i-1 +Z i *S i ,

[0087] Among them, U LZ 、U LR 、U LS Represents the weight connection matrix between the flattening layer and GRU, W HZ 、W HR 、W HS Represents the weight matrix b of the recursive GRU connection Z 、b R 、b S are the corresponding bias vectors, where:

[0088] Mish(x)=x*tanh(ln(1+e x )),

[0089] Among them, tanh() represents the tanh activation function, * represents the convolution operation, and x represents the signal feature, which is a matrix or vector.

[0090] In the above example, by defining the weight connection matrix between the flattening layer and the GRU, the weight matrix of the recursive GRU connection, and the bias vector, and combining activation functions with convolution operations, efficient feature vector fusion is achieved. This not only takes into account the linear and nonlinear relationships between features, but also prevents overfitting of the model through the dropout mechanism.

[0091] Preferably, the loss function Calculate the parameter gradient of the CNN-GRU model and update the parameters of the CNN-GRU prediction model in reverse, specifically:

[0092] The loss function The expression:

[0093]

[0094] Where M represents the number of time series. and are the true values ​​of acceleration and angular velocity, respectively. and They are respectively the acceleration prediction value (i.e. the acceleration change trend prediction value of the x / y / z axis) and the angular velocity prediction value (i.e. the angular velocity change trend prediction value of the x / y / z axis) output by the model.

[0095] In the above embodiment, the differences between the actual values ​​of acceleration and angular velocity and the model predicted values ​​are taken into account at the same time, which improves the overall performance of the model on multiple prediction tasks and enhances the generalization ability of the model, so that it can maintain stable prediction accuracy when facing new data.

[0096] like Figure 4 As shown, an embodiment of the present invention provides a device for predicting the change trend of a MEMS inertial sensor under high microgravity in space, comprising:

[0097] a preprocessing module, configured to collect raw MEMS time series data and perform data preprocessing on the raw MEMS time series data to obtain a plurality of initial time series segments, wherein the raw MEMS time series data includes time, temperature, acceleration, and angular velocity obtained under high microgravity in space;

[0098] a feature fusion module for performing block processing on the multiple initial time series segments, embedding the block sequence segments containing time data and temperature data, and fusing the embedded time series segments and temperature time series segments with the block acceleration time series segments and angular velocity time series segments respectively through a fully connected neural network to obtain fused features;

[0099] A weight distribution module, configured to construct a gated attention model based on the fused features, and distribute feature weights of acceleration and angular velocity through the gated attention model;

[0100] A model training module is used to construct a CNN-GRU multi-task prediction model, use the acceleration containing feature weights and the angular velocity containing feature weights as inputs of the CNN-GRU prediction model, perform result mapping restoration on the output features of the CNN-GRU prediction model, and update the network parameters of the CNN-GRU prediction model based on the restoration results, thereby completing the training of the CNN-GRU multi-task prediction model;

[0101] The change trend prediction module is used to predict the change trend of the to-be-processed acceleration and to-be-processed angular velocity of the MEMS three-axis inertial sensor through the trained CNN-GRU multi-task prediction model, and obtain the change trend prediction value corresponding to the to-be-processed acceleration and the change trend prediction value corresponding to the to-be-processed angular velocity.

[0102] An embodiment of the present invention also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are configured to execute the above-described method for predicting the change trend of a MEMS inertial sensor under high microgravity in space.

[0103] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the method for predicting the change trend of a MEMS inertial sensor under high microgravity in space as described above is implemented.

[0104] The advantages of the present invention are:

[0105] A CNN-GRU multi-task prediction model was constructed. The fusion features obtained by fusing time series segments and temperature series segments with the block-based acceleration time series segments and angular velocity time series segments were used to train the CNN-GRU multi-task prediction model. This model takes into account the effective interaction between the acceleration and angular velocity change trend prediction subtasks, the specific modeling between the subtasks, and the model inference accuracy, thus achieving accurate and rapid change trend prediction.

[0106] This method solves the problem of simultaneously modeling subtask interactions and specificity in the task of predicting acceleration and angular velocity trends, and the algorithm achieves high inference speed. Experimental results on a dataset of actual MEMS three-axis inertial sensors from a high-micro cabinet show that the mean square error (MSE) for acceleration trend prediction reaches 2.65e-5, and the mean square error for angular velocity trend prediction reaches 6.086. This achieves high-precision and rapid trend prediction in the field of MEMS inertial sensors, meeting the algorithmic requirements for practical engineering applications.

[0107] In some schemes, multiple embodiments of the present application can be combined and the combined scheme can be implemented. Optionally, some operations in the process of each method embodiment are optionally combined, and / or the order of some operations is optionally changed. In addition, the execution order between the steps of each process is only exemplary and does not constitute a limitation on the execution order between the steps. There can also be other execution orders between the steps. It is not intended to indicate that the execution order is the only order in which these operations can be performed. Ordinary technicians in this field will think of many ways to reorder the operations described herein. In addition, it should be noted that the process details involved in a certain embodiment of this article are also applicable to other embodiments in a similar manner, or different embodiments can be used in combination.

[0108] Furthermore, some steps in the method embodiments may be equivalently replaced with other possible steps. Alternatively, some steps in the method embodiments may be optional and may be deleted in certain usage scenarios. Alternatively, other possible steps may be added to the method embodiments.

[0109] Furthermore, the various method embodiments may be implemented separately or in combination.

[0110] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely schematic. For example, the division of the modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of systems or units, which can be electrical, mechanical or other forms.

[0111] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for predicting the change trend of MEMS inertial sensors under high microgravity in space, characterized by: include: S1. Collecting raw MEMS time series data and performing data preprocessing on the raw MEMS time series data to obtain multiple initial time series segments, wherein the raw MEMS time series data includes time, temperature, acceleration, and angular velocity obtained under high microgravity in space; S2. Blocking the multiple initial time series segments, embedding the block sequences containing time data and temperature data, and fusing the embedded time series segments and temperature series segments with the block acceleration time series segments and angular velocity time series segments through a fully connected neural network to obtain fused features; S3. Building a gated attention model based on the fusion features, and assigning feature weights of acceleration and angular velocity through the gated attention model; S4. Construct a CNN-GRU multi-task prediction model, use the acceleration containing feature weights and the angular velocity containing feature weights as inputs of the CNN-GRU prediction model, perform result mapping restoration on the output features of the CNN-GRU prediction model, and update the network parameters of the CNN-GRU prediction model according to the restoration results, thereby completing the training of the CNN-GRU multi-task prediction model; S5. Predict the change trend of the acceleration to be processed and the angular velocity to be processed of the MEMS inertial sensor through the trained CNN-GRU multi-task prediction model to obtain the change trend prediction value corresponding to the acceleration to be processed and the change trend prediction value corresponding to the angular velocity to be processed.

2. The method for predicting the change trend of MEMS inertial sensors under high microgravity in space according to claim 1, characterized in that: The original MEMS time series data is preprocessed to obtain multiple initial time series segments, specifically: The time, temperature, acceleration and angular velocity in the original MEMS timing data are normalized by a normalization formula, and the normalization formula is: Among them, xmean is the mean of different parameters in the original MEMS time series data, xvar is the variance of different parameters, and x i is the i-th value in the input window length L; The normalized raw MEMS time series data is decomposed into multiple time series samples according to a set fixed length. The multiple time series samples are clustered and analyzed using a clustering algorithm to obtain the cluster distance of each time series sample. The sample data with a cluster distance greater than a set cluster distance threshold is removed to obtain multiple initial time series segments.

3. The method for predicting the change trend of MEMS inertial sensors under high microgravity in space according to claim 1, characterized in that: The multiple initial time sequence segments are divided into blocks, specifically: The multiple initial time sequence segments are divided into blocks, and each block-processed time sequence segment is expressed as The time series segments include a time series segment containing time data, a temperature series segment containing temperature data, an acceleration series segment containing acceleration data, and an angular velocity series segment containing angular velocity data. Based on the expression of each block, the initial time series segment is expressed as: Where X is the time series segment after block division, R is the real number field, P×N represents a matrix with width P and length N, N is the number of blocks, L is the input window length, P is the length of the time series segment, and S is the block sliding distance.

4. The method for predicting the change trend of MEMS inertial sensors under high microgravity in space according to claim 3, characterized in that: The sequence containing time data and temperature data after segmentation is embedded and represented. The embedded time series segments and temperature series segments are fused with the segmented acceleration time series segments and angular velocity time series segments through a fully connected neural network to obtain fused features, specifically: Based on the expression of the initial time series segment, the temperature sequence segment is mapped into a temperature feature matrix Wtemp∈R D×P , and map the time series segments into a time feature matrix Wtime∈R D×P ; Construct the learning matrix W acc ∈R D×P and W ang ∈R D×P , by learning the matrix W acc ∈R D×P and W ang ∈R D×P The acceleration time series segment and the angular velocity time series segment are respectively mapped into acceleration feature matrices X acc ∈R D×N And the angular velocity characteristic matrix X ang ∈R D×N , where D×P represents a matrix with width D and length P; The temperature feature matrix and the time feature matrix are fused with the acceleration feature matrix and the angular velocity feature matrix respectively through a fully connected neural network to obtain the fused features corresponding to the acceleration and the fused features corresponding to the angular velocity: in, is the i-th variable in the initial acceleration time series segment, is the i-th variable in the initial angular velocity time series segment.

5. The method for predicting the change trend of MEMS inertial sensors under high microgravity in space according to claim 1, characterized in that: A gated attention model is constructed based on the fusion features, and feature weights of acceleration and angular velocity are allocated through the gated attention model, specifically: Constructing the mapping matrix And through the mapping matrix and And the attention score calculation formula for the fused acceleration and angular velocity information Calculate and obtain the attention score. The calculation formula of the attention score is as follows: in, By the mapping matrix get, By the mapping matrix get, By the mapping matrix get; Construct the mapping matrix W G ∈R D×D , through the mapping matrix W G For the initial acceleration time series segment X acc and the initial angular velocity time series segment X ang Perform mapping and splicing to obtain the gating matrix G=(x d ) T *W G , the weights of acceleration and angular velocity are matched through the gating matrix G to obtain the characteristic weight of acceleration and the characteristic weight of angular velocity.

6. The method for predicting the change trend of MEMS inertial sensors under high microgravity in space according to claim 1, characterized in that: Construct a CNN-GRU multi-task prediction model, use the acceleration with feature weights and the angular velocity with feature weights as the input of the CNN-GRU prediction model, perform result mapping restoration on the output features of the CNN-GRU prediction model, and update the network parameters of the CNN-GRU prediction model according to the restoration results, thereby completing the training of the CNN-GRU multi-task prediction model, specifically: Build feature extraction module, feature splicing module, feature fusion module and prediction module to obtain a preliminary CNN-GRU multi-task prediction model; Inputting the acceleration containing the feature weight and the angular velocity containing the feature weight into the feature extraction module to perform feature extraction to obtain a plurality of feature maps; Inputting the plurality of feature maps into the feature splicing module for flattening to obtain a one-dimensional feature vector; Inputting the one-dimensional feature vector into the feature fusion module for fusion processing to obtain a fusion vector; Inputting the fusion vector into the prediction module for prediction to obtain a prediction result, performing mapping restoration on the prediction result, and updating the network parameters of the CNN-GRU prediction model according to the restoration result; Through the loss function The parameter gradient of the CNN-GRU model is calculated, and the parameters of the CNN-GRU prediction model are updated in reverse, thereby completing the training of the CNN-GRU multi-task prediction model.

7. The method for predicting the change trend of MEMS inertial sensors under high microgravity in space according to claim 6, characterized in that: The feature extraction module includes a two-dimensional convolutional neural network, and the feature extraction expression of the two-dimensional convolutional neural network is: in, is the j-th input signal feature of the l-th layer, is the convolution kernel weight between the i-th output feature signal of the l-1th layer and the j-th input feature signal of the lth layer, k is the width of the lth layer, is the i-th output feature signal of the l-1th layer, is the bias of the j-th signal feature in the l-th layer, and * is the convolution operation.

8. The method for predicting the change trend of MEMS inertial sensors under high microgravity in space according to claim 6, characterized in that: The one-dimensional feature vector is input into the feature fusion module for fusion processing to obtain a fusion vector, specifically: The output of the feature fusion module is set using the dropout function, the dropout rate is set to 0.5, and the one-dimensional feature vector is input into the feature fusion module for fusion processing. The expression of the fusion processing is: R i =σ(U LR L i +W HR H i-1 +b R ), S i =Mish(U LS L i +W HS (R i *H i-1 )+b S ), H i =(1-Z i )*H i-1 +Z i *S i , Among them, U LZ 、U LR 、U LS are the weight connection matrices between the flattening layer and GRU, W HZ 、W HR 、W HS are the weight matrices b of the recursive GRU connections respectively Z 、b R 、b S are the corresponding bias vectors, where: Mish(x)=x*tanh(ln(1+e x )), Among them, tanh() is the tanh activation function, * is the convolution operation, and x is the signal feature, which is a matrix or vector.

9. The method for predicting the change trend of MEMS inertial sensors under high microgravity in space according to claim 6, characterized in that: The parameter gradient of the CNN-GRU model is calculated by the loss function L, and the parameters of the CNN-GRU prediction model are updated in reverse, specifically: The loss function The expression: Where M is the number of time series, and are the true values ​​of acceleration and angular velocity, respectively. and are the acceleration and angular velocity prediction values ​​output by the model respectively.

10. A device for predicting the change trend of a MEMS inertial sensor under high microgravity in space, characterized in that: include: a preprocessing unit, configured to collect raw MEMS time series data and perform data preprocessing on the raw MEMS time series data to obtain a plurality of initial time series segments, wherein the raw MEMS time series data includes time, temperature, acceleration, and angular velocity obtained under high microgravity in space; a feature processing unit configured to perform block processing on the plurality of initial time series segments, embed the block sequence segments containing time data and temperature data, and fuse the embedded time series segments and temperature series segments with the block acceleration time series segments and angular velocity time series segments through a fully connected neural network to obtain fused features; a weight allocation unit, configured to construct a gated attention model based on the fused features, and allocate feature weights of acceleration and angular velocity using the gated attention model; A model training unit is used to construct a CNN-GRU multi-task prediction model, use the acceleration containing feature weights and the angular velocity containing feature weights as inputs of the CNN-GRU prediction model, perform result mapping restoration on the output features of the CNN-GRU prediction model, and update the network parameters of the CNN-GRU prediction model according to the restoration results, thereby completing the training of the CNN-GRU multi-task prediction model; The change trend prediction unit is used to predict the change trend of the to-be-processed acceleration and to-be-processed angular velocity of the MEMS three-axis inertial sensor through the trained CNN-GRU multi-task prediction model, and obtain the change trend prediction value corresponding to the to-be-processed acceleration and the change trend prediction value corresponding to the to-be-processed angular velocity.

Images