Pressure response data calibration method, device and computer equipment

Abstract

The present invention relates to the field of high-precision pressure sensing technology, and discloses a pressure response data calibration method, device and computer equipment, the method comprising: collecting output response data of a pressure sensor array, and constructing a spatial feature matrix, a time feature matrix and an environmental feature matrix based on the output response data; performing trilinear least squares separation to obtain dynamic response data and time characteristic data; performing symmetric geometric modal decomposition to obtain linear response component data, periodic fluctuation component data and nonlinear characteristic component data; inputting a multi-branch gated recurrent unit network model to perform dynamic calibration strategy prediction to obtain a first dynamic calibration strategy; performing particle swarm iterative optimization to obtain a second dynamic calibration strategy, and generating a progressive calibration parameter set; performing static and dynamic pressure sensing calibration on the pressure sensor array to obtain a comprehensive pressure sensing calibration result, thereby improving the accuracy of the pressure sensing calibration.

Description

Technical Field

[0001] The present invention relates to the field of high-precision pressure sensing technology, and in particular to a pressure response data calibration method, device and computer equipment. Background Art

[0002] The dynamic response characteristics of the pressure sensor array are affected by multiple factors such as environmental noise, temperature and humidity changes, and nonlinear interference, which makes it difficult for traditional static calibration methods to meet the needs of high-precision dynamic measurement of pressure.

[0003] At present, the dynamic calibration of pressure sensors mainly adopts a single linear or nonlinear compensation method, which cannot effectively deal with the superposition effect of multi-source pressure signals and the coupling influence of environmental factors. At the same time, the existing calibration methods often simplify the dynamic characteristics into the superposition of static characteristics, ignoring the dynamic response differences of sensors under different frequency and amplitude conditions. Summary of the invention

[0004] The present invention provides a pressure response data calibration method, device and computer equipment, thereby improving the accuracy of pressure sensor calibration.

[0005] In a first aspect, the present invention provides a method for calibrating pressure response data, the method for calibrating pressure response data comprising:

[0006] Collecting output response data of the pressure sensor array, and constructing a spatial feature matrix, a temporal feature matrix and an environmental feature matrix based on the output response data;

[0007] Performing trilinear least square separation on the spatial feature matrix, the temporal feature matrix and the environmental feature matrix to obtain dynamic response data and temporal characteristic data;

[0008] Performing symmetrical geometric modal decomposition on the dynamic response data and the time characteristic data to obtain linear response component data, periodic fluctuation component data and nonlinear characteristic component data;

[0009] Inputting the linear response component data, the periodic fluctuation component data and the nonlinear characteristic component data into a multi-branch gated recurrent unit network model to perform dynamic calibration strategy prediction to obtain a first dynamic calibration strategy;

[0010] Performing particle swarm iterative optimization on the first dynamic calibration strategy to obtain a second dynamic calibration strategy, and generating a progressive calibration parameter set based on the second dynamic calibration strategy;

[0011] Static and dynamic pressure sensing calibration is performed on the pressure sensor array based on the progressive calibration parameter set to obtain a comprehensive pressure sensing calibration result.

[0012] In a second aspect, the present invention provides a pressure response data calibration device, the pressure response data calibration device comprising:

[0013] An acquisition module, used for acquiring output response data of the pressure sensor array, and constructing a spatial feature matrix, a temporal feature matrix and an environmental feature matrix based on the output response data;

[0014] A separation module, used for performing trilinear least square separation on the spatial feature matrix, the temporal feature matrix and the environmental feature matrix to obtain dynamic response data and temporal characteristic data;

[0015] A decomposition module, used for performing symmetrical geometric modal decomposition on the dynamic response data and the time characteristic data to obtain linear response component data, periodic fluctuation component data and nonlinear characteristic component data;

[0016] A prediction module, used for inputting the linear response component data, the periodic fluctuation component data and the nonlinear characteristic component data into a multi-branch gated recurrent unit network model to perform dynamic calibration strategy prediction to obtain a first dynamic calibration strategy;

[0017] an optimization module, configured to perform particle swarm iterative optimization on the first dynamic calibration strategy to obtain a second dynamic calibration strategy, and generate a progressive calibration parameter set based on the second dynamic calibration strategy;

[0018] The calibration module is used to perform static and dynamic pressure sensing calibration on the pressure sensor array based on the progressive calibration parameter set to obtain a comprehensive pressure sensing calibration result.

[0019] A third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned pressure response data calibration method.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned pressure response data calibration method.

[0021] In the technical solution provided by the present invention, the spatial characteristics, temporal characteristics and environmental characteristics of the pressure sensor are accurately extracted through parallel factor decomposition and trilinear least squares separation technology, the influence of environmental noise on dynamic calibration is reduced, and the sensor characteristics are decomposed into three components of linear response, periodic fluctuation and nonlinear characteristics by using a symmetric geometric modal decomposition method, thereby realizing the accurate identification of multi-dimensional features. A dynamic calibration strategy based on a multi-branch gated recurrent unit network model improves the timing performance of dynamic calibration through bidirectional feature extraction and a time node attention mechanism, and constructs a progressive calibration parameter system based on particle swarm optimization, realizing hierarchical optimization of linear calibration, periodic calibration and nonlinear calibration. Combined with the comprehensive test method of static and dynamic pressure calibration, a complete calibration evaluation system is established through multi-modal calibration characteristic analysis, thereby improving the accuracy of pressure sensor calibration. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 A schematic diagram of the steps of a method for calibrating pressure response data in an embodiment of the present invention;

[0024] Figure 2 It is a schematic diagram of the structure of a pressure response data calibration device according to an embodiment of the present invention;

[0025] Figure 3 It is a schematic block diagram of the structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] Embodiments of the present invention provide a method, apparatus, and computer device for calibrating pressure response data. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0027] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of a method for calibrating pressure response data in an embodiment of the present invention includes:

[0028] Step S1, collecting output response data of the pressure sensor array, and constructing a spatial feature matrix, a temporal feature matrix and an environmental feature matrix based on the output response data;

[0029] It is understandable that the execution subject of the present invention may be a pressure response data calibration device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0030] In this embodiment, the pressure sensor refers to a sensing element that can convert a pressure signal into an electrical signal, has the characteristics of high precision, high sensitivity, etc., and is used for pressure detection. The pressure sensor array is a sensor array system composed of multiple pressure sensors arranged in a specific spatial manner. Each sensor can independently sense the pressure changes at its location and output a corresponding electrical signal. This array arrangement not only improves the spatial resolution of pressure detection, but also improves the system's anti-interference ability and measurement reliability through multi-point collaborative measurement. The pressure sensor array needs to maintain stable measurement performance under different environmental conditions such as temperature and humidity, which requires high-precision dynamic calibration to ensure the accuracy and reliability of the measurement data.

[0031] Specifically, the standard pressure source is controlled at multiple frequencies to generate periodic pressure fluctuation signals containing different frequency and amplitude combinations. The control of the standard pressure source is of great significance. By applying pressure fluctuation signals of different frequencies and amplitudes, various dynamic pressure conditions that the sensor may encounter are fully simulated, thereby effectively evaluating the dynamic response characteristics of the sensor. At the same time, the temperature and humidity environmental parameters of the pressure sensor array are set to generate multiple sets of environmental condition combination data to examine the behavior of the sensor under the action of multiple environmental variables. The combination of different environmental conditions includes different temperature and humidity settings to simulate the complex environment faced by the pressure sensor in real application scenarios. Under multiple sets of environmental condition combination data, the pressure sensor array is collected based on the periodic pressure fluctuation signal to obtain the original response data. According to the spatial position, time sampling sequence and environmental condition sequence of the pressure sensor, the original response data is reconstructed in three dimensions to obtain three-dimensional tensor structure data. The output data of the sensor is arranged in order according to the three dimensions of space, time and environmental conditions to obtain an overall tensor structure data so that the feature decomposition can be effectively carried out.

[0032] The three-dimensional tensor structure data is input into the parallel factor decomposition model for feature decomposition, and useful feature information is extracted from these multidimensional data. The parallel factor decomposition model can decompose complex multidimensional data into a series of simple, physically meaningful components, which is easy to understand and analyze, and obtain the initial feature decomposition data. The minimum reconstruction error threshold is set for the initial feature decomposition data, and iterative calculation is performed on this basis to obtain the final tensor decomposition rank. The setting of the minimum reconstruction error threshold is to ensure the decomposition quality of the model for the tensor data, that is, to control the error between the reconstructed tensor and the original tensor within a reasonable range, and to ensure that the decomposed features can effectively reflect the characteristics of the original data. Based on the obtained tensor decomposition rank, the initial feature decomposition data is optimized by orthogonal constraints to obtain feature orthogonalized data. The purpose of orthogonal constraint optimization is to ensure that the features obtained by each decomposition are independent of each other. The orthogonalized feature data can reduce the redundant information between different features and improve the independence and effectiveness of the data. According to the spatial dimension, time dimension and environmental dimension, the feature orthogonalized data is reorganized to obtain the spatial feature matrix, the temporal feature matrix and the environmental feature matrix. These characteristic matrices are constructed to describe the output characteristics of the pressure sensor array at different spatial locations, different time nodes and different environmental conditions.

[0033] Step S2, performing trilinear least square separation on the spatial feature matrix, the temporal feature matrix and the environmental feature matrix to obtain dynamic response data and temporal characteristic data;

[0034] Specifically, the spatial feature matrix, the temporal feature matrix and the environmental feature matrix are concatenated to obtain a combined feature matrix. The combined feature matrix is ​​decomposed to obtain a left singular matrix, a singular value matrix and a right singular matrix. The singular value decomposition technique is used for decomposition. By decomposing the feature matrix, the main features and structures in the data can be effectively extracted, and the most important factors affecting the sensor response can be found. The singular value matrix is ​​sorted in descending order to ensure that the maximum amount of information is extracted. After sorting in descending order, the eigenvector corresponding to the maximum singular value is selected to obtain the main eigenvector data. The importance of different eigenvectors is measured by the size of the singular value, and the most representative eigenvector is extracted. The main eigenvector data can reflect the main components in the pressure sensor response, has strong representativeness and robustness, and can significantly simplify the complexity of subsequent models.

[0035] The transfer matrix of the main eigenvector data is constructed, and the transfer function matrix is ​​obtained through iterative calculation of trilinear least squares. The transfer matrix is ​​optimized through the trilinear least squares method so that it can effectively describe the relationship between the sensor input and output. The trilinear least squares method can simultaneously consider the data of three dimensions of space, time and environment, ensure the optimal fitting in multidimensional space, and accurately construct the transfer characteristics of the sensor. The transfer function matrix is ​​transformed into the frequency domain to obtain the frequency response function data. The frequency response function describes the response characteristics of the system at different frequencies, which helps to analyze the dynamic performance of the pressure sensor under different frequency inputs. Through frequency domain transformation, the complex changes in the time domain are converted into characteristic performance in the frequency domain.

[0036] In order to improve the validity of data and simplify processing, the frequency response function data is sparsely reconstructed based on the distributed compressed sensing algorithm to obtain the dynamic feature vector. The distributed compressed sensing algorithm can effectively utilize the sparse characteristics of data to extract the most representative features from a large amount of data, while compressing redundant information to achieve data simplification and effective reconstruction. The dynamic feature vector is analyzed in the time domain to obtain dynamic response data. The time domain response analysis can reveal the dynamic change characteristics of the pressure sensor in the time dimension, such as the response amplitude and phase changes at different time points. The dynamic response data is filtered based on the transfer function matrix to obtain the time characteristic data. The noise components in the dynamic response data are removed, making the final time characteristic data smoother and more accurate.

[0037] Step S3, performing symmetrical geometric modal decomposition on the dynamic response data and the time characteristic data to obtain linear response component data, periodic fluctuation component data and nonlinear characteristic component data;

[0038] Specifically, a symmetrical geometric transformation is performed on the dynamic response data and the time characteristic data to obtain a characteristic decomposition data group. By processing the data in a symmetrical manner, the dynamic response and the time characteristic can be expressed in a balanced manner in terms of geometric characteristics, thereby reducing errors caused by data skew. The characteristic decomposition data group is input into the maximum entropy spectrum analysis model for recursive calculation to obtain characteristic distribution data. Maximum entropy spectrum analysis is a spectrum estimation method that can extract the spectral characteristics of data to the maximum extent under the condition of limited data. Through recursive calculation, the spectrum estimation value is repeatedly corrected so that the obtained characteristic distribution data can accurately reflect the inherent regularity of the dynamic response data and the time characteristic data. The characteristic distribution data describes the distribution of the dynamic response of the pressure sensor at different frequencies and characteristic dimensions, reflecting the basic dynamic characteristics of the system.

[0039] The gain coefficient analysis is performed on the characteristic distribution data to obtain the linear dynamic parameter matrix. The gain coefficient analysis is used to measure the response gain of the system under different input conditions and can effectively extract the linear characteristics of the system. On this basis, zero drift feature extraction and deviation compensation are performed based on the linear dynamic parameter matrix to obtain linear response component data. The purpose of zero drift feature extraction is to eliminate the zero point offset caused by temperature drift, aging and other factors in the system, while the deviation compensation is to correct these offsets to ensure that the linear response component data can truly reflect the linear behavior of the sensor. At the same time, the characteristic distribution data is decomposed by wavelet to obtain the periodic signal feature vector. Wavelet decomposition is an effective time-frequency analysis method that can decompose the signal into components of different scales and frequencies and capture the periodic characteristics in the signal. The periodic signal feature vector is input into the harmonic component identification model for adaptive threshold segmentation to distinguish the main periodic components and obtain the periodic fluctuation component data. Adaptive threshold segmentation can automatically adjust the threshold according to the distribution characteristics of the characteristic signal, thereby ensuring the accuracy and adaptability of the segmentation, so that the periodic fluctuation component data can accurately describe the periodic influence on the sensor.

[0040] Based on the characteristic distribution data, the second-order and third-order Volterra nonlinear kernel functions are constructed and iteratively optimized to obtain the nonlinear characteristic coefficients. The Volterra kernel function is a generalized nonlinear system model. By constructing the second-order and third-order kernel functions, the nonlinear characteristics in the sensor response can be effectively captured. The iterative optimization process is used to repeatedly adjust the parameters of the kernel function to ensure that the kernel function can accurately fit the nonlinear behavior of the sensor and obtain stable nonlinear characteristic coefficients. The nonlinear characteristic coefficients are reconstructed by kernel function and nonlinearly mapped to obtain nonlinear characteristic component data. The purpose of kernel function reconstruction is to recombine the extracted nonlinear features to form a complete nonlinear response model, while nonlinear mapping is used to apply these features to the actual dynamic response to reveal the dynamic characteristics of the system under nonlinear effects.

[0041] Step S4, inputting the linear response component data, the periodic fluctuation component data and the nonlinear characteristic component data into a multi-branch gated recurrent unit network model to perform dynamic calibration strategy prediction to obtain a first dynamic calibration strategy;

[0042] Specifically, the linear response component data is input into the first bidirectional gated recurrent unit in the multi-branch gated recurrent unit (GRU) network model for bidirectional feature extraction. The first bidirectional GRU unit has a forward hidden layer and a reverse hidden layer, each of which contains an update gate, a reset gate, and a candidate hidden state. These structures work together to capture the time dependency and characteristic trend of the linear response. Through the bidirectional calculation of the forward hidden layer and the reverse hidden layer, the linear response data is fully feature extracted in the forward and reverse directions of the time axis, ensuring that the model can accurately understand the dynamic changes of the linear characteristics of the pressure sensor at different time nodes. The obtained linear branch feature data describes the complete characteristics of the linear response in time series, which helps to fine-tune the linear part in subsequent calibration.

[0043] At the same time, the periodic fluctuation component data is input into the second bidirectional gated recurrent unit in the multi-branch GRU network model for bidirectional feature extraction to obtain periodic branch feature data. The structure of the second bidirectional GRU is the same as that of the first bidirectional GRU, and both ensure the comprehensive capture of periodic fluctuation signals through forward and reverse feature extraction. When processing periodic signals, the bidirectional GRU helps to identify periodic patterns and trends in the signal, ensuring that periodic characteristics can be effectively captured at different time scales.

[0044] The nonlinear characteristic component data is input into the third bidirectional gated recurrent unit in the multi-branch GRU network model for bidirectional feature extraction to obtain nonlinear branch feature data. The structure of the third bidirectional GRU is also consistent with the previous two. Through the collaborative work of the forward hidden layer and the reverse hidden layer, the complex changing characteristics of the nonlinear response are accurately captured. The nonlinear response is caused by the non-ideal characteristics of the pressure sensor element in extreme environments. This complexity makes the bidirectional GRU an ideal tool for processing such data. It can extract the key features of nonlinearity from the data and ensure that the subsequent calibration strategy can effectively compensate and adjust these nonlinear errors.

[0045] The linear branch feature data, periodic branch feature data, and nonlinear branch feature data are subjected to time node attention calculation to obtain weighted feature data. The attention mechanism enables the model to dynamically assign attention weights at different time nodes, especially at specific time points, when some features are more important than other features. Through weighted calculation, the feature extraction process is ensured to be more flexible and efficient, focusing on feature data that are important to the calibration strategy and improving the model's responsiveness to complex pressure changes. The weighted feature data is input into the feature fusion layer for adaptive feature combination to obtain multi-branch fusion feature data. The features from the three branches of linear, periodic, and nonlinear are effectively combined. The adaptive feature combination method dynamically adjusts the weights of different features in the fusion process according to their importance in the current situation, achieving more refined and targeted feature synthesis.

[0046] Perform environmental variable feature correlation analysis on the multi-branch fusion feature data to obtain calibration feature data. Associate the dynamic response of the sensor with external environmental factors to better understand the impact of the environment on sensor performance and ensure that the calibration strategy can fully consider the factors affecting changes in the external environment. Through analysis, the impact characteristics of the environment on the dynamic response of the sensor are effectively extracted. Input the calibration feature data into the dynamic prediction layer for time series reasoning to obtain the first dynamic calibration strategy. The dynamic prediction layer predicts the future response of the sensor based on the calibration feature data. The process of time series reasoning considers the historical response of the sensor and the changes in the current environment, so as to make an accurate estimate of its future behavior. The first dynamic calibration strategy obtained provides the best calibration parameter settings for the use of pressure sensing elements in different dynamic environments, ensuring that it can maintain high-precision measurement performance under different conditions.

[0047] Step S5, performing particle swarm iterative optimization on the first dynamic calibration strategy to obtain a second dynamic calibration strategy, and generating a progressive calibration parameter set based on the second dynamic calibration strategy;

[0048] Specifically, the linear calibration parameters, periodic calibration parameters and nonlinear calibration parameters in the first dynamic calibration strategy are linearly transformed, and various parameters in the original calibration strategy are uniformly converted into vector form to obtain the initial optimization parameter vector. The size of the particle swarm is set to 100 based on the initial optimization parameter vector, and the search dimension is set to the dimension of the initial optimization parameter vector. The initial position matrix and velocity matrix of the particle swarm are obtained by random initialization of Gaussian distribution. The particle swarm size is set to 100 to ensure that there are a sufficient number of candidate solutions in the parameter search space to improve the global search capability, and the use of Gaussian distribution for random initialization gives the particle swarm a better distribution characteristic at the beginning, which helps to avoid falling into the local optimum. The initial position matrix and velocity matrix represent the position and movement speed of each particle in the particle swarm, respectively.

[0049] The initial position matrix and velocity matrix are input into the dynamic calibration objective function, and the initial fitness data is obtained by calculating the weighted sum of the calibration error and time cost. The calibration objective function is set to consider the balance between the calibration error and time cost, and the quality of each particle solution is evaluated by weighted sum. Based on the initial fitness data, the inertia weight coefficient, individual learning factor and social learning factor are set, and the speed and position are updated to obtain the updated position matrix. The inertia weight coefficient is used to control the search speed of the particle swarm to maintain the diversity of the population and avoid premature convergence; the individual learning factor and the social learning factor measure the ability of particles to learn from their own experience and the overall experience of the population, respectively. The reasonable setting of these factors can enable particles to achieve a balance between local search and global search. Through the setting and updating rules of these parameters, the speed and position of the particles are continuously adjusted and gradually approach the optimal solution. After each iteration, the speed and position are adjusted according to the fitness value of the particle to ensure that the particle can continue to move in the direction of reducing the calibration error and time cost.

[0050] The updated position matrix is ​​calculated for local and global optimal solutions to obtain the optimal parameter combination for the current iteration. The local optimal solution represents the best solution found by each particle in its historical position, while the global optimal solution represents the best solution found so far in the entire particle swarm. By combining the calculation of local and global optimal solutions, the particle swarm can be more effectively guided to search in the direction of the global optimal solution. The optimal parameter combination of the current iteration is inversely linearly transformed to obtain the second dynamic calibration strategy, ensuring that the optimized parameters can be remapped back to the original calibration parameter space.

[0051] Based on the second dynamic calibration strategy, a three-level calibration sequence including a linear calibration layer, a periodic calibration layer and a nonlinear calibration layer is constructed. Each layer of the calibration sequence corresponds to a different type of calibration task, and multi-level calibration parameters are obtained through parameter decoupling. The linear calibration layer is used to compensate for the linear error of the pressure sensor element, the periodic calibration layer is adjusted for periodic interference, and the nonlinear calibration layer is used to deal with the deviation of the nonlinear response. The parameters of these calibration layers are decoupled separately so that each layer can work independently, ensuring that different types of errors can be gradually eliminated during the calibration process. In the order of linear calibration, periodic calibration and nonlinear calibration, the multi-level calibration parameters are progressively combined to obtain a progressive calibration parameter set.

[0052] Step S6: Perform static and dynamic pressure sensing calibration on the pressure sensor array based on the progressive calibration parameter set to obtain a comprehensive pressure sensing calibration result.

[0053] Specifically, the progressive calibration parameter set is input into the static pressure calibration module. In the static pressure calibration module, the pressure sensor array is step-by-step applied with a standard pressure device within a preset range to obtain initial static response data. The step-by-step pressure application method accurately controls the applied pressure value, gradually increases it to fully cover the entire range, and can capture the static response characteristics of the pressure sensor at different pressure values.

[0054] The initial static response data is subjected to multiple cycles of pressurization and depressurization to obtain hysteresis characteristic data. Hysteresis is caused by energy loss inside the material or hysteresis effect of the physical structure, which has a significant impact on the measurement accuracy of the pressure sensor. Through multiple cycles of pressurization and depressurization, the hysteresis characteristics of the sensor are effectively captured and its impact is quantified.

[0055] Linearity compensation is performed based on the hysteresis characteristic data to obtain a static calibration coefficient group. The errors caused by the hysteresis effect and other static nonlinear characteristics are eliminated so that the output of the sensor can better correspond to the actual applied pressure value. The obtained static calibration coefficient group is used to correct the deviation in the static measurement process. At the same time, the progressive calibration parameter set is input into the dynamic pressure calibration module, and the sensor array is dynamically tested by a sinusoidal excitation signal to obtain frequency scanning response data. The purpose of dynamic pressure calibration is to examine the response characteristics of the sensor at different frequencies in order to accurately evaluate its dynamic performance. During the frequency scanning process, by gradually changing the frequency of the excitation signal, the amplitude and phase response of the pressure sensor at different frequencies are obtained, which fully reflects the performance characteristics of the sensor in a dynamic environment.

[0056] The frequency sweep response data is analyzed for amplitude-frequency characteristics, and the dynamic tracking error is calculated to obtain the dynamic amplitude calibration parameters. The amplitude-frequency characteristic analysis is used to examine the amplitude changes of the sensor at different frequencies, while the calculation of the dynamic tracking error can reflect the sensor's ability to respond to rapidly changing pressure signals. Based on these analysis results, the dynamic amplitude is calibrated to ensure that the amplitude response of the sensor in dynamic conditions has a high degree of accuracy.

[0057] The phase-frequency characteristic analysis of the frequency sweep response data is performed to identify the phase change of the pressure sensor at different frequencies, and based on this, phase delay compensation is performed to obtain dynamic phase calibration parameters. Through phase delay compensation, the phase offset is effectively reduced, making the phase response of the sensor to the dynamic pressure signal more accurate. Based on the static calibration coefficient group, dynamic amplitude calibration parameters and dynamic phase calibration parameters, multi-modal calibration characteristic data is generated. Through comprehensive analysis of the multi-modal calibration characteristic data, the results of static calibration and dynamic calibration are unified, and a comprehensive evaluation is performed in multiple dimensions to obtain the comprehensive pressure sensor calibration results. Comprehensive analysis evaluates the overall performance of the sensor by analyzing the static, amplitude and phase calibration errors to obtain the final calibration results.

[0058] In the embodiment of the present invention, the spatial characteristics, temporal characteristics and environmental characteristics of the pressure sensor are accurately extracted through parallel factor decomposition and trilinear least squares separation technology, and the influence of environmental noise on dynamic calibration is reduced. The symmetric geometric modal decomposition method is adopted to decompose the sensor characteristics into three components of linear response, periodic fluctuation and nonlinear characteristics, thereby realizing the accurate identification of multi-dimensional features. The dynamic calibration strategy based on the multi-branch gated recurrent unit network model improves the timing performance of dynamic calibration through bidirectional feature extraction and time node attention mechanism, and constructs a progressive calibration parameter system based on particle swarm optimization, realizing hierarchical optimization of linear calibration, periodic calibration and nonlinear calibration. Combined with the comprehensive test method of static and dynamic pressure calibration, a complete calibration evaluation system is established through multi-modal calibration characteristic analysis, thereby improving the accuracy of pressure sensor calibration.

[0059] In a specific embodiment, the process of executing step S1 may specifically include the following steps:

[0060] (1) Perform multi-frequency control on the standard pressure source to generate periodic pressure fluctuation signals containing different frequency and amplitude combinations, and set the temperature and humidity environmental parameters of the pressure sensor array to generate multiple sets of environmental condition combination data;

[0061] (2) Under multiple sets of environmental condition combination data, the pressure sensor array is collected based on the periodic pressure fluctuation signal to obtain the original response data;

[0062] (3) According to the spatial position of the pressure sensor, the time sampling sequence and the environmental condition sequence, the original response data is reconstructed into three-dimensional data to obtain three-dimensional tensor structure data;

[0063] (4) Inputting the three-dimensional tensor structure data into the parallel factor decomposition model for feature decomposition to obtain initial feature decomposition data, and setting a minimum reconstruction error threshold and iterative calculation for the initial feature decomposition data to obtain the tensor decomposition rank;

[0064] (5) Based on the tensor decomposition rank, the initial feature decomposition data is optimized with orthogonal constraints to obtain feature orthogonalized data. The feature orthogonalized data is then reorganized according to the spatial dimension, temporal dimension, and environmental dimension to obtain the spatial feature matrix, temporal feature matrix, and environmental feature matrix.

[0065] Specifically, the standard pressure source is subjected to multi-frequency control. By precisely adjusting the output of the pressure source, pressure fluctuation signals of different frequencies and amplitudes are generated. These signals are represented by the following mathematical expressions:

[0066] ;

[0067] in, Indicates time The pressure of the moment, It is The amplitude of the frequency component, Indicates frequency, Indicates The phase shift of each component is is the total number of frequencies. By adjusting each and The value of is used to generate a periodic pressure fluctuation signal containing multiple frequency and amplitude combinations. This signal is used to simulate various dynamic pressure scenarios that the pressure sensor may encounter, which helps to fully evaluate its response characteristics. While the pressure fluctuation signal is generated, the temperature and humidity environmental parameters of the pressure sensor array are set to generate multiple sets of environmental condition combination data. Assume that the parameter ranges of temperature and humidity are and By discretizing the ranges of these two parameters, we can obtain a variety of combination conditions. For example, the temperature is discretized into different values, discretizing the humidity into different values, there are a total of These combinations are used to simulate the response of different environmental conditions to the pressure sensor and ensure that the sensor can show stability under various environmental conditions. Under the action of the above multiple sets of environmental condition combination data, the pressure sensor array is collected to obtain the original response data under different frequency pressure signals and different environmental conditions. The position of each pressure sensor is represented by the coordinates is represented by, and the time sampling sequence is , the environmental conditions are The raw response data is represented as a three-dimensional data structure, defined as follows:

[0068] ;

[0069] in, Indicates location in space The pressure sensor at the time point and environmental conditions The response value under Represents the response of the sensor to the applied pressure signal. The original response data is reconstructed into three dimensions to obtain tensor structure data. According to the spatial position, time sampling sequence and environmental condition sequence of the pressure sensor, the original response data is reconstructed into three dimensions to construct a three-dimensional tensor R, whose structure is as follows:

[0070] ;

[0071] in, Indicates the number of spatial position dimensions of the sensor, represents the number of time sampling points, Represents the number of combinations of environmental conditions. A three-dimensional tensor It contains the response data of all sensors under different spatial, temporal and environmental conditions, and can fully describe the output characteristics of the sensors. In order to extract the main features in the tensor data, the three-dimensional tensor structure data is input into the parallel factor decomposition model for feature decomposition. Parallel factor decomposition is a tensor decomposition method that can decompose a multidimensional tensor into a combination of several low-rank matrices to reveal the intrinsic pattern of the data. Decomposed into three feature matrices:

[0072] ;

[0073] in, Represent spatial characteristics, temporal characteristics and environmental characteristics respectively, and the symbols represents the outer product operation, is the rank of decomposition. Through decomposition, the initial eigendecomposition data are obtained, which describe the feature information in the spatial, temporal and environmental dimensions respectively. The minimum reconstruction error threshold is set for the initial eigendecomposition data, and iterative calculation is performed to obtain the optimal rank of tensor decomposition. Reconstruction error Calculated by the following formula:

[0074] ;

[0075] in, Represents the Frobenius norm, which is used to measure the reconstruction error of the tensor. , so that the reconstruction error When the value is less than the preset minimum reconstruction error threshold, the optimal tensor decomposition rank is obtained. Based on the obtained tensor decomposition rank, the initial feature decomposition data is optimized by orthogonal constraints to obtain feature orthogonalized data. The purpose of orthogonal constraint optimization is to make each feature vector orthogonal to each other, so as to ensure the independence of the decomposed features, reduce feature redundancy, and improve the reliability of subsequent analysis. According to the spatial dimension, time dimension and environmental dimension, the feature orthogonalized data is reorganized to obtain the spatial feature matrix, time feature matrix and environmental feature matrix, which respectively describe the main feature information of the pressure sensor in the spatial, temporal and environmental dimensions.

[0076] In a specific embodiment, the process of executing step S2 may specifically include the following steps:

[0077] (1) Concatenate the spatial feature matrix, the temporal feature matrix, and the environmental feature matrix to obtain a combined feature matrix, and decompose the combined feature matrix to obtain a left singular matrix, a singular value matrix, and a right singular matrix;

[0078] (2) Arrange the singular value matrix in descending order and select the eigenvector corresponding to the largest singular value to obtain the main eigenvector data;

[0079] (3) Construct the transfer matrix of the main eigenvector data and obtain the transfer function matrix through trilinear least squares iteration calculation;

[0080] (4) Performing frequency domain transformation on the transfer function matrix to obtain frequency response function data, and sparsely reconstructing the frequency response function data based on a distributed compressed sensing algorithm to obtain a dynamic feature vector;

[0081] (5) Perform time domain response analysis on the dynamic feature vector to obtain dynamic response data, and filter the dynamic response data based on the transfer function matrix to obtain time characteristic data.

[0082] Specifically, the spatial feature matrix, the temporal feature matrix, and the environmental feature matrix are combined. These feature matrices describe the response characteristics of the pressure sensor array from the spatial, temporal, and environmental dimensions, respectively. Assume that the spatial feature matrix is , the time feature matrix is , the environmental feature matrix is ,in Represent the feature numbers of space, time and environment dimensions respectively, and Represents the rank of the feature. Perform row concatenation to obtain the combined feature matrix:

[0083] ;

[0084] in is the combined feature matrix obtained by splicing, which is used to capture the overall characteristics of the sensor. Perform singular value decomposition to obtain the left singular matrix , singular value matrix and the right singular matrix :

[0085] ;

[0086] in, Is a left singular matrix, containing the projection information of the eigenvector in the row direction; is a diagonal matrix, where the diagonal elements are matrices The singular values ​​of , which represent the strength of each feature in the matrix; is a right singular matrix, containing the projection information of the eigenvector in the column direction. The size of the singular value helps the system determine the importance of different features in the matrix. Arrange in descending order to obtain the arranged singular value sequence:

[0087] ;

[0088] in, Indicates singular values. Larger singular values ​​correspond to more important features in the matrix. Select the largest singular value The corresponding eigenvector is obtained to obtain the main eigenvector data. Assume that the corresponding eigenvector is , then the eigenvector contains the most important feature information in the combined feature matrix, representing the main response characteristics of the pressure sensor in three dimensions: space, time and environment. The transfer matrix is ​​constructed. The construction of the transfer matrix is ​​based on trilinear least squares iterative calculation, which is used to minimize the error of the eigenvector in different dimensions to ensure that the transfer matrix can accurately describe the relationship between the input and output of the pressure sensor. Assume that the transfer matrix is , and its construction process is expressed as:

[0089] ;

[0090] in, Indicates The principal eigenvector of samples, represents the input signal, represents the number of samples, is the optimal transfer matrix obtained through iterative optimization. Frequency domain transformation is performed to analyze the response characteristics of the pressure sensor at different frequencies. Frequency domain transformation uses Fourier transform to convert the time domain signal into a frequency domain signal. The frequency domain transformation formula is:

[0091] ;

[0092] in, represents the frequency response function data, is the transfer function in the time domain, is the frequency variable, is an imaginary unit. The frequency response function data describes the response of the sensor to input signals of different frequencies, reflecting the amplitude-frequency and phase-frequency characteristics of the system. Based on the distributed compressed sensing algorithm, the frequency response function data is sparsely reconstructed to obtain the dynamic feature vector. Compressed sensing is a method to reconstruct the original signal through a small amount of measurement data, which is suitable for the reconstruction of sparse signals. Assume that the frequency response function is , expressed as:

[0093] ;

[0094] in, is the observed measurement value, is the measurement matrix. Sparse reconstruction is achieved by solving the following optimization problem:

[0095] ;

[0096] in, express of norm, sparse reconstruction is achieved by minimizing The norm is used to find the sparsest solution and obtain the dynamic eigenvector. The dynamic eigenvector is subjected to time domain response analysis to obtain dynamic response data. Time domain response analysis is used to describe the behavior characteristics of the system over time, such as the response amplitude and change at different time points. Assume that the dynamic eigenvector is , the dynamic response data is obtained by the following formula:

[0097] ;

[0098] in, Indicates at time The dynamic response data below, is the previously constructed transfer matrix, which is combined with the dynamic eigenvector Multiply them to get the time domain response of the system. Filter the dynamic response data based on the transfer function matrix to get the time characteristic data. Remove the noise components in the signal to make the obtained time characteristic data smoother and more accurate. Assume that the transfer function of the filter is , then the filtering process is expressed as:

[0099] ;

[0100] in, represents the dynamic response data in the frequency domain, is the transfer function of the filter, Represents the frequency domain response data after filtering. Inverse Fourier transform back to the time domain to obtain the time characteristic data:

[0101] ;

[0102] in, The filtered time characteristic data is smoother and has less noise components, which can more accurately reflect the dynamic characteristics of the sensor in practical applications.

[0103] In a specific embodiment, the process of executing step S3 may specifically include the following steps:

[0104] (1) Perform symmetrical geometric transformation on the dynamic response data and time characteristic data to obtain a feature decomposition data group, and input the feature decomposition data group into the maximum entropy spectrum analysis model for recursive calculation to obtain feature distribution data;

[0105] (2) Performing gain coefficient analysis on the characteristic distribution data to obtain a linear dynamic parameter matrix, and performing zero drift feature extraction and deviation compensation based on the linear dynamic parameter matrix to obtain linear response component data;

[0106] (3) Perform wavelet decomposition on the characteristic distribution data to obtain the periodic signal feature vector, and input the periodic signal feature vector into the harmonic component recognition model for adaptive threshold segmentation to obtain the periodic fluctuation component data;

[0107] (4) Based on the characteristic distribution data, second-order and third-order Volterra nonlinear kernel functions are constructed and iteratively optimized to obtain nonlinear characteristic coefficients. The nonlinear characteristic coefficients are reconstructed by kernel functions and nonlinearly mapped to obtain nonlinear characteristic component data.

[0108] Specifically, a symmetrical geometric transformation is performed on the dynamic response data and the time characteristic data. Different types of data are symmetric in the geometric space so that they can describe the system characteristics more uniformly. Assume that the dynamic response data is , the time characteristic data is , then the symmetric geometric transformation is realized by the following formula:

[0109] ;

[0110] in, represents the characteristic decomposition data set obtained by symmetric geometric transformation, and are functions of dynamic response data and time characteristic data, respectively. Through geometric transformation, the dynamic response and time characteristics are combined to ensure that they are geometrically consistent, so that unified feature extraction can be performed on them. Input the maximum entropy spectrum analysis model, perform recursive calculations, and obtain characteristic distribution data. The purpose of maximum entropy spectrum analysis is to use the maximum entropy principle to estimate the spectrum of limited data to ensure that the resolution of spectrum information is maximized. The core of maximum entropy spectrum analysis is to estimate the parameters of the autoregressive model using the following recursive formula:

[0111] ;

[0112] in, Indicates frequency The power spectral density under are the parameters of the autoregressive model, is the order of the model, is an imaginary unit. Through recursive calculation, the distribution of the characteristic data group at different frequencies is estimated to obtain the characteristic distribution data , which can accurately describe the distribution of dynamic response and time characteristics in the frequency domain. Gain coefficient analysis is performed on the characteristic distribution data to obtain the linear dynamic parameter matrix. Gain coefficient analysis calculates the magnification of the characteristic distribution data to extract the dynamic characteristics of the linear part. is the gain function in the characteristic distribution data, then the linear dynamic parameter matrix is ​​expressed as:

[0113] ;

[0114] in, represents the linear dynamic parameter matrix, and are the lower and upper limits of the frequency range respectively. By integrating the gain coefficient, the overall gain of the system in a specific frequency band is obtained. This gain information is used to describe the linear characteristics. Zero drift feature extraction and deviation compensation are performed to obtain linear response component data. The purpose of zero drift feature extraction is to identify the reference drift caused by temperature change or component aging, while deviation compensation is used to eliminate these reference drifts to ensure the accuracy of system output. The compensation formula is expressed as:

[0115] ;

[0116] in, is the linear response data after compensation, Indicates the deviation value of the benchmark drift. At the same time, wavelet decomposition is performed on the characteristic distribution data to extract the characteristic vector of the periodic signal. Wavelet decomposition is a time-frequency analysis method that decomposes the signal into components of different scales and frequencies. For characteristic distribution data, through wavelet function Perform wavelet transform to obtain the periodic signal feature vector:

[0117] ;

[0118] in, is the wavelet coefficient, and are the scale and translation parameters, Represents the conjugate complex number of the wavelet function. Wavelet transform can analyze signals at different scales and extract periodic components. Input the periodic signal feature vector into the harmonic component recognition model for adaptive threshold segmentation to obtain periodic fluctuation component data. Adaptive threshold segmentation automatically adjusts the threshold according to the signal strength to ensure that the extracted periodic components are more accurate. Periodic fluctuation data after segmentation Used to describe the characteristics of periodic interference in the system. Based on the characteristic distribution data, the second-order and third-order Volterra nonlinear kernel functions are constructed to describe the nonlinear characteristics of the system. The Volterra series is a model used to describe nonlinear systems, and its kernel function is used to capture the high-order interaction effects in the system. The second-order and third-order Volterra kernel functions are expressed as:

[0119] ;

[0120] ;

[0121] in, and denote the second-order and third-order Volterra kernel functions, respectively. and are the second-order and third-order kernel coefficients, is the input signal, is the time delay. By iteratively optimizing the parameters of these kernel functions, nonlinear characteristic coefficients are obtained, which are used to describe the nonlinear characteristics of the system. The nonlinear characteristic coefficients are reconstructed by kernel functions and nonlinearly mapped to obtain nonlinear characteristic component data. Kernel function reconstruction forms a complete nonlinear characteristic model by weighted combination of second-order and third-order kernel functions. Nonlinear mapping is used to apply these characteristics to actual system inputs to obtain the nonlinear response of the system. The nonlinear mapping formula is:

[0122] ;

[0123] in, It is the nonlinear characteristic component data, which describes the nonlinear response of the system under different input conditions.

[0124] In a specific embodiment, the process of executing step S4 may specifically include the following steps:

[0125] (1) Inputting the linear response component data into the first bidirectional gated recurrent unit in the multi-branch gated recurrent unit network model for bidirectional feature extraction to obtain linear branch feature data, wherein the first bidirectional gated recurrent unit includes a forward hidden layer and a reverse hidden layer, and each hidden layer includes an update gate, a reset gate, and a candidate hidden state;

[0126] (2) inputting the periodic fluctuation component data into the second bidirectional gated recurrent unit in the multi-branch gated recurrent unit network model to perform bidirectional feature extraction to obtain periodic branch feature data, wherein the structure of the second bidirectional gated recurrent unit is the same as that of the first bidirectional gated recurrent unit;

[0127] (3) inputting the nonlinear characteristic component data into the third bidirectional gated recurrent unit in the multi-branch gated recurrent unit network model to perform bidirectional feature extraction to obtain nonlinear branch feature data, wherein the structure of the third bidirectional gated recurrent unit is the same as that of the first bidirectional gated recurrent unit;

[0128] (4) Perform time node attention calculation on the linear branch feature data, periodic branch feature data, and nonlinear branch feature data to obtain weighted feature data, and input the weighted feature data into the feature fusion layer for adaptive feature combination to obtain multi-branch fusion feature data;

[0129] (5) Performing environmental variable feature correlation analysis on the multi-branch fusion feature data to obtain calibration feature data, and inputting the calibration feature data into the dynamic prediction layer for time series reasoning to obtain the first dynamic calibration strategy.

[0130] Specifically, the linear response component data is input into the first bidirectional gated recurrent unit in the multi-branch gated recurrent unit (GRU) network model for bidirectional feature extraction to obtain linear branch feature data. Define a multi-branch GRU network model and configure the bidirectional gated recurrent unit structure therein. The input is processed in the first bidirectional GRU unit in the multi-branch network. The bidirectional GRU unit consists of a forward hidden layer and a reverse hidden layer, each of which consists of an update gate, a reset gate, and a candidate hidden state. These parts work together to effectively capture the characteristics of time series data and learn in both the forward and reverse time dimensions to gain a comprehensive understanding of the sequence data. In the first bidirectional GRU, the forward hidden layer is in the forward direction (from arrive ) processes the input data, while the reverse hidden layer processes it in reverse time (from arrive ) process the data, and finally combine the outputs of the two hidden layers to obtain the linear branch feature data. Update gate and reset gate It is the part used to control the hidden state and is calculated as follows:

[0131] ;

[0132] ;

[0133] in, Indicates at time Input data at time, are the weight matrices connected to the input, is the hidden state at the previous moment The weight matrix of the connection, is the bias term, is a Sigmoid activation function, which is used to map the input value to between (0,1). The candidate hidden state is controlled by the reset gate:

[0134] ;

[0135] in, represents the Hadamard product (element-wise multiplication), is the hyperbolic tangent function, are the weight matrices, is the bias term. The final hidden state The fusion of the new and old hidden states is controlled by the update gate:

[0136] ;

[0137] Bidirectional feature extraction means the combination of forward and reverse hidden states, and the resulting linear branch feature data is represented as ,in and Represent the forward and reverse hidden states respectively. Similarly, the periodic fluctuation component data The second bidirectional GRU unit in the multi-branch GRU network model is input for bidirectional feature extraction. The structure of the second bidirectional GRU is the same as the first bidirectional GRU. After similar update gates, reset gates and candidate hidden state calculations, the periodic branch feature data is obtained. . The nonlinear characteristic component data Input into the third bidirectional GRU unit, perform similar bidirectional feature extraction, and obtain nonlinear branch feature data Each bidirectional GRU unit can learn the features of different response characteristics independently by processing different types of input data. , periodic branch characteristic data and nonlinear branch characteristic data Perform attention calculations on the time nodes to obtain weighted feature data. The attention mechanism enables the network to automatically assign weights to each time node, thereby better identifying the features that are most important for the calibration strategy. Assume that the attention weight is , which is calculated as:

[0138] ;

[0139] ;

[0140] in, Indicates at time The characteristic data at and are the weight matrix and weight vector, is the bias term, is the attention score. The weighted feature data is expressed as:

[0141] ;

[0142] Weighted feature data Input feature fusion layer to perform adaptive feature combination to obtain multi-branch fusion feature data The function of the feature fusion layer is to adaptively combine data from different branches. Assume that the fused features are represented as:

[0143] ;

[0144] in, is the fusion weight matrix, is the bias term. The feature fusion layer combines the feature data of each branch in the best way by learning the correlation between different features, thereby capturing the overall characteristics of the sensor in various dynamic responses. The multi-branch fusion feature data is subjected to environmental variable feature association analysis to obtain calibration feature data. The purpose of environmental variable feature association analysis is to understand the relationship between sensor response and environmental conditions. Assume that the environmental variable is , the calculation of calibration characteristic data is expressed as:

[0145] ;

[0146] in, represents the weight matrix between environmental variables and feature data, By introducing environmental variables into feature analysis, we can better capture the situation where the sensor response is affected by the external environment and improve the accuracy of calibration. The input is fed into the dynamic prediction layer for time series reasoning to obtain the first dynamic calibration strategy. The dynamic prediction layer consists of a recurrent neural network, which is used to reason in the time dimension based on the input features and predict future dynamic calibration parameters. Assume that the output of the dynamic prediction layer is , then its calculation method is expressed as:

[0147] ;

[0148] in, represents the nonlinear activation function of the dynamic prediction layer, is the hidden state of the previous moment. The final output is Represents at time The calibration parameters under , are inferred over the entire time series to obtain a complete first dynamic calibration strategy.

[0149] In a specific embodiment, the process of executing step S5 may specifically include the following steps:

[0150] (1) performing linear transformation on the linear calibration parameters, periodic calibration parameters and nonlinear calibration parameters in the first dynamic calibration strategy to obtain an initial optimization parameter vector;

[0151] (2) Based on the initial optimization parameter vector, the particle swarm size is set to 100, the search dimension is the dimension of the initial optimization parameter vector, and the initial position matrix and velocity matrix of the particle swarm are obtained by random initialization of Gaussian distribution;

[0152] (3) The dynamic calibration objective function is input into the initial position matrix and velocity matrix of the particle swarm, and the initial fitness data is obtained by calculating the weighted sum of the calibration error and time cost;

[0153] (4) Based on the initial fitness data, the inertia weight coefficient, individual learning factor, and social learning factor are set, and the speed and position are updated to obtain the updated position matrix;

[0154] (5) Calculate the local and global optimal solutions for the updated position matrix to obtain the optimal parameter combination of the current iteration, and perform an inverse linear transformation on the optimal parameter combination to obtain the second dynamic calibration strategy;

[0155] (6) Based on the second dynamic calibration strategy, a three-level calibration sequence including a linear calibration layer, a periodic calibration layer and a nonlinear calibration layer is constructed. Multi-level calibration parameters are obtained through parameter decoupling. The multi-level calibration parameters are progressively combined in the order of linear calibration, periodic calibration and nonlinear calibration to obtain a progressive calibration parameter set.

[0156] Specifically, the linear calibration parameters, periodic calibration parameters and nonlinear calibration parameters in the first dynamic calibration strategy are linearly transformed. Assume that the linear calibration parameters are , the period calibration parameter is , the nonlinear calibration parameter is These parameters are in vector form. Through linear transformation, these parameters are integrated into an initial optimization parameter vector The mathematical expression of linear transformation is:

[0157] ;

[0158] in, are weight matrices used to adjust linear calibration parameters, periodic calibration parameters, and nonlinear calibration parameters, respectively. is the bias term, is the initial optimization parameter vector. Through linear combination, all calibration parameters are integrated into a unified optimization target, ensuring that three different types of parameters can be considered simultaneously during the optimization process. , set the size of the particle swarm to 100 and the search dimension to The particle swarm optimization is a global optimization method based on swarm intelligence. It simulates the mutual cooperation and competition of individuals in the swarm to find the global optimal solution. The particle swarm size is set to 100, which means that there are 100 candidate solutions at the same time during the optimization process. These candidate solutions are called "particles", which interact and update iteratively in the search space. The initial optimization parameter vector is distributed by Gaussian distribution. Perform random initialization to obtain the initial position matrix and velocity matrix of the particle swarm. Assume that the initial position matrix of the particle swarm is , the velocity matrix is , whose dimensions are ,in represents the optimization parameter vector The initial position and velocity are calculated as follows:

[0159] ;

[0160] ;

[0161] in, represents a Gaussian distribution with a mean of , the variance is are the mean and variance of the velocity respectively. Initializing the position and velocity by Gaussian distribution ensures that the particle swarm has a good initial distribution in the search space and avoids falling into the local optimum. and the velocity matrix Input the dynamic calibration objective function and obtain the initial fitness data by calculating the weighted sum of the calibration error and time cost. The purpose of the dynamic calibration self-calibration function is to evaluate the calibration effect of each particle at a given position. Let the objective function be , which consists of the weighted sum of calibration error and time overhead:

[0162] ;

[0163] in, Indicates The position of a particle, is the calibration error, For time expenditure, and are weighted coefficients, which measure the relative importance of calibration error and time cost respectively. Get the fitness value of each particle to guide the subsequent update of the particle swarm. Set the inertia weight coefficient based on the initial fitness data , Individual learning factor and social learning factors , and update the speed and position. Inertia weight coefficient Used to control the speed change of particles, individual learning factor and social learning factors They are used to measure the ability of particles to learn from their own experience and group experience respectively. The formulas for updating the particle's speed and position are:

[0164] ;

[0165] ;

[0166] in, and Respectively represent The particle in The velocity and position in the iteration, For particles In the best position in history, is the global optimal position of the group, is a random number between [0,1]. Through these update formulas, the particle position and velocity are continuously adjusted to approach the optimal solution. The local and global optimal solutions are calculated for the updated position matrix to find the optimal parameter combination for the current iteration. Local optimal solution represents the optimal position of each particle in its history, the global optimal solution Represents the optimal position encountered by all particles during the entire iteration process. After finding the optimal parameter combination, an inverse linear transformation is performed to remap the optimized parameters back to the original calibration parameter space to obtain the second dynamic calibration strategy. The formula for the inverse linear transformation is:

[0167] ;

[0168] in, are the optimized linear, periodic and nonlinear calibration parameters, is the weight matrix The inverse matrix of is the parameter vector obtained by optimization. Based on the second dynamic calibration strategy, a three-level calibration sequence including a linear calibration layer, a periodic calibration layer and a nonlinear calibration layer is constructed. Each calibration layer corresponds to different types of parameters so as to gradually eliminate various errors in the system. The linear calibration layer is used to process linear characteristic errors, mainly to adjust the linear response of the sensor. The periodic calibration layer is used to process periodic interference, and by compensating for periodic errors, the measurement accuracy in a dynamic environment is guaranteed. The nonlinear calibration layer is used to process nonlinear response characteristics to further improve the accuracy and robustness of the system. Through parameter decoupling, the various parameters in the second dynamic calibration strategy are separated to obtain multi-level calibration parameters, which are used for different calibration layers. The multi-level calibration parameters are expressed as:

[0169] ;

[0170] Among them, decouple represents the parameter decoupling operation, They represent linear calibration parameters, periodic calibration parameters and nonlinear calibration parameters respectively. According to the order of linear calibration, periodic calibration and nonlinear calibration, the multi-level calibration parameters are progressively combined to obtain a progressive calibration parameter set. The purpose of the progressive combination is to ensure that various types of calibration are carried out step by step, first eliminating linear errors, then compensating for periodic errors, and finally dealing with nonlinear errors, so as to achieve a comprehensive calibration effect.

[0171] In a specific embodiment, the process of executing step S6 may specifically include the following steps:

[0172] (1) Inputting the progressive calibration parameter set into the static pressure calibration module, applying pressure step by step within a preset range through a standard pressure device, and obtaining initial static response data;

[0173] (2) Performing multiple pressurization and depressurization cycles on the initial static response data to obtain hysteresis characteristic data, and performing linearity compensation based on the hysteresis characteristic data to obtain a static calibration coefficient group;

[0174] (3) Inputting the progressive calibration parameter set into the dynamic pressure calibration module, obtaining the frequency sweep response data through the sinusoidal excitation signal, and performing amplitude-frequency characteristic analysis and dynamic tracking error calculation on the frequency sweep response data to obtain the dynamic amplitude calibration parameters;

[0175] (4) Perform phase-frequency characteristic analysis and phase delay compensation on the frequency sweep response data to obtain dynamic phase calibration parameters;

[0176] (5) Generate multimodal calibration characteristic data based on the static calibration coefficient group, dynamic amplitude calibration parameters, and dynamic phase calibration parameters, and perform a comprehensive analysis on the multimodal calibration characteristic data to obtain a comprehensive pressure sensor calibration result.

[0177] Specifically, the progressive calibration parameter set is input into the static pressure calibration module, and the pressure is applied step by step within the preset range through the standard pressure device to obtain the initial static response data. A standard pressure device is used, which can accurately control the pressure applied to the pressure sensor within the set range. The progressive calibration parameter set contains different parameters for static and dynamic calibration. After these parameters are input into the static calibration module, they adjust the response of the system to ensure that its response to the static pressure signal is as linear and accurate as possible. Pressure is applied in a preset step-by-step manner, such as within the range. Every For a pressurization, the applied pressure is expressed as:

[0178] ;

[0179] in, It is The pressure applied per step, is the number of steps. Each pressure point applied by the standard pressure device will be recorded and the initial static response data will be obtained. The initial static response data will be tested by multiple cycles of pressurization and depressurization to obtain the hysteresis characteristic data. The hysteresis phenomenon refers to the inconsistency of the output value of the pressure sensor during the pressurization and depressurization process at the same pressure level, which is mainly caused by the internal friction of the material or other physical phenomena. Assume that The pressure sensor is pressurized and released at all times, and the response is expressed by the following formula:

[0180] ;

[0181] in, Indicates the response during pressurization, Indicates the response during the pressure relief process, is the hysteresis characteristic data. Through these hysteresis characteristic data, the deviation between the sensor output and the ideal situation is identified. Linearity compensation is performed based on the hysteresis characteristic data to eliminate the hysteresis effect and improve the linearity of the measurement. Linearity compensation is achieved by fitting and compensating the hysteresis data. Let the fitting function be , then the response after linearity compensation is expressed as:

[0182] ;

[0183] in, Represents the static response data after linearity compensation, is the uncompensated original response data. Through linearity compensation, the static calibration coefficient group is obtained , these coefficient groups are used to describe the linear characteristics of the sensor and ensure its measurement accuracy in static tests. The progressive calibration parameter set is input into the dynamic pressure calibration module, and a dynamic test is performed by applying a sinusoidal excitation signal to obtain frequency sweep response data. The sinusoidal excitation signal is expressed as:

[0184] ;

[0185] in, is the amplitude of the sinusoidal signal, is the frequency, By gradually changing the frequency , analyze the response characteristics of the pressure sensor at different frequencies and obtain the frequency scanning response data In order to analyze the dynamic characteristics of the sensor, the frequency scanning response data is subjected to amplitude-frequency characteristic analysis and dynamic tracking error calculation. Amplitude-frequency characteristic analysis is used to evaluate the amplitude response of the sensor to signals of different frequencies. Assuming that the amplitude-frequency response of the sensor is , then the amplitude-frequency characteristic is expressed as:

[0186] ;

[0187] in, For frequency The response amplitude under is the amplitude of the applied signal. Dynamic tracking error It is used to describe the tracking accuracy of the sensor for signals of different frequencies. The calculation formula is:

[0188] ;

[0189] The dynamic amplitude calibration parameters are obtained through amplitude-frequency characteristic analysis and calculation of dynamic tracking error. , these parameters are used to adjust the amplitude response of the sensor to ensure its dynamic accuracy at different frequencies. The phase-frequency characteristics of the frequency sweep response data are analyzed to evaluate the phase delay of the sensor at different frequencies. The phase-frequency characteristics describe the phase response of the system to signals of different frequencies. Assuming the phase delay is , then the phase-frequency characteristic is expressed as:

[0190] ;

[0191] in, and The input signal and output signal are By compensating the phase delay, the dynamic phase calibration parameter is obtained. These parameters are used to adjust the phase response of the sensor under dynamic conditions, ensuring that it can accurately synchronize with the input signal. , Dynamic Amplitude Calibration Parameters and dynamic phase calibration parameters Finally, multimodal calibration characteristic data is generated based on these parameters The multimodal calibration characterization data combines static and dynamic calibration information to fully describe the sensor's performance under different conditions. The multimodal calibration characterization data is generated by adding the calibration parameters:

[0192] ;

[0193] By integrating the data of static calibration, amplitude calibration and phase calibration, the measurement accuracy of the sensor in various usage environments is ensured. The multi-modal calibration characteristic data is comprehensively analyzed to obtain the comprehensive pressure sensor calibration results. The specific implementation process of comprehensive analysis of multi-modal calibration characteristic data is completed through multi-level data fusion and error evaluation. First, the static calibration coefficient group, dynamic amplitude calibration parameters and dynamic phase calibration parameters are weighted to construct a comprehensive calibration model, and then the residual error relative to the reference value after calibration of each mode is calculated respectively, including static pressure error, amplitude error and phase error. These errors are weighted and combined through the error comprehensive evaluation function to obtain the overall error index, and combined with the comprehensive calibration model, the final calibration result is obtained through the evaluation function. This fusion analysis method of multi-modal data realizes the comprehensive calibration of the static and dynamic characteristics of the pressure sensor, and effectively improves the accuracy and reliability of the calibration. To evaluate whether the calibrated pressure sensor achieves the expected accuracy and consistency under different test conditions, it is assumed that the comprehensive calibration error is , which is calculated by the following formula:

[0194] ;

[0195] in, denote the calibration errors of static, amplitude and phase respectively, is a weighting coefficient used to indicate the relative importance of different errors. By analyzing the comprehensive error, the overall performance of the sensor is evaluated and the final calibration result is obtained.

[0196] The above describes the pressure response data calibration method in the embodiment of the present invention. The following describes the pressure response data calibration device in the embodiment of the present invention. Figure 2 , an embodiment of a pressure response data calibration device in an embodiment of the present invention includes:

[0197] An acquisition module, used for acquiring output response data of the pressure sensor array, and constructing a spatial feature matrix, a temporal feature matrix and an environmental feature matrix based on the output response data;

[0198] A separation module is used to perform trilinear least square separation on the spatial feature matrix, the temporal feature matrix and the environmental feature matrix to obtain dynamic response data and temporal characteristic data;

[0199] A decomposition module is used to perform symmetrical geometric modal decomposition on the dynamic response data and the time characteristic data to obtain linear response component data, periodic fluctuation component data and nonlinear characteristic component data;

[0200] A prediction module, used for inputting the linear response component data, the periodic fluctuation component data and the nonlinear characteristic component data into the multi-branch gated recurrent unit network model to perform dynamic calibration strategy prediction to obtain a first dynamic calibration strategy;

[0201] An optimization module, configured to perform particle swarm iterative optimization on the first dynamic calibration strategy to obtain a second dynamic calibration strategy, and generate a progressive calibration parameter set based on the second dynamic calibration strategy;

[0202] The calibration module is used to perform static and dynamic pressure sensing calibration on the pressure sensor array based on a progressive calibration parameter set to obtain a comprehensive pressure sensing calibration result.

[0203] Through the collaborative cooperation of the above components, parallel factor decomposition and trilinear least squares separation technology are used to accurately extract the spatial, temporal and environmental characteristics of the pressure sensor, reduce the impact of environmental noise on dynamic calibration, and adopt the symmetric geometric modal decomposition method to decompose the sensor characteristics into three components: linear response, periodic fluctuation and nonlinear characteristics, so as to achieve accurate identification of multi-dimensional features. Based on the dynamic calibration strategy of the multi-branch gated recurrent unit network model, the timing performance of dynamic calibration is improved through bidirectional feature extraction and time node attention mechanism. A progressive calibration parameter system based on particle swarm optimization is constructed, which realizes the hierarchical optimization of linear calibration, periodic calibration and nonlinear calibration. Combined with the comprehensive test method of static and dynamic pressure calibration, a complete calibration evaluation system is established through multi-modal calibration characteristic analysis, thereby improving the accuracy of pressure sensor calibration.

[0204] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0205] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0206] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0207] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0208] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0209] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0210] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for calibrating pressure response data, characterized in that: The method comprises: Collecting output response data of the pressure sensor array, and constructing a spatial feature matrix, a temporal feature matrix and an environmental feature matrix based on the output response data; Performing trilinear least square separation on the spatial feature matrix, the temporal feature matrix and the environmental feature matrix to obtain dynamic response data and temporal characteristic data; Performing symmetrical geometric modal decomposition on the dynamic response data and the time characteristic data to obtain linear response component data, periodic fluctuation component data and nonlinear characteristic component data; Inputting the linear response component data, the periodic fluctuation component data and the nonlinear characteristic component data into a multi-branch gated recurrent unit network model to perform dynamic calibration strategy prediction to obtain a first dynamic calibration strategy; Performing particle swarm iterative optimization on the first dynamic calibration strategy to obtain a second dynamic calibration strategy, and generating a progressive calibration parameter set based on the second dynamic calibration strategy; Static and dynamic pressure sensing calibration is performed on the pressure sensor array based on the progressive calibration parameter set to obtain a comprehensive pressure sensing calibration result.

2. The pressure response data calibration method according to claim 1, characterized in that: The collecting the output response data of the pressure sensor array and constructing a spatial feature matrix, a temporal feature matrix and an environmental feature matrix based on the output response data include: Performing multi-frequency control on the standard pressure source to generate a periodic pressure fluctuation signal including different frequency and amplitude combinations, and setting the temperature and humidity environmental parameters of the pressure sensor array to generate multiple sets of environmental condition combination data; Under the multiple sets of environmental condition combination data, collecting data on the pressure sensor array based on the periodic pressure fluctuation signal to obtain original response data; According to the spatial position of the pressure sensor, the time sampling sequence and the environmental condition sequence, the original response data is reconstructed into three-dimensional data to obtain three-dimensional tensor structure data; Inputting the three-dimensional tensor structure data into a parallel factor decomposition model for eigendecomposition to obtain initial eigendecomposition data, and setting a minimum reconstruction error threshold and iterative calculation for the initial eigendecomposition data to obtain a tensor decomposition rank; The initial feature decomposition data is orthogonally constrained optimized based on the tensor decomposition rank to obtain feature orthogonalized data, and the feature orthogonalized data is feature reorganized according to the spatial dimension, time dimension and environmental dimension to obtain a spatial feature matrix, a temporal feature matrix and an environmental feature matrix.

3. The pressure response data calibration method according to claim 2, characterized in that: The performing trilinear least square separation on the spatial feature matrix, the temporal feature matrix and the environmental feature matrix to obtain dynamic response data and temporal characteristic data includes: Performing matrix concatenation on the spatial feature matrix, the temporal feature matrix, and the environmental feature matrix to obtain a combined feature matrix, and performing a decomposition operation on the combined feature matrix to obtain a left singular matrix, a singular value matrix, and a right singular matrix; Arrange the singular value matrix in descending order, and select the eigenvector corresponding to the largest singular value to obtain the main eigenvector data; Constructing a transfer matrix for the main eigenvector data, and obtaining a transfer function matrix through trilinear least squares iterative calculation; Performing frequency domain transformation on the transfer function matrix to obtain frequency response function data, and sparsely reconstructing the frequency response function data based on a distributed compressed sensing algorithm to obtain a dynamic feature vector; The dynamic characteristic vector is subjected to time domain response analysis to obtain dynamic response data, and the dynamic response data is subjected to filtering processing based on the transfer function matrix to obtain time characteristic data.

4. The pressure response data calibration method according to claim 3, characterized in that: The step of performing symmetrical geometric modal decomposition on the dynamic response data and the time characteristic data to obtain linear response component data, periodic fluctuation component data and nonlinear characteristic component data includes: Performing symmetrical geometric transformation on the dynamic response data and the time characteristic data to obtain a feature decomposition data group, and inputting the feature decomposition data group into a maximum entropy spectrum analysis model for recursive calculation to obtain feature distribution data; Performing gain coefficient analysis on the characteristic distribution data to obtain a linear dynamic parameter matrix, and performing zero drift feature extraction and deviation compensation based on the linear dynamic parameter matrix to obtain linear response component data; Performing wavelet decomposition on the characteristic distribution data to obtain a periodic signal characteristic vector, and inputting the periodic signal characteristic vector into a harmonic component recognition model for adaptive threshold segmentation to obtain periodic fluctuation component data; Based on the characteristic distribution data, second-order and third-order Volterra nonlinear kernel functions are constructed, and iterative optimization is performed to obtain nonlinear characteristic coefficients, and kernel function reconstruction and nonlinear mapping are performed on the nonlinear characteristic coefficients to obtain nonlinear characteristic component data.

5. The pressure response data calibration method according to claim 4, characterized in that: The step of inputting the linear response component data, the periodic fluctuation component data and the nonlinear characteristic component data into a multi-branch gated recurrent unit network model to perform dynamic calibration strategy prediction to obtain a first dynamic calibration strategy includes: Inputting the linear response component data into a first bidirectional gated recurrent unit in a multi-branch gated recurrent unit network model for bidirectional feature extraction to obtain linear branch feature data, wherein the first bidirectional gated recurrent unit includes a forward hidden layer and a reverse hidden layer, each hidden layer includes an update gate, a reset gate and a candidate hidden state; Inputting the periodic fluctuation component data into a second bidirectional gated recurrent unit in the multi-branch gated recurrent unit network model for bidirectional feature extraction to obtain periodic branch feature data, wherein the structure of the second bidirectional gated recurrent unit is the same as that of the first bidirectional gated recurrent unit; Inputting the nonlinear characteristic component data into a third bidirectional gated recurrent unit in the multi-branch gated recurrent unit network model to perform bidirectional feature extraction to obtain nonlinear branch feature data, wherein the structure of the third bidirectional gated recurrent unit is the same as that of the first bidirectional gated recurrent unit; Performing time node attention calculation on the linear branch feature data, the periodic branch feature data, and the nonlinear branch feature data to obtain weighted feature data, and inputting the weighted feature data into a feature fusion layer for adaptive feature combination to obtain multi-branch fusion feature data; Performing environmental variable feature association analysis on the multi-branch fusion feature data to obtain calibration feature data, and inputting the calibration feature data into a dynamic prediction layer for time series reasoning to obtain a first dynamic calibration strategy.

6. The method for calibrating pressure response data according to claim 5, characterized in that: The performing particle swarm iterative optimization on the first dynamic calibration strategy to obtain a second dynamic calibration strategy, and generating a progressive calibration parameter set based on the second dynamic calibration strategy, includes: Performing linear transformation on the linear calibration parameters, periodic calibration parameters and nonlinear calibration parameters in the first dynamic calibration strategy to obtain an initial optimization parameter vector; The particle swarm size is set to 100 based on the initial optimization parameter vector, the search dimension is the dimension of the initial optimization parameter vector, and the initial position matrix and velocity matrix of the particle swarm are obtained by random initialization of Gaussian distribution; Inputting a dynamic calibration objective function into the initial position matrix and the velocity matrix of the particle swarm, and obtaining initial fitness data by calculating a weighted sum of a calibration error and a time cost; Based on the initial fitness data, an inertia weight coefficient, an individual learning factor, and a social learning factor are set, and speed and position are updated to obtain an updated position matrix; Calculating local and global optimal solutions for the updated position matrix to obtain an optimal parameter combination for the current iteration, and performing an inverse linear transformation on the optimal parameter combination to obtain a second dynamic calibration strategy; Based on the second dynamic calibration strategy, a three-level calibration sequence including a linear calibration layer, a periodic calibration layer and a nonlinear calibration layer is constructed, multi-level calibration parameters are obtained through parameter decoupling, and the multi-level calibration parameters are progressively combined in the order of linear calibration, periodic calibration and nonlinear calibration to obtain a progressive calibration parameter set.

7. The pressure response data calibration method according to claim 6, characterized in that: The step of performing static and dynamic pressure sensing calibration on the pressure sensor array based on the progressive calibration parameter set to obtain a comprehensive pressure sensing calibration result includes: The progressive calibration parameter set is input into a static pressure calibration module, and pressure is applied stepwise within a preset range by a standard pressure device to obtain initial static response data; Performing multiple pressurization and depressurization cycles on the initial static response data to obtain hysteresis characteristic data, and performing linearity compensation based on the hysteresis characteristic data to obtain a static calibration coefficient group; Inputting the progressive calibration parameter set into a dynamic pressure calibration module, obtaining frequency sweep response data through a sinusoidal excitation signal, and performing amplitude-frequency characteristic analysis and dynamic tracking error calculation on the frequency sweep response data to obtain dynamic amplitude calibration parameters; Performing phase-frequency characteristic analysis and phase delay compensation on the frequency scanning response data to obtain dynamic phase calibration parameters; Multimodal calibration characteristic data is generated based on the static calibration coefficient group, the dynamic amplitude calibration parameter and the dynamic phase calibration parameter, and the multimodal calibration characteristic data is comprehensively analyzed to obtain a comprehensive pressure sensor calibration result.

8. A pressure response data calibration device, characterized in that: A method for calibrating pressure response data according to any one of claims 1 to 7, wherein the pressure response data calibration device comprises: An acquisition module, used for acquiring output response data of the pressure sensor array, and constructing a spatial feature matrix, a temporal feature matrix and an environmental feature matrix based on the output response data; A separation module, used for performing trilinear least square separation on the spatial feature matrix, the temporal feature matrix and the environmental feature matrix to obtain dynamic response data and temporal characteristic data; A decomposition module, used for performing symmetrical geometric modal decomposition on the dynamic response data and the time characteristic data to obtain linear response component data, periodic fluctuation component data and nonlinear characteristic component data; A prediction module, used for inputting the linear response component data, the periodic fluctuation component data and the nonlinear characteristic component data into a multi-branch gated recurrent unit network model to perform dynamic calibration strategy prediction to obtain a first dynamic calibration strategy; an optimization module, configured to perform particle swarm iterative optimization on the first dynamic calibration strategy to obtain a second dynamic calibration strategy, and generate a progressive calibration parameter set based on the second dynamic calibration strategy; The calibration module is used to perform static and dynamic pressure sensing calibration on the pressure sensor array based on the progressive calibration parameter set to obtain a comprehensive pressure sensing calibration result.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program executable on the processor, and wherein the processor implements the pressure response data calibration method according to any one of claims 1 to 7 when executing the computer program. 10 . A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor is caused to perform the pressure response data calibration method according to any one of claims 1 to 7 .

Images