Multi-source data joint measurement and analysis method based on physical consistency relationship

Abstract

The invention belongs to the technical field of multivariable measurement and data analysis, and relates to a multi-source data joint measurement and analysis method based on a physical consistency relationship, which comprises the following steps of: 1, acquiring and preprocessing multi-source data; 2, modeling physical characteristics; 3, constructing a physical consistency relationship; 4, multi-source joint feature extraction; 5, carrying out consistency evaluation and abnormity identification; and 6, result fusion and dynamic updating. According to the method, collaborative analysis and consistency feature extraction of different types of measurement signals are realized by establishing a physical constraint model among multi-source sensor data, so that the reliability and precision of a multivariable measurement system are improved, the robustness of the system to measurement errors, noise and abnormity under complex working conditions is effectively enhanced, and the measurement accuracy of the system is improved. The method is suitable for the fields of intelligent driving, industrial monitoring, multi-sensor fusion measurement and the like.

Description

Technical Field

[0001] This invention belongs to the field of measurement and signal processing technology, and in particular relates to a multi-source data joint measurement and analysis method based on physical consistency relationship. It is specifically applied to multi-sensor information fusion, physical constraint feature extraction and accuracy improvement of multivariable measurement systems. It can be widely used in fields such as intelligent driving, industrial process monitoring, structural health detection, unmanned system navigation, energy equipment measurement and control and multimodal perception systems in complex scenarios. Background Technology

[0002] With the rapid development of fields such as intelligent driving, industrial automation, and complex equipment monitoring, multi-source sensor collaborative measurement technology has become a crucial foundation for achieving high-precision perception and control. Currently, systems often employ multiple sensors, including LiDAR, cameras, millimeter-wave radar, and inertial measurement units (IMUs), to work collaboratively to acquire multi-dimensional information such as the target's spatial position, velocity, attitude, and acceleration. However, due to significant differences in measurement mechanisms, temporal resolution, data accuracy, and noise characteristics among various sensors, direct data fusion often leads to inconsistencies, feature shifts, and loss of physical meaning.

[0003] Existing multi-source data fusion methods are mainly based on statistical filtering or deep learning frameworks, such as Kalman filtering (KF), extended Kalman filtering (EKF), particle filtering (PF), and convolutional neural networks (CNN) or recurrent neural networks (RNN). These methods have achieved certain results in processing multimodal data, but still have the following limitations: (1) Lack of physical consistency constraints: Existing fusion algorithms mostly focus only on the correlation at the data level, while ignoring the physical dependencies between different sensor measurements, such as the coupling between velocity and position, acceleration and mechanical models, resulting in a lack of physical interpretability of the fusion results. (2) Inability to effectively identify anomalies and drift: When individual sensors are subjected to noise interference, signal obstruction or attack, traditional algorithms have difficulty identifying abnormal measurements through physical models, which can easily lead to deviations in system state estimation or even control failure. (3) Insufficient model update capability: In dynamic environments, sensor characteristics (such as calibration parameters and noise statistics) will change over time. Existing methods mostly rely on fixed models or static filters and lack an adaptive physical model update mechanism. (4) Limited ability to handle heterogeneity of multiple sources: The data dimensions, sampling frequency and noise model of different types of sensors are quite different. Existing algorithms are difficult to uniformly express these heterogeneous information, resulting in insufficient accuracy and real-time performance of joint measurement.

[0004] Therefore, there is an urgent need for a multi-source data joint measurement and analysis method that can introduce physical consistency constraints and take into account both data characteristics and physical semantics, so as to realize physical correlation modeling, anomaly detection and dynamic optimization between multimodal data, thereby improving the system's measurement accuracy, robustness and interpretability.

[0005] To address this, this invention proposes a multi-source data joint measurement and analysis method based on physical consistency relationships. By establishing a physical constraint model and consistency equation among multiple source sensors, it achieves collaborative measurement, feature extraction, and adaptive updating of multimodal data, overcoming the shortcomings of existing methods in terms of physical consistency, robustness, and real-time performance. Summary of the Invention

[0006] The purpose of this invention is to overcome the problems existing in the current multi-source sensor data fusion technology, such as lack of physical consistency constraints, insufficient anomaly identification capability, slow model update, and difficulty in uniformly representing multi-source heterogeneous data, and to propose a multi-source data joint measurement and analysis method based on physical consistency relationships.

[0007] This method establishes a physical constraint model between measurements from different sensors, introduces a physical consistency equation, and realizes collaborative feature extraction and joint measurement among multimodal data. Combined with a dynamic update mechanism, it enables the identification of abnormal measurements and adaptive optimization of the model, thereby significantly improving the accuracy, robustness, and interpretability of multivariate measurement systems.

[0008] This invention is achieved through the following method: a multi-source data joint measurement and analysis method based on physical consistency relationships, comprising the following steps:

[0009] Step 1) Multi-source data acquisition and preprocessing steps: Acquire measurement data from different types of sensors, and perform time synchronization, noise filtering, scale normalization and missing value compensation on the acquired data to obtain a preprocessed dataset with uniform spatiotemporal resolution.

[0010] The specific method in step 1) is as follows:

[0011] At any moment Assume the system includes Individual sensor sources Each sensor synchronously measures the physical state of the target.

[0012] Its original observation data is expressed as follows:

[0013]

[0014] in: : No. Each sensor at time The collected observation vectors; :sensor The measurement dimensions.

[0015] To ensure the comparability of multi-source data, the following preprocessing is performed:

[0016] Step 1.1) Time Synchronization: A timestamp correction algorithm is used to unify the sensor outputs from different sampling frequencies onto the time axis. ;

[0017] Step 1.2) Noise filtering: Apply a moving average or Kalman filter to the data from each channel.

[0018]

[0019] in This is the filter gain matrix;

[0020] Step 1.3) Scale normalization:

[0021]

[0022] in The first Mean and standard deviation of data from each sensor;

[0023] After processing, a multi-source dataset with a unified format is obtained:

[0024] .

[0025] Step 2) Physical feature modeling steps: Based on the kinematic and dynamic characteristics of the system, define a multi-dimensional feature space containing physical variables such as position, velocity, acceleration, and attitude, and establish a mapping model between the measurements of each sensor and the physical variables.

[0026] The specific method in step 2) is as follows:

[0027] Define the system's true physical state vector as:

[0028]

[0029] in: Spatial coordinates; : velocity component; : Acceleration component; No. The observation model of a single sensor can be expressed in linear or nonlinear form as follows:

[0030]

[0031] in: : No. Measurement functions of each sensor; : Measure noise, meet ;

[0032] If a linear approximation is used, then:

[0033]

[0034] in The measurement matrix describes the mapping relationship between physical quantities and observed values.

[0035] Step 3) Physical consistency relationship construction steps: Combining the mapping model constructed in Step 2), construct a physical consistency equation that reflects the coupling constraints between different physical variables, calculate the physical consistency deviation between the measurement results of each sensor, and form a joint constraint measurement model.

[0036] The specific method in step 3) is as follows:

[0037] Physical consistency refers to the relationship that measurement results from multiple sensors for the same physical state should satisfy under physical constraints. This is defined for any two sensors... and The consistency deviation is:

[0038]

[0039] Ideally, the following should be satisfied:

[0040]

[0041] in: : Permissible physical consistency tolerance threshold; Euclidean norm or weighted norm.

[0042] Construct a joint constraint measurement model:

[0043]

[0044] The optimization objective is to minimize measurement error while ensuring physical consistency.

[0045] Step 4) Joint feature extraction step: Under the constraint of the physical consistency equation, a deep feature extraction algorithm or a multi-source subspace mapping method is used to extract common and differential features in multimodal data to achieve unified feature representation of cross-source data.

[0046] The specific method in step 4) is as follows:

[0047] While maintaining consistency constraints, multi-source observations are mapped to a shared latent space, common-mode features are extracted, and a feature extraction function is defined:

[0048]

[0049] in: : has parameters Nonlinear mapping functions (such as autoencoders or graph convolutional networks); : No. The latent feature vectors of each sensor;

[0050] The joint optimization objective is:

[0051]

[0052] Among them: the first item guarantees semantic consistency; the second item guarantees physical consistency; These are the physical constraint weighting coefficients; This indicates the consistency deviation between sensor i and sensor j at time t calculated based on the physical constraint model;

[0053] After training, an interpretable set of multi-source features is obtained. .

[0054] Step 5) Consistency assessment and anomaly detection steps: Based on the constructed joint measurement model, the consistency residual of the measurement results of each sensor is calculated. When the physical consistency deviation of a sensor exceeds the set threshold, the sensor is determined to be abnormal, drifting or faulty.

[0055] The specific method in step 5) is as follows:

[0056] Consistency residuals are calculated for the measurement results of each sensor:

[0057]

[0058] in: : The state estimate obtained by joint optimization; : No. Consistency residuals of individual sensors;

[0059] Set threshold ,when

[0060]

[0061] This indicates that the sensor measurement is abnormal.

[0062] Calculate the consistency confidence level based on the residual statistics:

[0063]

[0064] in This represents the residual variance.

[0065] Step 6) Result fusion and model update steps: Weighted fusion or reconstruction of the measurement data with good consistency, output joint measurement results, and dynamically update and adaptively adjust the parameters of the physical constraint model based on the fusion results to achieve continuous optimization of the system.

[0066] The specific method in step 6) is as follows:

[0067] right The sensor data is weighted and fused to obtain joint measurement results:

[0068]

[0069] in: The merged system state estimate; Sensor confidence weights; : Measure the noise covariance matrix,

[0070] To maintain the model's adaptability, the physical parameters are dynamically updated:

[0071]

[0072] in: Learning rate; The local Jacobian matrix of the measurement function;

[0073] The update mechanism enables the system to automatically correct measurement deviations and physical model errors during long-term operation.

[0074] Compared with the prior art, the present invention has the following beneficial effects:

[0075] Enhanced Physical Interpretability: By introducing physical consistency constraints, the physical relationships between multi-source data are explicitly modeled, giving the fusion results clear physical meaning. Significantly Improved Measurement Accuracy: In noisy and drift-interference environments, consistency constraints and weighted fusion mechanisms significantly reduce system state estimation errors and improve position and velocity estimation accuracy to varying degrees. Anomaly Detection and Fault Tolerance Capabilities: Through a consistency residual detection mechanism, sensor anomalies and communication delays can be effectively identified, improving system security and robustness. Adaptive Update Capabilities: Measurement model parameters are automatically corrected through dynamic update formulas, enabling the system to adapt to environmental changes and sensor drift during long-term operation. Wide Applicability: This invention is not dependent on specific sensor types and is applicable to multi-source heterogeneous measurement scenarios such as intelligent driving vehicles, industrial equipment condition monitoring, structural health assessment, and robot navigation.

[0076] In summary, this invention achieves a deep integration of data-driven and physical modeling through a multi-source joint measurement and analysis mechanism driven by physical consistency, providing a high-precision, interpretable, and robust general solution for multivariable measurement systems. Attached Figure Description

[0077] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0078] The multi-source data joint measurement and analysis method based on physical consistency relationship of the present invention mainly includes the following steps: Step 1) Multi-source data acquisition and preprocessing; Step 2) Physical feature modeling; Step 3) Physical consistency relationship construction; Step 4) Multi-source joint feature extraction; Step 5) Consistency assessment and anomaly identification; Step 6) Result fusion and dynamic update.

[0079] The data flow relationship between each step is as follows: The acquisition module inputs multimodal data into the feature modeling module. After consistency analysis and feature extraction, the data is sent to the anomaly detection and fusion module to output the final measurement results. At the same time, the physical constraint model is updated through a feedback mechanism.

[0080] Step 1) Multi-source data acquisition and preprocessing

[0081] In an intelligent driving environment, the system consists of LiDAR, millimeter-wave radar, cameras, and an inertial measurement unit (IMU). All sensors synchronize their sampling via a unified time synchronization signal (e.g., GPS time reference or PTP precision time protocol).

[0082] At any moment , define the first The original observations from the sensors are:

[0083]

[0084] To ensure the comparability of measurement data from different channels, the present invention performs the following preprocessing on the raw data:

[0085] Time synchronization and interpolation:

[0086]

[0087] in, Indicates the amount of sensor delay compensation;

[0088] Noise filtering and normalization:

[0089]

[0090] in, These are the mean and standard deviation, respectively. This is the filter gain matrix.

[0091] This step effectively eliminates time drift and noise differences between different sensors, resulting in preprocessed data of a uniform scale.

[0092] Step 2) Physical Feature Modeling

[0093] This embodiment defines the state vector based on the system's dynamic characteristics:

[0094]

[0095] in: Target location coordinates; : velocity component; : Acceleration component. The observation equations for each sensor are:

[0096]

[0097] in, For the observation matrix, For measuring noise.

[0098] Step 3) Building Physical Consistency Relationships

[0099] To establish physical consistency between different sensors, the consistency deviation is defined as:

[0100]

[0101] And construct a joint optimization objective:

[0102]

[0103] Step 4) Multi-source joint feature extraction

[0104] This embodiment, based on the consistency model, employs a deep feature mapping function. Map the measurements from each sensor to a shared potential space:

[0105]

[0106] The training objective function is:

[0107]

[0108] After optimization, common mode features of each mode are obtained. This enables consistent representation of cross-source data.

[0109] The final state estimate is obtained through weighted fusion:

[0110]

[0111] Among them, weight Calculated from consistency confidence:

[0112]

[0113] For residuals, This represents the noise variance.

[0114] Step 5) Consistency assessment and anomaly identification

[0115] At each sampling time, calculate the consistency residual for each sensor:

[0116]

[0117] like (in If the observation result is positive, the sensor is determined to be faulty. The fault result is fed back to the model update module, which adaptively corrects the observation matrix.

[0118]

[0119] in This is the learning rate parameter, typically set to 0.01–0.1.

[0120] This mechanism enables the system to maintain measurement stability in the event of sensor calibration changes, environmental disturbances, or partial failure.

[0121] Step 6) Result fusion and dynamic update

[0122] Weighted fusion of sensor data with good consistency is performed to obtain joint measurement results:

[0123]

[0124] in: The merged system state estimate; Sensor confidence weights; : Measurement noise covariance matrix.

[0125] To maintain the model's adaptability, the physical parameters are dynamically updated:

[0126]

[0127] in: Learning rate; : The local Jacobian matrix of the measurement function.

[0128] This update mechanism ensures that the system automatically corrects measurement deviations and physical model errors during long-term operation, enabling self-learning and continuous optimization.

[0129] Through the above steps, the present invention realizes physical consistency modeling, joint measurement and anomaly identification of multi-source heterogeneous sensor data, ensuring the measurement accuracy and robustness of the system under complex working conditions.

Claims

1. A multi-source data joint measurement and analysis method based on physical consistency relationship, characterized in that, The method comprises the following steps: Step 1) Multi-source data acquisition and preprocessing step: collecting measurement data from different types of sensors, time synchronization, noise filtering, scale normalization and missing value compensation processing are performed on the collected data, and a unified spatiotemporal resolution preprocessed dataset is obtained; Step 2) Physical feature modeling step: based on the kinematics and dynamics characteristics of the system, a multi-dimensional feature space containing position, velocity, acceleration and attitude of physical variables is defined, and a mapping model between each sensor measurement and physical variable is established; Step 3) Physical consistency relationship construction step: combined with the mapping model constructed in step 2), a physical consistency equation reflecting the coupling constraint between different physical variables is constructed, the physical consistency deviation between the measurement results of each sensor is calculated, and a joint constraint measurement model is formed; Step 4) Joint feature extraction step: under the constraint of the physical consistency equation, the common mode feature and the difference feature in the multi-modal data are extracted by using a deep feature extraction algorithm or a multi-source subspace mapping method, and unified feature representation of cross-source data is realized; Step 5) Consistency evaluation and anomaly detection step: based on the constructed joint measurement model, the measurement results of each sensor are calculated for consistency residual, when the physical consistency deviation of a certain sensor exceeds the set threshold, it is determined that the sensor has anomaly, drift or failure; Step 6) Result fusion and model updating step: the measurement data with good consistency is weighted and fused or reconstructed, and the joint measurement result is output, and the physical constraint model is dynamically updated and the parameters are adaptively adjusted based on the fusion result, so that the system is continuously optimized. 2.The method of claim 1, wherein, The specific method in step 1) is: At time , let the system contain sensor sources , each of which measures the target physical state synchronously; The original observation data are represented as: ； wherein: : the th sensor collects an observation vector at time ; : the measurement dimension of the th sensor; In order to ensure the comparability of multi-source data, the following preprocessing is performed: Step 1.1) Time synchronization: Adopting time stamp correction algorithm to unify the sensor output of different sampling frequencies to time axis ; Step 1.2) Noise filtering: sliding average or Kalman filtering is used for each channel data: ； wherein is a filter gain matrix; Step 1.3) Scale normalization: ； wherein the first and second mean and standard deviation of the sensor data, respectively; the first and second mean and standard deviation of the sensor data, respectively; After processing, a multi-source dataset with unified format is obtained: 。 3.The method of claim 1, wherein, The specific method in step 2) is: The real physical state vector of the system is defined as: ； wherein: : spatial coordinate position; : velocity component; : acceleration component; the observation model of the sensor can be expressed in linear or non-linear form as: ； wherein: : the measurement function of the th sensor; : measurement noise, satisfying ; If a linear approximation form is used, then: ； wherein is a measurement matrix, describing the mapping of physical quantities to observations. 4.The method of claim 1, wherein, The specific method in step 3) is: Physical consistency refers to the relationship that the measurement results of the same physical state by multi-source sensors should satisfy under physical constraints. The consistency deviation of any two sensors and is defined as: ； Under ideal conditions, the following should be met: ； wherein: : allowed physical conformance tolerance threshold; : Euclidean norm or weighted norm; The joint constraint measurement model is constructed: ； The optimization objective minimizes the measurement error while ensuring physical consistency. 5.The method of claim 1, wherein, The specific method in step 4) is: Under the condition of maintaining consistency constraint, the multi-source observation is mapped to a shared latent space, the common mode feature is extracted, and a feature extraction function is defined: ； wherein: : a non-linear mapping function with parameters ; : a latent feature vector of the :th sensor; The joint optimization objective is: ； Wherein: the first term guarantees semantic consistency; the second term guarantees physical consistency; is a physical constraint weight coefficient; represents the consistency deviation of sensor i and sensor j at time t calculated based on a physical constraint model. An interpretable multi-source feature set is obtained after training . 6.The method of claim 1, wherein, The specific method in step 5) is: The consistency residual of the measurement result of each sensor is calculated: ； wherein: : state estimate obtained from joint optimization; : consistency residual of the i th sensor; Setting a threshold When ； If the sensor measurement is abnormal, According to the residual statistics, the consistency confidence is calculated: ； wherein is the residual variance. 7.The method of claim 1, wherein, The specific method in step 6) is: right The sensor data is weighted and fused to obtain joint measurement results: ； wherein: : fused system state estimate; : sensor trust weight; : measurement noise covariance matrix, In order to maintain the adaptability of the model, the physical parameters are dynamically updated: ； wherein: : learning rate; : local Jacobian matrix of the measurement function; The updating mechanism enables the system to automatically correct the measurement deviation and physical model error in long-term operation.

Images