A data assimilation method, device and equipment

Abstract

The present application relates to the technical field of data assimilation, and particularly relates to a data assimilation method, device and equipment, which comprises: collecting multi-beam scanning data according to a detection system to obtain a detection state data set; detecting and tracking a dynamic target according to the detection state data set to obtain detection target motion trajectory data; and analyzing spatial distribution characteristics of the detection target motion trajectory data to obtain a target group spatial distribution characteristic data set, wherein the target group spatial distribution characteristic data set comprises a spatial density distribution parameter and a velocity field gradient characteristic of the target group. Through multi-beam scanning data collection and dynamic target detection and tracking, the present application can more accurately obtain motion trajectory data of the target, and through analysis of the spatial distribution characteristics of the target group, including the spatial density distribution parameter and the velocity field gradient characteristic, it is helpful to more clearly understand the distribution and dynamic behavior of the target.

Description

Technical Field

[0001] This invention relates to the field of data assimilation technology, specifically to a data assimilation method, apparatus, and equipment. Background Technology

[0002] Traditional methods mostly rely on classical physics models, mainly including sounding wave propagation models and target motion models. These models assume that the target's motion is linear or quasi-linear and that wave propagation follows certain idealized rules. Among them, the sounding wave propagation model is used to assume that environmental factors (such as atmospheric refraction, ionosphere, etc.) have little influence on the signal when the sounding wave propagates, and ignores the correction for complex meteorological conditions and ionospheric disturbances. The target motion model is used to assume that the target moves in a certain known pattern (such as linear or uniform motion), and usually uses simple velocity and acceleration formulas to describe the target's trajectory.

[0003] Based on the above, it is clear that traditional methods often fail to accurately capture the trajectory and distribution characteristics of targets, especially in complex environments (such as atmospheric refraction and ionospheric interference). Due to the lack of in-depth analysis of the spatial distribution characteristics of target groups, traditional methods may lead to significant data deviations due to observation errors, thus affecting the accuracy of predictions. Furthermore, when processing probe state data, traditional methods typically lack corrections for factors such as atmospheric refractive index, meteorological changes, and ionospheric disturbances, which may result in errors in propagation paths and echo signals. In addition, the assimilation processing of probe data in traditional methods is relatively simplified, often neglecting the application of error compensation and correction factors. This makes it impossible to effectively refine the spatial distribution characteristics of target groups, resulting in high uncertainty in model results. Moreover, traditional methods have weak initial state optimization capabilities for numerical weather prediction systems, making it difficult to cope with rapid changes in weather patterns, leading to low prediction accuracy. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a data assimilation method, apparatus and equipment.

[0005] The technical solution adopted to solve the above-mentioned technical problems is: a data assimilation method, including:

[0006] The detection system performs multi-beam scanning data acquisition to obtain a detection status dataset; dynamic target detection and tracking are performed based on the detection status dataset to obtain the target motion trajectory data.

[0007] Spatial distribution feature analysis is performed on the motion trajectory data of the detected target to obtain a target group spatial distribution feature dataset, wherein the target group spatial distribution feature dataset includes the spatial density distribution parameters and velocity field gradient features of the target group;

[0008] Acquire real-time meteorological data, including atmospheric refractive index profile data and ionospheric disturbance parameters; perform electromagnetic propagation effect modeling on the detection state dataset based on the real-time meteorological data to obtain a propagation path correction factor;

[0009] The observation error of the target group spatial distribution feature dataset is compensated according to the propagation path correction factor to obtain the target group spatial distribution feature data after error correction.

[0010] An assimilation analysis model is constructed using the ensemble Kalman filter algorithm to analyze the spatial distribution characteristics of the target population after error correction, thereby obtaining a data assimilation model; the data assimilation model is then deployed to a numerical weather prediction system.

[0011] Preferably, dynamic target detection and tracking are performed based on the detection state dataset to obtain the target motion trajectory data, including:

[0012] The transmitted signal and the detection state dataset are subjected to temporal convolution to obtain high-resolution range image data; the high-resolution range image data are then filtered in the frequency domain according to the Doppler filter bank to separate moving target echoes from stationary clutter.

[0013] A constant false alarm rate (CFAR) detection threshold is set to detect target traces in the echoes of the moving target to obtain an initial set of target traces.

[0014] Based on the pointing parameters of the detection beam, a detection unit spatial index in polar coordinates is established. Based on the detection unit spatial index, the initial target point trace set is mapped to the gridded coordinate system of the detection scanning sector to obtain a spatiotemporally aligned initial target point trace set.

[0015] A time-series alignment model is constructed based on the detection scanning cycle parameters. The initial target point set obtained by different beam directions is projected to a unified Cartesian coordinate system through coordinate transformation based on the time-series alignment model to obtain a spatiotemporally aligned point set.

[0016] The trajectory of the newly emerging target is initialized to obtain a candidate trajectory set, wherein the candidate trajectory set includes unconfirmed trajectories and a first confirmed trajectory.

[0017] Preferably, the method of performing dynamic target detection and tracking based on the detection state dataset to obtain the target motion trajectory data further includes:

[0018] Interactive multi-model filtering is applied to the first confirmed trajectory to obtain the predicted state value of the target. The interactive multi-model filtering adopts a hybrid state transition model of uniform speed model, uniform acceleration model and cooperative turning model.

[0019] State prediction and updating are performed based on Kalman filtering, and the spatial coordinates, radial velocity and signal-to-noise ratio information of the spatiotemporal alignment point trace set are fused to obtain the optimized state prediction value of the appearing target.

[0020] The process noise covariance matrix and the measurement noise covariance matrix are dynamically corrected based on the target echo intensity to obtain the adaptive filtering parameter set; the trajectory confirmation threshold and deletion time limit are set to obtain the trajectory status label set.

[0021] Preferably, the method of performing dynamic target detection and tracking based on the detection state dataset to obtain the target motion trajectory data further includes:

[0022] An association matrix is ​​established based on the predicted state value of the target and the spatial proximity and Doppler frequency consistency of the spatiotemporal alignment point trace set. The association probability between the spatiotemporal alignment point trace set and the first confirmed trajectory is determined based on the association matrix.

[0023] The length of the first confirmed trajectory and the association probability are weighted and summed to obtain a trajectory confidence score; the first confirmed trajectory is then filtered based on the trajectory confidence score to obtain a second confirmed trajectory.

[0024] The spatial coordinate sequence of the second confirmed trajectory is converted into a geocentric fixed coordinate system representation to obtain the target motion trajectory data.

[0025] Preferably, the spatial distribution feature analysis is performed on the motion trajectory data of the detected target to obtain a target group spatial distribution feature dataset, including:

[0026] Based on the three-dimensional spatial coordinate sequence of the target motion trajectory data, the spatial distribution of the target points is modeled using a kernel density estimation algorithm to obtain a density distribution surface;

[0027] The local density extreme value regions in the density distribution surface are identified according to the density peak clustering algorithm to obtain the distribution region parameters of the target group, wherein the distribution region parameters include peak density, average density and area enclosed by contour lines;

[0028] The density distribution surface is classified into high-density and low-density regions based on the distribution region parameters of the target group. Noise filtering based on neighborhood connectivity is then performed on the low-density regions to obtain a set of spatial density distribution parameters.

[0029] Based on the velocity vector sequence of the target motion trajectory data, a three-dimensional velocity vector field is constructed, and the velocity gradient tensor of the three-dimensional velocity vector field is determined according to the five-point center difference method.

[0030] The velocity gradient tensor is convolved using the Sobel operator to obtain velocity field gradient features, which include the maximum gradient magnitude, gradient direction variance, and velocity field curvature.

[0031] Preferably, the electromagnetic propagation effect is modeled on the detection state dataset based on the real-time meteorological data to obtain a propagation path correction factor, including:

[0032] Based on the atmospheric refractive index profile data, a three-dimensional atmospheric refractive index field is constructed using the Kriging interpolation algorithm. Combined with the location of the detection station and the detection range, a spatially continuously distributed refractive index model is obtained.

[0033] Numerical integration was performed using the fourth-order Runge-Kutta method to obtain the region of drastic refractive index gradient change; the actual propagation path of electromagnetic waves in the region of drastic refractive index gradient change was determined using the three-dimensional ray tracing equation.

[0034] The geometric correction factor is determined based on the deviation between the actual propagation path and the apparent path of the electromagnetic wave, and the path delay compensation is determined based on the geometric correction factor.

[0035] Based on the ionospheric perturbation parameters, an equivalent phase screen model of the ionosphere is constructed, wherein the equivalent phase screen model of the ionosphere includes the total electron content and the scintillation index, and the spatial distribution of the phase perturbation is described by a piecewise linear method;

[0036] The phase delay and phase scintillation caused by the ionosphere are determined based on the spatial distribution of the phase perturbation and the equivalent phase screen model of the ionosphere, and the phase compensation factor is determined based on the phase delay and the phase scintillation.

[0037] The path delay compensation and the phase compensation factor are vector-superimposed to obtain the propagation path correction factor.

[0038] Preferably, an assimilation analysis model is constructed using an ensemble Kalman filter algorithm on the error-corrected spatial distribution characteristic data of the target group to obtain a data assimilation model, including:

[0039] The background field data of the numerical weather prediction model is projected onto the spatiotemporal coordinate system of the detection and observation space using a bilinear interpolation algorithm to obtain a spatiotemporally aligned background field dataset, wherein the background field data includes wind field, temperature field and pressure field.

[0040] The spatial distribution characteristics data of the target group after error correction are transformed to obtain the equivalent momentum flux field; the initial perturbation set is obtained by generating perturbations through the background field error covariance matrix based on singular value decomposition.

[0041] The disturbance magnitude of the initial disturbance set is matched with the uncertainty level of the background field to obtain the initial state set of the numerical weather prediction model; random error disturbances are applied to the detection and observation field to obtain the observation set.

[0042] Preferably, the assimilation analysis model is constructed by using an ensemble Kalman filter algorithm to analyze the spatial distribution characteristics of the target group after error correction, in order to obtain a data assimilation model, and further includes:

[0043] Numerical weather prediction model integration is performed on each member of the initial state set to predict the atmospheric state at future times, thereby obtaining a predicted state set; the predicted state set is mapped to the detection and observation space to determine the equivalent predicted observations, thereby obtaining a predicted observation set.

[0044] Determine the covariance matrix of the predicted state set and the covariance matrix of the predicted observation set; determine the correction weights of the observed data on the state of the numerical weather prediction model based on covariance analysis, so as to obtain the Kalman gain matrix;

[0045] The predicted state set is dynamically adjusted based on the Kalman gain matrix, and the constraint information of the detection and observation data is combined to obtain the analysis state set; the error covariance matrix is ​​updated based on the statistical characteristics of the analysis state set to obtain the data assimilation model.

[0046] The technical solution adopted to solve the above-mentioned technical problem is: a data assimilation device, which is applicable to the aforementioned data assimilation method, comprising:

[0047] The data acquisition module is used to acquire multi-beam scanning data from the detection system to obtain a detection state dataset; and to perform dynamic target detection and tracking based on the detection state dataset to obtain the target motion trajectory data.

[0048] The feature analysis module is used to perform spatial distribution feature analysis on the motion trajectory data of the detected target to obtain a target group spatial distribution feature dataset, wherein the target group spatial distribution feature dataset includes the spatial density distribution parameters and velocity field gradient features of the target group;

[0049] A propagation correction module is used to acquire real-time meteorological data, including atmospheric refractive index profile data and ionospheric disturbance parameters; and to model the electromagnetic propagation effect of the detection state dataset based on the real-time meteorological data to obtain a propagation path correction factor.

[0050] An error compensation module is used to compensate for the observation error of the target group spatial distribution feature dataset according to the propagation path correction factor, so as to obtain the target group spatial distribution feature data after error correction.

[0051] The data assimilation module is used to construct an assimilation analysis model for the error-corrected spatial distribution characteristic data of the target group using an ensemble Kalman filter algorithm, so as to obtain a data assimilation model; and to deploy the data assimilation model to the numerical weather prediction system.

[0052] The technical solution adopted to solve the above-mentioned technical problems is: a data assimilation device, comprising: a processor and a memory; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory, causing the processor to execute the aforementioned data assimilation method.

[0053] The beneficial effects of the present invention are as follows: (1) The present invention can more accurately obtain the motion trajectory data of the target by multi-beam scanning data acquisition and dynamic target detection and tracking. Moreover, by analyzing the spatial distribution characteristics of the target group, including spatial density distribution parameters and velocity field gradient characteristics, it helps to understand the distribution and dynamic behavior of the target more clearly, thereby reducing the data deviation caused by observation errors. Furthermore, by acquiring real-time meteorological data, especially atmospheric refractive index profile data and ionospheric disturbance parameters, and by modeling the electromagnetic propagation effect to correct the propagation path of the detection state data, it can effectively reduce the error caused by environmental factors (such as atmospheric refraction and ionospheric disturbance) and ensure the accuracy of the detection data; (2) The present invention utilizes Using a propagation path correction factor to compensate for errors in the spatial distribution characteristics of the target population can yield more accurate data, which is crucial for the assimilation of the probe data and helps improve the reliability and accuracy of the model. Furthermore, by performing assimilation analysis on the error-corrected spatial distribution characteristics of the target population using an ensemble Kalman filter algorithm, the initial state in the numerical weather prediction system can be optimized, reducing model uncertainty and thus improving prediction accuracy, especially when weather patterns change rapidly. Moreover, by effectively combining the data assimilation model with the numerical weather prediction system, the probe data can be used to optimize weather forecasts in real time, improving the ability to predict weather changes, especially in complex environments. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the overall method steps in one embodiment of the present invention;

[0055] Figure 2 This is a schematic diagram of the overall device architecture in one embodiment of the present invention;

[0056] Figure 3This is a schematic diagram of the overall device architecture in one embodiment of the present invention.

[0057] Reference numerals in the attached diagram: 1. Data acquisition module; 2. Feature analysis module; 3. Propagation correction module; 4. Error compensation module; 5. Data assimilation module; 6. Processor; 7. Memory. Detailed Implementation

[0058] Example 1, as Figure 1 As shown, the data assimilation method proposed in this invention is characterized by comprising:

[0059] S1. Perform multi-beam scanning data acquisition based on the detection system to obtain the detection status dataset; perform dynamic target detection and tracking based on the detection status dataset to obtain the target motion trajectory data;

[0060] S2. Perform spatial distribution feature analysis on the motion trajectory data of the detected target to obtain a target group spatial distribution feature dataset, which includes the spatial density distribution parameters and velocity field gradient features of the target group.

[0061] S3. Acquire real-time meteorological data, including atmospheric refractive index profile data and ionospheric disturbance parameters; model the electromagnetic propagation effect of the detection state dataset based on the real-time meteorological data to obtain the propagation path correction factor;

[0062] S4. Perform observation error compensation on the target group spatial distribution feature dataset according to the propagation path correction factor to obtain the target group spatial distribution feature data after error correction.

[0063] S5. Construct an assimilation analysis model for the spatial distribution characteristics of the target group after error correction using the ensemble Kalman filter algorithm to obtain a data assimilation model; deploy the data assimilation model to the numerical weather prediction system.

[0064] In this invention, observation error compensation is performed on the target group spatial distribution feature dataset based on the propagation path correction factor to obtain error-corrected target group spatial distribution feature data. Specifically: Spatiotemporal interpolation processing is performed on the propagation path correction factor. The Krugkin algorithm is used to improve the spatial resolution of the correction factor field to be consistent with the target group distribution dataset. The time dimension is synchronized at the second level through cubic spline interpolation, ensuring strict spatiotemporal correspondence between the correction factor and the target group trajectory points. A transformation matrix from the probe coordinate system to the geocentric fixed coordinate system (ECEF) is constructed. A seven-parameter coordinate transformation model is used to eliminate spatial reference differences between different data sources, achieving spatial registration between the correction factor and the target group trajectory points. The atmospheric refraction delay term and ionospheric projection deviation term in the analytical geometric correction factor are analyzed to establish a position correction model: For the path bending effect caused by atmospheric refraction, the target position is corrected through ray tracing inverse operation, compensating for distance deviation along the probe line of sight. For the position offset caused by ionospheric electron density inhomogeneity, the ionospheric puncture point (IPP) projection method is used for reverse correction to eliminate azimuth deviation caused by ionospheric refraction. The compensated position data is then subjected to RA (Radar Interference). The NSAC algorithm is used for filtering to remove position jumps caused by abnormal correction factors, retaining at least 95% of inliers to ensure the continuity of spatial distribution. Ionospheric delay and scintillation disturbance terms are extracted from the phase compensation factor to construct a velocity correction model. For velocity deviations caused by ionospheric dispersion, the influence of first-order terms is eliminated through carrier frequency normalization. For velocity fluctuations caused by phase scintillation, an adaptive threshold method is used to identify and filter high-frequency disturbance components, retaining effective velocity information. A Kalman smoother is used for time-domain filtering of the velocity correction, setting the process noise covariance matrix to suppress high-frequency noise while preserving the dynamic characteristics of small- and medium-scale weather systems. Based on the compensated position data, the spatial density distribution is recalculated using Adaptive Bandwidth Kernel Density Estimation (ABKDE). The bandwidth parameter is dynamically adjusted using Silverman rules to ensure a resolution better than 500 meters in high-density areas while smoothing noise in low-density areas. An improved DBSCAN algorithm is applied to perform cluster analysis on the density field, setting thresholds for neighborhood radius and core point number to identify the core distribution area of ​​the corrected target group and filter out false density peaks introduced by error compensation.

[0065] Example 2: The data assimilation method proposed in this invention, compared with Example 1, further includes: dynamic target detection and tracking based on the detection state dataset to obtain the target motion trajectory data, including:

[0066] A1. Perform temporal convolution on the transmitted signal and detection status dataset to obtain high-resolution range image data; perform frequency domain filtering on the high-resolution range image data according to the Doppler filter bank to separate moving target echoes from stationary clutter.

[0067] A2. Set a constant false alarm rate detection threshold to detect target traces on the echo of moving target in order to obtain an initial target trace set;

[0068] A3. Establish a detection unit spatial index in polar coordinate system based on the detection beam pointing parameters, and map the initial target point trace set to the gridded coordinate system of the detection scanning sector based on the detection unit spatial index to obtain the spatiotemporally aligned initial target point trace set.

[0069] A4. Construct a time series alignment model based on the detection scanning cycle parameters. Based on the time series alignment model, project the initial target point set obtained by different beam directions to a unified Cartesian coordinate system through coordinate transformation to obtain a spatiotemporally aligned point set.

[0070] A5. Initialize the trajectory of the newly appearing target to obtain a candidate trajectory set, which includes unconfirmed trajectory and the first confirmed trajectory.

[0071] In this embodiment, the transmitted signal and the received detection state data are convolved in the time domain. Time-domain convolution is a signal processing method used to extract features from the detection state data, ultimately obtaining high-resolution range image data. This data contains the range information of targets within the scanned area. Frequency-domain filtering is applied to the high-resolution range image data to separate moving target echoes from stationary clutter using the Doppler effect. The Doppler filter bank distinguishes the echo signals of moving targets and stationary objects based on different frequency changes, ensuring that only true moving targets are considered. A constant false alarm rate (CFAR) detection threshold is set to detect the presence of targets in the detection state. The CFAR method can automatically adjust the threshold value based on background clutter signals to avoid false detections of environmental noise. When the target echo signal is received, the target position can be confirmed, thus obtaining the initial target point set. An index for the detection unit space is established in polar coordinates using the beam pointing parameters of the phased array detection. Because the phased array detection scans the area... The domain is typically sector-shaped, so target points are spatially partitioned and indexed in polar coordinates for subsequent processing. The initial target point set is mapped to the gridded coordinate system of the probe scan to achieve spatiotemporal alignment. This step ensures that the target point set corresponds to the sector of the probe scan, facilitating further analysis. A time-series alignment model is constructed based on the probe scan cycle parameters. This means that as the probe scan cycle changes, the temporal information of the target points needs to be aligned with the points pointing under different beam directions. Through coordinate transformation, these target points are projected into a unified Cartesian coordinate system to obtain a spatiotemporally aligned point set. Track initialization is performed for newly appearing targets. This means that if the probe detects a new target, the system needs to make a preliminary estimate of its trajectory and initialize the trajectory. The candidate trajectory set includes two types of trajectories: one is unconfirmed trajectory (which may be a false detection or a temporary target); the other is the first confirmed trajectory (the confirmed target). The candidate trajectory set helps to determine the true trajectory of the target.

[0072] In an optional embodiment, dynamic target detection and tracking are performed based on the detection state dataset to obtain the target motion trajectory data, further comprising:

[0073] A6. Perform interactive multi-model filtering on the first confirmed trajectory to obtain the state prediction value of the target. The interactive multi-model filtering adopts a hybrid state transition model of uniform speed model, uniform acceleration model and cooperative turning model.

[0074] A7. Based on Kalman filtering, state prediction and update are performed by fusing the spatial coordinates, radial velocity and signal-to-noise ratio information of the spatiotemporal aligned point trace set to obtain the optimized state prediction value of the appearing target.

[0075] A8. Based on the target echo intensity, dynamically correct the process noise covariance matrix and the measurement noise covariance matrix to obtain the adaptive filtering parameter set; set the trajectory confirmation threshold and deletion time limit to obtain the trajectory status label set.

[0076] It should be noted that Interactive Multi-Model Filtering (IMM) is a filtering method based on multiple different motion models, capable of handling targets with varying dynamic behaviors. Here, IMM employs three different models: a uniform velocity model assuming the target moves at a constant speed; a uniform acceleration model assuming the target's motion is uniformly accelerated, meaning the target's speed changes continuously during motion; and a cooperative turning model used to describe situations where the target turns, suitable for the target's turning trajectory. These models are combined through a hybrid state transition model; the filter selects different models based on the target's motion to obtain more accurate state predictions. State prediction and updating based on Kalman filtering: Kalman filtering is a recursive filtering algorithm based on linear systems used to estimate the system's state. In this application, Kalman filtering incorporates the target's spatiotemporal alignment point trace set (including spatial coordinates, radial velocity, and signal-to-noise ratio information) for state prediction and updating. The Kalman filter is used to predict the target's state, i.e., predicting its future position and velocity based on the target's current state and motion model. The predicted values ​​are updated by comparing them with actual measurements to reduce errors. The acoustic covariance matrix is ​​used to describe the magnitude and relationship of process noise and measurement noise in the system. In this method, the noise covariance matrix is ​​dynamically adjusted by the target echo intensity to ensure that the filter can adapt to the target detection environment under different signal-to-noise ratio conditions. The process noise covariance matrix describes the uncertainty of the system itself, such as the dynamic model of the target; it describes the errors that may occur during the measurement process, such as detected noise or external interference; based on the dynamically adjusted noise covariance matrix, an adaptive set of filter parameters can be obtained; this means that the filter can self-adjust its parameters according to the actual echo intensity and environmental changes, thereby improving the accuracy of target state prediction; it is used to determine whether a target's trajectory is valid; when the target's state remains stable for a period of time and exceeds a certain threshold, the trajectory is considered valid and reliable; if the target does not update within a certain period of time or its state no longer meets expectations, the trajectory will be deleted according to the deletion time limit rule to avoid interference from false detections or invalid trajectories; finally, the system generates a trajectory state label set to mark the trajectory state of each target; these labels usually include information such as whether the target has been confirmed and whether the trajectory is valid, for subsequent processing and decision-making.

[0077] In an optional embodiment, dynamic target detection and tracking are performed based on the detection state dataset to obtain the target motion trajectory data, further comprising:

[0078] A8. Establish an association matrix based on the predicted state value of the target and the spatial proximity and Doppler frequency consistency of the spatiotemporal alignment point trace set. Determine the association probability between the spatiotemporal alignment point trace set and the first confirmed trajectory based on the association matrix.

[0079] A9. The length and association probability of the first confirmed trajectory are weighted and summed to obtain the trajectory confidence score; the first confirmed trajectory is filtered according to the trajectory confidence score to obtain the second confirmed trajectory.

[0080] A10. Convert the spatial coordinate sequence of the second confirmed trajectory into a geocentric fixed coordinate system representation to obtain the target motion trajectory data.

[0081] It should be noted that the target state prediction value predicts the target's state (position, velocity, etc.) through a dynamic model of the target (such as Kalman filtering); the spatiotemporally aligned point track set typically refers to a set of target point track data acquired through detection or other sensors, which, after spatiotemporal alignment processing, can provide specific location information of the target in time and space; the two parameters of spatial proximity and Doppler frequency consistency are used to evaluate the relationship between each point track and the target; spatial proximity refers to the degree of spatial closeness between the predicted target position and the point track, and Doppler frequency consistency refers to whether the target's frequency offset matches the predicted target frequency; this information is used to establish an association matrix, where each element represents the degree of association between the point track and the target trajectory; based on the established association matrix, the association probability between the point track and the confirmed trajectory (i.e., the first confirmed trajectory) can be calculated; this probability represents the association probability between a certain spatiotemporally aligned point track set and the confirmed trajectory. The correlation between trajectories; the length of a confirmed trajectory refers to the time span or distance of the target trajectory; a longer trajectory usually means that the target is moving stably within a certain period of time, with a higher confidence level; by weighted summation of the association probabilities with the points, the confidence score of the target trajectory can be obtained; this score measures the reliability of the target trajectory, and the higher the score, the more likely the trajectory is to be correct; based on the calculated confidence score, a new confirmed trajectory (i.e., the second confirmed trajectory) is selected; a high-confidence trajectory means that the target's movement trajectory is more stable and can continue to be tracked; a low-confidence trajectory may be discarded or requires further verification; finally, the spatial coordinate sequence of the second confirmed trajectory is converted into a geocentric fixed coordinate system, that is, the trajectory is transformed from a ground coordinate system to a coordinate system with the Earth's center as the origin; this transformation allows the trajectory data to more accurately represent the target's position and movement on the Earth's surface;

[0082] In an optional embodiment, spatial distribution feature analysis is performed on the motion trajectory data of the detected targets to obtain a target group spatial distribution feature dataset, including:

[0083] B1. Based on the three-dimensional spatial coordinate sequence of the target motion trajectory data, the spatial distribution of the target points is modeled using the kernel density estimation algorithm to obtain the density distribution surface;

[0084] B2. Identify local density extreme value regions in the density distribution surface using the density peak clustering algorithm to obtain the distribution region parameters of the target group. The distribution region parameters include peak density, average density, and area enclosed by contour lines.

[0085] B3. Based on the distribution area parameters of the target group, the density distribution surface is classified into high-density areas and low-density areas. Noise filtering based on neighborhood connectivity is implemented in the low-density areas to obtain the spatial density distribution parameter set.

[0086] B4. Based on the velocity vector sequence of the target's motion trajectory data, construct a three-dimensional velocity vector field, and determine the velocity gradient tensor of the three-dimensional velocity vector field using the five-point central difference method.

[0087] B5. Perform convolution operation on the velocity gradient tensor according to the Sobel operator to obtain the velocity field gradient features, which include the maximum gradient magnitude, gradient direction variance and velocity field curvature.

[0088] It should be noted that kernel density estimation is used to analyze the three-dimensional trajectory data of the target. It establishes a density distribution surface by calculating the spatial density distribution of the points. This surface shows the density of the target in different regions of space, reflecting the distribution characteristics of the target points. Density peak clustering is a clustering algorithm that determines the cluster centers by finding local density extrema in the data. Specifically, the algorithm identifies local density extrema regions in the density distribution surface, thereby dividing the distribution area of ​​the target group. Distribution area parameters include: peak density refers to the highest density value within the region; average density refers to the average density of the points within the region; and the area enclosed by contour lines refers to the size of the region surrounding the peak density, usually represented as an isodense region. Based on the distribution area parameters of the target group, the density distribution surface is classified into high-density and low-density regions. For low-density regions, noise filtering is performed using neighborhood connectivity methods to remove regions with excessively low density that may be noise. This step helps reduce erroneous target points, resulting in a more accurate target trajectory. The velocity vector field represents the direction and magnitude of motion at each point, reflecting the target's motion information. A three-dimensional velocity vector field is constructed by detecting the target's trajectory data. The velocity gradient tensor is an important quantity describing changes in the velocity field. The gradient of the velocity vector field is determined using the five-point center difference method (i.e., by calculating the difference between adjacent points). This captures the spatial characteristics of the target's velocity changes. The Sobel operator is a commonly used edge detection algorithm in image processing, capable of extracting gradient information from images. In this process, the Sobel operator is used to convolve the velocity gradient tensor to obtain the gradient features of the velocity field. The velocity field gradient features include: maximum gradient magnitude: the maximum change in gradient in the velocity field; gradient direction variance: the degree of change in the gradient direction, used to describe the irregularity of the velocity field; and velocity field curvature: the curvature of the velocity field, reflecting the degree of bending of the velocity field.

[0089] In an optional embodiment, electromagnetic propagation effects are modeled on the detection state dataset based on real-time meteorological data to obtain a propagation path correction factor, including:

[0090] C1. A three-dimensional atmospheric refractive index field is constructed based on atmospheric refractive index profile data using the Kriging interpolation algorithm. Combined with the location of the detection station and the detection range, a spatially continuous refractive index model is obtained.

[0091] C2. Numerical integration is performed using the fourth-order Runge-Kutta method to obtain the region of drastic refractive index gradient change; the actual propagation path of electromagnetic waves in the region of drastic refractive index gradient change is determined using the three-dimensional ray tracing equation.

[0092] C3. Determine the geometric correction factor based on the deviation between the actual propagation path and the apparent path of the electromagnetic wave, and determine the path delay compensation amount based on the geometric correction factor.

[0093] C4. Based on the ionospheric perturbation parameters, construct an equivalent phase screen model of the ionosphere. The equivalent phase screen model of the ionosphere includes the total electron content and the scintillation index. The spatial distribution of the phase perturbation is described using a piecewise linear method.

[0094] C5. Determine the phase delay and phase scintillation caused by the ionosphere based on the spatial distribution of phase perturbation and the equivalent phase screen model of the ionosphere, and determine the phase compensation factor based on the phase delay and phase scintillation.

[0095] C6. Vector superposition of path delay compensation and phase compensation factor to obtain propagation path correction factor.

[0096] It should be noted that Kriging interpolation is a statistical method used to estimate the values ​​of spatial data based on the known spatial distribution of data points. For atmospheric refractive index profile data, Kriging interpolation can construct a three-dimensional refractive index field based on the irregularly distributed refractive index data in the atmosphere. This refractive index field reflects the refractive index distribution in different regions of the atmosphere and is used to describe the propagation characteristics of electromagnetic waves in the atmosphere. Combined with the location and range of the detection station, this three-dimensional refractive index field provides a continuous spatial distribution of atmospheric refractive index, which helps predict the propagation behavior of electromagnetic waves in different atmospheric layers. The fourth-order Runge-Kutta method is used. The method involves numerical integration to calculate the gradient change of atmospheric refractive index. This method can handle regions with drastic refractive index changes in the atmospheric refractive index field, helping to locate areas of rapid refractive index variation, which has a significant impact on the propagation path of electromagnetic waves. Regions with drastic refractive index gradient changes are the most sensitive parts of electromagnetic wave propagation, often causing deviations in the signal propagation path. The three-dimensional ray tracing equation is used to calculate the actual propagation path of electromagnetic waves emitted from the detection station or satellite. Based on the refractive index field and gradient change region, the actual propagation path of the electromagnetic wave is determined, considering the influence of factors such as the atmosphere and ionosphere. The difference between these actual propagation paths and the ideal path reflects the propagation error caused by refractive index gradient changes, ionospheric disturbances, etc. By calculating the deviation between the actual propagation path and the apparent path of the electromagnetic wave, a geometric correction factor can be derived. This factor is used to correct for errors caused by atmospheric refraction and other environmental factors. The impact on the electromagnetic wave propagation path is considered. Furthermore, the path delay compensation is determined based on a geometric correction factor, which helps reduce time delay errors caused by refraction and irregular propagation. Ionospheric disturbance parameters are used to describe the influence of the ionosphere on electromagnetic wave propagation. Ionospheric disturbances include the total electron content (TEC) and the scintillation index (S4), which affect the phase of electromagnetic wave propagation. An equivalent phase screen model of the ionosphere is constructed, and the spatial distribution of phase disturbances is described using a piecewise linear method. This model can simulate the impact of ionospheric disturbances on the signal. By analyzing the spatial distribution of phase disturbances, the phase delay and phase scintillation caused by the ionosphere can be calculated, and the phase compensation factor can be further determined based on this information. Finally, the path delay compensation and the phase compensation factor are vector-superimposed to obtain the final propagation path correction factor. This correction factor can be used to correct the propagation path of electromagnetic waves, reduce signal propagation errors, and improve the accuracy of the system.

[0097] In an optional embodiment, an assimilation analysis model is constructed using an ensemble Kalman filter algorithm on the error-corrected spatial distribution characteristic data of the target population to obtain a data assimilation model, including:

[0098] D1. Project the background field data of the numerical weather prediction model onto the spatiotemporal coordinate system of the detection and observation space using the bilinear interpolation algorithm to obtain a spatiotemporally aligned background field dataset, which includes wind field, temperature field and pressure field.

[0099] D2. Perform variable transformation on the spatial distribution characteristic data of the target group after error correction to obtain the equivalent momentum flux field; generate perturbations through the background field error covariance matrix based on singular value decomposition to obtain the initial perturbation set.

[0100] D3. Match the disturbance magnitude of the initial disturbance set with the uncertainty level of the background field to obtain the initial state set of the numerical weather prediction model; apply random error disturbance to the detection and observation field to obtain the observation set.

[0101] It should be noted that background field data refers to the initial atmospheric data obtained through numerical weather prediction models, typically including wind, temperature, and pressure fields. The background field data is transformed from the original coordinate system (such as a three-dimensional grid of the atmosphere) to the spatiotemporal coordinate system of the observation space using bilinear interpolation algorithms. This transformation aims to ensure that the background data and the observed data are aligned in time and space during the observation process, guaranteeing data consistency and comparability. A new physical quantity, the equivalent momentum flux field, is obtained by transforming the error-corrected spatial distribution characteristics data of the target group. This is usually done to better match the original weather forecast data with the observed data and facilitate subsequent assimilation processes. Singular value decomposition (SVD): This involves using singular values... Decomposition allows for the extraction of key perturbation patterns from the background field error covariance matrix. These perturbations represent uncertainties in the background field. The perturbation set generated by the background field error covariance matrix describes the uncertainties in the initial state. This provides different initial conditions for weather forecasting, improving forecast robustness. The perturbation amplitude represents the magnitude of the perturbation and typically needs to match the magnitude of the uncertainty in the background field. This matching ensures that the perturbation remains within a certain range, avoiding excessive modification of the background field. To simulate errors in actual observations, random error perturbations are applied to the probe observation data, generating an observation set. This simulates errors that may occur in real-world probe observation data, such as noise or systematic errors. Ensemble Kalman Filter (EnKF) is used for assimilation analysis. Its basic idea is to combine multiple possible initial state sets and adjust them using Kalman filtering. By integrating each member of the initial state set using a numerical weather prediction model, the atmospheric state prediction for future times is obtained. These predicted states are mapped to the probe observation space, resulting in a predicted observation set.

[0102] In an optional embodiment, an assimilation analysis model is constructed using an ensemble Kalman filter algorithm on the error-corrected spatial distribution characteristic data of the target group to obtain a data assimilation model, which further includes:

[0103] D4. Perform numerical weather prediction model integration on each member of the initial state set to predict the atmospheric state at future times, thereby obtaining the predicted state set; map the predicted state set to the detection observation space to determine the equivalent predicted observations, thereby obtaining the predicted observation set.

[0104] D5. Determine the covariance matrix of the predicted state set and the covariance matrix of the predicted observation set; determine the correction weights of the observed data on the state of the numerical weather prediction model based on the covariance analysis, so as to obtain the Kalman gain matrix.

[0105] D6. Dynamically adjust the predicted state set based on the Kalman gain matrix, and combine it with the constraint information of the detection and observation data to obtain the analytical state set; update the error covariance matrix based on the statistical characteristics of the analytical state set to obtain the data assimilation model.

[0106] It should be noted that the covariance matrix is ​​used to describe the correlation between the predicted state set and the predicted observation set. Through covariance analysis, the weights of the observation data on the model state correction can be determined, i.e., the Kalman gain matrix. The role of the Kalman gain matrix is ​​to guide how to combine the observation data and the predicted state set to optimize weather forecasting. The predicted state set is dynamically adjusted using the Kalman gain matrix, thereby correcting the numerical weather forecast in conjunction with the observation data. Through this dynamic adjustment, the true atmospheric state can be better approximated. Finally, by updating the statistical properties of the analyzed state set, a new error covariance matrix is ​​obtained. This matrix is ​​used to describe the uncertainty of the model and ultimately yields the data assimilation model, i.e., the weather forecast model based on the observation data.

[0107] Example 3, as Figure 2 As shown, the present invention proposes a data assimilation apparatus, which is applicable to a data assimilation method, comprising:

[0108] Data acquisition module 1 is used to acquire multi-beam scanning data from the detection system to obtain a detection status dataset; and to perform dynamic target detection and tracking based on the detection status dataset to obtain the target motion trajectory data.

[0109] Feature analysis module 2 is used to perform spatial distribution feature analysis on the motion trajectory data of the detected target to obtain a target group spatial distribution feature dataset, wherein the target group spatial distribution feature dataset includes the spatial density distribution parameters and velocity field gradient features of the target group;

[0110] The propagation correction module 3 is used to acquire real-time meteorological data, including atmospheric refractive index profile data and ionospheric disturbance parameters; and to model the electromagnetic propagation effect of the detection state dataset based on the real-time meteorological data to obtain the propagation path correction factor.

[0111] Error compensation module 4 is used to compensate for the observation error of the target group spatial distribution feature dataset according to the propagation path correction factor, so as to obtain the target group spatial distribution feature data after error correction.

[0112] Data assimilation module 5 is used to construct an assimilation analysis model for the spatial distribution characteristics data of the target group after error correction using an ensemble Kalman filter algorithm, so as to obtain a data assimilation model; and to deploy the data assimilation model to the numerical weather prediction system.

[0113] Example 4, as Figure 3 The present invention proposes a data assimilation device, comprising: a processor 6 and a memory 7; the memory 7 stores computer execution instructions; the processor 6 executes the computer execution instructions stored in the memory 7, thereby enabling the processor 6 to perform a data assimilation method.

[0114] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A data assimilation method, characterized in that, include: The detection system performs multi-beam scanning data acquisition to obtain a detection status dataset; Dynamic target detection and tracking are performed based on the aforementioned detection state dataset to obtain the target motion trajectory data; Spatial distribution feature analysis is performed on the motion trajectory data of the detected target to obtain a target group spatial distribution feature dataset, wherein the target group spatial distribution feature dataset includes the spatial density distribution parameters and velocity field gradient features of the target group; Acquire real-time meteorological data, including atmospheric refractive index profile data and ionospheric disturbance parameters; perform electromagnetic propagation effect modeling on the detection state dataset based on the real-time meteorological data to obtain a propagation path correction factor; The observation error of the target group spatial distribution feature dataset is compensated according to the propagation path correction factor to obtain the target group spatial distribution feature data after error correction. An assimilation analysis model is constructed using the ensemble Kalman filter algorithm on the error-corrected spatial distribution characteristics data of the target group to obtain a data assimilation model; the data assimilation model is then deployed to a numerical weather prediction system. Specifically, electromagnetic propagation effect modeling is performed on the detection state dataset based on the real-time meteorological data to obtain a propagation path correction factor, including: Based on the atmospheric refractive index profile data, a three-dimensional atmospheric refractive index field is constructed using the Kriging interpolation algorithm. Combined with the location of the detection station and the detection range, a spatially continuously distributed refractive index model is obtained. Numerical integration was performed using the fourth-order Runge-Kutta method to obtain the region of drastic refractive index gradient change; the actual propagation path of electromagnetic waves in the region of drastic refractive index gradient change was determined using the three-dimensional ray tracing equation. The geometric correction factor is determined based on the deviation between the actual propagation path and the apparent path of the electromagnetic wave, and the path delay compensation is determined based on the geometric correction factor. Based on the ionospheric perturbation parameters, an equivalent phase screen model of the ionosphere is constructed, wherein the equivalent phase screen model of the ionosphere includes the total electron content and the scintillation index, and the spatial distribution of the phase perturbation is described by a piecewise linear method; The phase delay and phase scintillation caused by the ionosphere are determined based on the spatial distribution of the phase perturbation and the equivalent phase screen model of the ionosphere, and the phase compensation factor is determined based on the phase delay and the phase scintillation. The path delay compensation and the phase compensation factor are vector-superimposed to obtain the propagation path correction factor.

2. The data assimilation method according to claim 1, characterized in that, Dynamic target detection and tracking are performed based on the aforementioned detection state dataset to obtain the target motion trajectory data, including: The transmitted signal and the detection state dataset are subjected to temporal convolution to obtain high-resolution range image data; the high-resolution range image data are then filtered in the frequency domain according to the Doppler filter bank to separate moving target echoes from stationary clutter. A constant false alarm rate (CFAR) detection threshold is set to detect target traces in the echoes of the moving target to obtain an initial set of target traces. Based on the pointing parameters of the detection beam, a detection unit spatial index in polar coordinates is established. Based on the detection unit spatial index, the initial target point trace set is mapped to the gridded coordinate system of the detection scanning sector to obtain a spatiotemporally aligned initial target point trace set. A time-series alignment model is constructed based on the detection scanning cycle parameters. The initial target point set obtained by different beam directions is projected to a unified Cartesian coordinate system through coordinate transformation based on the time-series alignment model to obtain a spatiotemporally aligned point set. The trajectory of the newly emerging target is initialized to obtain a candidate trajectory set, wherein the candidate trajectory set includes unconfirmed trajectories and a first confirmed trajectory.

3. The data assimilation method according to claim 2, characterized in that, Dynamic target detection and tracking are performed based on the aforementioned detection state dataset to obtain the target motion trajectory data, and the method further includes: Interactive multi-model filtering is applied to the first confirmed trajectory to obtain the predicted state value of the target. The interactive multi-model filtering adopts a hybrid state transition model of uniform speed model, uniform acceleration model and cooperative turning model. State prediction and updating are performed based on Kalman filtering, and the spatial coordinates, radial velocity and signal-to-noise ratio information of the spatiotemporal alignment point trace set are fused to obtain the optimized state prediction value of the appearing target. The process noise covariance matrix and the measurement noise covariance matrix are dynamically corrected based on the target echo intensity to obtain the adaptive filtering parameter set; the trajectory confirmation threshold and deletion time limit are set to obtain the trajectory status label set.

4. The data assimilation method according to claim 3, characterized in that, Dynamic target detection and tracking are performed based on the aforementioned detection state dataset to obtain the target motion trajectory data, and the method further includes: An association matrix is ​​established based on the predicted state value of the target and the spatial proximity and Doppler frequency consistency of the spatiotemporal alignment point trace set. The association probability between the spatiotemporal alignment point trace set and the first confirmed trajectory is determined based on the association matrix. The length of the first confirmed trajectory and the association probability are weighted and summed to obtain a trajectory confidence score; the first confirmed trajectory is then filtered based on the trajectory confidence score to obtain a second confirmed trajectory. The spatial coordinate sequence of the second confirmed trajectory is converted into a geocentric fixed coordinate system representation to obtain the target motion trajectory data.

5. The data assimilation method according to claim 4, characterized in that, Spatial distribution feature analysis is performed on the motion trajectory data of the detected targets to obtain a target group spatial distribution feature dataset, including: Based on the three-dimensional spatial coordinate sequence of the target motion trajectory data, the spatial distribution of the target points is modeled using a kernel density estimation algorithm to obtain a density distribution surface; The local density extreme value regions in the density distribution surface are identified according to the density peak clustering algorithm to obtain the distribution region parameters of the target group, wherein the distribution region parameters include peak density, average density and area enclosed by contour lines; The density distribution surface is classified into high-density and low-density regions based on the distribution region parameters of the target group. Noise filtering based on neighborhood connectivity is then performed on the low-density regions to obtain a set of spatial density distribution parameters. Based on the velocity vector sequence of the target motion trajectory data, a three-dimensional velocity vector field is constructed, and the velocity gradient tensor of the three-dimensional velocity vector field is determined according to the five-point center difference method. The velocity gradient tensor is convolved using the Sobel operator to obtain velocity field gradient features, which include the maximum gradient magnitude, gradient direction variance, and velocity field curvature.

6. The data assimilation method according to claim 5, characterized in that, An assimilation analysis model is constructed using the ensemble Kalman filter algorithm on the error-corrected spatial distribution characteristics data of the target group to obtain a data assimilation model, including: The background field data of the numerical weather prediction model is projected onto the spatiotemporal coordinate system of the detection and observation space using a bilinear interpolation algorithm to obtain a spatiotemporally aligned background field dataset, wherein the background field data includes wind field, temperature field and pressure field. The spatial distribution characteristics data of the target group after error correction are transformed to obtain the equivalent momentum flux field; the initial perturbation set is obtained by generating perturbations through the background field error covariance matrix based on singular value decomposition. The disturbance magnitude of the initial disturbance set is matched with the uncertainty level of the background field to obtain the initial state set of the numerical weather prediction model; random error disturbances are applied to the detection and observation field to obtain the observation set.

7. The data assimilation method according to claim 6, characterized in that, An assimilation analysis model is constructed using the ensemble Kalman filter algorithm on the error-corrected spatial distribution characteristic data of the target group to obtain a data assimilation model, which also includes: Numerical weather prediction model integration is performed on each member of the initial state set to predict the atmospheric state at future times, thereby obtaining a predicted state set; the predicted state set is mapped to the detection and observation space to determine the equivalent predicted observations, thereby obtaining a predicted observation set. Determine the covariance matrix of the predicted state set and the covariance matrix of the predicted observation set; determine the correction weights of the observed data on the state of the numerical weather prediction model based on covariance analysis, so as to obtain the Kalman gain matrix; The predicted state set is dynamically adjusted based on the Kalman gain matrix, and the constraint information of the detection and observation data is combined to obtain the analysis state set; the error covariance matrix is ​​updated based on the statistical characteristics of the analysis state set to obtain the data assimilation model.

8. A data assimilation apparatus, applicable to the data assimilation method according to any one of claims 1-7, characterized in that, include: A data acquisition module is used to acquire multi-beam scanning data according to the detection system to obtain a detection status dataset; Dynamic target detection and tracking are performed based on the aforementioned detection state dataset to obtain the target motion trajectory data; The feature analysis module is used to perform spatial distribution feature analysis on the motion trajectory data of the detected target to obtain a target group spatial distribution feature dataset, wherein the target group spatial distribution feature dataset includes the spatial density distribution parameters and velocity field gradient features of the target group; A propagation correction module is used to acquire real-time meteorological data, including atmospheric refractive index profile data and ionospheric disturbance parameters; and to model the electromagnetic propagation effect of the detection state dataset based on the real-time meteorological data to obtain a propagation path correction factor. An error compensation module is used to compensate for the observation error of the target group spatial distribution feature dataset according to the propagation path correction factor, so as to obtain the target group spatial distribution feature data after error correction. The data assimilation module is used to construct an assimilation analysis model for the error-corrected spatial distribution characteristic data of the target group using an ensemble Kalman filter algorithm, so as to obtain a data assimilation model; and to deploy the data assimilation model to the numerical weather prediction system.

9. A data assimilation device, characterized in that, include: Processor and memory; The memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory, causing the processor to perform a data assimilation method as described in any one of claims 1-7.

Images