An artificial intelligence-based real-time flight trajectory tracking system and method

Abstract

The present invention discloses a real-time flight ballistic tracking system and method based on artificial intelligence, which relates to the field of ballistic tracking technology. The present invention effectively solves the problems of insufficient real-time performance and limited accuracy existing in traditional methods by dynamically adjusting the time window length. When the target moves smoothly, the time window remains short, reducing redundant calculations and improving the tracking efficiency. When the target's motion changes violently, the time window automatically extends to effectively cover sufficient historical data to support high-precision prediction. In addition, a dynamic characteristic index is constructed by comprehensively considering the speed change rate and acceleration change trend of the target. Compared with the traditional fixed-time window method, this method has stronger robustness in the data interruption scenario. By automatically detecting data interruption and extending the time window to the maximum value, the historical data input required for trajectory completion is ensured, thereby improving the trajectory continuity and tracking reliability.

Description

Technical Field

[0001] The present invention relates to the technical field of trajectory tracking, and in particular to a real-time flight trajectory tracking system and method based on artificial intelligence. Background Art

[0002] Real-time flight trajectory tracking relies on a variety of sensors to provide high-frequency, continuous measurement data to achieve accurate positioning and dynamic updating of the target trajectory; however, in severe weather, signal obstruction or high-interference electromagnetic environments, the sensor's measurement capabilities will be limited, and the data stream may be interrupted, mismeasured or delayed; in addition, when the acceleration of the target changes dramatically during the flight phase, the measurement error is easily amplified, making it difficult for the tracking system to capture the target's trajectory changes in real time and accurately.

[0003] To solve the above problems, traditional real-time flight trajectory tracking methods usually use data smoothing, interpolation and extrapolation prediction within a fixed time window to make up for short-term data loss. These methods are based on the assumption that the flight trajectory has continuity and low dynamic changes in a short period of time, and use historical data to infer the current trajectory points. However, the fixed time window strategy has obvious shortcomings when dealing with complex trajectories of high-speed and nonlinear flight: when the target motion pattern changes suddenly or the measurement signal is interrupted for a long time, the prediction error will increase significantly. In addition, the fixed window size often needs to be statically set before the task starts, and lacks the ability to dynamically adapt to the real-time data status and change rate, further limiting its applicability in complex environments. Therefore, there is an urgent need for a real-time flight trajectory tracking system and method based on artificial intelligence to solve such problems. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] The present invention provides a real-time flight trajectory tracking system and method based on artificial intelligence to solve the problem that the fixed time window adopted by the traditional tracking method cannot adjust the calculation range according to the real-time data quality, and the accuracy and stability are insufficient when the measured data fluctuates greatly or the interruption time is prolonged.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a real-time flight trajectory tracking system based on artificial intelligence, which includes:

[0008] Ballistic data acquisition module, used to collect ballistic data from radar, optoelectronic measurement equipment, inertial navigation system and global satellite navigation GNSS system;

[0009] A real-time data integrity monitoring module, which is used to monitor the integrity of ballistic data and automatically mark data missing or abnormal points;

[0010] A dynamic time window management module, which adjusts the length of the time window in real time according to the dynamic characteristics of the target movement;

[0011] A trajectory state update and prediction module, which combines historical data and real-time ballistic data within the adjusted time window and uses a multi-modal prediction algorithm to update the current trajectory state of the target in real time;

[0012] A model calibration and error compensation module, which compares the updated trajectory state with the real-time ballistic data, calculates the trajectory deviation, and dynamically adjusts the parameters of the tracking model, including filter covariance and deep learning weights. When the sensor is interrupted for a short time, the prediction result is used to maintain the continuity of trajectory tracking;

[0013] A trajectory state output module, which based on the optimized model, real-time feedbacks the current position, speed and movement trend of the target and outputs the ballistic tracking result.

[0014] In a second aspect, the present invention provides a real-time flight ballistic tracking method based on artificial intelligence, including,

[0015] Step S1, real-time collect ballistic data from radar, optoelectronic measurement, inertial navigation sensors and satellite navigation data to form a multi-source heterogeneous time-series data stream, and perform data preprocessing, including noise filtering, format standardization and coordinate transformation; identify the spatio-temporal relationship of the target ballistic through trajectory association analysis to form a preliminary trajectory model; at the same time, perform integrity evaluation on the collected data stream and mark missing data;

[0016] Step S2, according to the ballistic data output in step S1, real-time evaluate the dynamic characteristics of the target movement, including the rate of change of speed and the trend of acceleration change, and define the length of the dynamic time window based on the dynamic characteristics;

[0017] Step S3, within the adjusted time window, combine historical data and real-time ballistic data characteristics, and use a multi-modal prediction algorithm to update the current state of the target in real time;

[0018] Step S4, compare the updated trajectory state in step S3 with the real-time ballistic data, calculate the trajectory deviation, and dynamically adjust the tracking model parameters in step S3;

[0019] Step S5, based on the adjusted model, dynamically update the trajectory state.

[0020] As a preferred solution of the real-time flight trajectory tracking method based on artificial intelligence according to the present invention, wherein: the step of collecting trajectory data in real time from radar, optoelectronic measurement, inertial navigation sensors and satellite navigation data to form a time-series data stream of multi-source heterogeneity is as follows.

[0021] Let the collected data stream be D j (t) = {x j，t , y j，t , z j，t , v j，t , a j，t}, j = 1, 2,..., M, where D j (t) is the set of trajectory measurement data provided by sensor j at time t, x j，t , y j，t , z j，t are the three-dimensional space coordinates provided by sensor j at time t, v j，t is the speed information provided by sensor j at time t, a j，t is the acceleration information provided by sensor j at time t, and M is the total number of sensors.

[0022] The fusion formula is:

[0023]

[0024] where D(t) is the time-series data stream after multi-source fusion, and Φ is a function for preprocessing the data of sensor j;

[0025] Denoise and standardize the fused data D(t) to generate data in a unified format

[0026] The denoising formula is: where is the data after denoising, n(t) = [n x，t , n y，t , n z，t , n v，t , n a，t is the noise vector

[0027] The standardization formula is:

[0028]

[0029] where is the data after standardization, μ = [μ x , μ y , μ z , μ v , μ a is the mean vector of the data, and σ = [σ x , σ y, σ z , σ v , σ a is the standard deviation vector of the data.

[0030] As a preferred solution of the real-time flight trajectory tracking method based on artificial intelligence according to the present invention, wherein: the step of identifying the spatio-temporal relationship of the target trajectory through trajectory correlation analysis and forming a preliminary trajectory model is as follows,

[0031] Use multi-dimensional Kalman filtering to perform trajectory correlation on the fusion data and predict and generate a unified trajectory of the target.

[0032] The state prediction formula is:

[0033] Wherein, is the predicted target state vector, F is the state transition matrix, G is the control input matrix, and u t is the external control input;

[0034] The state update formula is:

[0035]

[0036] K t = P t|t-1 H T (HP t|t-1 H T + R) -1 ,

[0037] Wherein, is the updated target state vector, z t = [z x，t , z y，t , z z，t , z v，t , z a，t is the fused observation vector, H is the observation matrix, K t is the Kalman gain matrix, P t|t-1 is the predicted error covariance matrix, and R is the observation noise covariance matrix;

[0038] Use the output of the trajectory correlation analysis to construct a preliminary trajectory model of the target, and the model formula is:

[0039] Wherein, is the preliminary trajectory model at time t, is the updated target position, is the updated target speed, is the updated target acceleration;

[0040] The steps of evaluating the integrity of the collected data stream and marking the missing data are as follows:

[0041] Define the integrity index of the time-series data as κ(t):

[0042]

[0043] where κ(t) is the data integrity index at time t, M is the total number of sensors, and δ j (t) is the integrity flag of sensor j at time t;

[0044] If κ(t) < κ min , then mark the data at time t as missing data, marked as: where is the set of time points of the missing data, and κ min is the integrity threshold.

[0045] As a preferred solution of the real-time flight ballistic tracking method based on artificial intelligence according to the present invention, wherein: in step S2, adjust the time window range according to the change of the ballistic data tracked in real time:

[0046] When the target motion is stable, the time window remains short;

[0047] When a drastic change in motion or data interruption is detected, the time window automatically extends.

[0048] As a preferred solution of the real-time flight ballistic tracking method based on artificial intelligence according to the present invention, wherein: the steps of evaluating the dynamic characteristics of the target motion in real time according to the ballistic data output in step S1, including the rate of change of speed and the trend of acceleration change, and defining the length of the dynamic time window based on the dynamic characteristics are as follows

[0049] Based on the ballistic data output in step S1 Evaluate the rate of change of speed and the trend of acceleration change of the target, and define the rate of change of speed of the target as Δv:

[0050]

[0051] where Δv(t) is the rate of change of speed at time t, and v x (t), v y (t), v z (t) are the speed components of the target in the x, y, and z directions, sourced from the speed information in, t - Δt is the time of the previous moment, and Δt is the time interval;

[0052] Define the trend of acceleration change as Δa:

[0053]

[0054] Among them, Δa(t) is the acceleration change trend at time t, a x (t), a y (t), a z (t) are the acceleration components of the target in the x, y, and z directions, sourced from the acceleration information, t - Δt is the time of the previous moment, and Δt is the time interval;

[0055] Combining the velocity change rate and the acceleration change trend, the dynamic characteristic index Θ is defined as:

[0056] Θ(t) = w v ·Δv(t) + w a ·Δa(t),

[0057] Among them, Θ(t) is the dynamic characteristic index of the target at time t, w v , w a are the weight factors;

[0058] According to the dynamic characteristic index Θ(t), the length of the time window is adjusted in real time. The time window length formula is:

[0059] If Θ(t) ≤ Θ low , then T w (t) = T min ,

[0060] If Θ(t) ≥ Θ high , then T w (t) = T max ,

[0061] If Θ low <Θ(t)<Θ high , then

[0062] Among them, T w (t) is the dynamic time window length at time t, T min is the shortest time window length, T max is the longest time window length,

[0063] Θ low , Θ high , are the thresholds of the dynamic characteristic index, corresponding to the boundaries of smooth and drastic changes in motion respectively;

[0064] If a data interruption is detected, the time window is automatically extended to T max .

[0065] As a preferred embodiment of the real-time flight ballistic tracking method based on artificial intelligence according to the present invention, wherein: the current state of the target is updated as follows,

[0066] For a linear trajectory, the Kalman filter is used to update the target position and velocity,

[0067] For a non-linear trajectory, the LSTM deep learning method is introduced to update the target position and velocity,

[0068] Meanwhile, for the missing marked data, a multi-model predictor dynamically trained by the variational Bayesian algorithm, the ARIMA+LSTM hybrid model, is used to complete the missing data.

[0069] As a preferred embodiment of the real-time flight ballistic tracking method based on artificial intelligence according to the present invention, wherein: in the adjusted time window, combining the historical data and the characteristics of the real-time ballistic data, the steps of using the multi-modal prediction algorithm to update the current state of the target are as follows,

[0070] The length T of the adjusted time window w (t) provides the input data range of the trajectory data from t - T w (t) to t Select the corresponding prediction method according to the trajectory characteristics:

[0071] For a linear trajectory, the Kalman filter is used to update the target state, and the prediction formula is:

[0072] x k|k-1 = F k x k-1|k-1 + G k u k ,

[0073]

[0074] Update formula:

[0075]

[0076] x k|k = x k|k-1 + K k (z k - H k x k|k-1 ),

[0077] P k|k = (I - K k H k )P k|k-1 ,

[0078] where, x k|k-1 = [x k|k-1, y k|k-1 , Z k|k-1 , v x，k|k-1 , v y，k|k-1 , v z，k|k-1 is the target state prediction at time k, F k is the state transition matrix, G k is the control input matrix, u k is the control input, P k|k-1 is the state prediction error covariance matrix, Q k is the process noise covariance matrix, K k is the Kalman gain matrix, z k is the observation at time k, H k is the observation matrix, R k is the observation noise covariance matrix, I is the identity matrix;

[0079] Use LSTM to update the target state for non-linear trajectories. Use the historical data within the adjusted time window as the LSTM input to update the target state. The input time series is The structure of LSTM is:

[0080] h t = f(W h x t + U h h t-1 + b h ),

[0081] c t = g(W c x t + U c h t-1 + b c ),

[0082]

[0083] where, x t = [x t , y t , z t , v x，t , v y，t , v z，t , a x，t , a y，t , a z，t is the input state vector at time t, h t , c t are the hidden state and memory state at time t, is the predicted target state, W h , W c , W o , U h , Uc is the weight matrix, b h , b c , b o is the bias vector, and f, g, o are activation functions.

[0084] As a preferred solution of the real-time flight trajectory tracking method based on artificial intelligence according to the present invention, wherein: in step S3, the missing marked data is completed, and the steps are as follows

[0085] The missing data is completed by a multi-model predictor ARIMA+LSTM dynamically trained by the variational Bayesian algorithm:

[0086] The ARIMA model is used to perform differencing and regression modeling on the time series to capture the linear trend. The ARIMA formula is:

[0087]

[0088] wherein, is the predicted value at time t, φ i is the autoregressive parameter i = 1, 2,..., p, θ j is the moving average parameter j = 1, 2,..., q, ∈ t is the error term at time t;

[0089] The LSTM model is used to supplement and capture the non-linear pattern. The LSTM structure is the same as the LSTM structure of the non-linear trajectory update target;

[0090] The fusion weights of ARIMA and LSTM are dynamically adjusted by the variational Bayesian method. The variational Bayesian formula is:

[0091]

[0092] wherein, is the posterior distribution of the weight w, is the data 's likelihood function, P(w) is the prior distribution of the weight, is the marginal probability of the data,

[0093] The completed result after fusion is:

[0094] α + β = 1, wherein α and β are the fusion weights of ARIMA and LSTM.

[0095] As a preferred solution of the real-time flight trajectory tracking method based on artificial intelligence according to the present invention, wherein: the step of comparing the updated trajectory state in step S3 with the real-time ballistic data, calculating the trajectory deviation, and dynamically adjusting the tracking model parameters in step S3 is as follows

[0096] Record the updated trajectory status output from step S3 as Include the predicted target status:

[0097]

[0098] Record the ballistic data collected in real time from the sensor as

[0099]

[0100] Define the trajectory deviation as the difference between the predicted value and the observed value, and the calculation formula is:

[0101]

[0102] Wherein, is the trajectory deviation set, Δx t , Δy t , Δz t is the position deviation, Δv x，t , Δv y，t , Δv z，t is the velocity deviation;

[0103] Calculate the weighted norm of the trajectory deviation, and the calculation formula is:

[0104]

[0105] Wherein, is the weighted norm, is the weighted factor of position and velocity;

[0106] For the tracking of the linear trajectory, adjust the state noise covariance matrix Q k and the observation noise covariance matrix R k ,

[0107] State noise covariance adjustment formula:

[0108]

[0109] Observation noise covariance adjustment formula:

[0110]

[0111] Wherein, Q k is the state noise covariance matrix, R k is the observation noise covariance matrix, α Q , α R is the adjustment factor, and I is the identity matrix;

[0112] Tracking of non - linear trajectories, adjusting the learning rate η of LSTM based on trajectory deviation t And the weight update rule, dynamic learning rate adjustment formula:

[0113] Where η t Is the learning rate at time t, η init Is the initial learning rate, β is the adjustment factor,

[0114] Weight update formula:

[0115]

[0116] Where W t Is the weight matrix at time t, Is the loss function at time t, Is the gradient of the loss function.

[0117] The beneficial effects of the present invention are as follows: By dynamically adjusting the time - window length, the present invention effectively solves the problems of insufficient real - time performance and limited accuracy existing in traditional methods. When the target motion is stable, the time - window remains short, reducing redundant calculations and improving the tracking efficiency; while when the target motion changes drastically, the time - window automatically extends, effectively covering enough historical data to support high - precision prediction, thus achieving a balance between real - time performance and accuracy; in addition, the dynamic characteristic index constructed by integrating the speed change rate and acceleration change trend of the target can comprehensively reflect the motion state of the target, enabling the system to flexibly respond to the motion characteristics in different scenarios; compared with the traditional fixed - time - window method, this method has stronger robustness in the data - interruption scenario. By automatically detecting data interruption and extending the time - window to the maximum value, it ensures the input of historical data required for trajectory completion, thereby improving the trajectory continuity and tracking reliability; at the same time, introducing the dynamic time - window mechanism effectively avoids unnecessary resource waste, reducing the computational burden during stable motion and making full use of motion information to support prediction in high - dynamic scenarios, significantly improving the intelligent level and adaptability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0118] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0119] Figure 1 It is a schematic structural diagram of the real - time flight ballistic tracking system in Embodiment 1. DETAILED DESCRIPTION OF THE INVENTION

[0120] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification.

[0121] In the following description, numerous specific details are set forth to facilitate a thorough understanding of the present invention. However, the present invention may be practiced in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0122] Secondly, as used herein, "an embodiment" or "embodiments" refers to specific features, structures, or characteristics that may be included in at least one implementation of the present invention. The appearances of "in an embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that are mutually exclusive of other embodiments.

[0123] Example 1, referring to Figure 1 This example provides a real-time flight ballistic tracking system based on artificial intelligence:

[0124] A ballistic data acquisition module for collecting ballistic data from radar, optoelectronic measurement equipment, inertial navigation systems, and the global satellite navigation GNSS system;

[0125] A real-time data integrity monitoring module for monitoring the integrity of ballistic data and automatically marking data missing or abnormal points;

[0126] A dynamic time window management module that adjusts the length of the time window in real time according to the dynamic characteristics of the target's movement;

[0127] A trajectory state update and prediction module that combines historical data and real-time ballistic data within the adjusted time window and uses a multi-modal prediction algorithm to update the current trajectory state of the target in real time;

[0128] A model calibration and error compensation module that compares the updated trajectory state with real-time ballistic data, calculates the trajectory deviation, and dynamically adjusts the parameters of the tracking model, including filter covariance and deep learning weights. When the sensor is interrupted for a short time, the prediction result is used to maintain the continuity of trajectory tracking;

[0129] A trajectory state output module that based on the optimized model, real-time feedbacks the current position, speed, and movement trend of the target and outputs the ballistic tracking result.

[0130] This example also provides a real-time flight ballistic tracking method based on artificial intelligence, including:

[0131] Step S1: Real-time collect ballistic data from radar, optoelectronic measurement, inertial navigation sensors, and satellite navigation data to form a multi-source heterogeneous time-series data stream, and perform data preprocessing, including noise filtering, format standardization, and coordinate transformation; identify the spatio-temporal relationship of the target ballistic trajectory through trajectory correlation analysis to form a preliminary trajectory model; at the same time, evaluate the integrity of the collected data stream and mark the missing data;

[0132] The step of real-time collecting ballistic data from radar, optoelectronic measurement, inertial navigation sensors, and satellite navigation data to form a multi-source heterogeneous time-series data stream is as follows:

[0133] Let the collected data stream be D j (t) = {x j，t , y j，t , z j，t , v j，t , a j，t}, j = 1, 2,..., M, where D j (t) is the set of ballistic measurement data provided by sensor j at time t, x j，t , y j，t , z j，t are the three-dimensional space coordinates provided by sensor j at time t, v j，t is the velocity information provided by sensor j at time t, a j，t is the acceleration information provided by sensor j at time t, and M is the total number of sensors.

[0134] The fusion formula is:

[0135]

[0136] where D(t) is the multi-source fused time-series data stream, and Φ is the function used to preprocess the data of sensor j;

[0137] Denoise and standardize the fused data D(t) to generate data in a unified format

[0138] The denoising formula is: where is the denoised data, n(t) = [n x，t , n y，t , n z，t , n v，t , n a，t is the noise vector

[0139] The standardization formula is:

[0140]

[0141] where The data after standardization, μ = [μ x , μ y , μ z , μ v , μ a is the mean vector of the data, and σ = [σ x , σ y , σ z , σ v , σ a is the standard deviation vector of the data;

[0142] The steps to identify the spatio-temporal relationship of the target trajectory through trajectory association analysis and form a preliminary trajectory model are as follows:

[0143] Use multi-dimensional Kalman filtering for trajectory association of the fused data to predict and generate a unified trajectory of the target.

[0144] The state prediction formula is:

[0145] Among them, is the predicted target state vector, F is the state transition matrix, G is the control input matrix, and u t is the external control input;

[0146] The state update formula is:

[0147]

[0148] K t = P t|t-1 H T (HP t|t-1 H T + R) -1 ,

[0149] Among them, is the updated target state vector, z t = [z x，t , z y，t , z z，t , z v，t , z a，t is the fused observation vector, H is the observation matrix, K t is the Kalman gain matrix, P t|t-1 is the predicted error covariance matrix, and R is the observation noise covariance matrix;

[0150] Use the output of trajectory association analysis to construct a preliminary trajectory model of the target. The model formula is:

[0151] Among them, is the preliminary trajectory model at time t, is the updated target position, is the updated target speed, is the updated target acceleration;

[0152] The steps for evaluating the integrity of the collected data stream and marking missing data are as follows,

[0153] Define the integrity index of the time series data as κ(t):

[0154]

[0155] where κ(t) is the data integrity index at time t, M is the total number of sensors, and δ j (t) is the integrity flag of sensor j at time t;

[0156] If κ(t) < κ min , then mark the data at time t as missing data, marked as: where, is the set of time points of missing data, and κ min is the integrity threshold,

[0157] Specifically, in this step, after multi-sensor data fusion, denoising and standardization are performed uniformly, and then a preliminary trajectory model is constructed through multi-dimensional Kalman filtering, and the integrity index is calculated to mark missing data points.

[0158] Step S2: According to the ballistic data output in step S1, evaluate the dynamic characteristics of the target movement in real time, including the rate of change of speed and the trend of acceleration change, and define the dynamic time window length based on the dynamic characteristics;

[0159] In step S2, adjust the time window range according to the change of the ballistic data tracked in real time:

[0160] When the target movement is stable, the time window remains short;

[0161] When a drastic change in movement or data interruption is detected, the time window automatically extends;

[0162] The steps for evaluating the dynamic characteristics of the target movement in real time, including the rate of change of speed and the trend of acceleration change, and defining the dynamic time window length based on the dynamic characteristics according to the ballistic data output in step S1 are as follows,

[0163] Based on the ballistic data output in step S1 Evaluate the rate of change of speed and the trend of acceleration change of the target, and define the rate of change of speed of the target as Δv:

[0164]

[0165] where Δv(t) is the rate of change of velocity at time t, v x (t), v y (t), vz(t) are the velocity components of the target in the x, y, and z directions, sourced from the velocity information therein, t - Δt is the time at the previous moment, and Δt is the time interval;

[0166] The acceleration change trend is defined as Δa:

[0167]

[0168] where Δa(t) is the acceleration change trend at time t, a x (t), a y (t), a z (t) are the acceleration components of the target in the x, y, and z directions, sourced from the acceleration information therein, t - Δt is the time at the previous moment, and Δt is the time interval;

[0169] Combining the rate of change of velocity and the acceleration change trend, the dynamic characteristic index is defined as Θ:

[0170] Θ(t) = w v ·Δv(t) + w a ·Δa(t),

[0171] where Θ(t) is the dynamic characteristic index of the target at time t, w v , w a are the weight factors;

[0172] According to the dynamic characteristic index Θ(t), the length of the time window is adjusted in real time, and the time window length formula is:

[0173] If Θ(t) ≤ Θ low , then T w (t) = T min ,

[0174] If Θ(t) ≥ Θ high , then T w (t) = T max ,

[0175] If Θ low <Θ(t)<Θ high , then

[0176] where T w (t) is the length of the dynamic time window at time t, T min is the shortest time window length, and T max is the longest time window length,

[0177] Θ low ,Θ high , which are the thresholds of the dynamic characteristic indexes, corresponding to the boundaries of stable and drastic changes in motion respectively;

[0178] If a data interruption is detected, automatically extend the time window to T max ;

[0179] Specifically, through the above steps, the adjustment of the time window can dynamically adapt to the motion state of the target.

[0180] Step S3, within the adjusted time window, combine the historical data and the characteristics of the real-time ballistic data, and use a multimodal prediction algorithm to update the current state of the target in real time;

[0181] The update method of the current state of the target is

[0182] For a linear trajectory, use the Kalman filter to update the target position and speed,

[0183] For a non-linear trajectory, introduce the LSTM deep learning method to update the target position and speed,

[0184] At the same time, for the missing marked data, use a multi-model predictor dynamically trained by the variational Bayesian algorithm, the ARIMA+LSTM hybrid model, to complete the missing data;

[0185] The steps of combining the historical data and the characteristics of the real-time ballistic data and using a multimodal prediction algorithm to update the current state of the target within the adjusted time window are

[0186] The length T of the adjusted time window w (t) provides the input data range from t-T w (t) to the trajectory data of t Select the corresponding prediction method according to the trajectory characteristics:

[0187] Use the Kalman filter to update the target state for a linear trajectory, and the prediction formula is:

[0188] x k|k-1 = F k x k-1|k-1 + G k u k ,

[0189]

[0190] Update formula:

[0191]

[0192] x k|k = xk|k-1 +K k (z k -H k x k|k-1 ),

[0193] P k|k =(I-K k H k )P k|k-1 ,

[0194] where x k|k-1 =[x k|k-1 ,y k|k-1 ,Z k|k-1 ,v x，k|k-1 ,v y，k|k-1 ,v z，k|k-1 is the predicted target state at time k, F k is the state transition matrix, G k is the control input matrix, u k is the control input, P k|k-1 is the state prediction error covariance matrix, Q k is the process noise covariance matrix, K k is the Kalman gain matrix, z k is the observation at time k, H k is the observation matrix, R k is the observation noise covariance matrix, and I is the identity matrix;

[0195] The LSTM is used to update the target state for the non-linear trajectory. The historical data within the adjusted time window is used as the LSTM input to update the target state. The input time series is The structure of the LSTM is:

[0196] h t =f(W h x t +U h h t-1 +b h ),

[0197] c t =g(W c x t +U c h t-1 +b c ),

[0198]

[0199] where x t =[x t ,y t ,z t ,vx，t , v y，t , v z，t , a x，t , a y，t , a z，t is the input state vector at time t, h t , c t is the hidden state and memory state at time t, is the predicted target state, W h , W c , W o , U h , U c are weight matrices, b h , b c , b o are bias vectors, f, g, o are activation functions;

[0200] In step S3, the marked missing data is completed. The steps are as follows:

[0201] The missing data is completed by a multi-model predictor ARIMA+LSTM dynamically trained by the variational Bayesian algorithm:

[0202] The ARIMA model is used to perform differencing and regression modeling on the time series to capture the linear trend. The ARIMA formula:

[0203]

[0204] Among them, is the predicted value at time t, φ i are autoregressive parameters i = 1, 2,..., p, θ j are moving average parameters j = 1, 2,..., q, ∈ t is the error term at time t;

[0205] The LSTM model is used to supplement and capture the non-linear pattern. The LSTM structure is the same as the LSTM structure of the non-linear trajectory update target;

[0206] The fusion weights of ARIMA and LSTM are dynamically adjusted by the variational Bayesian method. The variational Bayesian formula is:

[0207]

[0208] Among them, is the posterior distribution of the weight w, is the data 's likelihood function, P(w) is the prior distribution of the weight, is the marginal probability of the data,

[0209] The completed result after fusion is:

[0210] α + β = 1, where α and β are the fusion weights of ARIMA and LSTM;

[0211] Specifically, through the above steps, multi-modal prediction and data completion are performed.

[0212] Step S4: Compare the updated trajectory state in step S3 with the real-time ballistic data, calculate the trajectory deviation, and dynamically adjust the tracking model parameters in step S3;

[0213] The steps of comparing the updated trajectory state in step S3 with the real-time ballistic data, calculating the trajectory deviation, and dynamically adjusting the tracking model parameters in step S3 are as follows:

[0214] Denote the updated trajectory state output by step S3 as Including the predicted target state:

[0215]

[0216] Denote the ballistic data collected in real time from the sensor as

[0217]

[0218] Define the trajectory deviation as the difference between the predicted value and the observed value, and the calculation formula is:

[0219]

[0220] where is the trajectory deviation set, Δx t , Δy t , Δz t are the position deviations, and Δv x，t , Δv y，t , Δv z，t are the velocity deviations;

[0221] Calculate the weighted norm of the trajectory deviation, and the calculation formula is:

[0222]

[0223] where is the weighted norm, are the weighted factors of position and velocity;

[0224] For the tracking of the linear trajectory, adjust the state noise covariance matrix Q k and the observation noise covariance matrix Rk ,

[0225] State noise covariance adjustment formula:

[0226]

[0227] Observation noise covariance adjustment formula:

[0228]

[0229] Where, Q k is the state noise covariance matrix, R k is the observation noise covariance matrix, α Q , α R are adjustment factors, and I is the identity matrix;

[0230] For the tracking of non - linear trajectories, based on the trajectory deviation, adjust the learning rate η of LSTM t and the weight update rule, dynamic learning rate adjustment formula:

[0231] Where, η t is the learning rate at time t, η init is the initial learning rate, and β is the adjustment factor.

[0232] Weight update formula:

[0233]

[0234] Where, W t is the weight matrix at time t, is the loss function at time t, is the gradient of the loss function;

[0235] Specifically, this step compares the updated trajectory state with the real - time ballistic data, calculates the trajectory deviation, and dynamically adjusts the tracking model parameters in step S3.

[0236] Step S5, based on the trajectory state dynamically updated by the adjusted model.

[0237] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A real-time flight trajectory tracking system based on artificial intelligence, characterized by: include, Ballistic data acquisition module, used to collect ballistic data from radar, optoelectronic measurement equipment, inertial navigation system and global satellite navigation GNSS system; Real-time data integrity monitoring module, used to monitor the integrity of ballistic data and automatically mark missing or abnormal data points; Dynamic time window management module, which adjusts the length of the time window in real time according to the dynamic characteristics of the target movement; In the dynamic time window management module, the time window range is adjusted by real-time tracking of trajectory data changes: When the target motion is stable, the time window is kept short; When a dramatic change in motion or data interruption is detected, the time window automatically extends; In the dynamic time window management module, according to the trajectory data collected by the trajectory data collection module, the dynamic characteristics of the target motion, including the velocity change rate and the acceleration change trend, are evaluated in real time, and the dynamic time window length is defined based on the dynamic characteristics, the steps are as follows; Based on ballistic data Evaluate the velocity change rate and acceleration change trend of the target, and define the velocity change rate of the target as Δv: Where Δv(t) is the velocity change rate at time t, v x (t),v y (t),v z (t) is the velocity component of the target in the x, y, and z directions, which comes from The speed information in, t-Δt is the time of the previous moment, Δt is the time interval; The acceleration change trend is defined as Δa: Among them, Δa(t) is the acceleration change trend at time t, a x (t),a y (t),a z (t) is the acceleration component of the target in the x, y, and z directions, which comes from The acceleration information in , t-Δt is the time of the previous moment, Δt is the time interval; Based on the velocity change rate and acceleration change trend, the dynamic characteristic index is defined as Θ: Θ(t)=w v ·Δv(t)+w a ·Δa(t), Among them, Θ(t) is the dynamic characteristic index of the target at time t, w v ,w a is the weight factor; According to the dynamic characteristic index Θ(t), the length of the time window is adjusted in real time. The formula for the length of the time window is: If Θ(t)≤Θ low , then T w (t) = T min , If Θ(t)≥Θ high , then T w (t) = T max , If θ low <Θ(t)<Θ high ,but Among them, T w (t) is the length of the dynamic time window at time t, T min is the shortest time window length, T max is the maximum time window length, Θ low ,Θ high , are the thresholds of the dynamic characteristic index, corresponding to the limits of stable and drastic changes in motion respectively; If a data interruption is detected, the time window is automatically extended to T max ; The trajectory status update prediction module combines the historical data and real-time trajectory data within the adjusted time window and uses a multi-modal prediction algorithm to update the current trajectory status of the target in real time; The model correction and error compensation module compares the updated trajectory state with the real-time trajectory data, calculates the trajectory deviation, and dynamically adjusts the parameters of the tracking model, including the filter covariance and deep learning weights; The trajectory status output module provides real-time feedback on the target’s current position, speed, and motion trend based on the optimized model, and outputs the trajectory tracking results.

2. A real-time flight trajectory tracking method based on artificial intelligence, based on the real-time flight trajectory tracking system based on artificial intelligence according to claim 1, characterized in that: include: Step S1, collecting trajectory data in real time from radar, optoelectronic measurement, inertial navigation sensor and satellite navigation data to form a multi-source heterogeneous time series data stream, and performing data preprocessing, including noise filtering, format standardization and coordinate conversion; Through trajectory correlation analysis, the temporal and spatial relationship of the target trajectory is identified to form a preliminary trajectory model; at the same time, the integrity of the collected data stream is evaluated and missing data is marked; Step S2, based on the trajectory data outputted in step S1, real-time evaluation of the dynamic characteristics of the target motion, including the velocity change rate and the acceleration change trend, and definition of the dynamic time window length based on the dynamic characteristics; Step S3, within the adjusted time window, combining historical data and real-time trajectory data characteristics, using a multi-modal prediction algorithm to update the current state of the target in real time; Step S4, comparing the trajectory state updated in step S3 with the real-time trajectory data, calculating the trajectory deviation, and dynamically adjusting the tracking model parameters in step S3; Step S5, dynamically updating the trajectory state based on the adjusted model.

3. A real-time flight trajectory tracking method based on artificial intelligence as claimed in claim 2, characterized in that: The step of collecting trajectory data from radar, photoelectric measurement, inertial navigation sensor and satellite navigation data in real time to form a multi-source heterogeneous time series data stream is as follows: Assume that the collected data stream is D j (t) = {x j,t ,y j,t ,z j,t ,v j,t ,a j,t },j=1,2,…,M, where D j (t) is the ballistic measurement data set provided by sensor j at time t, x j,t ,y j,t ,z j,t is the three-dimensional spatial coordinate provided by sensor j at time t, v j,t is the speed information provided by sensor j at time t, a j,t is the acceleration information provided by sensor j at time t, M is the total number of sensors, The fusion formula is: Where D(t) is the time series data stream after multi-source fusion, Φ is the function used to preprocess the data of sensor j; De-noise and standardize the fused data D(t) to generate data in a unified format The denoising formula is: in, is the denoised data, n(t)=[n x,t ,n y,t ,n z,t ,n v,t ,n a,t ] is the noise vector The normalization formula is: in, is the standardized data, μ=[μ x ,μ y ,μ z ,μ v ,μ a ] is the mean vector of the data, σ=[σ x ,σ y ,σ z ,σ v ,σ a ] is the standard deviation vector of the data.

4. A real-time flight trajectory tracking method based on artificial intelligence as claimed in claim 3, characterized in that: The steps of identifying the spatiotemporal relationship of the target trajectory through trajectory association analysis and forming a preliminary trajectory model are: Use multidimensional Kalman filtering to correlate the fusion data and predict the unified trajectory of the target. The state prediction formula is: in, is the predicted target state vector, F is the state transfer matrix, G is the control input matrix, u t It is the external control input; The state update formula is: K t =P t|t-1 H T (HP t|t-1 H T +R) -1 , in, is the updated target state vector, z t =[z x,t ,z y,t ,z z,t ,z v,t ,z a,t ] is the fused observation vector, H is the observation matrix, K t is the Kalman gain matrix, P t|t-1 is the prediction error covariance matrix, R is the observation noise covariance matrix; Using the output of trajectory association analysis, a preliminary trajectory model of the target is constructed. The model formula is: in, is the preliminary trajectory model at time t, is the updated target position, is the updated target speed, is the updated target acceleration; The steps of performing integrity assessment on the collected data stream and marking missing data are as follows: The integrity index of time series data is defined as κ(t): Where κ(t) is the data integrity index at time t, M is the total number of sensors, and δ j (t) is the integrity flag of sensor j at time t; If κ(t)<κ min , then the data at time t is marked as missing data, marked as: in, is the set of time points with missing data, κ min is the completeness threshold.

5. A real-time flight trajectory tracking method based on artificial intelligence as claimed in claim 4, characterized in that: The current state of the target is updated as follows: For linear trajectories, the Kalman filter is used to update the target position and velocity. For nonlinear trajectories, the LSTM deep learning method is introduced to update the target position and speed. At the same time, for the missing labeled data, a multi-model predictor dynamically trained by a variational Bayesian algorithm, an ARIMA+LSTM hybrid model, is used to complete the missing data.

6. A real-time flight trajectory tracking method based on artificial intelligence as claimed in claim 5, characterized in that: The step of updating the current state of the target in real time by using a multi-modal prediction algorithm in the adjusted time window in combination with the historical data and the real-time trajectory data characteristics is as follows: Adjusted time window length T w (t) Provide input data range tT w (t) to t trajectory data Select the corresponding prediction method according to the trajectory characteristics: The Kalman filter is used to update the target state for the linear trajectory, and the prediction formula is: x k|k-1 =F k x k-1|k-1 +G k u k , Update formula: x k|k =x k|k-1 +K k ( z k-H k x k|k-1 ), P k|k =(I-K k H k )P k|k-1 , Among them, x k|k-1 =[x k|k-1 ,y k|k-1 ,z k|k-1 ,v k,k|k-1 ,v y,k|k-1 ,v z,k|k-1 ] is the target state prediction at time k, F k is the state transfer matrix, G k is the control input matrix, u k is the control input, P k|k-1 is the state prediction error covariance matrix, Q k is the process noise covariance matrix, K k is the Kalman gain matrix, z k is the observed value at time k, H k is the observation matrix, R k is the observation noise covariance matrix, I is the identity matrix; Use LSTM to update the target state for nonlinear trajectories, use the historical data within the adjustment time window as LSTM input, update the target state, and the input time series is The structure of LSTM is: h t =f(W h x t +U h h t-1 +b h ), c t =g(W c x t +U c h t-1 +b c ), Among them, x t =[x t ,y t ,z t ,v x,t ,v y,t ,v z,t ,a x,t ,a y,t ,a z,t ] is the input state vector at time t, h t ,c t are the hidden state and memory state at time t, is the predicted target state, W h ,W c ,W o ,U h ,U c is the weight matrix, b h ,b c ,b o is the bias vector, and f, g, o are activation functions.

7. A real-time flight trajectory tracking method based on artificial intelligence as claimed in claim 6, characterized in that: In step S3, the missing data of the tag is completed, and the steps are as follows: The missing data is completed by the multi-model predictor ARIMA+LSTM dynamically trained by the variational Bayes algorithm: The ARIMA model is used to perform difference and regression modeling on the time series to capture the linear trend. The ARIMA formula is: in, is the predicted value at time t, φ i is the autoregressive parameter i=1,2,…,p,θ j is the moving average parameter j=1,2,…,q,∈ t is the error term at time t; The LSTM model is used to supplement the capture of nonlinear patterns. The LSTM structure is the same as the LSTM structure of the nonlinear trajectory update target. The fusion weights of ARIMA and LSTM are dynamically adjusted through the variational Bayesian method. The variational Bayesian formula is: in, is the posterior distribution of weight w, For data The likelihood function of , P(w) is the prior distribution of weights, is the marginal probability of the data, The fused completion result is: Among them, α and β are the fusion weights of ARIMA and LSTM.

8. A real-time flight trajectory tracking method based on artificial intelligence as claimed in claim 7, characterized in that: The step of comparing the trajectory state updated in step S3 with the real-time trajectory data, calculating the trajectory deviation, and dynamically adjusting the tracking model parameters in step S3 is as follows: The updated trajectory state output in step S3 is recorded as The target state containing the prediction: The trajectory data collected from the sensor in real time is recorded as The trajectory deviation is defined as the difference between the predicted value and the observed value, and the calculation formula is: in, is the trajectory deviation set, Δx t ,Δy t ,Δz t is the position deviation, Δv x,t ,Δv y,t ,Δv z,t is the speed deviation; Calculate the weighted norm of the trajectory deviation, the calculation formula is: in, is the weighted norm, w x ,w y ,w z , is the weighting factor of position and velocity; For linear trajectory tracking, adjust the state noise covariance matrix Q of the Kalman filter k and the observation noise covariance matrix R k , State noise covariance adjustment formula: Observation noise covariance adjustment formula: Among them, Q k is the state noise covariance matrix, R k is the observation noise covariance matrix, α Q ,α R is the adjustment factor, I is the unit matrix; Tracking of nonlinear trajectories, adjusting the learning rate η of LSTM based on trajectory deviation t And weight update rules, dynamic learning rate adjustment formula: Among them, η t is the learning rate at time t, η init is the initial learning rate, β is the adjustment factor, Weight update formula: Among them, W t is the weight matrix at time t, is the loss function at time t, is the gradient of the loss function.

Images