Method for predicting motion state of intermittent lost target by aircraft in complex environment

Abstract

The invention relates to a method for predicting the motion state of an intermittent lost target by an aircraft in a complex environment. Comprising the following steps: acquiring a target motion state signal; decomposing the signal into a plurality of intrinsic mode functions (IMF); extracting instantaneous frequency characteristics of each IMF to obtain target historical motion state sequence characteristics, calculating a correlation coefficient and an energy retention rate of a reconstructed signal and an original signal, and analyzing spectral characteristics of each IMF component; training the converted target historical motion state sequence features, outputting a prediction result, and generating a prediction error; a dual-threshold iterative prediction mechanism is set, the size of a prediction window is dynamically adjusted, and the maximum number of iterations is set; and calculating a prediction trajectory confidence interval, dynamically adjusting parameters based on historical prediction errors input circularly, and introducing smooth constraints at the same time. The method combines signal decomposition and model prediction to improve prediction accuracy, and is suitable for high-precision trajectory prediction of a complex dynamic system.

Description

Technical Field

[0001] The present invention belongs to the field of target motion state prediction, and particularly relates to a method for predicting the motion state of an intermittently lost target by an aircraft in a complex environment. Background Art

[0002] In recent years, with the development of artificial intelligence technology, the intelligence level of unmanned aerial vehicles (UAVs) has been continuously improved. Among the numerous applications of UAVs, a sensor system equipped for quickly predicting and tracking ground moving targets is a common requirement for many application tasks, such as disaster rescue and battlefield situation awareness. When a UAV flies in a complex interference environment, due to various environmental interferences and the hiding performance of moving targets, the estimation errors of the motion states of moving targets, such as the moving speed, relative distance, and relative angle, by detection devices are relatively large. Therefore, it is usually challenging for a UAV to obtain an accurate prediction of target position information. Thus, it is necessary to design a high-precision and high-efficiency method for predicting the motion state of a target in a complex flight environment.

[0003] In the past few decades, scholars have proposed many methods to solve the problems in the design process of a method for predicting the motion state of an intermittently lost target in a complex environment. The filtering-based prediction method performs state prediction based on the target motion state and model at the previous moment, corrects the prediction result using the observation data at the current moment, and then obtains the optimal estimate of the current state after fusion according to a certain weight to achieve the prediction of the future motion state. The neural network-based prediction method designs a suitable neural network structure according to the characteristics of the prediction task, and realizes the minimization of the prediction error by continuously training and adjusting the network weights, and finally outputs the prediction result of the future motion state of the target. The multi-modal-based prediction method preprocesses, extracts features from, and fuses data of different modalities (multiple sensors or information sources), constructs a prediction model (deep learning, probability model, etc.) to train the fused data to obtain the mapping relationship between the input features and the target motion state, and finally obtains the prediction result of the future state of the target after iterative cycles.

[0004] However, there are still different defects in the above methods. The filtering-based prediction method has a poor prediction effect on the target motion state with strong nonlinearity and insufficient adaptability to the sudden change and intermittently lost target motion state. The neural network-based prediction method has complex model structure and parameter selection, poor interpretability, is prone to overfitting, and at the same time, the generalization ability of simply using a neural network is insufficient when facing abnormal target motion states. The multi-modal-based prediction method has poor multi-source data synchronization, complex feature fusion between different modalities, and is sensitive to abnormal modality data in actual flight applications. Summary of the Invention

[0005] The object of the present invention is to provide a method for predicting the motion state of an intermittently lost target by an aircraft in a complex environment.

[0006] The technical solution for achieving the object of the present invention is as follows: A method for predicting the motion state of an intermittently lost target by an aircraft in a complex environment, comprising the following steps:

[0007] Step (1): Obtain the target motion state signal;

[0008] Step (2): Decompose and reconstruct the target motion state signal with complex multi-frequency interference through the optimized improved signal reconstruction algorithm RBMO-ISVMD, and decompose the signal into a number of intrinsic mode functions IMF;

[0009] Step (3): Extract the instantaneous frequency characteristics of each IMF through Hilbert transform to obtain the target historical motion state sequence characteristics, calculate the correlation coefficient and energy retention rate of the reconstructed signal and the original signal, analyze the spectral characteristics of each IMF component, and eliminate the mode mixing phenomenon;

[0010] Step (4): Train and output the prediction result for the converted target historical motion state sequence characteristics through the RBMO-ELM algorithm, and generate a prediction error;

[0011] Step (5): Establish a double-threshold iterative prediction mechanism, dynamically adjust the prediction window size, and set the maximum number of iterations to avoid excessive calculation;

[0012] Step (6): Calculate the confidence interval of the prediction trajectory, dynamically adjust the parameters based on the cyclically input historical prediction errors to ensure the accuracy of the prediction result, and introduce a smoothing constraint to ensure the continuity of the trajectory.

[0013] Further, step (2) is specifically as follows:

[0014] Step (21): Introduce a quadratic penalty term and a penalty function to constrain the decomposition accuracy, and the improved function expression is as follows:

[0015]

[0016] In the formula, f(t) represents the original signal, v k (t) represents the kth modal component (IMF) after decomposition, α represents the L 2 penalty term weight based on the time gradient, and β represents the L 1 penalty term weight;

[0017] Step (22): Construct a fitness function using spectral entropy and decomposition balance factor, and optimize the decomposition modal number and penalty parameter of ISVMD through the RBMO algorithm:

[0018] {α, β, K} t+1 = argmax[Balance Factor(B(v k )) + Spectral Entropy(H(v k ))]

[0019] Where ω k represents the center frequency of the k-th modal component, H(v k ) represents the spectral entropy, which is used to measure the spectral complexity of the mode, and B(v k ) represents the decomposition balance factor;

[0020] Step (23): Construct a constraint equation, that is, construct a Lagrange equation to ensure that the sum of the modal components approximates the original signal after summation. The formula is as follows:

[0021]

[0022] Where λ(t) represents the Lagrange multiplier;

[0023] Step (24): Use the Alternating Direction Method of Multipliers (ADMM) to decompose the iteratively solvable problem into multiple sub-problems, gradually update the modal components and frequencies in the sub-problems to approach the global optimal solution step by step, and at the same time feedback to the constraint equation to obtain the optimized IMF set and the corresponding center frequency set:

[0024] Modal component update formula:

[0025] Center frequency update formula:

[0026] Lagrange multiplier update formula:

[0027] Where ρ represents the step size factor in the ADMM update.

[0028] Furthermore, step (3) is specifically as follows:

[0029] Step (31): Based on the IMF set decomposed in step (2), extract the instantaneous frequency characteristics of each IMF through Hilbert transform. The extraction formula is as follows:

[0030]

[0031] Step (32): Recombine the processed IMFs into a reconstructed signal and calculate the correlation coefficient to evaluate the accuracy of the decomposition and reconstruction signals. The formula is as follows:

[0032]

[0033]

[0034] In the formula, Cov represents covariance, and σ f and are the standard deviations of f(t) and respectively;

[0035] Step (33): Statistically analyze the energy of each IMF, calculate the retention rate of the total energy and the energy of the original signal, and ensure that there is no significant energy loss during the reconstruction process:

[0036]

[0037] Step (34): Analyze the spectral characteristics of each IMF using the fast Fourier transform FFT to solve the mode mixing phenomenon.

[0038] Furthermore, step (4) is specifically as follows:

[0039] Step (41): Through the frequency-distance conversion formula, decompose each moment, extract the frequency of the IMF component, calculate the distance of the target at each moment, and construct the input feature vector:

[0040] X = [s(t - nΔt) s(t - (n - 1)Δt) … s(t)] T

[0041] In the formula, s(t) represents the current state, Δt represents the time step, n represents the number of historical steps, and s(t + Δt) represents the future state of the target;

[0042] Step (42): Define the objective function and use the RBMO algorithm to optimize the weights, bias parameters, and the number of hidden layer nodes of the ELM;

[0043] Step (43): Use the optimized ELM to predict the motion state of the target, and obtain the motion state of the target in the future time period, as shown in the following formula:

[0044]

[0045] In the formula, X ′ is the new input feature;

[0046] Step (44): Use the prediction algorithm evaluation index to verify the accuracy of the prediction result output by the model training and the generalization ability of the model.

[0047] Furthermore, step (42) is specifically:

[0048] The output of the ELM is expressed as:

[0049]

[0050] Among them, \(H = h(Xw + b)\), and \(\beta\) is the output weight matrix;

[0051] The optimization objective function of RBMO is shown as the following formula:

[0052]

[0053] In the formula, represents the prediction error, and \(\|\beta\|\) 2 represents the \(L\) 2 norm of the output weight;

[0054] By using the fitness function of RBMO in combination with the objective function of ELM, the optimal parameter combination is searched, and the formula is expressed as follows:

[0055]

[0056] \(\{w, b, N\}\) * \(=\underset{w}{\mathrm{argmin}}\text{ Fitness}\)

[0057] \(w\) t+1 \(= w\) t \(+\Delta w\) t , \(b\) t+1 \(= b\) t \(+\Delta b\) t

[0058] \(\Delta w\) t , \(\Delta b\) t are the increments determined by the RBMO algorithm.

[0059] Furthermore, step (44) is specifically as follows:

[0060] The mean absolute error MAE is shown as the following formula:

[0061]

[0062] The mean square error MSE is shown as the following formula:

[0063]

[0064] The coefficient of determination \(R\) 2 is shown as the following formula:

[0065]

[0066] The present invention decomposes and reconstructs the detection signal through an improved VMD algorithm optimized by RBMO, solves the problem of inaccurate extraction of motion features caused by the complexity and non-stationarity of the target motion state signal, and introduces a quadratic penalty term and a penalty function to constrain the decomposition accuracy of traditional VMD. By combining the RBMO-ELM algorithm, it non-linearly models and efficiently trains the time-varying complex target motion history features, and solves the problem of prediction accuracy of targets with complex motion states in traditional prediction methods. At the same time, the overall method evaluates the prediction reliability through the confidence interval, dynamically adjusts the model parameters to improve the prediction accuracy, and ensures the continuity of the trajectory through smoothing constraints, finally forming a dynamic and adaptive iterative prediction framework. The present invention is applicable to the high-precision trajectory prediction of complex dynamic systems.

[0067] Compared with the prior art, the present invention has the following remarkable advantages:

[0068] (1) The present invention introduces an input signal decomposition and reconstruction part at the front end of the target prediction algorithm, which can cope with signal sources of different complexities;

[0069] (2) The present invention designs an improved signal decomposition algorithm optimized by an intelligent algorithm as the front-end signal processor of the prediction, realizes precise division and reconstruction, and ensures the accurate recognition of the target historical features by the prediction algorithm;

[0070] (3) The present invention designs an ELM optimized by an intelligent algorithm as the target motion state predictor to realize the precise tracking of the real-time target motion state;

[0071] (4) The present invention has a double-threshold iterative prediction mechanism, which can realize the fast and efficient decomposition of signals and the prediction output of the target motion state, and can cope with real-time flight tasks. Description of the Drawings

[0072] Figure 1 is a schematic diagram of the application scenario of the method for predicting the motion state of an intermittently lost target by an aircraft in a complex flight environment according to the present invention;

[0073] Figure 2 is a flowchart of the method for predicting the motion state of an intermittently lost target by an aircraft in a complex flight environment according to the present invention;

[0074] Figure 3 is the time-domain diagram after processing of the prediction method according to the present invention;

[0075] Figure 4 is the frequency spectrum diagram after processing of the prediction method according to the present invention;

[0076] Figure 5 is the Hilbert frequency spectrum diagram after processing of the prediction method according to the present invention;

[0077] Figure 6 is the processed instantaneous energy diagram of the prediction method of the present invention;

[0078] Figure 7 is the result diagram of the adaptability degree of the signal decomposition algorithm parameters in the prediction method of the present invention;

[0079] Figure 8 is the iterative result diagram of the maximum value of the signal decomposition algorithm parameters in the prediction method of the present invention;

[0080] Figure 9 is the prediction performance result diagram of the prediction algorithm in the prediction method of the present invention;

[0081] Figure 10 is the iterative performance result diagram of the prediction algorithm in the prediction method of the present invention. Detailed implementation manners

[0082] The present invention will be further described in detail below with reference to the accompanying drawings.

[0083] The application scenario of the present invention is as Figure 1 shown. Referring to Figure 2 , the present invention provides a method for predicting the motion state of an intermittently lost target by an unmanned aerial vehicle in a complex flight environment, mainly including the following steps:

[0084] Step A1: The on-board detection device of the aircraft obtains the motion state information at the target moment;

[0085] Step A2: Decompose and reconstruct the target motion state signal with complex multi-frequency interference through the optimized improved signal reconstruction algorithm (RBMO-ISVMD), and decompose the complex signal into several intrinsic mode functions (IMFs);

[0086] Step A2 specifically includes:

[0087] First, introduce the quadratic penalty term and the penalty function to constrain the decomposition accuracy of the traditional signal decomposition method. The improved function expression is as follows, optimize the original objective function, and solve the pseudo-dimension problem in the traditional VMD decomposition algorithm.

[0088]

[0089] f(t) represents the original signal, v k (t) represents the kth modal component (IMF) after decomposition, α represents the L 2 penalty term weight based on the time gradient, which is used to enhance the signal smoothness, and β represents the L 1 penalty term weight based on the time gradient, which is used to improve the anti-noise ability.

[0090] Then, a fitness function is constructed using spectral entropy and decomposition balance factor, and the number of decomposition modes and penalty parameters of ISVMD are optimized through the RBMO algorithm.

[0091] {α,β,K} t+1 =argmax [Balance Factor (B(v k ))+ Spectral Entropy (H(v k ))](2)

[0092] In the above, ω k represents the central frequency of the k-th modal component, H(v k ) represents spectral entropy, which is used to measure the spectral complexity of the mode, and B(v k ) represents the decomposition balance factor, which is used to measure the stability of mode decomposition. Then, a constraint equation is constructed, that is, a Lagrange equation is constructed to ensure that the sum of the modal components approximates the original signal after summation. The formula is expressed as follows:

[0093]

[0094] In the formula, λ(t) represents the Lagrange multiplier.

[0095] Finally, the iteratively solvable problem is decomposed into multiple sub-problems through the Alternating Direction Method of Multipliers (ADMM), and the modal components and frequencies in the sub-problems are gradually updated to approach the global optimal solution step by step. At the same time, it is fed back to the constraint equation to obtain the optimized IMF set and the corresponding central frequency set.

[0096]

[0097]

[0098] Equation 4 represents the modal component update formula, Equation 5 represents the central frequency update formula, and Equation 6 represents the Lagrange multiplier update formula. In the formula, ρ represents the step size factor in ADMM update.

[0099] Step A3: Extract the instantaneous frequency characteristics of each IMF through Hilbert transform, calculate the correlation coefficient and energy retention rate between the reconstructed signal and the original signal, and analyze the spectral characteristics of each IMF component to ensure that there is no aliasing phenomenon in the reconstructed signal;

[0100] Step A3 specifically includes:

[0101] First, after obtaining the IMF set based on the decomposition in Step A2, extract the instantaneous frequency characteristics of each IMF through Hilbert transform. The extraction formula is as follows:

[0102]

[0103] Then, the processed IMF is recombined into a reconstructed signal, and the correlation coefficient is calculated to evaluate the accuracy of the decomposition and reconstruction signals. The formula is as follows.

[0104]

[0105] In the formula, Cov represents covariance, and σ f and are the standard deviations of f(t) and respectively. The energy of each IMF is statistically analyzed, and the retention rate of the total energy and the energy of the original signal is calculated to ensure that there is no significant energy loss during the reconstruction process.

[0106]

[0107] Finally, the fast Fourier transform (FFT) is used to analyze the spectral characteristics of each IMF to solve the mode mixing phenomenon.

[0108] Step A4: Train and output predictions for the characteristics of the converted target historical motion state sequence through the RBMO-ELM algorithm;

[0109] Step A4 specifically includes:

[0110] First, for the corresponding decomposed and reconstructed signal frequencies at each moment obtained in Step A3, through the frequency-distance conversion formula of the corresponding detection sensor, the measured values of each IMF component decomposed at each moment are converted, the detection distance of the target at each moment is calculated, and the historical input feature vector of the target is constructed using the detection distance.

[0111] X = [s(t - nΔt) s(t - (n - 1)Δt) … s(t)] T (12)

[0112] In the formula, s(t) represents the current state, Δt represents the time step, n represents the number of historical steps, and s(t + Δt) represents the future state of the target.

[0113] Define the objective function, and use the RBMO algorithm to optimize the weights, bias parameters, and the number of hidden layer nodes of the ELM. The output of the ELM can be expressed as:

[0114]

[0115] Among them, H = h(Xw + b), and β is the output weight matrix.

[0116] In the present invention, the RBMO is used to optimize the weights and bias parameters of the ELM network, and the constructed optimization objective function can be expressed as:

[0117]

[0118] In the formula, represents the prediction error, and ‖β‖ 2 represents the L 2 norm of the output weight. By combining the fitness function of RBMO with the objective function of ELM, the optimal parameter combination is searched, and the formula is expressed as follows:

[0119]

[0120] {w, b, N} * = argmin Fitness(16)

[0121] w t+1 = w t + △w t , b t+1 = b t + △b t (17)

[0122] Formula 15 represents the objective function after combining the fitness function of RBMO and ELM. Formula 16 represents the formula for adaptively searching the optimized parameter set of ELM using the RBMO algorithm. Formula 17 represents the optimization update formula. △w t , △b t are the increments determined by the RBMO algorithm.

[0123] Then, the optimized ELM model is used to predict the target motion state. By training the ELM model with the optimized parameter set {w, b, N}, the historical motion state is fitted to the output motion state. Using the trained ELM model, the new input set is predicted to obtain the motion state of the target in the future time period, as shown in the following formula.

[0124]

[0125] In the formula, X′ is the new input feature.

[0126] Finally, to ensure the accuracy of the predicted results, the prediction algorithm evaluation index is used to verify the accuracy of the predicted results output by the model training and the generalization ability of the model.

[0127] The mean absolute error (MAE) is shown in the following formula:

[0128]

[0129] The mean square error (MSE) is shown in the following formula:

[0130]

[0131] Coefficient of determination (R2 )As shown in the following formula:

[0132]

[0133] Step A5: Establish a dual-threshold iterative prediction mechanism, dynamically adjust the size of the prediction window, and set the maximum number of iterations to avoid excessive calculation;

[0134] Step A6: Calculate the confidence interval of the prediction trajectory, dynamically adjust the model parameters based on the historical prediction errors of the cyclic input, ensure the accuracy of the prediction results, and introduce a smoothing constraint to ensure the continuity of the trajectory.

[0135] The present invention decomposes and reconstructs the detection signal through the improved VMD algorithm optimized by RBMO, solves the problem of inaccurate extraction of motion characteristics caused by the complexity and non-stationarity of the target motion state signal, and introduces a quadratic penalty term and a penalty function to constrain the decomposition accuracy of the traditional VMD. By combining the RBMO-ELM algorithm, non-linear modeling and efficient training are carried out on the time-varying complex target motion history characteristics, and the problem of prediction accuracy of targets with complex motion states in traditional prediction methods is solved. At the same time, the reliability of the prediction is evaluated through the confidence interval as a whole, the model parameters are dynamically adjusted to improve the prediction accuracy, and the continuity of the trajectory is ensured through the smoothing constraint, finally forming a dynamic and adaptive iterative prediction framework. The present invention is applicable to high-precision trajectory prediction of complex dynamic systems.

[0136] In order to verify the performance of the present invention, a performance verification experiment is carried out using the signal dataset received by the millimeter-wave radar actually carried on the unmanned aerial vehicle. In the experiment, the algorithm of the present invention is encapsulated into a library, and the millimeter-wave radar module installed on the unmanned aerial vehicle is used as the detection device. This module has a compact structure, is easy to deploy, has a stable signal transmission power, and the signal is adjustable. A real detection environment with obstacles is established in the actual experiment, and a remote-controlled car traveling according to the planned points is selected as the detection target, and the unmanned aerial vehicle hovers at a height of 1.5 meters from the ground.

[0137] The experimental results are as Figures 3 - 9 shown. From Figures 3 - 7 which represents the result calculated when the algorithm processes the target passing through a planned point, it can be seen from the figure that the signal processing algorithm in the present invention can completely adaptively decompose and reconstruct the signal modulus when processing the actual signal containing environmental noise. Figure 3 IMF3 in Figure 5 is completely separated from other IMFs. From Figure 6It can be seen that the fitness function gradually decreases as the number of iterations increases, reaches the minimum value after 8 iterations and remains stable. At this time, the optimization algorithm has calculated the optimal parameter values, demonstrating the fast convergence of the algorithm.

[0138] Using the iterative mechanism in the method, the present invention can calculate the relative distance of the detected target at each moment, and the relative position of the target can be obtained by defining the relative detection angle of the detection sensor. In order to verify the superior performance of the prediction network in the present invention, different prediction methods are used to predict the historical motion characteristics of the target. Figure 8 and Figure 9 illustrate the prediction accuracy and performance of four algorithms. At the beginning, due to the target deviating from its historical trajectory, the RBMO-ELM prediction algorithm has significant prediction errors. However, compared with other algorithms, the error of RBMO-ELM is still the smallest and it can quickly adapt to these changes. Similarly, in the second stage when the target transitions from a curved motion pattern to a linear motion pattern, although the algorithm has prediction errors, it can converge quickly. In both cases involving changes in target motion, RBMO-ELM shows the best performance, followed by SSA-ELM and PSO-ELM, and the traditional ELM shows the worst performance. Figure 9 Shows the changes in the fitness function of three optimization algorithms. Although the convergence speed of RBMO is not as fast as that of PSO, its final convergence effect is better, reaching the minimum and most stable value. Experiments prove that the prediction accuracy of the present invention can achieve the best prediction of the future state with intermittent target motion.

[0139] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting the motion state of an aircraft for intermittently lost targets in a complex environment, characterized in that: The steps include: Step (1): obtaining a target motion state signal; Step (2): Decompose and reconstruct the target motion state signal with complex multi-frequency interference through the optimized improved signal reconstruction algorithm RBMO-ISVMD, and decompose the signal into several intrinsic mode functions IMF; Step (3): Extract the instantaneous frequency characteristics of each IMF through Hilbert transform, obtain the target historical motion state sequence characteristics, calculate the correlation coefficient and energy retention rate between the reconstructed signal and the original signal, analyze the spectral characteristics of each IMF component, and eliminate the modal aliasing phenomenon; Step (4): Train the converted target historical motion state sequence features through the RBMO-ELM algorithm, output the prediction results, and generate the prediction error; Step (5): Establish a dual-threshold iterative prediction mechanism, dynamically adjust the prediction window size, and set the maximum number of iterations to avoid over-calculation; Step (6): Calculate the confidence interval of the predicted trajectory and dynamically adjust the parameters based on the historical prediction error of the loop input to ensure the accuracy of the prediction results. At the same time, introduce smoothness constraints to ensure the continuity of the trajectory.

2. The method according to claim 1, characterized in that Step (2) is specifically as follows: Step (21): Introduce the quadratic penalty term and penalty function to constrain the decomposition accuracy. The improved function expression is as follows: Where f(t) represents the original signal, v k (t) represents the kth modal component (IMF) after decomposition, α represents the weight of the L2 penalty term based on the time gradient, and β represents the weight of the L1 penalty term based on the time gradient; Step (22): Use spectral entropy and decomposition balance factor to construct a fitness function, and use the RBMO algorithm to optimize the number of decomposition modes and penalty parameters of ISVMD: {α,β,K} t+1 =argmax[Balance Factor(B(v k ))+Spectral Entropy(H(v k ))] In the formula, ω k represents the center frequency of the kth modal component, H(v k ) represents the spectral entropy, which is used to measure the spectral complexity of the mode. k ) represents the decomposition balance factor; Step (23): Construct a constraint equation, that is, construct a Lagrange equation to ensure that the modal components approximate the original signal after summing. The formula is as follows: Where λ(t) is the Lagrange multiplier; Step (24): Decompose the iterative solution problem into multiple sub-problems through the alternating direction multiplier method ADMM, gradually update the modal components and frequencies in the sub-problems to gradually approach the global optimal solution, and feed back to the constraint equation at the same time to obtain the optimized IMF set and the corresponding center frequency set: Modal component update formula: Center frequency update formula: Lagrange multiplier update formula: Where ρ represents the step size factor in ADMM update.

3. The method according to claim 2, characterized in that Step (3) is specifically as follows: Step (31): Based on the decomposition of step (2), the IMF set is obtained, and the instantaneous frequency characteristics of each IMF are extracted by Hilbert transform. The extraction formula is as follows: Step (32): The processed IMFs are reassembled into reconstructed signals and the correlation coefficient is calculated to evaluate the accuracy of the decomposed and reconstructed signals. The formula is as follows: In the formula, Cov represents covariance, σ f and are f(t) and The standard deviation of Step (33): Count the energy of each IMF and calculate the retention rate of the total energy and the original signal energy to ensure that there is no significant energy loss during the reconstruction process: Step (34): Use Fast Fourier Transform (FFT) to analyze the frequency spectrum characteristics of each IMF to resolve the modal aliasing phenomenon.

4. The method according to claim 3, characterized in that Step (4) is as follows: Step (41): Decompose each moment using the frequency-distance conversion formula to obtain the IMF component for frequency extraction, calculate the distance of the target at each moment, and construct the input feature vector: X=[s(t-n△t)s(t-(n-1)△t)…s(t)] T In the formula, s(t) represents the current state, Δt represents the time step, n represents the number of historical steps, and s(t+Δt) represents the target future state; Step (42): define the objective function and use the RBMO algorithm to optimize the weight, bias parameters and number of hidden layer nodes of the ELM; Step (43): Use the optimized ELM to predict the target motion state and obtain the target motion state in the future period, as shown in the following formula: Where X′ is the new input feature; Step (44): Use the prediction algorithm evaluation index to verify the accuracy of the prediction results output by the model training and the generalization ability of the model.

5. The method according to claim 4, characterized in that Step (42) is specifically: The ELM output is expressed as: Where H = h(Xw+b), β is the output weight matrix; The RBMO optimization objective function is shown as follows: In the formula, represents the prediction error, ‖β‖2 represents the L2 norm of the output weight; The fitness function of RBMO is combined with the objective function of ELM to search for the optimal parameter combination. The formula is as follows: {w,b,N} * =argmin Fitness w t+1 =w t +△w t ,b t+1 =b t +△b t Δw t , △b t The increment determined by the RBMO algorithm.

6. The method according to claim 5, characterized in that Step (44) is as follows: The mean absolute error MAE is shown as follows: The mean square error MSE is as follows: Coefficient of determination R 2 As shown below:

Images