Data interpolation method and device in radar performance evaluation data missing state

Abstract

The invention discloses a data interpolation method and device in a radar performance evaluation data missing state, and the method comprises the steps: obtaining an original sample data set which comprises N samples; processing the original sample data set to obtain an index sample; preprocessing the original sample data set to obtain a preprocessed sample set; solving a preset first differential equation model by using the pre-processing sample set and the index sample to obtain a channel expansion training sample set under different scales of noise; training a preset score neural network model by using the channel extension training sample set to obtain an optimized score neural network model; processing the optimized score neural network model to obtain a second differential equation model; and processing the original data to be interpolated by using the second differential equation model to obtain complete data after interpolation. According to the invention, the integrity of radar performance evaluation data is improved.

Description

Technical Field

[0001] The present invention relates to the field of big data processing technology, and in particular to a data interpolation method and device in a state where radar effectiveness evaluation data is missing. Background Art

[0002] As a high-precision detection and sensing device, radar is currently widely used in various fields of society, including autonomous driving, weather forecasting, mapping and navigation, surveillance and detection, etc. How to predict and evaluate the expected performance of radar in specific environments and specific tasks in order to design or improve and enhance the functions, performance, working mode and technical system of radar in a targeted manner has become a focus of radar designers, producers and users. Radar performance evaluation usually requires the comprehensive use of data from different sources, including data generated by the radar itself, environmental data, target data, and data generated by the platform carrying the radar. However, in reality, due to the influence of various factors such as data acquisition conditions, acquisition accuracy, and transmission channels, the data used for radar performance evaluation will be partially lost and incomplete, which in turn affects the accuracy of the radar performance evaluation results. It is necessary to perform high-precision interpolation on the missing data to form an accurate and complete data set as much as possible, laying the foundation for subsequent radar performance evaluation. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a data interpolation method and device in the state of missing radar effectiveness evaluation data, obtain the original sample data set, and construct a corresponding indication sample for each sample in the data set; construct a forward time stochastic differential equation model, use the original sample as the initial value to solve the equation under multi-scale noise conditions, cascade the solution result of each scale with the indication sample to generate a channel expansion training sample set under different scale noises; construct a score neural network of data distribution, and train the score neural network using the channel expansion training sample set under different scale noises; use the trained score neural network to construct a reverse time stochastic differential equation model; input the original data to be interpolated into the reverse time stochastic differential equation model to generate complete data after interpolation. The technical solution provided by the present invention improves the completeness of radar effectiveness evaluation data, and can realize missing data interpolation without a complete data set in the model training stage, which can be adapted to a wider range of applications.

[0004] In order to solve the above technical problems, a first aspect of an embodiment of the present invention discloses a data interpolation method in a state where radar effectiveness evaluation data is missing, the method comprising:

[0005] S1, obtaining an original sample data set, wherein the original sample data set includes N samples, where N is a positive integer;

[0006] S2, processing the original sample data set to obtain an index sample;

[0007] S3, processing the original sample data set and the indicator sample to obtain complete data after interpolation.

[0008] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the processing of the original sample data set and the indicator sample to obtain the interpolated complete data includes:

[0009] S31, preprocessing the original sample data set to obtain a preprocessed sample set;

[0010] S32, using the preprocessing sample set and the index sample, solving a preset first differential equation model to obtain a channel expansion training sample set under noise of different scales;

[0011] S33, using the channel expansion training sample set to train the preset score neural network model to obtain an optimized score neural network model;

[0012] S34, processing the optimized score neural network model to obtain a second differential equation model;

[0013] S35, using the second differential equation model to process the original data to be interpolated to obtain complete data after interpolation.

[0014] As an optional implementation manner, in the first aspect of the embodiment of the present invention, preprocessing the original sample data set to obtain a preprocessed sample set includes:

[0015] S311, performing data standardization on the original sample data set to obtain a standardized data set;

[0016] S312, denoising the standardized data set to obtain a preprocessed sample set.

[0017] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the first differential equation model expression is:

[0018] dy(t)=f(y(t),t)dt,g(t)dw

[0019] Among them, y(t) is the observed variable, corresponding to the sample in the original sample data set, t is the time variable, f(y(t), t) is the drift function, T is the total time, g(t) is a single-valued diffusion function, w is a standard Wiener random process variable.

[0020] As an optional implementation, in the first aspect of the embodiment of the present invention, the method of solving a preset first differential equation model using the original sample data set and the index sample to obtain a channel expansion training sample set under noise of different scales includes:

[0021] S321, using the original sample data set, solving a preset first differential equation model to obtain a solution to the first differential equation;

[0022] The expression of the solution of the first differential equation is:

[0023] y(t n+1 )=y(t n )+f(y(t n ),t n )Δt+g(t n )Δw n

[0024] Among them, y(t n+1 ) is t n+1 The solution of the first differential equation at time y(t n ) is t n The solution of the first differential equation at time Δt = t n+1 -t n , Δw n ~Ν(0,τ n I) is the Gaussian distribution sampling value, I is the unit vector, τ n is the noise scale parameter, N 1 is the number of noise scales, T is the total time, n=0,1,2,…,N 1 ;

[0025] S322, processing the solution of the first differential equation and the index sample to obtain a channel expansion training sample set under noises of different scales;

[0026] The channel expansion training sample set expression under different scale noise is:

[0027] X(t n )=((y(t n ) 1 ,D 1 ),(y(t n ) 2 ,D 2 ),…,(y(t n ) d ,D d ))

[0028] Among them, X(t n ) is tn The channel expansion training sample at time t, D 1 ,D 2 ,…,D d is the indicator sample, d is the number of indicator samples, y(t n ) 1 ,y(t n ) 2 ,…,y(t n ) d is the sample in the original sample dataset corresponding to the indicator sample.

[0029] As an optional implementation, in the first aspect of the embodiment of the present invention, the use of the channel expansion training sample set to train a preset score neural network model to obtain an optimized score neural network model includes:

[0030] S331, setting the scoring neural network model parameters;

[0031] S332, extracting the channel expansion training sample set to obtain small batch training samples;

[0032] S333, using the small batch training samples, updating the score neural network model parameters to obtain an optimized score neural network model.

[0033] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the second differential equation model expression is:

[0034]

[0035] in, To optimize the score neural network model, g(t) is a single-valued diffusion function, t is a time variable, w is a standard Wiener random process variable, and x(t) is a channel expansion training sample set.

[0036] A second aspect of an embodiment of the present invention discloses a data interpolation device in a state where radar effectiveness evaluation data is missing, the device comprising:

[0037] A data acquisition module is used to acquire an original sample data set, where the original sample data set includes N samples, where N is a positive integer;

[0038] An indicator sample construction module, used to process the original sample data set to obtain an indicator sample;

[0039] The data interpolation module is used to process the original sample data set and the indicator sample to obtain complete data after interpolation.

[0040] As an optional implementation manner, in the second aspect of the embodiment of the present invention, the processing of the original sample data set and the indicator sample to obtain the interpolated complete data includes:

[0041] S31, preprocessing the original sample data set to obtain a preprocessed sample set;

[0042] S32, using the preprocessing sample set and the index sample, solving a preset first differential equation model to obtain a channel expansion training sample set under noise of different scales;

[0043] S33, using the channel expansion training sample set to train the preset score neural network model to obtain an optimized score neural network model;

[0044] S34, processing the optimized score neural network model to obtain a second differential equation model;

[0045] S35, using the second differential equation model to process the original data to be interpolated to obtain complete data after interpolation.

[0046] As an optional implementation manner, in the second aspect of the embodiment of the present invention, the preprocessing of the original sample data set to obtain the preprocessed sample set includes:

[0047] S311, performing data standardization on the original sample data set to obtain a standardized data set;

[0048] S312, denoising the standardized data set to obtain a preprocessed sample set.

[0049] As an optional implementation manner, in the second aspect of the embodiment of the present invention, the first differential equation model expression is:

[0050] dy(t)=f(y(t),t)dt,g(t)dw

[0051] Among them, y(t) is the observed variable, corresponding to the sample in the original sample data set, t is the time variable, f(y(t), t) is the drift function, T is the total time, g(t) is a single-valued diffusion function, w is a standard Wiener random process variable.

[0052] As an optional implementation, in the second aspect of the embodiment of the present invention, the method of solving a preset first differential equation model using the original sample data set and the index sample to obtain a channel expansion training sample set under noise of different scales includes:

[0053] S321, using the original sample data set, solving a preset first differential equation model to obtain a solution to the first differential equation;

[0054] The expression of the solution of the first differential equation is:

[0055] y(t n+1 )=y(t n )+f(y(t n ),t n )Δt+g(t n )Δw n

[0056] Among them, y(t n+1 ) is t n+1 The solution of the first differential equation at time y(t n ) is t n The solution of the first differential equation at time Δt = t n+1 -t n , Δw n ~Ν(0,τ n I) is the Gaussian distribution sampling value, I is the unit vector, τ n is the noise scale parameter, N 1 is the number of noise scales, T is the total time, n=0,1,2,…,N 1 ;

[0057] S322, processing the solution of the first differential equation and the index sample to obtain a channel expansion training sample set under noises of different scales;

[0058] The channel expansion training sample set expression under different scale noise is:

[0059] X(t n )=((y(t n ) 1 ,D 1 ),(y(t n ) 2 ,D 2 ),…,(y(t n ) d ,D d ))

[0060] Among them, X(t n ) is t n The channel expansion training sample at time t, D 1 ,D 2 ,…,D d is the indicator sample, d is the number of indicator samples, y(t n ) 1,y(t n ) 2 ,…,y(t n ) d is the sample in the original sample dataset corresponding to the indicator sample.

[0061] As an optional implementation, in the second aspect of the embodiment of the present invention, the use of the channel expansion training sample set to train a preset score neural network model to obtain an optimized score neural network model includes:

[0062] S331, setting the scoring neural network model parameters;

[0063] S332, extracting the channel expansion training sample set to obtain small batch training samples;

[0064] S333, using the small batch training samples, updating the score neural network model parameters to obtain an optimized score neural network model.

[0065] As an optional implementation, in the second aspect of the embodiment of the present invention, the second differential equation model expression is:

[0066]

[0067] in, To optimize the score neural network model, g(t) is a single-valued diffusion function, t is a time variable, w is a standard Wiener random process variable, and x(t) is a channel expansion training sample set.

[0068] A third aspect of the present invention discloses another data interpolation device in a state where radar effectiveness evaluation data is missing, the device comprising:

[0069] A memory storing executable program code;

[0070] a processor coupled to the memory;

[0071] The processor calls the executable program code stored in the memory to execute part or all of the steps in the data interpolation method in a state where radar effectiveness evaluation data is missing disclosed in the first aspect of an embodiment of the present invention.

[0072] The fourth aspect of the present invention discloses a computer-storable medium, which stores computer instructions. When the computer instructions are called, they are used to execute some or all of the steps in the data interpolation method in the case of missing radar effectiveness evaluation data disclosed in the first aspect of an embodiment of the present invention.

[0073] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0074] To address the problem of missing data interpolation, a multi-scale noisy stochastic differential equation neural network was designed to estimate the scores of the data probability distribution. Based on this, a data interpolation model was constructed. Compared with other neural network structures and single-scale neural network interpolation methods such as autoencoder networks and generative adversarial networks, it can more accurately fit the probability distribution of missing data. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0076] Figure 1 It is a flow chart of a data interpolation method in a state where radar effectiveness evaluation data is missing, disclosed in an embodiment of the present invention;

[0077] Figure 2 is an original sample example disclosed in the embodiment of the present invention;

[0078] Figure 3 is an example of an indication sample corresponding to the original sample disclosed in the embodiment of the present invention;

[0079] Figure 4 This is a schematic diagram of randomly selecting a region from a feature map and randomly assigning 0 to coordinate points therein, as disclosed in an embodiment of the present invention;

[0080] Figure 5 It is a schematic diagram of constructing a dropout layer output for coordinate points assigned 0 according to an embodiment of the present invention;

[0081] Figure 6 is a schematic diagram of a score neural network model disclosed in an embodiment of the present invention;

[0082] Figure 7 It is a flow chart of another data interpolation method in a state where radar effectiveness evaluation data is missing, disclosed in an embodiment of the present invention;

[0083] Figure 8 It is a structural schematic diagram of a data interpolation device in a state where radar effectiveness evaluation data is missing, disclosed in an embodiment of the present invention;

[0084] Fig. 9 It is a structural schematic diagram of another data interpolation device in a state where radar effectiveness evaluation data is missing, disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0085] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0086] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, device, product or equipment that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or equipment.

[0087] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0088] The present invention discloses a data interpolation method and device in a state where radar effectiveness evaluation data is missing, the method comprising: obtaining an original sample data set, the original sample data set comprising N samples; processing the original sample data set to obtain an index sample; preprocessing the original sample data set to obtain a preprocessed sample set; solving a preset first differential equation model using the preprocessed sample set and the index sample to obtain a channel extension training sample set under noises of different scales; training a preset score neural network model using the channel extension training sample set to obtain an optimized score neural network model; processing the optimized score neural network model to obtain a second differential equation model; processing the original data to be interpolated using the second differential equation model to obtain complete data after interpolation. The present invention improves the integrity of radar effectiveness evaluation data. The following are detailed descriptions respectively.

[0089] Embodiment 1

[0090] See also Figure 1 , Figure 1 1 is a flow chart of a method for data interpolation in the absence of radar effectiveness evaluation data disclosed in an embodiment of the present invention. Figure 1The data interpolation method described in the radar effectiveness evaluation data missing state is applied to the field of big data processing technology to implement missing data interpolation, which is not limited in the embodiments of the present invention. Figure 1 As shown, the data interpolation method for radar effectiveness evaluation in the missing data state may include the following operations:

[0091] S1, obtaining an original sample data set, wherein the original sample data set includes N samples, where N is a positive integer;

[0092] S2, processing the original sample data set to obtain an index sample;

[0093] S3, processing the original sample data set and the indicator sample to obtain complete data after interpolation.

[0094] Figure 2 is an original sample example disclosed in the embodiment of the present invention; Figure 3 is an example of an indication sample corresponding to the original sample disclosed in the embodiment of the present invention;

[0095] Optionally, the processing of the original sample data set and the indicator sample to obtain complete data after interpolation includes:

[0096] S31, preprocessing the original sample data set to obtain a preprocessed sample set;

[0097] S32, using the preprocessing sample set and the index sample, solving a preset first differential equation model to obtain a channel expansion training sample set under noise of different scales;

[0098] S33, using the channel expansion training sample set to train the preset score neural network model to obtain an optimized score neural network model;

[0099] S34, processing the optimized score neural network model to obtain a second differential equation model;

[0100] S35, using the second differential equation model to process the original data to be interpolated to obtain complete data after interpolation.

[0101] Optionally, preprocessing the original sample data set to obtain a preprocessed sample set includes:

[0102] S311, performing data standardization on the original sample data set to obtain a standardized data set;

[0103] Data standardization is to perform operations such as data alignment and data filling on the original sample data set, which is a prior art in the field and is not limited in this embodiment.

[0104] S312, denoising the standardized data set to obtain a preprocessed sample set.

[0105] The method for denoising the standardized data set is:

[0106] Perform time-frequency analysis on the standardized data set to obtain a time-frequency image;

[0107] Processing the time-frequency image to obtain a real component and an imaginary component;

[0108] Performing wavelet transform on the real component to obtain first wavelet parameter information;

[0109] Performing wavelet transform on the imaginary component to obtain second wavelet parameter information;

[0110] Performing filtering processing on the first wavelet parameter information to obtain third wavelet parameter information;

[0111] Performing filtering processing on the second wavelet parameter information to obtain fourth wavelet parameter information;

[0112] Reconstructing the third wavelet parameter information and the fourth wavelet parameter information to obtain a reconstructed time-frequency image;

[0113] Performing an inverse time-frequency transform on the reconstructed time-frequency image to obtain a preprocessed sample set.

[0114] In the above steps, the filtering method is:

[0115] Set the initial threshold δ th The target interval is [α, β], where 0≤α<β≤max{S(m,n)}, S(m,n) is the wavelet parameter information;

[0116] Divide the target interval [α, β] into L equal parts;

[0117] The initial threshold δ th Set to α, according to The step size is incremented and iterated to obtain the threshold

[0118] Utilizing Thresholds Calculate the wavelet parameter information after filtering for each iteration

[0119]

[0120] Among them, S(m,n) represents the original wavelet parameter information, It represents the wavelet parameter information after filtering, and a and b represent the time and frequency parameters in the threshold;

[0121] Calculate the difference between each iteration and the wavelet parameter information obtained in the previous iteration is the number of iterations;

[0122] The difference Perform curve fitting to obtain curve C. When curve C has an inflection point, the corresponding threshold is the optimal threshold.

[0123] The optimal threshold is used to perform optimal threshold filtering on the wavelet parameter information S(m,n) to obtain the wavelet parameter information after filtering.

[0124] Optionally, the first differential equation model expression is:

[0125] dy(t)=f(y(t),t)dt,g(t)dw

[0126] Among them, y(t) is the observed variable, corresponding to the sample in the original sample data set, t is the time variable, f(y(t), t) is the drift function, T is the total time, g(t) is a single-valued diffusion function, w is a standard Wiener random process variable.

[0127] Optionally, using the original sample data set and the index sample to solve a preset first differential equation model to obtain a channel expansion training sample set under noise of different scales includes:

[0128] S321, using the original sample data set, solving a preset first differential equation model to obtain a solution to the first differential equation;

[0129] The expression of the solution of the first differential equation is:

[0130] y(t n+1 )=y(t n )+f(y(t n ),t n )Δt+g(t n )Δw n

[0131] Among them, y(t n+1 ) is t n+1 The solution of the first differential equation at time y(t n ) is t n The solution of the first differential equation at time Δt = t n+1 -t n , Δw n ~Ν(0,τ n I) is the Gaussian distribution sampling value, I is the unit vector, τ n is the noise scale parameter, N 1 is the number of noise scales, T is the total time, n=0,1,2,…,N 1 ;

[0132] S322, processing the solution of the first differential equation and the index sample to obtain a channel expansion training sample set under noises of different scales;

[0133] The channel expansion training sample set expression under different scale noise is:

[0134] X(t n )=((y(t n ) 1 ,D 1 ),(y(t n ) 2 ,D 2 ),…,(y(t n ) d ,D d ))

[0135] Among them, X(t n ) is t n The channel expansion training sample at time t, D 1 ,D 2 ,…,D d is the indicator sample, d is the number of indicator samples, y(t n ) 1 ,y(t n ) 2 ,…,y(t n ) d is the sample in the original sample dataset corresponding to the indicator sample.

[0136] Optionally, the using the channel to expand the training sample set to train a preset score neural network model to obtain an optimized score neural network model includes:

[0137] S331, setting the scoring neural network model parameters;

[0138] S332, extracting the channel expansion training sample set to obtain small batch training samples;

[0139] The extraction method may be a random extraction method or an optimization algorithm, which is not limited in this embodiment.

[0140] S333, using the small batch training samples, updating the score neural network model parameters to obtain an optimized score neural network model.

[0141] Optionally, the method of using the channel to expand the training sample set to train the preset score neural network model to obtain the optimized score neural network model may be as follows:

[0142] (1) Denoising preprocessing: A portion of the channel expansion training sample set is used to train the CNN denoising model. The remaining data is transformed through STFT and then preprocessed using the trained CNN denoising model. The denoised STFT spectrum is output as the training set of the preset score neural network model, and the remaining channel expansion training sample set for training the CNN denoising model is used as the validation set.

[0143] The CNN denoising model is a fully convolutional network consisting of 16 convolutional layers. The size of the convolution kernel of the first layer is 9×8, with 18 convolution kernels. The 2nd to 13th convolutional layers are a group of 3 layers repeated 4 times, with convolution kernel widths of 5, 9, and 9 respectively, and the number of convolution kernels is 64. The convolution kernel width of the last convolution layer is 129 and there is only 1 convolution kernel. In this network, convolution is performed only in the frequency dimension, and for all layers except the first layer, the convolution kernel width along the time dimension is set to 1. The convolution layer passes through a batch normalization layer and a ReLU activation function layer, and is finally output through a regression layer.

[0144] (2) Setting the fitness function: Use the training set to train the preset score neural network model, and the inverse of the accuracy of the score neural network model that reaches the maximum number of training times on the validation set is used as the fitness function.

[0145] (3) Initializing the parameters of the quantum sailfish algorithm and the quantum positions of the quantum sailfish and quantum sardines. Each quantum position is a three-dimensional vector, which represents the number of hidden layer neurons, the initial learning rate, and the number of training times of the preset score neural network model.

[0146] (4) Iterative update of quantum positions: Update the quantum positions of quantum swordfish and quantum sardines, calculate the fitness values, determine the optimal fitness values ​​of individuals, and iteratively update the population.

[0147] (5) After the quantum sailfish algorithm is iterated, the optimal position is taken as the hyperparameter of the preset score neural network model to obtain the optimized score neural network model.

[0148] The quantum sailfish algorithm combines the advantages of the quantum evolution theory of sailfish and quantum computing. In the quantum sailfish algorithm, quantum mechanisms design a new way to update the positions of sailfish and sardines, which can improve the global search ability of sailfish.

[0149] The quantum sailfish algorithm evolves through the quantum states of sailfish and sardine schools. There is a sailfish population consisting of N qEach sailfish has its own quantum position and corresponding position. Therefore, when solving a d-dimensional problem, the quantum position of the i-th sailfish is defined as q i =[q i,1 ,q i,2 ,...,q i,d ], where 0<q i,j <1, i = 1, 2, ..., N q , j = 1, 2, ..., d. There is a sardine school with an initial population size of Where p is the ratio of the initial swordfish population to the sardine population, and the quantum position of the kth sardine is defined as s k =[s k,1 ,s k,2 ,...,s k,d ], where 0<s k,j <1,

[0150] Map it to a position in the solution space, and the mapping relationship is defined as follows:

[0151]

[0152] Among them, x min =[x min,1 ,x min,2 ,...,x min,d ] is the lower boundary of the problem, x max =[x max,1 ,x max,2 ,...,x max,d ] is the upper bound of the problem, and “e” represents the Hadamard product, which is the multiplication of the elements in the corresponding dimensions of the two vectors.

[0153] Each position of the swordfish and sardines is a feasible solution to the problem, and the quality of the feasible solution can be calculated by the fitness function f(·). The minimum fitness value of the swordfish population in the g-th iteration is the optimal fitness value. Determine its quantum position as the optimal quantum position of the current iteration The optimal fitness value in the entire swordfish population up to the gth iteration is the global optimal fitness value Its location is the elite sailfish quantum location The minimum fitness value of the sardine population in the gth iteration is the optimal fitness value Determine its quantum position as the injured sardine quantum position of the current iteration

[0154] The sailfish population, led by the elite sailfish, attacks the sardine school and approaches the injured sardines to complete the predation. Therefore, the next generation quantum position of each sailfish is affected by the elite sailfish and the injured sardines. The jth quantum bit of the i-th sailfish in the g+1th iteration is updated using a simulated quantum revolving gate.

[0155]

[0156] Among them, r 1 is a uniform random number between [0,1], r 2 To satisfy the Gaussian distribution random number with mean 0 and variance 1, λ g+1 A coefficient is defined as follows

[0157] λ g+1 =2×rand×P g+1 -P g+1

[0158] in is the density of sardines, indicating that the number of sardines will decrease significantly with the predation of swordfish in each iteration. is the population size of sardines at the g+1th iteration, and rand is a uniform random number between [0,1].

[0159] As the sailfish continue to attack the sardines, the sardines' energy storage will be reduced, reducing their ability to detect the sailfish's location information, which will affect the sardines' escape, and eventually they will be attacked and injured by the sailfish, becoming the sailfish's prey. Therefore, the jth qubit of the kth sardine in the g+1th iteration is updated as follows:

[0160]

[0161] Among them, r 3 is a uniform random number between [0,1], r 4 A is a Gaussian random number with a mean of 0 and a variance of 1. g+1 is the energy of the sardines at the g+1th iteration, and its expression is

[0162] A g+1 =E×[1-2×(g+1)×ξ]

[0163] Where E is the total energy of sardines and ξ is the energy attenuation coefficient.

[0164] Sardines injured in the final stage of predation will be caught by the swordfish. When the fitness value of the sardine is better than that of the swordfish, it will be caught, and each swordfish will catch at most one sardine in each iteration. In this case, the quantum position of the swordfish is replaced by the quantum position of the captured sardine. but The caught sardines are removed from the sardine population, so the sardine population is Where D g+1 is the number of sardines caught by swordfish.

[0165] The algorithm terminates when the maximum number of iterations is reached or all sardines are caught, and finally outputs the position of the elite swordfish. The algorithm time complexity is O(n 3 ).

[0166] Optionally, the second differential equation model expression is:

[0167]

[0168] in, To optimize the score neural network model, g(t) is a single-valued diffusion function, t is a time variable, w is a standard Wiener random process variable, and x(t) is a channel expansion training sample set.

[0169] Figure 4 This is a schematic diagram of randomly selecting a region from a feature map and randomly assigning 0 to coordinate points therein, as disclosed in an embodiment of the present invention; Figure 5 It is a schematic diagram of constructing a dropout layer output for coordinate points assigned 0 according to an embodiment of the present invention; Figure 6 It is a schematic diagram of a score neural network model disclosed in an embodiment of the present invention.

[0170] It can be seen that the present invention specifically designs a multi-scale noise random differential equation neural network to estimate the score of the data probability distribution for the interpolation problem of missing data, and constructs a data interpolation model based on this. Compared with other neural network structures and single-scale neural network interpolation methods such as autoencoder networks and generative adversarial networks, it can more accurately fit the probability distribution of missing data. A data missing indication mechanism is introduced in the neural network training process, and a channel expansion training sample set is constructed, so that the neural network can pay more attention to the missing sample data when learning the data probability distribution, thereby further improving the fitting accuracy of the data probability distribution and the accuracy of interpolation data generation. The present invention can interpolate missing data without a complete data set, and has high interpolation accuracy and a wide range of applications.

[0171] Embodiment 2

[0172] See also Figure 7 , Figure 7 1 is a flow chart of another method for data interpolation in the absence of radar effectiveness evaluation data disclosed in an embodiment of the present invention. Figure 7 The data interpolation method described in the radar effectiveness evaluation data missing state is applied to the field of big data processing technology to implement missing data interpolation, which is not limited in the embodiments of the present invention. Figure 7 As shown, the data interpolation method for radar effectiveness evaluation in the missing data state may include the following operations:

[0173] Step 1: Obtain the original sample data set and construct a corresponding indicator sample for each sample in the data set;

[0174] Step 2, constructing a forward time stochastic differential equation model, solving the equation under multi-scale noise conditions with the original sample as the initial value, cascading the solution result of each scale with the indicator sample to generate a channel expansion training sample set under different scale noises;

[0175] Step 3, constructing a score neural network of data distribution, and training the score neural network using the channel expansion training sample set under the noise of different scales;

[0176] Step 4, using the trained score neural network to construct a reverse time stochastic differential equation model;

[0177] Step 5: Input the original data to be interpolated into the reverse time stochastic differential equation model to generate complete data after interpolation.

[0178] Optionally, the original sample data set is obtained, and a corresponding indicator sample is constructed for each sample in the data set. For each sample Y=(Y 1 ,…,Y d ), a binary data D with the same dimension as Y = (D 1 ,…,D d ), D i ∈{0,1}, when the i-th dimension Y in the sample Y i When missing, the corresponding dimension D in D i The value is 0 if yes, otherwise it is 1.

[0179] Optionally, a forward time stochastic differential equation model is constructed, and the equation is solved under multi-scale noise conditions with the original sample as the initial value. The solution result of each scale is cascaded with the indicator sample to generate a channel expansion training sample set under different scale noises, specifically:

[0180] The forward time stochastic differential equation model (first differential equation model) is specifically:

[0181] dy(t)=f(y(t),t)dt,g(t)dw

[0182] Among them, y(t) is the observed variable, corresponding to the sample in the original sample data set, t is the time variable, f(y(t), t) is the drift function, T is the total time, g(t) is a single-valued diffusion function, w is a standard Wiener random process variable.

[0183] The equation is solved under multi-scale noise conditions with the original sample as the initial value, specifically:

[0184] y(t n+1 )=y(t n )+f(y(t n ),t n )Δt+g(t n )Δw n

[0185] Among them, y(t n+1 ) is t n+1 The solution of the first differential equation at time y(t n ) is t n The solution of the first differential equation at time Δt = t n+1 -t n , Δw n ~Ν(0,τ n I) is the Gaussian distribution sampling value, I is the unit vector, τ n is the noise scale parameter, N 1 is the number of noise scales, T is the total time, n=0,1,2,…,N 1 ;

[0186] The solution result of each scale is cascaded with the indicator sample to generate a channel expansion training sample set under different scale noises. Specifically, the indicator sample D is used as the scale noise τ n The sample y(t n ) the second channel, generating sample data X(t n )=((y(t n ) 1 ,D 1 ),(y(t n ) 2 ,D 2 ),…,(y(t n ) d ,D d )), X(t n ) is the channel expansion training sample.

[0187] The score neural network is specifically:

[0188] The input of the score neural network is x(t) and t, and the output is an approximation of the derivative of the log-likelihood function

[0189] The score neural network structure is a residual U-type neural network, comprising a downsampling module, an upsampling module and a skip connection module; the upsampling module and the downsampling module each comprise H sub-blocks, each of which comprises a convolutional layer, a dropout layer, a pooling layer, a batch normalization layer, and an activation function layer; the skip connection module comprises H sub-blocks, each of which comprises an identity mapping layer and a dropout layer between the corresponding sub-blocks of the downsampling module and the upsampling module;

[0190] The convolution layer of the residual U-type neural network is specifically: first, the convolution kernel is used to y(t n ) performs a convolution operation, and then processes the convolution result and the indication sample D described in claim 2 using a reversible 1×1 convolution, and the result is used as the output of the convolution layer;

[0191] The dropout layer of the residual U-type neural network is specifically:

[0192] The input of the dropout layer is the output feature map F of the convolutional layer;

[0193] A region is randomly selected in the feature map F, and each coordinate point in the region generates 0 with probability p and generates 1 with probability 1-p;

[0194] Taking the coordinate point where 0 is generated as the center, expand it into an L×L area, set the elements of the feature map to 0 in the directions of 0°, 45°, 90°, 135°, 180°, 215°, 270°, and 315° at the center point of the area, and keep the other elements unchanged; the feature map processed by the above steps is used as the output of the dropout layer;

[0195] The probability of the coordinate point generating 0 or 1 follows a Bernoulli distribution with parameter p.

[0196] Optionally, the channel expansion training sample set under the noise of different scales is used to train the score neural network S θ (x(t), t) is trained, specifically:

[0197] Set up the scoring neural network S θ The training parameters of (x(t), t) are: the number of samples in the mini-batch sample set, Batchsize, the learning rate η, the number of iterations epoch, and the Bernoulli distribution parameter p;

[0198] Training score neural network S θ (x(t),t): From the training sample data set x(t 0 ) and the noise sample set x(t n ) Extract small batch samples and optimize the loss function:

[0199]

[0200] in, represents the weighting function, and U[0,t]] represents the uniform distribution.

[0201] The Adam algorithm is used to update the parameter θ of the score neural network:

[0202]

[0203] The parameter of the optimized score neural network is θ * .

[0204] Optionally, the data interpolation method and device for radar effectiveness evaluation in the absence of data is characterized in that the trained score neural network is used to construct a reverse time stochastic differential equation model, specifically:

[0205] The optimized score neural network is

[0206] The backward time stochastic differential equation (the second differential equation model) is specifically:

[0207]

[0208] in, To optimize the score neural network model, g(t) is a single-valued diffusion function, t is a time variable, w is a standard Wiener random process variable, and x(t) is a channel expansion training sample set.

[0209] Optionally, the original data to be interpolated is input into the reverse time stochastic differential equation model, and Gaussian noise is randomly generated at the missing position of the data to be interpolated y′ to fill the missing position to form the complete data y, which is input into the reverse time stochastic differential equation;

[0210] Optionally, the original data to be interpolated is input into a reverse time stochastic differential equation model, where the reverse time stochastic differential equation model is a discretized stochastic differential equation, specifically:

[0211]

[0212] Optionally, the interpolated data is generated by taking the complete data y as x(t N ) Input the discretized stochastic differential equation, and after N steps of iteration, we get x(t 0 ), and finally x(t 0 ) fills in the missing positions of y′ to generate interpolated data.

[0213] It can be seen that the present invention specifically designs a multi-scale noise random differential equation neural network to estimate the score of the data probability distribution for the interpolation problem of missing data, and constructs a data interpolation model based on this. Compared with other neural network structures and single-scale neural network interpolation methods such as autoencoder networks and generative adversarial networks, it can more accurately fit the probability distribution of missing data. A data missing indication mechanism is introduced in the neural network training process, and a channel expansion training sample set is constructed, so that the neural network can pay more attention to the missing sample data when learning the data probability distribution, thereby further improving the fitting accuracy of the data probability distribution and the accuracy of interpolation data generation. The present invention can interpolate missing data without a complete data set, and has high interpolation accuracy and a wide range of applications.

[0214] Embodiment 3

[0215] See also Figure 8 , Figure 8 The present invention discloses a data interpolation device for radar effectiveness evaluation in the absence of data. Figure 8 The data interpolation device described in the radar effectiveness evaluation data missing state is applied to the field of big data processing technology to implement missing data interpolation, which is not limited in the embodiments of the present invention. Figure 8 As shown, the data interpolation device in the radar effectiveness evaluation data missing state may include the following operations:

[0216] S301, a data acquisition module, used to acquire an original sample data set, where the original sample data set includes N samples, where N is a positive integer;

[0217] S302, an indicator sample construction module, used to process the original sample data set to obtain an indicator sample;

[0218] S303, a data interpolation module, is used to process the original sample data set and the indicator sample to obtain complete data after interpolation.

[0219] Embodiment 4

[0220] See also Fig. 9 , Fig. 9 1 is a schematic diagram of the structure of another data interpolation device for radar effectiveness evaluation in the absence of data disclosed in an embodiment of the present invention. Fig. 9 The data interpolation device described in the radar effectiveness evaluation data missing state is applied to the field of big data processing technology to implement missing data interpolation, which is not limited in the embodiments of the present invention. Fig. 9 As shown, the data interpolation device in the radar effectiveness evaluation data missing state may include the following operations:

[0221] A memory 401 storing executable program codes;

[0222] a processor 402 coupled to the memory 401;

[0223] The processor 402 calls the executable program code stored in the memory 401 to execute the steps of the data interpolation method in the case of missing radar effectiveness evaluation data described in the first and second embodiments.

[0224] Embodiment 5

[0225] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program enables a computer to execute the steps of the data interpolation method in the case of missing radar effectiveness evaluation data described in Embodiments 1 and 2.

[0226] The device embodiments described above are only illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, i.e., they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative work.

[0227] Through the specific description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution can be essentially or partly contributed to the prior art in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0228] Finally, it should be noted that the data interpolation method and device for radar effectiveness evaluation in a data missing state disclosed in the embodiment of the present invention only discloses a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for data interpolation in the absence of radar effectiveness evaluation data, characterized in that: The method comprises: S1, obtaining an original sample data set, wherein the original sample data set includes N samples, where N is a positive integer; S2, processing the original sample data set to obtain an index sample; S3, processing the original sample data set and the indicator sample to obtain complete data after interpolation.

2. The data interpolation method for radar effectiveness evaluation in the absence of data according to claim 1, characterized in that: The processing of the original sample data set and the indicator sample to obtain the complete data after interpolation includes: S31, preprocessing the original sample data set to obtain a preprocessed sample set; S32, using the preprocessing sample set and the index sample, solving a preset first differential equation model to obtain a channel expansion training sample set under noise of different scales; S33, using the channel expansion training sample set to train the preset score neural network model to obtain an optimized score neural network model; S34, processing the optimized score neural network model to obtain a second differential equation model; S35, using the second differential equation model to process the original data to be interpolated to obtain complete data after interpolation.

3. The data interpolation method in the case of missing radar effectiveness evaluation data according to claim 2, characterized in that: The preprocessing of the original sample data set to obtain a preprocessed sample set includes: S311, performing data standardization on the original sample data set to obtain a standardized data set; S312, denoising the standardized data set to obtain a preprocessed sample set.

4. The data interpolation method for radar effectiveness evaluation in the absence of data according to claim 2, characterized in that: The first differential equation model expression is: dy(t)=f(y(t),t)dt,g(t)dw Among them, y(t) is the observed variable, corresponding to the sample in the original sample data set, t is the time variable, f(y(t), t) is the drift function, T is the total time, g(t) is a single-valued diffusion function, w is a standard Wiener random process variable.

5. The data interpolation method in the case of missing radar effectiveness evaluation data according to claim 2, characterized in that: The method of solving a preset first differential equation model by using the original sample data set and the index sample to obtain a channel expansion training sample set under noise of different scales includes: S321, using the original sample data set, solving a preset first differential equation model to obtain a solution to the first differential equation; The expression of the solution of the first differential equation is: y(t n+1 )=y(t n )+f(y(t n ),t n )Δt+g(t n )Δw n Among them, y(t n+1 ) is t n+1 The solution of the first differential equation at time y(t n ) is t n The solution of the first differential equation at time Δt = t n+1 -t n , Δw n ~Ν(0,τ n I) is the Gaussian distribution sampling value, I is the unit vector, τ n is the noise scale parameter, N1 is the number of noise scales, T is the total time, n=0,1,2,…,N1; S322, processing the solution of the first differential equation and the index sample to obtain a channel expansion training sample set under noises of different scales; The channel expansion training sample set expression under different scale noise is: X(t n )=((y(t n )1,D1),(y(t n )2,D2),…,(y(t n ) d ,D d )) Among them, X(t n ) is t n The channel expansion training samples at time, D1,D2,…,D d is the indicator sample, d is the number of indicator samples, y(t n )1,y(t n )2,…,y(t n ) d is the sample in the original sample dataset corresponding to the indicator sample.

6. The data interpolation method in the case of missing radar effectiveness evaluation data according to claim 2, characterized in that: The method of using the channel to expand the training sample set to train the preset score neural network model to obtain an optimized score neural network model includes: S331, setting the scoring neural network model parameters; S332, extracting the channel expansion training sample set to obtain small batch training samples; S333, using the small batch training samples, updating the score neural network model parameters to obtain an optimized score neural network model.

7. The data interpolation method for radar effectiveness evaluation in the absence of data according to claim 2, characterized in that: The second differential equation model expression is: dx(t)=[f(x(t),t)-g 2 (t)S θ* (x(t),t)]dt+g(t)dw Among them, S θ* (x(t), t) is the optimized score neural network model, g(t) is the single-valued diffusion function, t is the time variable, w is the standard Wiener random process variable, and x(t) is the channel expansion training sample set.

8. A data interpolation device for radar effectiveness evaluation in the absence of data, characterized in that: The device comprises: A data acquisition module is used to acquire an original sample data set, where the original sample data set includes N samples, where N is a positive integer; An indicator sample construction module, used to process the original sample data set to obtain an indicator sample; The data interpolation module is used to process the original sample data set and the indicator sample to obtain complete data after interpolation.

9. A data interpolation device for radar effectiveness evaluation in the absence of data, characterized in that: The device comprises: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the data interpolation method in a state where radar effectiveness evaluation data is missing as described in any one of claims 1 to 7.

10. A computer storable medium, characterized in that: The computer storable medium stores computer instructions, and when the computer instructions are called, they are used to execute the data interpolation method in a radar effectiveness evaluation data missing state as described in any one of claims 1 to 7.

Images