CN-121765652-B - Unmanned aerial vehicle group space structure regularity assessment method based on brain-like calculation

Abstract

The invention discloses a brain-like calculation-based unmanned aerial vehicle group space structure regularity assessment method, which belongs to the technical field of unmanned aerial vehicle group space structure assessment, and comprises the steps of calculating optimal pose estimation of an unmanned aerial vehicle group under a global coordinate system, extracting a global position set at a selected moment, constructing three variant matrixes with various distance deviations, calculating a correlation coefficient matrix and an autocorrelation value of each variant matrix, mapping the autocorrelation value into structure regularity, combining a consistency index to calculate a confidence coefficient, constructing an heterogeneous integrated learning framework based on a pulse neural network and a gradient lifting decision tree model, inputting the autocorrelation value into the pulse neural network, inputting the autocorrelation value and the structure regularity into the gradient lifting decision tree model, executing a heterogeneous model fusion strategy to obtain an integrated assessment result, and generating a core output parameter set of an integrated system through system performance verification and assessment.

Inventors

• YAN QIANG
• ZHANG PING
• ZHANG RUI
• ZHOU TONGYU
• Luo Mennan
• Zhou Huaifang
• TAN LIGUO

Assignees

• 西南科技大学

Dates

Publication Date

20260508

Application Date

20260303

Claims (10)

1. 1. The unmanned aerial vehicle group space structure regularity evaluation method based on brain-like calculation is characterized by comprising the following steps of: S1, calculating optimal pose estimation of an unmanned aerial vehicle cluster under a global coordinate system according to multi-source sensor data fusion and a factor graph optimization model, and extracting a global position set at a selected moment; S2, according to the global position set, three variant matrixes of various distance deviations are constructed through Euclidean distance calculation, mean value removal processing and differential operation, and the space structure of the unmanned aerial vehicle group is represented by multiple angles; s3, defining a Pearson correlation coefficient standard, and calculating a correlation coefficient matrix and an autocorrelation value of each variant matrix; S4, mapping the autocorrelation value into structural regularity through an improved Chaddock scale function, calculating confidence coefficient by combining the consistency index, and outputting a core parameter set and a detailed data set for judging the structural regularity; s5, constructing a heterogeneous integrated learning framework based on a pulse neural network and a gradient lifting decision tree model, cooperatively optimizing a training framework, inputting an autocorrelation value into the pulse neural network, inputting an autocorrelation value and structural regularity into the gradient lifting decision tree model, and executing a heterogeneous model fusion strategy to obtain an integrated evaluation result; and S6, based on the integrated evaluation result, generating a core output parameter set of the integrated system through system performance verification and evaluation.
2. 2. The method for evaluating the regularity of the space structure of the unmanned aerial vehicle group based on brain-like calculation according to claim 1, wherein S1 comprises the following sub-steps: s11, setting the cluster scale and the observation time sequence of the unmanned aerial vehicle, defining a single machine state vector and a system full state vector, and integrating multi-source sensor data, wherein the multi-source sensor data comprises odometer observation, RTK global positioning observation and relative observation; S12, constructing objective functions of odometer observation, RTK global positioning observation and relative observation, and building a factor graph model by integrating the objective functions; S13, linearizing an error function at the current estimation position by utilizing a sparse structure of the factor graph model, and solving a normal equation by adopting an incremental optimization algorithm to obtain optimal pose estimation of all unmanned aerial vehicles at all moments; s14, extracting a global position set at a selected moment according to the optimal pose estimation of all unmanned aerial vehicles at all moments.
3. 3. The brain-like calculation-based space structure regularity evaluation method of Unmanned Aerial Vehicle (UAV) according to claim 2, wherein in S11, a single machine state vector is defined And system full state vector The expression of (2) is specifically: in the formula, For the horizontal axis coordinates of the drone i at time t, For the vertical axis coordinates of the drone i at time t, For the vertical axis coordinates of the unmanned aerial vehicle i at time t, For the roll angle of the drone i at time t, For the pitch angle of the drone i at time t, For the yaw angle of the drone i at time t, In order to transpose the symbol, For the size of the cluster of unmanned aerial vehicles, Is the observation time; integrating multisource sensor data, which specifically includes odometer observation Representing unmanned aerial vehicle From the moment of time By the time Relative motion observation of (2), RTK global positioning observation Representing unmanned aerial vehicle At the moment of time Absolute global position coordinates, relative observations of (a) Representing unmanned aerial vehicle Unmanned plane At the moment of time Is a relative position vector of (a); In S12, the expression of the factor graph model is specifically: in the formula, For optimal pose estimation of all unmanned aerial vehicles at all times, As an objective function of the odometer observation, For the RTK to globally localize the observed objective function, As a function of the relative observed object, As a function of the relative observed errors, For the error function of the RTK global positioning observation, As a function of the error observed by the odometer, 、 And The distance of the mahalanobis is indicated, For the covariance matrix of the odometer observation noise, For the covariance matrix of the RTK observation noise, As a covariance matrix with respect to observed noise, Is unmanned plane There is a set of moments of the RTK observations, A set of unmanned aerial vehicle pairs for which there is a relative observation at time t; in the formula, A relative transformation matrix for adjacent moments predicted by the state; in the formula, Is composed of The transformed transformation matrix of the transformation is used, Is composed of A transformed transformation matrix; in the formula, Is composed of The rotation matrix is formed by the components, Is unmanned plane Is provided with a three-dimensional position coordinate of (c), , Representing a group formed by all rigid body transformation matrixes in a three-dimensional space for a three-dimensional special Euclidean group; in the formula, Is unmanned plane Is a three-dimensional position coordinate of (2); In S13, the expression of the normal equation is specifically: in the formula, In the form of a jacobian matrix, As an increment of the parameter(s), For the block diagonal noise covariance matrix, , Constructing a function for the block diagonal matrix for arranging the input matrix parameters along the main diagonal to form the block diagonal matrix, To at the same time An error vector calculated according to the odometer, the RTK and the relative observation, The current estimated value; where e is the total error vector, As a component of the state error, Is an observed error component; The method for solving the normal equation by adopting the incremental optimization algorithm comprises the following steps: s131, establishing a hessian matrix according to the sparse structure of the factor graph model ; In the formula, Corresponding unmanned aerial vehicle And unmanned aerial vehicle Is used to determine the state variable of (1), Is unmanned plane The state variable of the device itself is, By unmanned aerial vehicle The self odometer observation and the RTK global positioning observation are formed, Only when unmanned aerial vehicle And unmanned aerial vehicle There is a relative view between only non-zero during time; S132, establishing a first linear equation by adopting a Levenberg-Marquardt algorithm, wherein the expression of the first linear equation is specifically as follows: in the formula, To take the main diagonal elements of the matrix to construct a diagonal matrix, Is a damping factor; S133, solving the first linear equation to obtain an increment According to the increment by Iterative updating is carried out to obtain the optimal pose estimation of all unmanned aerial vehicles at all moments ; In the formula, As an addition operation on the manifold, Is the first The system full state vector estimate at the time of the iteration, Is the first Estimating a system full state vector value in the next iteration; S14, extracting the selected time Global position set of (a) The expression of (2) is specifically: in the formula, Is unmanned plane At the moment of time Is provided with a three-dimensional position coordinate of (c), , At the moment for unmanned plane i Is set at the transverse axis coordinates of (c), At the moment for unmanned plane i Is defined by the longitudinal axis coordinates of (c), At the moment for unmanned plane i Is a vertical axis coordinate of (c).
4. 4. A method for evaluating the regularity of the space structure of an unmanned aerial vehicle group based on brain-like computation according to claim 3, wherein S2 comprises the following sub-steps: s21, calculating Euclidean distance according to the global position set, extracting ordered neighbor distance, and constructing a first variant matrix; S22, calculating a row average value of the first variant matrix, and constructing a second variant matrix for measuring the distance deviation through mean removal processing; s23, carrying out differential processing on the ordered neighbor distances, and constructing a third variant matrix for enhancing the sensitivity of the local dense structure.
5. 5. The brain-like calculation-based space structure regularity evaluation method of unmanned aerial vehicle group according to claim 4, wherein S21 specifically is: S211, calculating Euclidean distance between any unmanned aerial vehicles according to the global position set, and constructing a symmetrical distance matrix R; in the formula, Is the euclidean distance between unmanned aerial vehicle k and unmanned aerial vehicle l; S212, acquiring distance sets from any unmanned aerial vehicle to all other unmanned aerial vehicles based on a symmetric distance matrix, carrying out ascending arrangement on elements in the distance sets through ordered neighbor distance extraction operation, selecting m values to generate neighbor distance sets, wherein the neighbor distance sets of unmanned aerial vehicle k are obtained through the ordered neighbor distance extraction operation The expression of (2) is specifically: in the formula, For the set of distances of drone k to all other drones, , For an in-order neighbor distance extraction operation, For the distance of drone k to its first nearest neighbor, For the distance of drone k to its second nearest neighbor, To unmanned plane k The distance of the nearest neighbor, , To take the minimum value; S213, constructing a first variant matrix according to the neighbor distance set ; In the formula, Is a relative position vector of the drone k, To unmanned plane N Distance of the near neighbor; In S22, a second variant matrix The expression of (2) is specifically: in the formula, For the first variant matrix Is a line 1 average value of (c), For the first variant matrix Is a line 2 average value of (c), For the first variant matrix Is the first of (2) Line mean, first variant matrix Is the first of (2) Line average The expression of (2) is specifically: in the formula, For the first variant matrix Is the first of (2) Line 1 The column elements are arranged in a row, , Is unmanned plane To the first thereof Distance of the near neighbor; in S23, a third variant matrix The expression of (2) is specifically: in the formula, For the first variant matrix Is the first of (2) A differential vector of column ordering distances; in the formula, For the first variant matrix Is the first of (2) The difference in the rows is used to determine, Is unmanned plane To the first thereof The distance of the nearest neighbor, Is unmanned plane To the first thereof Distance of the near neighbor.
6. 6. The brain-like calculation based space structure regularity evaluation method of unmanned aerial vehicle group according to claim 5, wherein S3 comprises the following sub-steps: s31, calculating a correlation coefficient matrix corresponding to each variant matrix according to the Pearson correlation coefficient based on the three constructed variant matrices, and quantifying the pairwise correlations inside the clusters; S32, aggregating the correlation coefficient matrixes, and calculating the autocorrelation value of each variant matrix to obtain a macroscopic regularity index; S33, analyzing the characteristic of the autocorrelation value, and outputting an autocorrelation value set and a correlation coefficient matrix set; S31, calculating a correlation coefficient matrix corresponding to each variant matrix, and elements of the correlation coefficient matrix The expression of (2) is specifically: in the formula, Is the first The ith row and kth column elements in the variant matrix, Is the first The ith row and first column elements in the variant matrix, Is the first The mean value of the kth column vector in the variant matrix, Is the first The mean value of the first column vector in the variant matrix, m is the number of neighbors; in S32, the autocorrelation values of each variant matrix are calculated The expression of (2) is specifically: in S33, the range of the autocorrelation value is The physical meaning is defined as follows: the autocorrelation value is about equal to 1, and a space structure with a high rule is represented, namely unmanned aerial vehicles are uniformly and orderly distributed; the autocorrelation value is about equal to 0, the random distribution is represented, and no obvious regularity exists; The autocorrelation value is approximately equal to-1, the highly inverse regular distribution is represented, and obvious rejection or aggregation phenomenon exists; And obtaining an autocorrelation value set according to the autocorrelation values of each variant matrix, and obtaining a correlation coefficient matrix set according to the correlation coefficient matrix corresponding to each variant matrix.
7. 7. The brain-like calculation based space structure regularity evaluation method of unmanned aerial vehicle group according to claim 6, wherein S4 comprises the following sub-steps: S41, calculating structural regularity of each variant matrix based on an improved Chaddock scale function, and fusing to obtain comprehensive structural regularity and consistency indexes; s42, determining cluster structure grades according to the comprehensive structure regularity, calculating confidence coefficient according to the comprehensive structure regularity and the consistency index, and outputting a core parameter set and a detailed data set for judging the structure regularity; in S41, structural regularity of each variant matrix is calculated based on the improved Chaddock scale function The expression of (2) is specifically: in the formula, Is a modified Chaddock scale function; Calculating weighted composite structural regularity The expression of (2) is specifically: in the formula, For the first variant matrix structure regularity, For the second variant matrix structure regularity, For the third variant matrix structure regularity, As the regular weight of the basic distance structure, In order to correct the distance structure regularity weight, Regular weights are structured for neighbor distance differences; Calculating consistency index of integrated structure regularity The expression of (2) is specifically: in the formula, As the mean value of the regularity of the variant structure, ; S42, determining the cluster structure level according to the integrated structure regularity The expression of (2) is specifically: Calculating confidence according to the integrated structure regularity and consistency index The expression of (2) is specifically: core parameter set for structural regularity judgment And detailed data set The expression of (2) is specifically: in the formula, In order to be a time stamp, As a result of the first auto-correlation value, As a result of the second autocorrelation value, Is the third autocorrelation value.
8. 8. The brain-like computation based space structure regularity assessment method of unmanned aerial vehicle according to claim 7, wherein in S5, a pulse neural network based on a small world topology is constructed, and a neuron dynamics model of the pulse neural network is defined to simulate a biological neuron discharge characteristic; The pulse neural network comprises an input layer, a hidden layer and an output layer which are sequentially connected, wherein the input layer is provided with 3 neurons and is used for inputting a first autocorrelation value to a third autocorrelation value, the hidden layer comprises a first hidden layer and a second hidden layer which are mutually connected, the first hidden layer is provided with 128 LIF neurons, the second hidden layer is provided with 64 LIF neurons, and the output layer is provided with 5 neurons and is used for outputting a core parameter set for brain-like calculation; The hidden layer adopts a small world network connection mode, and the connection probability is high The expression of (2) is specifically: in the formula, Is a neuron And neurons The functional distance between the two layers, In order for the decay constant to be a function of, As a small-world parameter, the device is provided with a small-world parameter, Is the average connectivity; The expression of the neuron dynamics model is specifically as follows: in the formula, Is the membrane potential, the membrane is a membrane, In order to accommodate the current flow, In the form of a film capacitor, In order to leak the electrical conductance, As a slope factor of the slope of the signal, In order to leak the reverse potential, For the threshold potential to be a threshold potential, In the event of a synaptic current, In order to accommodate the time constant of the current, In order to adapt to the sensitivity coefficient of the current, To accommodate the increase in current after the pulse triggering, For the moment of the release of the pulse, Is a dirac function, used for representing A time-of-day pulse; the working flow of the impulse neural network specifically comprises the following steps: A1, inputting the autocorrelation value into a pulse neural network, and executing a space-time coding mechanism to convert the continuous autocorrelation value into a discrete pulse sequence mode; A2, dynamically updating synaptic weights by adopting a multi-factor pulse time sequence dependent plasticity learning rule so as to extract deep features; A3, performing brain-like decision by combining multi-time scale integration and an attention mechanism, and mapping decision output vectors into category probability distribution through a Softmax function And define the multi-scale feature vector as brain-like features Simultaneously calculating a pulse activity matrix and multi-index confidence coefficient; A4, optimizing network performance based on energy efficiency constraint and noise robustness, and outputting a core evaluation parameter set of brain-like calculation; In A1, the method for executing the space-time coding mechanism specifically comprises the following steps: Three parallel coding strategies are adopted to code the autocorrelation value, the coding strategies comprise rate coding, time coding and phase coding, the coded characteristics are integrated by adopting a distributed group coding strategy, and a pulse sequence mode is output ; In the formula, As a nonlinear mapping function of features to pulse patterns, For the number of encoded neurons per feature, As an item of background noise, In order to encode the weights of the neurons, Is the autocorrelation value of the input; In A2, the method for dynamically updating the synaptic weight by adopting the multi-factor pulse time sequence dependent plasticity learning rule specifically comprises the following steps: calculating a synaptic weight delta, updating the synaptic weight by the synaptic weight delta, wherein the synaptic weight delta The expression of (2) is specifically: in the formula, For the global learning rate of the device, For the pulse timing to depend on the fusion coefficient of the plasticity factor, Is the fusion coefficient of the steady-state plasticity factors, In order to reward the fusion coefficient of the modulation factor, For the STDP weights, Is a steady-state plastic weight that is a weight of plasticity, Modulating weights for rewards; in the formula, For the STDP enhanced amplitude value, For the STDP suppression magnitude value, In order to enhance the time constant of the device, In order to suppress the time constant of the time, As an upper threshold for synaptic weights, In order to update the pre-update synaptic weight, Is the difference between the front neuron pulse time and the back neuron pulse time; in the formula, Is a steady-state plasticity adjustment coefficient, Is a neuron Is used for the optical fiber, the emissivity of (a), Parameters for maintaining a target emissivity; in the formula, A dynamic reward signal based on classification accuracy; a3 is specifically: (1) Performing multi-time scale integration on pulse activity of output layer neurons to obtain output vector of brain-like decision ; In the formula, And The short-term and long-term integration windows respectively, For the short-term integration weight, For the long-term integration weights, For the purpose of issuing a function of the pulses, Is an integral variable; (2) Output vector pair using Softmax function Normalization processing is carried out, and class probability distribution is output ; In the formula, Representing the class probability distribution of the impulse neural network output, Representing the normalized exponential function of the sample, As a function of Softmax (r), To the output layer The integrated output of the individual neurons, To the output layer The integrated output of the individual neurons, The number of neurons in the output layer; Output vector by maximum value indexing method Decoding to obtain the structure level of the output of the impulse neural network ; In the formula, Extracting an index j at which the function takes a maximum value; (3) Pulse activity on hidden layer neurons using a multiscale integration strategy Respectively performing pulse counting integration on time windows with different lengths to obtain Integration results for different time scales, integration results for the kth time scale The expression of (2) is specifically: in the formula, In order to correspond to hidden layer neuron at moment in the kth time scale Is used to pulse the state vector, , At time of day for hidden layer 1 neuron Is used to pulse the state vector, At time of day for hidden layer 2 neurons Is used to pulse the state vector, At time of day for hidden layer H neuron Is used to pulse the state vector, Is the first Integrating window widths corresponding to the time scales; vector stitching is carried out on all integration results to obtain a multi-scale feature vector ; In the formula, As a result of the integration on the 1 st time scale, As a result of the integration on the 2 nd time scale, Is the first Integration results on a time scale; (4) Dynamically constructing a pulse activity matrix through an attention nerve modulation mechanism for realizing the selective enhancement of specific characteristic channels and constructing the pulse activity matrix The expression of (2) is specifically: in the formula, Is the first Individual neurons are in The pulse activity at the moment modulates the intensity, Is the first Individual neurons are in The pulse activity at the moment modulates the intensity, Is the first Individual neurons are in The pulse activity at the moment modulates the intensity, Is the first Individual neurons are in The pulse activity at the moment modulates the intensity, In order to perform the diagonalization operation, For Sigmoid activation functions, the calculation result is mapped to the (0, 1) interval, representing the modulation intensity, To pay attention to a weight matrix that can be learned in a subnetwork, To pay attention to the learnable bias in the subnetwork, Is the first Average firing rate of individual neurons over a recent time window; in the formula, , At time for hidden layer ith neuron A pulse issuing state vector of (a); (5) Calculating multi-index confidence including synchronicity, entropy and stability The expression of (2) is specifically: in the formula, As an index of the synchronicity of the data, As an index of the entropy of the liquid, Is a stability index; in the formula, As a reference signal, a reference signal is provided, In order to output the signal, In order to output the standard deviation of the signal, Is the standard deviation of the reference signal; in the formula, In order to output the shannon entropy of the distribution, Is a category probability distribution vector; in the formula, Is the output probability distribution at the time t, The probability distribution is output at the previous moment; in A4, the method for optimizing the network performance based on energy efficiency constraint and noise robustness specifically comprises the following steps: constructing an optimized objective function based on energy efficiency constraint, which is used for improving network performance and optimizing the objective function The expression of (2) is specifically: in the formula, As a result of the total energy consumption, For the energy consumption of a single pulse, Is a neuron Is used for the optical fiber, the emissivity of (a), In order to achieve a leakage current energy consumption rate, For the transmission of the energy consumption coefficient for the synapse, As a result of the synaptic weight, Is a front neuron Is a ratio of emissivity of (2); noise robustness processing is introduced to enhance stability of the network in an uncertain environment, noise robustness processing The expression of (2) is specifically: in the formula, For the original input to be entered, As a result of the gaussian noise, In order to distribute the noise uniformly, In order to even the amplitude of the noise, Is the standard deviation of Gaussian noise; Core evaluation parameter set for output brain-like calculation The expression of (2) is specifically: in the formula, For the structural level of the SNN output, In the case of a multi-scale feature vector, Is a pulse active matrix; In S5, the workflow of the gradient lifting decision tree model is specifically: b1, collecting autocorrelation values And structural regularity determination parameters Performing dimension alignment and splicing to construct an input feature vector of a gradient lifting decision tree model The expression is specifically as follows: B2, constructing decision trees sequentially, and each new tree is specially used for correcting the residual error of the previous tree, so as to optimize an objective function The expression of (2) is specifically: in the formula, Is unmanned plane Is used for the prediction of the loss of (a), Is the first The regularized term of the tree is then generated, Unmanned plane for integrated model Is provided with a prediction output of (c) for the prediction, Is unmanned plane K is the total number of decision trees, Is unmanned plane Is a feature vector of the input of the (a); the passing degree lifting decision tree model adopts an ordered lifting strategy to avoid target leakage, and a gradient estimation formula is as follows: in the formula, Is unmanned plane Is used for the prediction of the loss of (a), Unmanned plane for integrated model Is provided with a prediction output of (c) for the prediction, Is unmanned plane Is used to determine the input feature vector of (a), Is unmanned plane Is a real tag of the (c) in the (c), For the purpose of the gradient operator, And A random arrangement sequence of training samples; B3, inputting feature vector Input to the slave Decision tree In the integrated model formed, for each decision tree Defining mapping functions by utilizing internal splitting criteria and topological structure Mapping a continuous high-dimensional feature space to a discrete leaf node index space Wherein Is the first The total number of leaf nodes of the tree is input with feature vectors Leaf node index activated in each tree Uniquely determined by the following formula: B4, generating corresponding leaf node indexes output by each decision tree Sparse-dimensional single-hot coding vector The first of the vectors Individual elements The value rule of (2) is defined as if and only if In the time-course of which the first and second contact surfaces, Otherwise ; B5, will Single-hot coding vector generated by decision tree Longitudinal vector splicing is carried out according to the generation sequence of the decision tree, and final statistical characteristics are synthesized ; In the formula, Is the first The one-hot encoding vector of the decision tree, Is the first The one-hot encoding vector of the decision tree, Is the first And (5) a single-hot coding vector of the decision tree.
9. 9. The unmanned aerial vehicle group space structure regularity assessment method based on brain-like calculation according to claim 8, wherein in S5, a heterogeneous model fusion strategy is executed through a heterogeneous integrated learning framework, for the brain-like features output by a pulse neural network and the statistical features output by a gradient lifting decision tree model, fusion feature vectors are generated by splicing the brain-like features and the statistical features at feature levels, a decision level fusion mechanism is executed at a decision level, the output probabilities of the pulse neural network and the gradient lifting decision tree are subjected to weighted fusion to obtain final probability distribution, and the fusion feature vectors and the final probability distribution are used as integrated assessment results; wherein feature vectors are fused The expression of (2) is specifically: in the formula, Is a brain-like feature, and is characterized by, Is a statistical feature; Final probability distribution The expression of (2) is specifically: in the formula, For the class probability distribution of the impulse neural network output, The class probability distribution output for the gradient-lifting decision tree model, Is a dynamic fusion weight; in the formula, In order to adjust the parameters of the device, For the classification accuracy of the impulse neural network on the validation set, The classification accuracy of the decision tree model on the verification set is improved for the gradient; In S5, the method for collaborative optimization of the training framework specifically comprises the following steps: (1) The method comprises a training strategy of two stages, an independent pre-training stage, a combined fine tuning stage, a gradient lifting decision tree model and a verification set performance adjustment stage, wherein the training strategy of the two stages is that a pulse neural network performs non-supervision pre-training through an STDP rule, and a gradient lifting decision tree model performs supervision training on traditional characteristics; (2) Defining an overall loss function for collaborative optimization training ; In the formula, Predictive classification loss for gradient-enhanced decision tree models, In order to pulse the composite loss of the neural network, Outputting a consistency penalty term of probability distribution for the heterogeneous model for restricting the consistency of the SNN branch and CatBoost branch decision logic, The weights are adjusted for the SNN losses, The weights are adjusted for consistency loss.
10. 10. The method for evaluating the regularity of the space structure of the unmanned aerial vehicle group based on brain-like calculation according to claim 9, wherein S6 comprises the following sub-steps: S61, according to the confidence level And final probability distribution Calculating a final confidence level The expression of (2) is specifically: in the formula, For the set weight coefficient(s), , To take the maximum probability value in the vector; S62, calculating a final cluster structure grade according to the final probability distribution, wherein the final cluster structure grade is obtained The expression of (2) is specifically: in the formula, For the final probability distribution Corresponds to the first The probability components of the individual structure levels, As a function of the index of the largest element, In the form of a discrete level space, Represents a high degree of structuring of the structure, It is indicated that a good structuring is achieved, Represents a medium level of structuring and, Indicating that the structure is weak and that the structure is weak, Representing no structure; S63, based on the final cluster structure level, verifying and calculating the overall accuracy through comparing the true value label and the system accuracy The expression of (2) is specifically: in the formula, In order to test the total number of samples, To test a sample Is used to determine the final cluster structure level of the (c), To test a sample Is used to compare the true value label of the (c) with the reference value label of the (c), Is a kronecker function; s64, in order to quantify the determination degree of the evaluation result, the final probability distribution Performing shannon entropy calculation to calculate shannon entropy value The expression of (2) is specifically: S65, introducing variance into the input data as Gaussian noise of (a) Observing the fluctuation condition of the final confidence coefficient, and the robustness stability index The expression of (2) is specifically: in the formula, To introduce a final confidence in the integrated system after gaussian noise, Final confidence level of the integrated system under the original input data; S66, when the whole accuracy rate Entropy value of shannon And robustness stability index When the preset threshold value is above, judging that the performance verification is passed, and further generating a core output parameter set of the integrated system ; 。

Description

Unmanned aerial vehicle group space structure regularity assessment method based on brain-like calculation Technical Field The invention belongs to the technical field of space structure evaluation of unmanned aerial vehicle groups, and particularly relates to a space structure regularity evaluation method of an unmanned aerial vehicle group based on brain-like calculation. Background Under a complex environment, the unmanned aerial vehicle group becomes a core strength for target acquisition and situation awareness by virtue of the distributed cooperation advantage and flexible formation characteristic. However, the existing unmanned aerial vehicle cluster cooperation technology still faces the following significant drawbacks in practical application: First, there is a lack of stable global spatial references in GNSS rejection environments. The conventional model is highly dependent on absolute position information provided by the global positioning system (GPS/GNSS). Under urban canyons, underground spaces or strong electronic interference environments, satellite signal loss causes that unmanned aerial vehicle clusters cannot acquire accurate coordinates, and a cooperative control strategy fails due to the loss of uniform space references, so that situation perception sharing and decision consistency of large-scale clusters are severely restricted. Second, the prior art lacks an objective assessment means for the regularity of the spatial structure of the clusters. At present, the evaluation of the uniformity, regularity and repeatability of the unmanned aerial vehicle group configuration is still mostly carried out in the stage of artificial vision judgment or simple geometric centroid calculation. The method is strong in subjectivity, single in dimension, free of scale, rotation and translation invariance, and difficult to quantitatively describe the spatial arrangement quality of the clusters in a complex maneuvering process, so that the control system lacks an effective closed loop feedback index. Thirdly, the accuracy and the robustness of the multi-source observation data fusion are insufficient. In processing high-speed, non-cooperative targets, stand-alone sensor observations are subject to occlusion and noise interference. When the traditional algorithm processes highly nonlinear motion, the problems of space-time dislocation and observation heterogeneity are often faced, so that the estimation deviation of the target state is large, and the high-precision collaborative capturing task is difficult to support. Fourth, decision algorithm real-time and security under complex constraint are unbalanced. The traditional method generally adopts a step-by-step optimization or rule base, and when multiple conflict constraints such as rapid surrounding, obstacle avoidance, collision avoidance, target visibility and the like are faced, the calculated amount grows exponentially along with the cluster scale, so that the unmanned aerial vehicle is extremely easy to fall into a decision-making dead office of 'oscillating loitering' or 'path blocking', and millisecond-level real-time response cannot be realized at an embedded terminal. Disclosure of Invention Aiming at the defects in the prior art, the unmanned aerial vehicle group space structure regularity evaluation method based on brain-like calculation solves the following technical problems faced by the existing unmanned aerial vehicle group cooperation technology in practical application: (1) A stable global spatial reference is lacking in GNSS rejection environments. (2) There is no objective assessment means for the regularity of the spatial structure of clusters. (3) The accuracy and robustness of the multi-source observation data fusion are insufficient. (4) The real-time performance and the safety of the decision algorithm under the complex constraint are unbalanced. In order to achieve the aim of the invention, the technical scheme adopted by the invention is that the method for evaluating the regularity of the space structure of the unmanned aerial vehicle group based on brain-like calculation comprises the following steps: S1, calculating optimal pose estimation of an unmanned aerial vehicle cluster under a global coordinate system according to multi-source sensor data fusion and a factor graph optimization model, and extracting a global position set at a selected moment; S2, according to the global position set, three variant matrixes of various distance deviations are constructed through Euclidean distance calculation, mean value removal processing and differential operation, and the space structure of the unmanned aerial vehicle group is represented by multiple angles; s3, defining a Pearson correlation coefficient standard, and calculating a correlation coefficient matrix and an autocorrelation value of each variant matrix; S4, mapping the autocorrelation value into structural regularity through an improved Chaddock scale function, calculating