50 results about "Matrix iteration" patented technology

The invention discloses a harmonic traceability method based on a harmonic power direction of an electric power system, which belongs to the technical field of electric power quality of the electric power system and comprises the following steps of: constructing matrixes of fundamental waves and harmonic signals, initializing active and reactive power deviation values of nodes, and adopting a decoupling algorithm combined with a Jacobian matrix iteration method to carry out harmonic traceability. And sequentially calculating the fundamental wave voltage and each harmonic current, and finally obtaining the power flow direction and judging the harmonic source attribute. When harmonic power flows from a load to a power grid, the node is identified as a harmonic source; and when the harmonic power flows to the load from the power grid, identifying the node as a harmonic receiving point. An equivalent circuit model is used for separating harmonic influences of a linear load and a non-linear load, and a real-time monitoring technology is combined to realize dynamic tracking and accurate positioning of a harmonic source. The method is low in calculation complexity, excellent in convergence performance and suitable for a modern power distribution network scene where distributed energy and power electronic equipment are widely connected, the harmonic suppression efficiency is remarkably improved, and the safety and stability of power grid operation are ensured.

The invention relates to the technical field of aircraft simulation training, and discloses a multi-mode man-machine interaction aircraft simulation training system. A multi-modal feature extraction module of the system generates a multi-modal feature map including a flight environment situation map and an equipment operation health map through a hierarchical feature extraction network; the incremental analysis module divides an incremental data set, updates flight state prediction model parameters through an incremental clustering algorithm and outputs an incremental state prediction result; the parameter optimization layer initializes a tabu search population according to a prediction result, and iteratively optimizes parameters of an operation rule base through a dynamic efficiency evaluation matrix; the decision generation module fuses the optimized rule base parameters and the real-time multi-modal feature map, and generates a flight operation instruction sequence based on dynamic confidence score; and the execution optimization layer analyzes the instruction sequence, satisfies constraint conditions through a constraint projection algorithm, and completes instruction sequence integer programming in combination with the operation portrait deviation. The system improves the interaction accuracy and scene adaptability of simulation training.

The invention discloses an underwater target tracking method based on multivariable deflection Laplace distribution modeling, and the method comprises the steps: collecting the motion state data of an underwater target through a sonar sensor, and obtaining the target position measurement information through a coordinate system conversion mode; the method comprises the following steps: establishing a motion model of an underwater target, determining a state-space equation of the underwater target, analyzing mathematical characteristics of underwater noise, and establishing a measurement model of the target under non-Gaussian noise; the method comprises the following steps: modeling non-Gaussian noise on the basis of multivariable skew lapace distribution, solving a mixed parameter, a shape parameter and a scale matrix of the noise under a variational Bayesian framework, iteratively updating a target state and a noise covariance matrix, iteratively updating an estimated target motion state by using Kalman filtering, and after the number of iterations is reached, calculating the target motion state according to the estimated target motion state. And outputting the estimated values of the underwater target position and speed and the estimated value of the covariance matrix. The method has better robustness and estimation precision, does not need to select a degree-of-freedom parameter, and can be better applied to target tracking.

The invention discloses a multi-robot path planning method for an electric arc additive manufacturing system. The multi-robot path planning method comprises the specific steps that a three-dimensional model is obtained, slicing and layering are conducted, environment information is obtained, and environment modeling is conducted; determining an ideal task load; performing matrix iteration, and optimizing the current task load; constructing a minimum spanning tree, and generating a two-dimensional optimal task path of each robot; and performing three-dimensional conversion and asynchronous execution. Compared with the prior art, the method has the advantages that the form requirement of an actual additive manufacturing path is fully considered, the process defects of air holes, insufficient fusion, slag inclusion and the like in the manufacturing process are effectively reduced, and the problems of region segmentation and deposition path planning in an electric arc additive manufacturing (WAAM) task cooperatively executed by multiple robots are successfully solved; and therefore, high-quality additive manufacturing of the component is realized. Meanwhile, the path design strategy proposed in the method avoids the start-stop operation of the welding gun, ensures the parallelism and independence of each robot path in the three-dimensional space, and further improves the operation efficiency and stability of the system.

The invention is suitable for the technical field of load prediction, and provides an electric vehicle charging load spatio-temporal distribution prediction method considering multi-factor interaction influence, and the method comprises the steps: obtaining the initial state information of a preset number of electric vehicles; calling a preset travel probability matrix corresponding to the vehicle type of the electric vehicle, and generating a travel target position of each vehicle by using a Monte Carlo method; calculating the shortest path, the driving speed and the driving time of each vehicle from the initial position to the travel target position; determining the residual electric quantity, the current position and the current time of each vehicle; when the residual electric quantity is lower than a preset threshold value, selecting an optimal charging station and recording charging information in combination with charging pile distribution at the current position, real-time capacity of power grid nodes and charging waiting time; generating a charging power space-time distribution matrix; and outputting a final charging load space-time distribution prediction result when iteration meets a convergence condition. And through dynamic coupling modeling and Monte Carlo simulation, the prediction precision of the time-space distribution of the charging load is improved.

The invention relates to the field of precision machining, and discloses a thin-walled workpiece machining deformation prediction and optimization method under multiple working conditions. Comprising the following steps: setting a fixed constraint and an initial environment temperature by constructing a finite element model containing an initial residual stress field; the cutting effect of the tool is equivalent to a milling force-temperature load, a stress and a temperature field of a tool-workpiece contact area are calculated, and material removal is simulated by using a life-death unit technology; based on a Python script, establishing a space mapping algorithm of tool location point coordinates and grid units to obtain a tool path file; a working condition combination matrix of a cutting mode-machining sequence is constructed, and multi-working-condition simulation is conducted on nine machining strategies; a deformation prediction matrix of multiple working conditions is constructed, and a minimum deformation processing strategy is output through iterative optimization; the machining deformation can be obviously reduced by preferably selecting the following peripheral mode and the cross machining sequence. The method solves the problems of prediction and optimization of machining deformation of the thin-walled workpiece under multiple working conditions, and contributes to key technical support for the field.

The invention discloses a partial mark feature selection method based on particle ball prototype disambiguation and a related device, and the method comprises the steps: obtaining a data set, constructing initial coarse-grained particle balls, dynamically splitting the coarse-grained particle balls, and calculating the feature value of each divided particle ball and the mark of each particle ball; calculating a prototype vector of each mark of all the pellets under the mark, and calculating the similarity between the feature vector of each sample and the prototype vector; determining a weight voting matrix based on the similarity ratio, and iteratively updating the marked space confidence coefficient based on the weight voting matrix to obtain a marked space confidence coefficient matrix; introducing a global mark distribution matrix, establishing a disambiguation-feature selection bidirectional optimization model, and constructing constraint terms to be embedded into the model; and carrying out end-to-end global optimization on the model by adopting an alternating iteration updating strategy, and outputting a feature sorting result to obtain a feature subset with the highest identification capability.

The invention discloses a pulse signal source tracking method based on a robust Kalman filter. The method comprises the following steps: reading a target state vector and error covariance matrix at a previous moment, and observing a noise covariance matrix; a variational Bayesian parameter before iteration, a target state vector and an error covariance matrix are initialized; starting iteration, and calculating the quadratic form sufficient statistics of the observation error and the forecast error; updating variational Bayesian parameters; updating the forecast error covariance matrix and the observation noise covariance matrix; calculating a target state vector and an error covariance matrix of the iteration; when the variation of the target state vector of the current iteration and the target state vector of the last iteration is smaller than the tolerance, ending the iteration, and outputting the target state vector at the current moment, the error covariance matrix and the observation noise covariance matrix; compared with a deterministic integral sampling method, the method provided by the invention adopts adaptive volume sampling to realize higher tracking precision under lower calculation cost.

The embodiment of the invention provides an unknown cross-site script security detection method and device generated based on a test script, and intelligent construction of the test script is realized through innovatively constructing a script generation mechanism driven by a Markov chain and through atomic element sequence analysis and a random walk strategy. And designing a DOM tree difference-based filtering analysis mechanism, and establishing a filtering rule identification strategy in combination with multi-dimensional mark extraction and node positioning. And a matrix iterative optimization mechanism is introduced, and self-adaptive optimization of the test script is realized through dynamic updating and convergence calculation of a state transition matrix. According to the method, the defects of the traditional technology in the aspects of script generation, filtering analysis, optimization adjustment and the like are effectively overcome, and the unknown cross-site script detection effect is remarkably improved.

The present invention discloses a method and system for blur removal and jitter correction in visual monitoring of super high-rise buildings, comprising the following steps: a visual sensor acquires a video stream at a fixed frame rate, and locates an initial region of interest (ROI) by mapping prior information; the spatial gradient field is calculated for the ROI image of the current frame, and a Gaussian-weighted two-dimensional gradient structure tensor matrix is ​​constructed; for the clear image sequence that passes blur detection, variational mode decomposition is then used in the time domain to decouple the displacement time series into eigenmode functions of different frequencies, and the low-frequency components are reconstructed to preserve the true deformation of the building; the reconstructed low-frequency displacement signal is output as the final monitoring result. The present invention employs a lightweight algorithm design throughout, significantly reducing the matrix operation dimension of subsequent processing through a dynamic ROI clipping mechanism; the structural tensor eigenvalues ​​are solved using a direct algebraic analytical method, avoiding complex matrix iterative decomposition.

The application discloses a self-iteration-based deep subspace training method and device, which comprises the following steps: a pre-training step, in which a first optimization function is minimized to train a deep subspace network, so as to initialize a similarity matrix; a training step, in which when the iteration number t is less than a first iteration number, a self-expression coefficient matrix is fixed, the second iteration number is iterated, the deep subspace network parameters are updated by minimizing a second optimization function; the deep subspace network parameters are fixed, the third iteration number is iterated, the self-expression coefficient matrix is updated by minimizing a third optimization function, the similarity matrix generated in the tth iteration is updated according to the updated self-expression coefficient matrix, and finally the update of the similarity matrix in the first t iterations is realized; and the iteration number t is increased by 1, and the training step is repeated. The application introduces the idea of self-iteration, supervises the training of the model through the results of the previous training, and can discard the previous training results, so that the memory and the calculation resources are saved.

This application discloses a method, device, and medium for reconstructing the absorption coefficient of biological tissue, relating to the field of tissue optical parameter measurement. The method includes: segmenting a time-resolved analog signal and a measured signal according to a preset time window segmentation rule; constructing a depth-time window coupling weight matrix based on the time window and photon path information; determining the overall error based on the simulated value sequence, the measured value sequence, and the depth-time window coupling weight matrix; determining whether the overall error meets a preset iteration termination condition; if not, updating the sensitivity matrix based on the photon path information, and iteratively updating the current absorption coefficient using the depth-time window coupling weight matrix and the sensitivity matrix; determining a new time-resolved analog signal based on the updated absorption coefficient and photon path information; and continuing until the preset iteration termination condition is met. This application can improve the reliability, accuracy, and practicality of absorption coefficient reconstruction.

The invention discloses a pile head spring stiffness matrix iteration method for jacket platform seismic analysis, and the method comprises the steps: building a three-dimensional structure model of a jacket platform, simulating a seismic load in a manner of applying an acceleration field, carrying out the structural static analysis calculation, building a jacket structure analysis model copy, and judging the reasonability of a pile head spring stiffness matrix. After a reasonable pile head spring stiffness matrix is obtained, the model can be adopted to complete the subsequent structural strength check of the jacket platform under the action of the seismic load. According to the method, in the process of determining the pile head spring stiffness matrix through jacket platform earthquake analysis, designers do not need to determine the size and the application position of a concentrated point load according to experience, and the problems of calculation complexity, efficiency and the like are solved. According to the method, the seismic load is simulated by applying the acceleration field to the platform model, iterative calculation can be completed more conveniently, more accurately and more quickly, a reasonable and accurate pile head spring stiffness matrix is generated, and then seismic analysis calculation of the jacket platform is achieved.

Embodiments of the present disclosure disclose a self-adaptive single-bit quantization SAR imaging method and device, a storage medium and a product. The method comprises: iteratively reconstructing an image matrix, in each iteration, transforming the image matrix to an echo domain based on a threshold matrix, an original echo signal matrix and an inverse imaging operator, performing single-bit quantization on the image matrix in the echo domain to obtain an echo residual; performing matched filtering on the echo residual using an imaging operator and then performing gradient descent to obtain a gradient descent result matrix; performing threshold shrinkage on the gradient descent result matrix based on a weight matrix and a fractional order sparse norm to obtain an updated image matrix; updating the weight matrix based on the updated image matrix; updating the threshold matrix using a delta modulation algorithm; and obtaining a SAR imaging result when a maximum number of iterations is reached. The method can compensate for the loss of amplitude information and improve the SAR imaging accuracy by adaptively updating the threshold and the fractional order sparse norm constraint while only retaining the signs of the echo signals.

The invention discloses a multi-constraint-based anti-parallel DNA triplex design method and system, and relates to the technical field of DNA triplex targeting genes, and the method comprises the following steps: constructing an mCGR initialization model, adjusting a matrix iteration rule according to a sequence length and a basic constraint, and generating all possible sequence sets satisfying the constraint; mapping the sequence set into flow network nodes, screening out approximate nodes through similarity search, constructing to-be-screened edges, then performing sequence comparison on the to-be-screened edges, only reserving the edges of the sequence within a preset stem length range, and deleting the edges beyond the range, so as to overcome the problem of high complexity of mCGR during combined constraint calculation; and finally, outputting a DNA triplex coding sequence meeting all constraint conditions. According to the method, the anti-parallel DNA triplex TFO sequence meeting multiple constraints can be efficiently generated, the sequence set scale is large, the stability is high, the crosstalk rate is low, and a key technical support is provided for research and development of nucleic acid drugs in the field of targeted gene therapy.

The invention discloses a variable cycle combination method for multiple oil displacement systems based on a gradient boosting tree algorithm. The method comprises the following steps: S1, constructing a variable cycle combination large-scale learning sample of the oil displacement system; and S2, establishing a gradient boosting tree prediction model. According to the gradient boosting tree algorithm, a plurality of weak decision trees are integrated to gradually optimize model errors, flexibility and interpretability are kept, meanwhile, high robustness and over-fitting resistance are achieved, a plurality of weak learners can be weighted and combined in each step, and high prediction capacity and precision are achieved. According to the invention, on the basis of the gradient boosting tree algorithm, mining learning is carried out on large-scale learning samples by designing an appropriate deep neural network, the internal law of the numerical calculation process can be captured, and then time-consuming large-scale matrix iteration in numerical calculation is replaced; accurate prediction solving of the variable circulation oil displacement process of multiple oil displacement systems is achieved while calculation consumption is remarkably reduced.

The present invention proposes a hybrid precoding full connection structure and its beamforming method; the present invention proposes a new hybrid precoding structure, which can further utilize the advantages of the system structure, simplify the hardware structure, and compared with the relatively mature MO iterative optimization method, does not require complex matrix iterative optimization, and the algorithm is simple; the solution of the present inventor is not limited to the performance index of the system spectral efficiency, and can directly approximate the target vector, having more general versatility.

The application relates to the technical field of aircraft simulation training and discloses a multi-modal man-machine interaction aircraft simulation training system. A multi-modal feature extraction module of the system generates a multi-modal feature map containing a flight environment situation graph and a device operation health graph through a hierarchical feature extraction network; an incremental analysis module divides an incremental data set, updates flight state prediction model parameters through an incremental clustering algorithm, and outputs an incremental state prediction result; a parameter optimization layer initializes a tabu search population according to the prediction result, iteratively optimizes operation rule library parameters through a dynamic performance evaluation matrix, a decision generation module fuses the optimized rule library parameters and real-time multi-modal feature maps, generates a flight operation instruction sequence based on a dynamic confidence score, an execution optimization layer analyzes the instruction sequence, satisfies constraint conditions through a constraint projection algorithm, and completes integer programming of the instruction sequence in combination with operation image deviation. The system improves the interaction accuracy and scene adaptability of simulation training.

The invention discloses an array mutual coupling self-correction DOA method based on sparse off-network, and belongs to the field of array signal processing. The method comprises the following steps: constructing a sparse off-network DOA estimation system model considering an array mutual coupling effect; initializing a mutual coupling coefficient vector and an off-grid parameter vector, and reconstructing a signal sparse matrix by using a smooth norm according to a system model; estimating an off-network parameter vector according to the reconstructed signal sparse matrix and a known initialized (or after last iteration) mutual coupling matrix; estimating a mutual coupling matrix from a system model by using a method based on gradient descent by using the reconstructed signal sparse matrix and the estimated off-network parameter vector; and after iteration meets a stop condition, obtaining DOA estimation of the signal according to the signal sparse matrix, the mutual coupling coefficient matrix and the off-network parameter vector after iteration. According to the method, a better DOA estimation effect can be obtained on the premise that the array antenna has an unknown mutual coupling effect.

This disclosure relates to the field of multiple-input multiple-output (MIMO) systems, and proposes a signal detection method, apparatus, and device for MIMO systems. The method includes: determining an iterative weight matrix based on an iterative reference variance, a system channel matrix, and system noise power; performing Newton iteration on the iterative weight matrix to generate an iterative covariance matrix; determining the iterative distribution variance based on the iterative covariance matrix; determining the iterative distribution expectation based on the iterative covariance matrix, an iterative reference expectation, the system output signal, the channel matrix, and noise power; updating the iterative reference expectation and iterative reference variance according to the iterative distribution expectation and iterative distribution variance, and proceeding to the next iteration cycle; when the iteration reaches a preset cycle, determining the iterative distribution expectation as the target distribution expectation, which is used to estimate the system's input signal. The technical solutions provided by one or more embodiments of this disclosure can reduce the computational complexity of the detection process and ensure the accuracy of the signal detection results.

This invention discloses a five-dimensional seismic data reconstruction method based on preconditioned Riemann gradient descent, belonging to the field of seismic data processing technology. The method involves performing a Hankel transform on fixed-frequency four-dimensional seismic data to construct a fourth-order block Hankel matrix; performing hard thresholding on the fourth-order block Hankel matrix to obtain a low-rank approximation matrix, and calculating the Euclidean gradient at the low-rank approximation matrix in the current iteration; calculating a preconditioner based on the diagonal part of the Euclidean gradient outer product; calculating the projection of the Euclidean gradient onto the tangent space of the Riemann manifold based on the Euclidean gradient and the preconditioner, and updating the low-rank approximation matrix based on the projected gradient to obtain the low-rank approximation matrix for the next iteration; iterative updates continue until a termination condition is met; the low-rank approximation matrix from the last iteration is inversely transformed into a fourth-order tensor, and the reconstructed five-dimensional seismic data is obtained based on the fourth-order tensor and the frequency components in the five-dimensional seismic data. This method can improve the reconstruction efficiency and quality of five-dimensional seismic data.

A method for iteratively generating a matrix of base elements that includes forming at least one base matrix, applying modification operators iteratively, which are able to transform the base matrix or any matrix arising therefrom into a modified matrix, applying expansion operators iteratively, which form a larger matrix from a plurality of optionally modified smaller matrices from the preceding iteration by copying, rotation or reflection by virtue of parts of the larger matrix being filled with the optionally modified smaller matrices, performing virtual experiments within the scope of which the properties of a created matrix are examined by systematic creation of values deliberately containing errors from which an error signal is derivable, and in order to create a complex matrix, forming all permutations of next larger matrices by applying expansion operators and then evaluating by means of virtual experiments, the next larger matrices selected with the smallest error signals, and then forming the next larger matrices successively therefrom by applying expansion operators and evaluating by means of virtual experiments until the error signal drops below a given limit.

The invention relates to a mathematical modeling method for FPC contact deformation based on segment analysis and solution of nonlinear equation groups, which is used for analyzing the deformation of a flexible printed circuit (FPC) when it contacts with other components. The core steps include: dividing the FPC into different segments based on contact points; linearizing the relationship between the curvature of the FPC and the length of a path along the surface of the FPC in different FPC segments, and obtaining a series of unknowns, such as the slope and intercept in the linear relationship; in addition, the specific coordinates of the contact points and the length of each FPC in different segments are also used as unknown parameters to be solved; the known relationships include the coordinate difference and angle difference at the two ends of different segments, and the curvature at the connecting points of the head and the tail of different segments must be equal; and at the same time, using the Jacobian matrix iteration method and the size iteration method to solve all unknowns, and finally obtaining the deformation curves of the FPC under different contact conditions with other components that need to be solved.

The application discloses a SOMP off-grid direct positioning method with dictionary matrix iterative optimization, and belongs to the technical field of passive positioning. The application uses a uniform linear array as a base station receiving device, monitors a signal source by using a distributed base station structure, and collects signal source receiving data. The application carries out noise reduction preprocessing on the signals received by the array, and then uses the noise-reduced signals to carry out matching search on a spatial grid to obtain the position of the signal source. The application uses a Taylor iterative optimization dictionary matrix in a matching process, and reduces the compensation deviation of a traditional off-grid algorithm. Compared with the traditional off-grid algorithm, the estimation result of the application is more accurate, and the application has important application value.

The present invention discloses a beam tracking method and system based on an orthogonal dual-frequency uniform linear array. The method comprises: initializing base station parameters and kinematic parameters, wherein the kinematic parameters include the velocity vector and position vector of the target, and the base station includes an orthogonal dual-frequency linear antenna array; performing two-dimensional projection on the kinematic parameters to calculate the echo signal reflected by the target received by the base station; performing matrix modeling on the echo signal received by the base station in each coherent processing interval to obtain a multi-dimensional matrix of the echo signal; iteratively solving the multi-dimensional matrix of the echo signal in each coherent processing interval to obtain a target velocity prediction value and a target position prediction value. The present invention avoids co-frequency interference between signals by allocating different carrier frequencies to the two linear subarrays. The projection transformation improves the perception accuracy of the target motion state and enhances the joint estimation capability of the multi-dimensional velocity components of the moving target, thereby improving the overall performance of the system.

The invention discloses a network event aggregation method and device based on non-negative matrix factorization. The network event aggregation method comprises the steps that network flow data are acquired and preprocessed; performing analysis processing on the preprocessed network flow data to generate a basic network event set; clustering the basic network event set through a clustering algorithm, and extracting and generating a key event description set; and constructing a pattern matrix for the key event description set and carrying out matrix iterative decomposition processing to obtain an aggregation event set. According to the invention, event aggregation and network threat discovery can be realized, the network behavior analysis accuracy is effectively improved, and the network security is guaranteed.

This application provides a method and apparatus for detecting unknown cross-site scripting (XSS) based on test script generation. It innovatively constructs a Markov chain-driven script generation mechanism and achieves intelligent construction of test scripts through atomic element sequence analysis and random walk strategies. A filtering analysis mechanism based on DOM tree difference is designed, combining multi-dimensional tag extraction and node localization to establish a filtering rule recognition strategy. A matrix iterative optimization mechanism is introduced, achieving adaptive optimization of the test script through dynamic updating and convergence calculation of the state transition matrix. This method effectively addresses the shortcomings of traditional techniques in script generation, filtering analysis, and optimization adjustment, significantly improving the detection effect of unknown XSS.

The invention discloses an intelligent agricultural product sorting method based on physical feature iterative optimization and a related device, and belongs to the technical field of agricultural product sorting. The method comprises the following steps: collecting a surface reflection image and a near-infrared transmission image of an agricultural product to be sorted; using homomorphic filtering to carry out anti-wrinkle preprocessing on the surface reflection image; extracting physical feature vectors of the two images; iterative optimization is carried out based on the physical characteristic matrix, an optimal judgment threshold parameter combination is obtained, feature extraction and judgment are carried out on agricultural product images collected in real time, and an execution mechanism is controlled to complete the sorting action according to encoder signals and a delay compensation mechanism. According to the method, morphological interference is solved through homomorphic filtering, interpretability detection of quality is achieved through white-box physical characteristic quantification, top-speed self-adaption of parameters is achieved through matrix iteration, and therefore precise and intelligent sorting of complex agricultural products is achieved.

This invention discloses a method and system for generating transmission patterns for highly sparse multi-user transmission, primarily addressing the problem of low system error rate performance in existing ODMA random access models when the number of users is large. The implementation includes: dividing the binary information bits transmitted by the user into two information bit sequences; converting the first information bit sequence into a decimal number; encoding and modulating the second information bit sequence to obtain a modulation symbol sequence; generating an all-zero matrix based on the known lengths of the first information bit sequence and the modulation symbol sequence, as well as a pre-determined number of transmission time slots; iteratively updating this matrix by continuously finding the update positions of matrix elements through path tree generation; and finally updating the initial all-zero matrix into a binary sparse matrix, which is the multi-user transmission pattern. This invention reduces the system error rate and improves system performance when handling the access demands of more active users, and can be used for highly sparse multi-user transmission in massive machine-type communication.

The invention discloses an electromagnetic gradient joint data high-resolution variable-dimension inversion method, and belongs to the technical field of electromagnetic signal inversion, and the method comprises the steps: constructing a joint residual target function based on a conventional inversion error and a gradient data residual; establishing an initial two-dimensional inversion model; defining a variable-dimension mapping operator, and mapping two-dimensional inversion model parameters to a three-dimensional space; a three-dimensional Jacobian matrix is calculated in the forward modeling process; based on the three-dimensional Jacobian matrix, calculating a quasi-two-dimensional inversion gradient data Jacobian matrix; and iteratively updating the two-dimensional inversion model to obtain an inversion result. According to the variable-dimension ground-air electromagnetic signal inversion method provided by the invention, three-dimensional forward modeling simulation is carried out under a two-dimensional inversion parameter scale by utilizing a data residual term under the combined action of an electromagnetic field and a gradient field of the electromagnetic field and introducing a variable-dimension mapping mechanism, conventional electromagnetic field data and gradient data are effectively fused, the calculation efficiency is ensured, and meanwhile, the calculation efficiency is improved. And the precision and the boundary resolution of an inversion result are remarkably improved.