103 results about "Triangular matrix" patented technology

The invention discloses an electromagnetic field simulation optimization method and system based on deep learning. The method comprises the following steps: discretizing a calculation area, and dividing the calculation area into a limited number of units; constructing a finite element matrix for the electric field of each unit; the finite element matrix is subjected to block processing, and two sub-block matrixes describing the electric field gradient and the electric field rotation of the triangular surface area are decomposed into an upper triangular matrix and a lower triangular matrix through an incomplete-like Cholesky decomposition model; inversing the upper triangular matrix and the lower triangular matrix to obtain approximate inverse of the finite element matrix; substituting the approximate inverse of the finite element matrix into a conjugate gradient algorithm, and solving to obtain an electric field of each unit; and obtaining a corresponding magnetic field according to the electric field of each unit so as to obtain an electromagnetic simulation result. According to the method, through a technical path of block dimension reduction-graph structure learning-physical constraint enhancement, the bottleneck problem that preprocessing efficiency and adaptability in electromagnetic field simulation are difficult to consider at the same time is solved.

The invention discloses an imitated radio frequency division multiplexing signal detection method based on low-complexity optimization expected propagation, and belongs to the signal detection technology of multi-carrier communication. According to the method, information is updated through iteration to approach an ML result, and the complexity is reduced from an exponential level to a cubic level; in order to reduce the complexity of the EP, sparse and quasi-banded channel structures in the DAFT domain are utilized. Through matrix block decomposition, upper and lower triangular decomposition and upper triangular matrix forward substitution, low-complexity matrix inversion is successfully realized, and the complexity of the EP algorithm is reduced to a linear level. In addition, the EP algorithm performance is further improved through a deep learning optimization moment matching strategy, so that the process becomes more adaptive through trainable positive and negative component parameters. In addition, DAFT domain iteration detection and decoding can be achieved based on the method, external information of a decoder is fed back to a detector, and the reliability of the whole system is remarkably improved. According to the invention, effective balance between performance improvement and calculation complexity is realized.

A computer implementation method for obtaining a lower triangular matrix for a matrix multiplication result value, the method comprising: a process of allocating elements of each of a first operand matrix pair and a second operand matrix pair input into a memory to calculation units of an accelerator; and a process of the calculation units of the accelerator performing a matrix multiplication calculation on the elements of the first operand matrix pair and the second operand matrix pair, wherein the process of allocating the elements thereof to the calculation units of the accelerator comprises: allocating elements of the first operand matrix pair to first calculation units corresponding to elements of the lower triangular matrix for the matrix multiplication result value of the first operand matrix pair; and allocating elements of the second operand matrix pair to second calculation units corresponding to elements of the lower triangular matrix for the matrix multiplication result value of the second operand matrix pair.

The invention relates to the technical field of array signal processing, in particular to a space-frequency power inversion anti-interference method fusing SMI-LDL decomposition. The problems that an existing space-time anti-interference technology is high in calculation complexity, and real-time processing is difficult to achieve in limited resources of a navigation receiver are mainly solved. The method comprises the following steps: firstly, converting a radio frequency signal received by an array into a digital intermediate frequency signal and carrying out sliding window partitioning processing; windowing the data block and performing FFT (Fast Fourier Transform) to obtain a frequency domain array; then calculating an auto-covariance matrix R of other array elements except the reference array element and a cross-covariance matrix r of the reference array element and other array elements; lDL decomposition is carried out on the auto-covariance matrix to obtain a lower triangular matrix L and a diagonal matrix D, and inversion is carried out on the lower triangular matrix L and the diagonal matrix D respectively, so that huge operand caused by direct inversion of the covariance matrix is avoided; and finally, calculating an optimal weighting vector according to a power inversion criterion, and performing spatial filtering on the frequency domain data.

The embodiments relate to a method for encoding / decoding, the method comprising obtaining (310) a set of input samples; obtaining (320) a triangular matrix, a cross-correlation vector and a scaling vector by utilizing the set of input samples; determining (330) an intermediate vector with a first back-substitution using the triangular matrix and the cross-correlation vector; scaling (340) the intermediate vector using the scaling vector; determining (350) a filter coefficient vector with a second back-substitution using the triangular matrix and the scaled intermediate vector; and determining (360) predicted sample values for a block of samples based on the filter coefficient vector. The embodiments also relate to an apparatus and a computer program product for implementing the method.

The invention relates to a mechanical arm model parameter identification method and system based on deep learning, and the method comprises the following steps: 1, converting a parameter identification problem of a space mechanical arm kinetic equation into a deep learning parameter identification problem, and designing a feedforward network approximate mass matrix; step 2, constructing a loss function to optimize a model parameter deep learning process of a mechanical arm kinetic equation; 3, realizing structure description of a network layer in deep learning; 4, performing fractional order optimization on the triangular matrix obtained in the step 1, and designing an activation function for the deep learning part; and 5, evaluating the performance of the deep learning algorithm on the mechanical arm parameter identification problem. According to the method, a system identification parameter set is simplified, and the number of identification parameters is reduced, so that the convergence speed of deep learning training is improved, the learning efficiency is enhanced, the high efficiency and precision of an algorithm are ensured, and the identification precision and generalization ability are improved.

The invention relates to a dynamic increment calculation method for a power flow Jacobian matrix and power unbalance, which comprises the following steps of: redesigning a calculation process of the Jacobian matrix, storing a lower triangular admittance value of a target column into a circular queue, and dynamically updating a power accumulated value into a diagonal register at the same time; and the time for searching the admittance matrix during Jacobian lower triangular matrix calculation is greatly shortened. According to the method, the parallel pipeline dynamic increment calculation of the Jacobian matrix and the power unbalance is realized, and the calculation time and storage resource consumption of the Jacobian matrix of a large-scale network are reduced.

The invention relates to the technical field of artificial intelligence, in particular to an aspect triple extraction method based on a multi-channel enhanced multi-head graph convolutional network, which comprises the following steps of: encoding an input text by using a pre-training model to obtain context representation; analyzing the text to extract language features, and constructing a multi-language feature map; probability distribution between context representations is obtained through a double-affine attention mechanism, and context semantic enhancement features are calculated through a multi-head self-attention mechanism; then, features are aggregated by using a multi-head graph convolutional network, and global semantic feature representation is obtained; combining the global semantic features with the multi-language feature map, judging word pair matching and calculating emotion polarity probability distribution; and finally, analyzing the aspect emotion triad by adopting a triangular matrix structure, and performing association verification through emotion polarity probability distribution to finally obtain the aspect emotion triad. According to the method, multi-source features and an attention mechanism are effectively fused, and the accuracy and efficiency of aspect emotion triple extraction are improved.

The application discloses a multiple wave prediction method and device, a computing device and a storage medium. The method comprises the following steps: acquiring frequency domain data of any shot point at any seismic trace; generating a seismic data matrix according to the frequency domain data, the seismic data matrix being an upper triangular matrix with a main diagonal line of 0, and any non-first column of the seismic data matrix containing frequency domain data of at least one seismic trace of the same shot point; constructing a multiple wave matrix, the multiple wave matrix being an upper triangular matrix with a main diagonal line of 0, and any non-first column of the multiple wave matrix containing a multiple wave model of at least one seismic trace of the same shot point; establishing a functional relationship between the multiple wave matrix and the seismic data matrix; and obtaining model frequency domain data corresponding to any multiple wave model in the multiple wave matrix according to the functional relationship. According to the scheme, the prediction efficiency of the multiple wave can be improved under the premise of ensuring the prediction accuracy of the multiple wave.

An LDPC-code-based communication method and communication apparatus. In the method, a device can perform encoding or decoding on the basis of an LDPC matrix, wherein the LDPC matrix is determined on the basis of an LDPC base matrix, the base matrix comprises an extended check matrix, the extended check matrix is a lower triangular matrix with diagonal elements all being 1, the extended check matrix comprises a plurality of lower triangular matrices, diagonal elements of each of the plurality of lower triangular matrices are located on the diagonal of the extended check matrix, the positions of the diagonal elements of the plurality of lower triangular matrices do not overlap, and the elements in the extended check matrix other than the diagonal elements of the extended check matrix and the diagonal elements of the plurality of lower triangular square matrices are all zero elements. The method can increase the column weight of a check region, reduce the occurrence probability of an error plane, and also support retransmissions at some specific columns, and thus the method is compatible with an IR-HARQ scheme of 5G LDPC codes.

This application relates to the field of plasma physics technology and provides an asymmetric inversion method and apparatus based on a matching mechanism for plasma diagnostics. The method acquires plasma magnetic flux data, observation lines, and measurement integral values. The magnetic flux data is preprocessed to obtain preprocessed magnetic flux data. Based on the preprocessed magnetic flux data, a closed magnetic surface uniquely matched to each observation line is determined. A lower triangular matrix is ​​constructed based on each observation line and the closed magnetic surface. An inversion equation is constructed based on this lower triangular matrix and the measurement integral values. Finally, the inversion equation is solved to determine the parameter distribution values ​​of the plasma. This method achieves three-dimensional spatial projection of the observation lines and magnetic flux, rapidly determining the position of the closed magnetic surface using geometric relationships, thus improving computational efficiency. Simultaneously, it accurately establishes a one-to-one correspondence between the closed magnetic surface and the observation lines, making the ill-posed problem in solving the matrix equations a unique solution, improving the success rate of solving using the stacking method, and significantly reducing computational errors.

The present invention discloses a method for quickly calculating the frequency response characteristics of a hollow winding coil of arbitrary shape, belonging to the technical field of electrical and electronic equipment components and sensors. The method comprises: establishing a three-dimensional rectangular reference coordinate system with a certain point in space as the origin; calculating the geometric center spatial coordinates of each turn based on the geometric centerline parameter equation of the skeleton, thereby obtaining the spatial parameter equation of each turn; combining the Newman formula to calculate the mutual inductance of any two turns and the self-inductance coefficient of the turns, the capacitance, resistance and skin depth between the turns, and the electrical relationship equation between any turn and other turns; writing the electrical relationship equation in matrix form, converting the coefficient matrix of the equation into an upper triangular matrix, and then converting the matrix into a coil port voltage and current relationship equation to obtain the coil port frequency response characteristics. The present invention has strong versatility, convenient calculation, and is easy to program for calculation, saving design time. No empirical formula is required during the calculation process, and the calculation accuracy is high.

This disclosure provides systems, methods and apparatuses for encoding a channel. The method includes mapping a plurality of information bits and a plurality of frozen bits to a vector based on a reliability order. The method includes generating a resulting vector by multiplying the vector by a transformed matrix to. The transformed matrix is defined by: a first identity matrix with a size equal to a first quantity of frozen bits in the vector that are located before a first information bit in the input vector, an upper triangle matrix, and a second identity matrix with a size equal to a second quantity of information bits in the vector that are located after a last frozen bit in the input vector. The method includes multiplying the resulting vector by a polar encoder matrix.

The present invention relates to the technical fields of wireless communication and industrial wireless network, and particularly relates to a data security transmission method based on coding and encryption. The method includes solidifying a key inside the chip physical layer, where each key corresponds to a configured key serial number; converting a coding check matrix into an approximate lower triangular matrix; obtaining a conjugate matrix of the coding check matrix; multiplying the right of the codeword row sequence by the conjugate matrix transpose matrix and setting it to 0 and solving to obtain a coding row sequence; recording the length of the information sequence as a sequence of a specified number of digits and performing k-bit data grouping alignment; then performing data encryption to obtain an encrypted sequence; splicing the k-bit aligned information sequence, the encrypted sequence, and the coding row sequence to obtain a transmitted coding row sequence and transmitting it through a channel; and the receiving end receives and performs decoding and decryption. The method solves the problems of data transmission, anti-interference, and ensuring the security of physical data and keys in the current secure encrypted communication of the application layer and the MAC layer.

The invention discloses a joint sorting QR decomposition method and related device for multi-antenna detection, and belongs to the technical field of multi-input multi-output signal detection, and the method comprises the steps: initializing an orthogonal matrix and an upper triangular matrix; acquiring a channel matrix H; extracting candidate columns from residual columns in the channel matrix H based on the sum of residual energy and self energy joint weighting, updating an orthogonal matrix and an upper triangular matrix according to the candidate columns until all columns in the channel matrix H are traversed, and obtaining a final orthogonal matrix and a final upper triangular matrix; and taking the final orthogonal matrix and the upper triangular matrix as a joint sorting QR decomposition result for multi-antenna detection, the method and the related device can shear out the nodes with the super radius as early as possible, reduce access nodes, and improve the speed of multi-antenna detection while ensuring the detection performance.

The application provides a pathological section recognition neural network model weight encryption method, including: acquiring a bit string as a pathological section recognition neural network model weight key; writing the bit string into a lower triangular matrix according to 8 bits as a number to obtain a first data matrix; performing digital conversion on the first 8 bits in the bit string to obtain first element data; reading a weight matrix of the pathological section recognition neural network model into a blank matrix to obtain a weight vector matrix; combining the first element data matrix and the weight vector matrix together to obtain a first data vector; obtaining an intermediate ciphertext vector by using a chaotic mapping algorithm according to the first data matrix and the first data vector; performing element data processing on the intermediate ciphertext vector to obtain a weight vector intermediate encryption matrix; and performing position disordering on the weight vector intermediate encryption matrix to obtain an encrypted weight vector matrix. The application can improve the safety coefficient of the neural network weight.

The invention discloses a robot teaching obstacle avoidance method based on a neural network and a dynamic system model. The method comprises the following steps: acquiring a demonstration data set; coding the current robot state into a high-dimensional feature vector; nonlinear feature components are extracted from the encoded features, weighted summation is carried out on the features, quadratic terms are added, and an energy function is constructed; constructing a decision function, and multiplying the state and environment information of the robot by corresponding weights to obtain the output of the decision function; inputting the output of the decision function into a multi-task hybrid neural network, generating a parameter vector, constructing a lower triangular matrix based on the parameter vector, and constructing a modulation matrix through cholosky decomposition; constructing a loss function, and obtaining an optimal model parameter through a nonlinear programming method; and taking the output of the neural dynamic system model as a robot control strategy to complete a specified task. According to the invention, by introducing the dynamic system model and the neural network, the bottleneck problems of time dependence and dynamic obstacle avoidance are solved.

The application discloses a text reasoning method, product, equipment and storage medium, and relates to the technical field of computers, and comprises the following steps: through the orthogonal triangular decomposition mode and in combination with the judgment mechanism whether the pre-trained language model has been trained on the fine-tuning data set, if the pre-trained language model has been trained on the fine-tuning data set, the orthogonal triangular decomposition is performed on the to-be-fine-tuned parameter matrix, the triangular matrix obtained after the decomposition is fine-tuned and trained on the fine-tuning data set, and the fine-tuned language model is obtained; if the language model has not been trained on the fine-tuning data set, the orthogonal fine-tuning is performed on the language model starting from a preset unit matrix, the fine-tuned language model is obtained, and the text reasoning is performed by using the fine-tuned language model, through the above mode, the robustness of the language model to different data sets is improved, and the accuracy of the text reasoning result is improved, and meanwhile, the problem that the robustness of the traditional pre-trained language model parameter after fine-tuning is poor and the performance is reduced is solved.

To provide a selection device, an estimation system, etc. that can appropriately select a center frequency for which a trained model should be obtained before collecting data from a real environment.SOLUTION: The selection device includes: signal generation means for generating signal data simulating an arrival signal based on parameters including at least a center frequency and arrival angle information; correlation matrix calculation means for calculating a correlation matrix based on the signal data; inference means for generating an evaluation spectrum based on a lower triangular matrix extracted from the correlation matrix using a trained model; performance evaluation means for evaluating the performance of the trained model based on an estimated arrival angle estimated based on the evaluation spectrum and a correct arrival angle of the arrival angle information; and next frequency determination means for determining a center frequency to be evaluated next based on a result of the performance evaluation. An estimation system according to the present disclosure includes the selection device and an estimation device for estimating the arrival direction of an arriving wave.SELECTED DRAWING: Figure 1

The application provides a distributed security outsourcing computing processing method based on sparse matrix transformation, and relates to the technical field of computers, and the method comprises the following steps: acquiring a first matrix and a second matrix; randomly generating a key set comprising a binary permutation matrix, a sparse triangular matrix and a random vector; performing primary encryption on the first matrix and the second matrix based on the binary permutation matrix; performing secondary encryption on the primary encrypted matrix based on the sparse triangular matrix; distributing the secondary encrypted first secondary encrypted matrix to each edge end in an integrated manner, distributing the second secondary encrypted matrix to each edge end in an even manner after dividing the second secondary encrypted matrix into column vectors according to columns, and enabling the edge end to perform a subtask calculation operation by using the received matrix and return the result to the user end; receiving the calculation results of the edge ends, and aggregating the results; verifying the reliability of the aggregated results by using the random vector, and decrypting the final calculation result. The scheme can improve the security and timeliness of outsourcing computing.

The invention provides a terminal initialization method and device, electronic equipment and a storage medium. The terminal initialization method comprises the following steps: acquiring initialization parameters, inertial constraints and visual constraints of a terminal; constructing an information matrix according to the initialization parameters, inertial constraints and visual constraints; performing Schel compensation elimination on the information matrix, and eliminating the feature points from the information matrix to obtain an eliminated information matrix; performing Cholesky decomposition on the eliminated information matrix to obtain an upper triangular matrix and a lower triangular matrix; performing inversion on the image frame state quantity corresponding to the nearest preset moment in the upper triangular matrix to obtain a square root of a covariance, and further determining the covariance; and determining the initialization state of the terminal according to the covariance and the initialization parameter. According to the method disclosed by the invention, the covariance can be quickly solved, and the state quantity fluctuation amplitude is reduced.

The MIMO equalizer circuit (300) includes: a controller (339); a calculation circuit (302) generates and outputs an autocorrelation matrix W (308) in response to receiving the covariance matrix RI (123) and the channel estimation matrix H (121). A matrix calculation circuit (316) that receives and combines the inverse autocorrelation matrix (315) and the channel estimation matrix H (121), and outputs an equalization gain P matrix (324); an inverse autocorrelation matrix (315), a received signal vector y (301), and a channel estimation matrix H (121) are received and combined, and a vector'f '(323) is output. A successive interference cancellation (SIC) detection circuit (325) converts the received P matrix (324) into a first orthogonal unitary matrix Q (328) and a first upper triangular matrix R (327), and converts the matrix Q (328), the matrix R (327), and the f vector (323) into an SIC estimated equalization symbol Z2 vector (330), which is output.

A communication method and a communication device based on LDPC codes, in the method, a device can perform encoding or decoding based on an LDPC check matrix, the LDPC check matrix is determined based on an LDPC basis matrix, the basis matrix comprises an extended check matrix, the extended check matrix is a lower triangular matrix with diagonal elements being 1, the extended check matrix comprises a plurality of lower triangular matrixes, and the lower triangular matrixes are arranged in the extended check matrix. The diagonal elements of each lower triangular matrix in the plurality of lower triangular matrixes are located on the diagonal of the expanded check matrix, the positions of the diagonal elements of the plurality of lower triangular matrixes are not overlapped, and the remaining elements in the expanded check matrix except the diagonal elements of the expanded matrix and the plurality of lower triangular matrixes are all 0 elements. According to the method, the column weight of the check area can be improved, the probability of error floor occurrence can be reduced, retransmission in some specific columns can be supported, and therefore the method can be compatible with an IR-HARQ scheme of LDPC codes of 5G.

The application discloses a method for calculating equivalent resistance based on a many-to-many mode, which comprises: filling a conductance matrix of a corresponding resistance network to obtain an extended matrix based on a node to be solved of an equivalent resistance, preordering the extended matrix to determine a permutation matrix P; performing LDL T decomposition on the extended matrix to calculate a lower triangular matrix L and a diagonal matrix D required for calculating the equivalent resistance and determining a non-zero element position in a sparse matrix F; calculating a matrix element of a target matrix Z according to the non-zero element position in the sparse matrix F and the lower triangular matrix L and the diagonal matrix D; and calculating the equivalent resistance according to the target matrix Z. For this purpose, the application does not need to solve all matrix elements, but only needs to calculate matrix elements consistent with the non-zero element position distribution of F, i.e. Z, so as to effectively reduce the calculation amount. Since Z is a sparse matrix and a general dense matrix, the total calculation amount is greatly reduced due to the fewer non-zero elements of Z.

The invention relates to a positioning method and device, electronic equipment and a computer storage medium. The method comprises the following steps: acquiring a variance-covariance matrix of a carrier phase observation value acquired by the positioning device; performing LD transformation on the matrix to obtain a lower triangular matrix and a diagonal matrix; and storing the effective elements in the lower triangular matrix and the effective elements in the diagonal matrix into the RAM, loading the effective elements in the lower triangular matrix and the diagonal matrix and the unit matrix from the RAM by adopting the DAM, and calculating according to the effective elements in the lower triangular matrix and the diagonal matrix and the unit matrix by adopting a decorrelation algorithm to obtain a positioning result. According to the method, only the effective elements in the lower triangular matrix and the diagonal matrix are stored, storage is reduced, reading and calculation are facilitated, meanwhile, the effective elements in the lower triangular matrix, the effective elements in the diagonal matrix and the unit matrix are stored in the RAM through the DAM, system control of a CPU is reduced, calculation power is increased, and efficiency is improved.

The invention provides a signal processing method and device, electronic equipment, a chip and a storage medium, and relates to the field of communication, and the method comprises the steps: splicing a received signal vector and a channel matrix of a receiving end, so as to obtain a to-be-processed matrix; wherein the channel matrix is used for indicating the channel gain between the transmitting stream of the transmitting end and the plurality of receiving antennas of the receiving end, and the receiving channel vector is used for indicating the signals received by the plurality of receiving antennas; performing multi-round iteration on the to-be-processed matrix based on the target systolic array to obtain an upper triangular matrix and an equivalent receiving vector; and recovering the physical air interface signal sent by the sending end according to the upper triangular matrix and the equivalent receiving vector. Therefore, matrix decomposition is realized by executing at least one round of iteration by adopting an efficient parallel computing architecture, namely the target systolic array, so that the computing complexity can be simplified, the computing efficiency can be improved, high-throughput and low-delay computing can be realized, and the quality and the reliability of physical air interface signals recovered by a receiving end can also be improved.

The embodiment of the invention discloses a matrix operation method and device, computer equipment and a storage medium, and belongs to the technical field of computers.The method comprises the steps that block processing is conducted on a transformation matrix to obtain a plurality of transformation blocks, and block processing is conducted on a triangular matrix to obtain a plurality of initial full-element blocks and a plurality of initial triangular blocks; obtaining a transposed full-element block corresponding to each initial full-element block, and obtaining a restored full-element block corresponding to each initial triangular block; according to a preset matrix block position index, determining restored full-element blocks and initial full-element blocks with block positions having an index row-column symmetric relationship, and transposing the full-element blocks into the same target processing group; and performing matrix operation on each transformation block and the processing block at the corresponding block position in the corresponding target processing group, and determining a target matrix operation result according to a matrix block operation result. According to the invention, the matrix operation efficiency, especially the matrix operation efficiency for the symmetric matrix, can be improved.

The application relates to the field of high-performance computing and computing power optimization, and relates to a method for accelerating triangular matrix multiplication based on an atlas 800 platform, which comprises the following steps: cutting a result matrix into multiple task blocks, and presetting an equal division strategy based on the multiple task blocks; generating a mask matrix on a vector calculation unit, and storing the mask matrix back to a global memory; reading the matrix and the mask matrix from the global memory, implementing triangular masking of the matrix by using the vector calculation unit, and obtaining a triangular matrix A after triangular masking; based on the triangular matrix A after triangular masking, calling a cube calculation unit to complete matrix multiplication operation of all the task blocks according to the preset equal division strategy, and obtaining a result matrix. The mix method combining the vector calculation unit and the cube calculation unit is used to complete the triangular matrix multiplication, the calculation efficiency of the triangular matrix multiplication on the NPU is improved, the computing power of the Huawei atlas 800 NPU can be fully utilized, the triangular matrix multiplication calculation is accelerated, and computing power support is provided for the fields of deep learning and scientific calculation.

A computer-implemented method for memory management in a computer system configured for generating a signature of a binary data message m using a key B of a predetermined lattice-based structure is proposed, which comprises: determining coefficients of a 2×2 signature generation matrix SG, wherein the non-diagonal coefficients of the signature generation matrix SG are complex polynomials with a non-zero imaginary part, and the diagonal coefficients of the signature generation matrix SG are real polynomials; and determining a LDL representation of the signature generation matrix SG according to which SG is represented by a matrix product L.D.L*, wherein L is a 2×2 lower triangular matrix with ones on the diagonal, D is a 2×2 diagonal matrix, and L* is the adjoint of L; wherein the storing in a memory buffer of the computer system of the coefficients of the signature generation matrix SG and the coefficients of the matrices of the LDL representation is managed based on that the signature generation matrix SG has real diagonal coefficients, and the memory buffer is used alternatively to store the coefficients of the signature generation matrix SG or the coefficients of the matrix D of the LDL representation.

The present invention belongs to the technical field of traffic prediction, and specifically discloses a traffic prediction method and system based on a spatiotemporal fusion graph neural network, which utilizes the historical traffic flow data of each traffic node; calculates the similarity of the historical data sequences between each traffic node, and constructs a spatiotemporal adjacency matrix; forms a lower triangular matrix based on the original road network adjacency matrix and the spatiotemporal adjacency matrix; inputs the historical traffic flow data into a fully connected layer for feature dimension upgrading, and sets a spatiotemporal convolution reconstruction layer with multiple sublayers, and uses the dimension upgraded data and the lower triangular matrix as inputs to the spatiotemporal convolution reconstruction layer; aggregates the output of each sublayer of the spatiotemporal convolution reconstruction layer, and inputs the aggregated data into the output layer to obtain a prediction result. The adoption of this technical solution can more completely retain the spatiotemporal characteristics of traffic data and optimize the accuracy of traffic data prediction.