94results about How to "Improve convergence efficiency" patented technology

The invention provides a method for carrying out blind source separation on convolutionary aliasing voice signals. Firstly, a time domain convolutionary aliasing model is converted into a frequency domain multi-channel linear instantaneous convolutionary aliasing model, which can be realized by the following steps: firstly, converting convolutionary aliasing time domain signals into a frequency domain; then carrying out relatively independent ICA operations on each channel to obtain independent components. Next, the independent components are rearranged by an MSBR algorithm, which specificallycomprising the following steps: firstly, classifying signals of different frequency bands; then progressively obtaining transposed matrixes according to different object functions step by step, wherein the steps of rearrangement are mutually complementary. The MSBR algorithm utilizes the strong relevance of harmonic frequency to improve the iteration accuracy and solves the residual uncertainty of residual frequency bands according to the continuity of adjacent frequency bands and corresponding reference frequencies, and the computational complexity of the MSBR algorithm is approximately in direct proportion to the number of reference frequency bands. The invention improves the convergence efficiency and the accuracy, is more suitable for real-time processing, has good separation performance of convolutionary mixed voice signals and can also be applied to real phonetic environment.

The invention relates to an extra-high-voltage direct-current transmission line fault location method based on a wavelet transformation transient state energy spectrum, and belongs to the technical field of power system relay protection. When a direct-current line breaks down, according to two-pole direct-current voltage collected at a protection installation position (img file='dest_path_dest_path_image001. TIF'wi= '40' he='24' / ) and (img file='dest_path_dest_path_image002. TIF'wi= '40' he='24' / ), line mode voltage (img file='dest_path_dest_path_image003. TIF'wi= '48' he='24' / ) is obtained, wavelet decomposition is carried out on line mode voltage (img file='dest_path_855728dest_path_image003. TIF'wi= '48' he='24' / ) to obtain a high-frequency coefficient of the wavlet decomposition, the high-frequency signal wavelet energy sum (img file='dest_path_dest_path_image004. TIF'wi= '21' he='25' / ) is solved by means of the high-frequency coefficient, normalization processing is carried out on (img file='dest_path_594008dest_path_image004. TIF'wi= '21' he='24' / ) to obtain an input sample of a neural network, and the input sample in the neural network is trained to obtain a fault location result. By means of characteristics of a wavelet energy frequency band whose appearance is obvious and whose position is easy to determine, a fault position is looked for, and precision of fault location is improved; by means of non-linear fitting capacity of the neural network, extra-high-voltage direct-current grounded transmission line fault locating is carried out, the property of the sample is clear, the scale of a sample set is small, convergence efficiency is high, and the method is not prone to being influenced by system parameter conversion and transition resistance.

The invention discloses an improved quantum genetic algorithm-based micro-grid energy storage locating and sizing optimization method. The method comprises the following steps: establishing an energy storage locating and sizing optimization model, wherein the energy storage locating and sizing optimization model comprises a target function formula and a constraint formula; improving a quantum genetic algorithm; and solving the energy storage locating and sizing optimization model by using the improved quantum genetic algorithm. According to the method, the energy storage locating and sizing optimization model is established, energy storage whole life cycle period cost, peak clipping-valley filling earning and grid loss earning are taken as targets, and the trend, energy storage charge-discharge and energy storage charge-discharge energy balance are constrained and considered; the quantum genetic algorithm is corrected, the dynamic adjustment strategy of a quantum revolving door revolving angle is used for improving the search efficiency, and a selection operation implemented by a simulated annealing method and a good point set cross operation can avoid local optimum; a 34-node micro grid is adopted to carry out verification so as to indicate that the disclosed algorithm is feasible, and the convergence efficiency of the quantum genetic algorithm and the ability of jumping out of local optimum are effectively improved.

A pulse width modulated pulse signal is generated and its duty cycle adjusted according to at least two different schedules, for example, a coarse tuning cycle and a fine tuning cycle. The target duty cycle is derived according to an offset between output voltage and target voltage, as well as the rate of changes in the offset. In the coarse tuning cycle, large changes in the control signal pulse width is used to drive output voltage to target at a fast pace. As output voltage approaches target voltage, smaller adjustment are made to the duty cycle of the control pulse width in the fine tuning cycle. The rate of pulse width change is also taken into consideration, to further improve operation of the DC-DC converter.

The invention provides a beam design method based on safety energy efficiency in a satellite-ground integrated network, which is used for solving the low system safety energy efficiency of a downlinkof the existing satellite-ground integrated network. The method comprises the following steps: firstly, forming a high-gain directional beam by adopting analog precoding; secondly, designing digital precoding to eliminate inter-user interference; and finally, proposing a joint optimization problem of transmitting power and a power splitting coefficient, so as to maximize the safety energy efficiency of the system and meet the service quality constraint of each user at the same time, and the interference limitation of a satellite ground station and power consumption limitation of a base station. In order to solve the problem, an iterative algorithm based on Dinkelbach and continuous convex approximation SCA is provided to obtain the solution of the problem, and the convergence efficiency ofthe safe transmission rate of the system is improved. According to the method, a millimeter wave large-scale MIMO NOMA system is combined with the SWIPT technology in the ground network, and the basestation provides services for a plurality of users and protects the satellite ground station from interference.

The invention discloses a big data analysis method adopting a grey wolf optimization algorithm, each piece of data of an applied big data set is used as a feasible solution, each wolf represents a solution, and a single iteration alpha wolf decides a local optimal solution. And after multiple iterative calculations, the selected alpha wolf represents a globally optimal solution. The method mainlycomprises the steps of setting initial parameters of a wolf algorithm; randomly generating a first-generation wolf group, dividing the first-generation wolf group according to the fitness values, andselecting three optimal wolves as alpha, beta and delta wolves, with all the rest being omega wolves; and when the maximum number of iterations is reached, the output alpha wolf represents the globaloptimal solution. The method has higher performance of searching for the globally optimal solution and is high in calculation speed.

The invention discloses a gridding method for one-line diagram layout and optimization of a zone electric distribution system. The method comprises the following steps of: gridding a layout space, forming a grid lattice of a node layout solution space and a line corridor space of lines, and limiting the node layout solution space in the grid lattice of nodes; describing each grid state by adopting a relevant enumeration variable; performing node layout, establishing a node layout model, and adopting a shortest Manhattan distance of the nodes as a target function for node layout and optimization; performing mapping and initialization, determining the position of a transformer substation in grids according to relative geographic information by utilizing hierarchical branch attributes of a power distribution line, and arranging branch contact lines of a distribution network by selecting a hierarchical layout algorithm; introducing a self-adaptive crossover probability, and introducing different crossover probabilities for different fitness values; and avoiding overlapping and crossing situations of the lines. The gridding method disclosed by the invention effectively assists in node layout and optimization of a one-line diagram and arrangement of the lines in the zone electric distribution system.

The invention provides a shuffled frog-leaping based division method of software and hardware. The method comprises the steps of determining the maximum iterations of the algorithm at the initial; calculating the area and time cost of the hardware in the corresponding software and hardware division scheme corresponding to each frog; descending and sequencing the whole frog group according to the adaptive fitness of individual fog; updating the fog groups in each group at a plurality of times to enable the adaptive fitness of the worst frog in the group to be prior to that of the previous frog as well as meeting the system limitation condition; updating the optimal frog in the group by the random updating method in case of failing to update; shuffling the whole frog groups; finding the global optimal solution as the optimal software and hardware division scheme after the iteration is done. According to the method, the sub-task of one embedded composite system is divided through a processor, and meanwhile, the searching and the optimizing capacity of the system are improved; the optimal software and hardware division scheme can be found, and as a result, the task performing efficiency of the system can be increased.

A group robot cooperation search method based on an improved particle swarm algorithm belongs to the group robot technology field and is used to improve problems that an exploration range is too concentrated and a remote area is easy to omit. The method comprises the following steps of (1) initializing a robot system state; (2) using an artificial potential field improvement strategy to carry outroaming search and obtaining the rough position of a target; (3) detecting a target signal by a robot, and then determining whether to obtain a target search right according to a result, and decidingwhether to continue the roaming search or enter a cooperation search phase; (4) adopting the improved particle swarm algorithm by a robot group searching for a same target, carrying out fine search onthe target and improving search efficiency; and (5) discovering all the targets by roaming robots, and carrying out fine positioning through the particle swarm algorithm and ending the task. In the method, the task can be effectively completed during multi-robot cooperation target search; and a PSO algorithm based on an assisted orientation technology can be used to improve the convergence efficiency of the later period of the particle swarm algorithm.

The invention relates to a power distribution network optimization scheduling method and system considering dynamic reconstruction of a network frame. Firstly, carrying out active power distribution network element time sequence modeling is carried out; then, establishing an active double-layer optimization scheduling model considering network dynamic reconstruction with the economy of power distribution network dispatching and the stability of fast voltage as objectives; and performing decimal encoding on branches in each basic loop, setting an iterative condition of a network frame population, adding a radial constraint condition of a power distribution network frame into an iterative strategy, and solving the active double-layer optimization scheduling model. The method can effectivelyreduce the operation cost of the power distribution network, smooth the voltage level of the power distribution system and improve the system voltage stability.

The invention brings forward a general airplane air-material demand prediction method based on an MPSO-BP network with the combination of a particle swarm algorithm and a BP network. The method includes: firstly, analysis and research of main influence factors of the general airplane air-material demand are conducted, then the basic principle of the BP network, the PSO algorithm, and the improved strategy of the PSO algorithm are introduced, and a prediction model for improving the particle swarm algorithm and optimizing the BP network is built. Five main influence factors of the general airplane air-material demand are regarded as the input of the network, the demand quantity of the general airplane air-material is regarded as the output, and a nonlinear mapping relation between the input and the output is established. According to the method, accurate prediction of the general airplane air-material demand quantity can be realized, the non-linear fitting capability and generalization capability are very good, the convergence efficiency is improved, the possibility of falling to a local minimum value is reduced, the prediction precision is higher, and the application effect in the general airplane air-material demand prediction is good.

The invention relates to a crop plantation space layout optimization method based on an intelligent optimization algorithm, in particular to a crop plantation space layout optimization method based onan improved Cuckoo algorithm, and belongs to the field of geographic space layout planning. The method aims at a crop plantation layout optimization problem to redefine an individual coding way, an initial layout scheme generation method and a multi-objective function and constraint condition processing method in the Cuckoo algorithm on the basis of a Cuckoo algorithm framework, a one-to-one mapping relationship of the problem and the Cuckoo algorithm is established, and a cross replacement strategy and a local update strategy, which aim at the space layout, are put forward so as to establishthe information exchange and update mechanism of a Cuckoo individual again to meet crop space layout optimization requirements. Meanwhile, the method is combined with a Pareto multi-objective processing technology, and various candidate schemes can be generated to provide decision support for decision makers. By use of the cross replacement and local update strategy put forward by the invention,the convergence efficiency and the solving accuracy of the algorithm are improved.

The invention discloses a multi-body-dynamics-equation solving method based on a Bathe integration strategy. The method includes: establishing a corresponding coordinate system in a multi-body dynamics system, and using a relative-coordinate method and a Lagrangian-multiplier method to obtain dynamics equations according to a virtual-power principle; seeking a second derivative of a constraint equation with respect to time to obtain differential-algebraic equations of an index -1, and adding constraint stability terms to form a general form of the dynamics equations with complete constraints;carrying out integration iteration solving on the general form of the dynamics equations with the complete constraints on a time step [t, t+h / 2] to obtain motion parameters of a moment t+h / 2; carryingout integration iteration solving on a time step [t+h / 2, t+h] to obtain the motion parameters of a moment t+h; and carrying out repeated iteration calculation, outputting values of generalized displacement, velocity and acceleration, which are of a system and are changed over time, when calculation time reaches set total simulation time, and thus analyzing motion status of the multi-body process-closed-loop dynamics system in certain time. The method can also achieve better precision in a case of larger integration step length.

The invention belongs to the communication signal detection field and the non-linear field and particularly relates to an adaptive stochastic resonance system and method based on an artificial fish swarm algorithm. The system comprises a normalized stochastic resonance module, an artificial fish swarm adaptive module, a step adjustment module, a noise adjustment module and an iteration control module. Through the system, not only can a problem of difficulty in setting an optimization searching initial value of an adaptive system be solved through normalized stochastic resonance, adaptability is improved, no stochastic resonance, excessive stochastic resonance and insufficient resonance caused by parameter selection are considered, relatively strong search capability is kept through a step function adjusted according to resonance convergence conditions, but also local optimum can be properly discovered and jumped out from, and optimal value convergence efficiency is accelerated. An input signal to noise ratio is optimized through the noise adjustment module, and system signal to noise ratio gain is substantially improved.

The invention discloses a pre-conditioned time domain elastic medium multi-parameter full-waveform inversion method, which comprises the following steps of: S1, selecting a plurality of scale frequency groups to pre-process observation data and seismic wavelets; S2, calculating target seismic source wavelet and a plurality of parameter initial models through staggered grid finite difference forward modeling to obtain a simulated observation record, and acquiring a wave field residual error by subtracting an observation record under each scale from the simulated observation record under the corresponding scale; S3, solving a conjugate operator of a positive operator through adopting an adjoint state method, and acquiring a residual wave field through utilizing the wave field residual error;S4, calculating parameter gradients of a parameter initial model through using the residual wave field and the simulated observation record, and preprocessing all the parameter gradients; S5, optimizing the initial iteration step length through employing a parabola fitting method; and S6, updating a formula and all parameter models through adopting an LBFGS method model to obtain a multi-parameter inversion result. According to the pre-conditioned time domain elastic medium multi-parameter full-waveform inversion method, the efficiency, effect and stability of multi-parameter full-waveform inversion are effectively improved.

The invention discloses an intrusion detection system and an intrusion detection method for an energy Internet. A feature selection module is used for forming a reduced data set by removing features which have no contribution or low contribution to an intrusion detection model from an original data set; a data preprocessing module is used for carrying out further data preprocessing on the reduceddata set and processing the reduced data set into a data set which can be received and processed by a deep learning model; and an intrusion detection module is used for training, optimizing and testing the deep learning model by using the data set obtained by the data preprocessing module, and then detecting intrusion flow by using the deep learning model. According to the intrusion detection system for the energy Internet, through testing on a real data set of the energy Internet, the numerical result shows that the intrusion detection system disclosed by the invention is very effective in detection of various network threats in the energy Internet, and is superior to most of existing intrusion detection system schemes.

The invention discloses a large-scale unit combination optimization method for a multi-stack fuel cell power generation system, which comprises the steps of clustering fuel cell stacks according to the scene through a K-means clustering method and a Monte Carlo simulation method, and building a probability distribution model; enabling each fuel cell stack to search and follow the current optimal solution in a solution space according to the probability distribution model by applying an adaptive inertia weight algorithm, and achieving the optimal distribution of the output power of each fuel cell stack. According to the invention, the reasonable output of large-scale units of a fuel cell power generation array system is ensured, the on-off of each stack and the shifting between the stacks are controlled through the adaptive inertia weight algorithm, so that the service life of fuel cells is prolonged, the system stability is improved, the system cost is reduced, and a purpose of fuel economy is achieved.

The invention discloses an antenna selection method for a united receiving and transmitting end of an MIMO system and belongs to the field of wireless communication. The method of the invention can be used for exporting channel capacity, which is expressed by binary-encoded strings, via analyzing a channel capacity formula of the MIMO system by using a thought based on binary system particle swarm optimization, as fitness function of the particle swarm optimization, which simplifies the calculated amount, and the method of the invention simultaneously improves search efficiency by performing population initialization in a cyclic shift way so as to improve the channel capacity. The method of the invention can be used for obtaining higher channel capacity than other suboptimal algorithms while reducing the computation complexity.

The invention relates to a semi-active fuzzy control method, and aims at solving the existing problems that the selection of parameters of a control system is difficult and parameter optimization is not comprehensive of fuzzy control in semi-active control based on a magnetorheological damper. The invention provides the semi-active control algorithm which combines and improves a crow searching algorithm and the fuzzy control, seismic response is taken as the input of improving a fuzzy controller, and therefore the fuzzy controller is allowed to directly output control signal. The comprehensiveoptimization of key modules of the fuzzy control system is achieved on the basis that an original crow algorithm is improved. The method has the advantages that a voltage signal of the magnetorheological damper can be accurately controlled to achieve the full shock-absorbing performance of the magnetorheological damper so that the vibration of a building structure under seismic excitation can besignificantly inhibited, and the seismic response can still be effectively reduced when a seismic wave and structural parameters change.

The invention discloses a construction pipe pile based on a wet land and a preparation method thereof. The building tubular pile based on the wet land comprises a tubular pile body, tubular pile endsare fixedly installed at the top end and the bottom end of the tubular pile body correspondingly, tubular pile connecting ribs are evenly embedded in the top end of the upper tubular pile end, tubularpile connecting grooves are evenly formed in the bottom end of the lower tubular pile end, a steel reinforcement cage is installed in the inner part of the pipe pile body in an embedded mode, and reinforcing ribs are evenly embedded in positions, corresponding to inner sides of steel reinforcement cages, in the inner part of the pipe pile body. According to the construction pipe pile based on thewet land and the preparation method thereof, the construction pipe pile based on the wet land is scientific and reasonable in structure and safe and convenient to use, through the arranged reinforcing ribs, the overall strength of the pipe pile body can be improved, the waterproof performance of the pipe pile body can be improved, the applicability of the pipe pile body is improved; and through the arrangement of a collecting baffle plate, a conduction cavity, a driving shaft and a spiral conveying blade, rapidly collecting concrete overflowing in the pouring process can be carried out, so that the collecting efficiency of the overflowing concrete is improved, and waste of concrete resources is reduced.

The invention discloses an unstructured grid direction template selection method and a flow field solving method, and the corresponding direction template selection method for any grid unit j in an unstructured grid comprises the steps: A1, taking the grid unit j as a central unit, and taking the grid unit, sharing the vertex with the central unit, in the unstructured grid as a template unit; A2,making two straight lines passing through the geometric center of the central unit, wherein one straight line is parallel to the normal direction of the wall surface, and the other straight line is parallel to the flow direction; and A3, taking the template unit, intersected with one of the two straight lines in the step A2, in the template units obtained in the step A1 as a direction template corresponding to the grid unit j. The method is simple and effective, the geometric center of the grid unit is corrected through an area weighting method, and the problem of grid deflection on the triangular grid unit with the large compression ratio is effectively solved; the method is high in calculation precision, can effectively reduce the number of template units, is controllable in number of template units, is short in calculation time, is high in convergence efficiency, and can accurately capture the flow field features and flow anisotropy.

The invention discloses a multi-workflow scheduling method based on multi-population hybrid intelligent optimization in a cloud environment, and the method comprises the steps: respectively optimizing the total execution time and cost of a multi-workflow scheduling scheme through employing two populations, employing a third population, carrying out the proper balance of the optimization of the two targets; By focusing on searching different types of non-dominant solutions through the three populations, the diversity of elite solutions is improved, the defect that an existing intelligent optimization method is large in searching randomness is effectively overcome, and an optimal workflow scheduling scheme set can be found under the condition that the deadline condition of a user is met.

The invention discloses a boundless crossed line distribution method of a power distribution feeder line single line diagram. The boundless crossed line distribution method comprises the following steps: step 1, solving a node layout solution with minimum edge crossing of a network related connection line by utilizing a genetic algorithm; step 2, quantitatively describing and judging the edge crossing between non-related connection lines on the basis of a node layout basis, and finally, carrying out adjustment elimination; and finally, carrying out simple and regular topology connection line on the node which is arranged after optimization to form the power distribution feeder line single line diagram with clear logic. According to the method, the diagram layout of the power distribution feeder line single line diagram containing annular power supply equipment is effectively assisted; in an adopted node layout optimization algorithm, an established target function is easy to realize; meanwhile, a node layout position is limited by a grid, and geographic information of a power station is introduced to be used as constraint, so that the optimization solution space is reduced and the convergence efficiency of the algorithm is improved.

The invention relates to a planet gravity-assist low-thrust trajectory optimization method based on a decomposition and coordination strategy and belongs to the technical field of aerospace technologies. The planet gravity-assist low-thrust trajectory optimization method based on the decomposition and coordination strategy comprises firstly decomposing a planet gravity-assist low-thrust transfer trajectory optimization problem into a system-level optimization problem and two second-level optimization sub-problems with planet gravity-assist positions serving as nodes; and then utilizing the system-level optimization problem solving as a main iteration for coordinating status matching of the planet gravity-assist positions of the optimization sub-problems, and utilizing the second-level optimization sub-problem solving as an auxiliary iteration for determining thrust control laws of every subsection trajectory until the main iteration and the auxiliary iteration are both converged. The planet gravity-assist low-thrust trajectory optimization method based on the decomposition and coordination strategy reduces the complexity of an optimization model and the sensitivity of the trajectory constraint to optimized parameters, and improves the convergence efficiency of the transfer trajectory optimization problem.

The invention provides a plant oil classifying method based on combination of a GC-MS technology and a PSO-SVM algorithm. The method comprises the following steps: using the GC-MS technology to obtain the qualitative and quantified results of fatty acids in six different kinds of plant oils and establish the fatty acid fingerprint database of the plant oils; adopting the POS optimal parameter-based SVM classification algorithm to classify samples and establish a PSO-SVM model; and substituting data to the PSO-SVM model, and determining the class of the samples according to a calculation result. The classifying method has the advantages of simple pretreatment operation, high sensitivity and high accuracy.

The invention discloses a solder paste printing quality prediction method and system based on IGA-DNN, and the method comprises the steps: collecting solder paste printing key process parameters and solder paste printing relative volume original data, carrying out the preprocessing, determining a deep neural network topological structure according to the types of input and output variables, the solder paste printing relative volume predicted by the deep neural network is used as a fitness function of the genetic algorithm; finally, a genetic algorithm is used for optimizing and solving the weight and the threshold value of the deep neural network, algorithm model training is completed through a training sample. However, with a traditional neural network which adopts a gradient descent method to update hyper-parameters of the network, the network search speed is low, and local optimum is extremely prone to occurring. According to the method, the hyper-parameters of the deep neural network are initialized based on the genetic algorithm, the convergence speed and precision of the neural network are improved, high precision and stability of SMT solder paste printing quality prediction are achieved, and meanwhile corresponding technical method support is provided for SMT solder paste printing process parameter optimization.

The invention provides a power distribution network source network load storage collaborative optimization scheduling model and method considering carbon emission, which are used for collaborative scheduling of a load part and an energy storage part of a power distribution network. The method comprises the following steps: S1, selecting a load storage element of a load storage collaborative scheduling model by taking an optimized carbon emission level as a standard; s2, taking the carbon emission level of the power distribution network, the scheduling operation cost and the voltage stability as optimization objectives, and taking the active output of the distributed power supply, the reactive output of the distributed power supply, the input level of a switchable capacitor, the tap adjustment of an on-load tap-changing transformer and the network frame of the power distribution network as decision variables, establishing a power distribution network source network load storage collaborative optimization scheduling model considering carbon emission; s3, solving the scheduling model established in the step S2 by using a multi-target particle swarm algorithm with elite strategy rapid non-dominated sorting; and carrying out scheduling according to a solving result; The method can effectively reduce the carbon emission level of the power distribution network, improves the operation economy of the power distribution network, and improves the voltage stability of the system.

The invention belongs to the technical field of maintenance equipment inventory data processing, and discloses a maintenance equipment site selection and inventory method under a multi-stage supply chain and an inventory warehouse, and the method comprises the steps of carrying out the optimization and updating of a constructed warehouse site selection and inventory collaborative optimization system of maintenance equipment under the multi-stage supply chain through employing a hybrid fruit fly optimization algorithm; and determining the opening position of the base warehouse, assigning a maintenance point to the base warehouse, and making an inventory control decision, so that the sum of the base warehouse site selection cost, the ordering cost and the storage cost is minimum. According to the invention, the opening position of the base warehouse can be determined, and the sum of the site selection cost, the ordering cost and the storage cost is minimized through the assignment from the maintenance point to the base warehouse and the inventory control decision. According to the invention, the JRD model structure based on the RAND approximation algorithm is derived and arranged again, so that the HFOA algorithm is designed to be very suitable for solving the model, the algorithm is not prone to falling into local optimum and is stable in convergence, and the invention has higher solving efficiency for the large-scale site selection inventory problem.

The invention relates to a public opinion information classification method and device, a computer device and a storage medium. The method comprises the steps that a classification model is established, wherein the classification model includes word vector models and multiple layers of cyclic neural networks; public opinion information is acquired, wherein the public opinion information includes multiple sentences; training of the word vector models is utilized to obtain sentence vectors corresponding to the multiple sentences, and the sentence vectors corresponding to the multiple sentences are utilized to generate weight matrixes; codes corresponding to the multiple sentences respectively are acquired, and the codes of the multiple sentences are input to the multiple layers of trained cyclic neural networks; based on the codes of the multiple sentences and the weight matrixes, calculation is performed through the multiple layers of trained cyclic neural networks, and categories of the multiple sentences are output; according to the categories of the multiple sentences, the categories corresponding to the public opinion information are determined. By adopting the method, a large amount of public opinion information can be effectively classified.

The invention provides an elastic wave least squares inverse time migration method based on an acoustic-elastic coupling equation. The method comprises the steps of: firstly, inputting an initial longitudinal and horizontal wave velocity and a density initial model, synthesizing a shot end forward propagation wave field and a detection end reverse propagation wave field by using an acoustic-elastic coupling propagation equation and an accompanying equation thereof, and figuring out a multi-parameter image according to the imaging conditions of the longitudinal and horizontal wave velocity anddensity; then, synthesizing reflected wave data by back migration according to a first-order Bonn scattering equation of the acoustic-elastic coupling equation; and evaluating the imaging result according to a four-component waveform residual objective function, if the imaging result is not satisfied, iteratively updating the longitudinal and horizontal wave velocity and density imaging results byusing a CGLS algorithm of a diagonal Hessian precondition, until the condition is satisfied, and returning a final result.