658results about How to "Guaranteed convergence" patented technology

The invention provides a hardware neural network conversion method which converts a neural network application into a hardware neural network meeting the hardware constraint condition, a computing device, a compiling method and a neural network software and hardware collaboration system. The method comprises the steps that a neural network connection diagram corresponding to the neural network application is acquired; the neural network connection diagram is split into neural network basic units; each neural network basic unit is converted into a network which has the equivalent function with the neural network basic unit and is formed by connection of basic module virtual bodies of neural network hardware; and the obtained basic unit hardware networks are connected according to the splitting sequence so as to generate the parameter file of the hardware neural network. A brand-new neural network and quasi-brain computation software and hardware system is provided, and an intermediate compiling layer is additionally arranged between the neural network application and a neural network chip so that the problem of adaptation between the neural network application and the neural network application chip can be solved, and development of the application and the chip can also be decoupled.

The invention relates to a three-dimensional entity model surface finite element mesh automatic generation method which has the characteristics of good dimension control performance of mesh quality, density and dimension, good convergence in a dividing process, high calculation efficiency, stability and reliability. The method comprises the following steps: (1.1) establishing a three-dimensional geometrical model, reading an STL neutral file of a model surface, and reconstructing a topology relation of triangle patches; (1.2) according to a relation between a triangle patch and an adjacent patch, grouping surface triangle patches, wherein each group of triangle patches is a characteristic face; (1.3) determining boundaries of each characteristic face, and determining characteristic edges on each characteristic face boundary; (1.4) according to mesh density distribution, generating mesh nodes on each characteristic edge, and expressing boundaries of the characteristic face with mesh node rings on the boundaries; (1.5) selecting a cut surface and an optimal dissection surface, and generating mesh nodes on cut lines and dissection lines; (1.6) generating a surface mesh on each characteristic face, and carrying out smoothing process.

The invention provides a robot adaptive grabbing method based on deep reinforcement learning. The method comprises the following steps: when distanced a certain distance away from an object to be grabbed, a robot obtaining a picture of a target through a pick-up head in the front, then according to the picture, calculating position information of the target by use of a binocular distance measurement method, and applying the calculated position information to robot navigation; when the target goes into the grabbing scope of a manipulator, taking a picture of the target through the pick-up head in the front again, and by use of a DDPG-based deep reinforcement learning network trained in advance, performing data dimension reduction feature extraction on the picture; and according to a feature extraction result, obtaining a control strategy of the robot, and the robot controlling a movement path and the posture of the manipulator by use of a control strategy so as to realize adaptive grabbing of the target. The grabbing method can realize adaptive grabbing of objects which are in different sizes and shapes and are not fixedly positioned and has quite good market application prospect.

The invention discloses a three-dimensional multi-missile cooperative guidance method with finite time convergence. The method comprises the steps that for an unknown target, a multi-missile cooperative guidance model with attacking angle constraint in three-dimensional space is established; by adaptive control, integral sliding model control and a finite time consistency protocol, an acceleration guidance law in the sight direction is established to make all missiles hit the target at the same time; through the adaptive control and non-singular fast terminal sliding model control, and the acceleration guidance law in the sight direction is designed to make the finite time convergence of a sight angular velocity rate and a sight angle of each missile achieved. According to the three-dimensional multi-missile cooperative guidance method with the finite time convergence, all the missiles can hit the target at a desired angle at the same time on condition that no maneuvering information of the target is known, and at the same time the occurrence of buffeting is avoided.

The invention discloses an optimization method for solving an omni-channel logistics distribution problem. In a first stage, a Lagrange relaxation technology is used to solve a LAP (Location-allocation Problem) problem. In a second stage, adaptive large-scale neighborhood search is used for solving a multi-vehicle-type vehicle path problem, a feasible solution can be searched in a large range in a solution space through one group of simple destruction and reconstruction algorithms, and a situation of falling into local optimum can be effectively avoided. Meanwhile, a simulated annealing acceptance criteria realized in the adaptive large-scale neighborhood search can guarantee the quality of a solution and the convergence of the algorithm, a result is output after the algorithm is executed for an appointed iteration number, a time constraint requirement can be met, and a good vehicle distribution scheme is solved for enterprises in limiting time. The simpleness of the Lagrange relaxation technology is combined with the efficiency of the adaptive large-scale neighborhood search, so that the integral solving efficiency of a mixed algorithm is high, and the problem of omni-channel logistics distribution can be effectively solved.

The invention discloses an in-orbit moving target detecting method with space-based radar and infrared data fusion. The method includes the steps that when an infrared detecting system is used for detection, background suppression is carried out on each frame in an infrared image sequence by adopting a filtering method based on the morphology, self-adaptive threshold segmentation is carried out on images obtained after background suppression, a single frame detection result is extracted, a multi-frame target is determined through a tracking algorithm, and whether the detection result is a real target or not is judged to obtain the detection result of the multi-frame target; during radar detection, pulse doppler (PD) is adopted for processing. After time calibration and space calibration are carried out on the obtained infrared and radar target information, data fusion is carried out on the information by adopting a measurement fusion method, and track prediction is carried out on fused data to obtain the estimated position of the target.

The invention discloses a multi-objective deep convolution generative adversarial network model and a learning method thereof so that problems that the exiting deep convolution generative adversarialnetwork model can not be converged easily and training is not stable can be solved. A target deep convolution generative adversarial network model is provided; on the basis of a group search strategy,multiple groups of deep convolution generative adversarial networks are trained simultaneously to realize coevolution of multiple individuals, thereby ensuring stability of model training; with a Pareto dominant mechanism, a potential optimal combination unit of generative networks and adversarial networks is selected for follow-up training in each iteration so as to ensure convergence of the model training; on the basis of the characteristics of the deep convolution network, a cross operator is designed to realize interaction of parameter information between different networks; and because of combination of the global search ability of the evolutionary algorithm and fast local search ability of the gradient descent algorithm, the accuracy and effectiveness of a learning frame put forwardnewly are ensured.

A camera with dynamic calibration and a method thereof is provided. The camera is first subject to an initial calibration. Then, a motion amount of the camera is calculated, and a plurality of motion amount estimation samples of the camera is generated according to the motion amount. Then, a weight of each of the motion amount estimation samples is calculated. Thereafter, the plurality of motion amount estimation samples is re-sampled based on the weights, and the camera is calibrated by the re-sampled estimated motion samples.

The invention discloses a multi-agent formation tracking control method and system. The method comprises the steps that the position of each agent is obtained; according to the position of each agent,the position status of a navigator, the position statuses of followers and the position status of each follower relative to the navigator are determined, thereby obtaining an agent formation; an augmented object is introduced, and a fault and saturation model of the agent formation is established; the status of the navigator is taken as reference to determine an objective function of time-varyingformation tracking based on the fault and saturation model; an adaptive fault-tolerant control law under saturation constraints is established by using an output signal of a first-order filter and anoutput signal of the augmented object, a parameter of the control law is determined, and an adaptive fault-tolerant and anti-saturation control law is obtained; the objective function is taken as a control target, and each agent is controlled according to the adaptive fault-tolerant and anti-saturation control law. The multi-agent formation tracking control method and system have the advantages that the convergence of the multi-agent formation tracking error is ensured, and even if in the case of the fault, a control signal is within the saturation boundary of an actuator.

The invention relates to a cable net reflecting surface antenna surface accuracy and tension state simultaneous design method which is characterized by at least including the steps of firstly, determining an initial balance reference configuration, namely the configuration when k=0, marking the corresponding node position as X<(0)>, and marking a cable section lofting length as L0<(0)>; secondly, determining the node displacement Us<(k)> required by the cable net during the kth iteration; thirdly, on the basis of the kth balance, determining the sensitivity matrix T / L0 of the cable section tension to the cable length and the sensitivity matrix X / L0 of the node position to the cable length during the kth iteration; fourthly, using a optimization model to solve the kth cable length correction amount; fifthly, correcting the cable section lofting length to be L0<(k+1)>=L0<(0)>+delta L0<(k)> after the kth iteration; sixthly repeating the process until the given convergence condition is satisfied. By the method, cable net tension is allowed to be even while the antenna surface accuracy is guaranteed.

The invention discloses a preset performance ocean bottom flying node trajectory tracking control method based on a disturbance observer, relates to a preset performance ocean bottom flying node trajectory tracking control method, and aims to solve the problems that an existing method does not consider modeling uncertainty and ocean environment disturbance and propeller faults affect the OBFN (Ocean Bottom Flying Node). The method comprises the following steps that: 1: establishing a Fossen outline six-degree-of-freedom nonlinear kinetic model; 2: carrying out OBFN kinetic model transformationon the nonlinear kinetic model established in S1 to obtain an OBFN kinetic model, and determining an OBFN tracking error equation according to the OBFN kinetic model; 3: establishing a performance function; 4: carrying out error transformation on the tracking error in the S3 to obtain a transformed error; and 5: according to the transformed error obtained in S4, designing an OBFN system total uncertainty observer and a preset performance trajectory tracking controller. The method is used in the field of trajectory tracking control.

The invention discloses an internal temperature estimation method for a lithium battery based on a discrete sliding-mode observer. The method comprises the following steps that a lithium battery discrete state space model for internal temperature estimation is established; a working voltage, a working current, a surface temperature and an external environment temperature of the lithium battery under a working condition are collected in real time; parameters of the lithium battery discrete state space model are identified online; and the discrete sliding-mode observer is constructed to estimate the internal temperature of the lithium battery in real time. The method has a good temperature estimation effect, can ensure the convergence of the observer strictly, and is highly robust for modeling errors, model parameter perturbation and external disturbance of the lithium battery thermal model.

The present invention discloses a robust adjusting method of surface precision on a cable network reflective surface based on finite element model correction. The method comprises: building a reference model of an antenna structure on a cable network reflective surface, calculating a sensitivity matrix relating to a cable length in a net surface node position, measuring a position of each node on an entity reflective surface, calculating a desired node displacement when a reflective node is adjusted to an optimal coinciding paraboloid, acquiring adjusting quantity of an optimal robust cable length, exerting the adjusting quantity to an entity reflective surface, measuring to obtain node position of entity reflective surface after being adjusted, simultaneously exerting adjusting quantity of an optimal robust cable length to the reference model, acquiring node position of the reflective surface after the reference mode being adjusted, and implementing correction to the reference model; and repeating the steps until the precision does not increase, that is, finishing robust adjustment for surface precision on an antenna cable network reflective surface. According to the present invention, the influence of error factors and robustness of the adjusting quantity are considered, which ensures astringency during the process of adjusting, and significantly increases the efficiency of reflector surface adjusting.

The invention relates to a method for registering a complex curved surface. A system in the method comprises a curved surface CAD design model, a three-coordinate measuring machine and a registering solver, wherein the curved surface CAD design model comprises geometric information of the curved surface; after the geometric information of the curved surface is extracted, a curved surface NURBS model is constructed; the three-coordinate measuring machine detects a curved surface real object prototype, and acquires measuring point data; and the registering solver registers the measurement data of the curved surface NURBS model and the curved surface real object prototype by adopting a genetic algorithm. Compared with the prior art, the method has the advantages of accurate theory model, high registering stability, good registering accuracy, capability of achieving multidimensional variable registrations, and wide application range.

Provided is a boiler combustion optimization air distribution method based on online model prediction. The method includes: constructing a boiler combustion optimization air distribution model by employing an improved BP neural network algorithm and model online update calculation; and performing optimization calculation on parameters such as the optimal oxygen amount, the primary air amount, each secondary air door and a burnout air door etc. with the combination of an adaptive genetic algorithm and dynamic optimization boundary so that a multi-target function constructed by the boiler efficiency and the pollutant discharge capacity reaches the optimal combustion range.

The invention provides a multi-unmanned aerial vehicle cooperative task allocation method based on an improved genetic algorithm. The method comprises the following three steps: establishing constraint equations such as the minimum turning radius of an unmanned aerial vehicle and the number of tasks required by a target, and a multi-unmanned aerial vehicle cooperative task allocation model based on Dubins flight path cost; generating an initial population of a predetermined scale conforming to model constraint conditions; taking the Dubins track path cost of the unmanned aerial vehicle as a fitness function, and iteratively updating the initial population by using genetic operations such as elite strategy, selection, crossover, variation and the like of an improved genetic algorithm to generate a feasible solution which minimizes the target function in fixed iteration times, and taking the feasible solution as a result of multi-unmanned aerial vehicle cooperative task allocation and route planning. The method has wide application value in multi-unmanned aerial vehicle cooperative task combat, is beneficial to implementation of multi-unmanned aerial vehicle multi-target cooperativetask execution, and improves the task completion efficiency. The method has important significance in the field of multi-unmanned aerial vehicle cooperative control.

The invention discloses a processing method for correcting error of a low-gradient optical mirror surface, which comprises the following steps: firstly acquiring a removal function through experiment, then using a wave surface interferometer to acquire a surface shape error function of an element, dispersing the surface shape error function, programming a processing path at the same time, extending the surface shape error function, establishing a model and solving processing residence time, carrying out numerical control processing by using the solved residence time, identifying error and controlling accuracy for the processed result, so as to finish processing. The processing method has the advantages of high process efficiency, low processing cost and good error correction effect, and can effectively correct the surface shape error of the low-gradient optical mirror surface.

The invention discloses a self-adaptive robust control method for a nonlinear bilateral teleoperation system. For a communication time delay problem of the bilateral teleoperation system, by online estimating a non-power environmental parameter of a slave end via a radial basis function neural network model and transmitting the non-power environmental parameter back to a master end to reconstructan environmental force, the passivity of the teleoperation system as well as the balancing problems of stability and transparency is prevented. For problems of nonlinearity and uncertainty of the bilateral teleoperation system, a track generator, and a nonlinear self-adaptive robust controller based on a radial basis function neural network are respectively provided at the master and slave ends, and based on a Lyapunov theory, a self-adaptive rate for online adjusting a neural network parameter is designed, so that the system is guaranteed to have good asymptotic stability and convergence. Theself-adaptive robust control method can guarantee the global robust stability of the nonlinear bilateral teleoperation system well, implements the position tracking of a robot at the slave end, and provides a vivid force feedback for an operator.

The invention aims at providing a novel CKF(Crankshaft Fluctuation Sensor)-based SINS (Ship Inertial Navigation System) large misalignment angle initially-aligning method comprising the following steps of: determining an initial position parameter of a carrier by using a GPS (Global Position System); acquiring data output by an optical fiber gyroscope and a quartz accelerometer; finishing the coarse alignment of the system by using an analysis method; preliminarily determining the posture information of the carrier; establishing an initial aligning nonlinear model of a strapdown inertial navigation system; establishing a CKF filtering state equation by taking the speed error as the state variable and a measuring equation by taking the speed error as the measurement quantity under a static base; carrying out filtering estimation by using a CKF filtering method to estimate the misalignment angle of the platform; and obtaining an accurate strapdown initial posture matrix by using a strapdown initial posture matrix of a platform misalignment angle correcting system, thereby finishing the accurate initial alignment. The invention can greatly improve the aligning precision of the strapdown inertial navigation system at the large misalignment angle and provide the accurate strapdown initial posture matrix for the navigation process.

The invention discloses an iterative learning-based subway train automatic running speed control method. The method comprises the following steps of: 1, establishing a train running dynamic model for an urban railway transit train automatic running speed control system; 2, automatically adjusting a learning gain through output errors and modified functions in an iteration process, and using the learning gain to update the input of a speed controller; and 3, in order to ensure the robustness, for the initial-state errors, of an algorithm, learning an iteration initial state while a controlled quantity is learnt to ensure that the system can restrain to an expected track under any initial condition without requiring the iteration initial state to be accurately located on an expected initial state, so as to finally realize the accurate tracking, for a target speed curve and a target displacement curve, of the train. According to the method, a learning gain initial value and a system state and tracing error of the last iteration are utilized to correct the system initial state of the current iteration, and an iteration initial state correction algorithm is given, so that the convergence, for any system initial state, of a law of learning is ensured.

The invention discloses a discrete variable structure observer-based lithium battery SOC (state of charge) estimation method. The method includes the following steps that: a rapid calibration test is performed on a lithium battery, a relation curve of SOC and open-circuit voltage (OCV) is obtained; a lithium battery discrete state space model for SOC estimation is established; a pulse discharge test is performed on the lithium battery, the parameters of the lithium battery model are identified; the terminal voltage and charge-discharge current of the lithium battery under a real-time collection condition are obtained; and a discrete variable structure observer is constructed to estimate the SOC of the lithium battery accurately. The method not only has a good SOC estimation effect, but also can ensure convergence strictly, and exhibits strong robustness to lithium battery modeling error, internal parameter perturbation and external disturbance.

The invention relates to a fixed time cooperative tracking control method of a second-order nonlinear multi-agent system, which comprises the steps of: designing a fixed time convergence nonsingular terminal sliding mode surface; designing an RBF neural network approximation system dynamics nonlinear term; designing a multi-agent system fixed time cooperative tracking controller; and enabling thedesigned controller to act on the system to realize the control of a multi-agent system for cooperatively tracking a target agent within fixed time. The fixed time cooperative tracking control methodhas strong robustness to external disturbance, has strong adaptability to unknown nonlinearity of a model, and can ensure the convergence of the system within fixed time, and enables the convergence time to be irrelevant to the initial state of the system. The fixed time cooperative tracking control method has wide application range and can be applied to systems with high requirement on convergence time.

The invention provides a method of predicting reservoir fractures based on offset vector tile seismic pre-stack gather data. The reservoir fracture development density and fracture azimuth are directly inverted based on offset vector tile seismic data containing azimuth information. The defects of the conventional narrow-azimuth pre-stack seismic fracture prediction technology, such as low inversion accuracy and low spatial resolution, are overcome. More comprehensive and complete offset and azimuth information can be retained. More suitable seismic data is provided for azimuth anisotropy research. The method is more advantageous to the identification of fault, fracture and reservoir air water variation characteristics.

The invention discloses a transverse federated learning system optimization method and device, equipment and a readable storage medium. The method comprises the steps that a target type of local modelparameter updating needing to be sent by each participating device in each round of model updating is determined from parameter updating types according to a preset strategy, and the parameter updating types comprise model parameter information and gradient information; sending indication information for indicating the target type to each participation device, so that each participation device performs local training according to the indication information and returns local model parameter update of the target type; and fusing the local model parameter updates of the target type received fromeach participating device, and sending the global model parameter updates obtained by fusion to each participating device, so that each participating device performs model updating according to the global model parameter updates. According to the method, the advantages of a gradient averaging algorithm and a model averaging algorithm are combined, and a hybrid federated averaging mechanism is realized.

A nonsingular fixed time adaptive attitude control method of a rigid aircraft aims at an aircraft attitude stabilization problem having the centralized uncertainty, utilizes a sliding mode control method, and combines the adaptive control, so that a nonsingular fixed time adaptive controller is designed. The design of a nonsingular fixed time sliding mode surface guarantees the fixed time convergence of the state of a system, but also solves a singular value problem. In addition, an adaptive updating law is used to estimate the upper bound of the uncertainty and interference of the system, sothat the upper bound information does not need to be known in advance. According to the present invention, on the condition that the system has the uncertainty and the interference, a control method that the fixed time of the state of the system is uniformly ultimately bounded is realized.

The invention discloses a multi-unmanned aerial vehicle distributed type contract auction online task planning method. The method comprises the following steps that 1, a multi-unmanned aerial vehiclecooperative patrol route planning model is set up; 2, the patrol route planning model is solved through a distributed type contract auction online task planning algorithm. According to the method, theperformance and task executing needs of multiple heterogeneous unmanned aerial vehicles are fully considered, a problem model is set up, the completely distributed type contract auction online task planning algorithm is provided for solving a problem, and multi-unmanned aerial vehicle fast task planning is achieved.

The invention discloses a user random access method and device based on ephemeris broadcast auxiliary positioning, which are applied to mobile communication system for satellite. A ground terminal receives a downlink synchronization sequence and an ephemeris periodically broadcasted by a satellite; in a downlink time-frequency synchronization stage, a signal arrival time difference and a Doppler frequency offset sequence are obtained, the Doppler frequency offset sequence is converted into another group of timing information, and in combination with a satellite ephemeris table, a satellite-to-ground distance is estimated and satellite-to-ground propagation time delay is predicted to serve as a random access uplink timing advance. According to the invention, a positioning algorithm based ontime arrival difference is used; the established model can cope with various satellite-ground distance measurement conditions in a satellite scene, so that the timing advance error of each terminal signal can be in a cyclic prefix, and the orthogonality between uplink signals of different terminal users in a cell range when orthogonal frequency division multiple access transmission is adopted isensured. Compared with a random access timing advance scheme redesigned by a leader sequence, the method does not need to increase access power consumption and is easy to implement.

The invention provides an equivalent coordinative optimization method for a large-scale power grid dispatching plan. The optimization method comprises the steps as follows: (1) a large power grid is divided into a plurality of regional power grids, and the regional power grids are taken as equivalent nodes; (2) the regional power grids submit information of the equivalent nodes to the whole grid; (3) the whole grid determines a generation dispatching plan under the equivalent regional power grids, and issues a connecting line plan between the regions, planned total output of the regional power grids and a plan of a direct dispatching set; (4) the regional power grids calculate and determine output plans of regional sets according to the planned total output of the regional power grids, and calculate locational marginal cost; and (5) whether full-grid optimization is converged or not is judged through the marginal cost of two ends of a connection line, if so, calculation is ended, if not, the set which affects the grid security constraint corrects output bounds according to the optimization result and turns to the step (2). The iteration times can be controlled; the convergence is rapid; and rapid determination of the large power grid dispatching plan can be achieved.

The invention discloses a method for controlling the random vibration of an electrohydraulic servo system based on RLS filters, which comprises the following steps: converting an acceleration power spectral density reference signal into an acceleration time domain driving signal; taking an acceleration input signal of the electrohydraulic servo system as that of an RLS filter I; online regulating weights of the RLS filter I and an RLS filter II in real time by using an RLS adaptive filter algorithm and performing identification on a frequency response function of the electrohydraulic servo system and an impedance function of the RLS filter I respectively; and constructing a filter III based on the identification result of the impedance function of the RLS filter II, filtering the acceleration time domain driving signal and taking an output signal of the filtration as the acceleration input signal of the electrohydraulic system. The convergence property of the RLS adaptive filter algorithm adopted by the method guarantees the convergence property of the real-time online iteration of acceleration power spectral density when the characteristics of the electrohydraulic servo system and a test piece are changed.

The invention discloses an autonomous navigation method of an AUV (Autonomous Underwater Vehicle) based on a FastSLAM (Simultaneous Localization and Mapping) algorithm. The autonomous navigation method comprises the steps that 1) the AUV acquires initial pose and position information through the GPS and a navigation sensor on the water surface; 2) predicting the pose and position and an environmental road sign of the AUV by adopting unscented particle filtering according to latest control variables inputted into the AUV and observation variables of the sensor; 3) generating a proposal distribution function for parameter adaptive adjustment by adopting fading adaptive unscented particle filtering, and sampling in the proposal distribution function; 4) associating the latest observation environment information according to each particle, and updating estimation for each characteristic by adopting unscented Kalman filtering; 5) performing resampling on a particle set by adopting an adaptive partial system resampling method; and 6) performing AUV positioning and map building. The autonomous navigation method can improve the particle sampling efficiency of the Unscented FastSLAM algorithm and reduce the degradation degree of the particles through improving the proposal distribution function and the resampling process of the Unscented FastSLAM algorithm, thereby enabling the consistency of AUV pose and position estimation and the accuracy of autonomous navigation to be greatly improved.