274results about How to "Convergent stability" patented technology

The embodiment of the invention provides a neural network training and construction method and device for object detection, and an object detection method and device based on a neural network. The neural network training and construction method for object detection comprises the steps: inputting a training image comprising a training object into the neural network, so as to obtain a prediction boundary frame of the training object; obtaining a first loss function according to the ratio of the intersection area and union area of the prediction boundary frame and a real boundary frame, wherein the real boundary frame is the boundary frame of the training object marked in the training image in advance; and adjusting the parameters of the neural network at least through the first loss function, so as to carry out the training of the neural network. According to the invention, the first loss function is employed for the regression of a target boundary frame into an integrated unit, and the object detection precision of the neural network is remarkably improved. Moreover, the training and detection efficiency of the neural network can be effectively improved through two branch structures of the neural network.

A training circuit for training a tap coefficient of an adaptive equalizer that performs error calculations in a training process and updating of the tap coefficient in a time area and thereby makes the tap coefficient converge stably in a short time.

The invention provides an underwater autonomous robot fixed depth control method based on reinforcement learning and belongs to the underwater robot control field. The method comprises steps that firstly, a Markoff decision process model for underwater autonomous robot fixed depth control is constructed, and expressions of a state variable, a control variable, a transfer model and a one-step lossfunction for underwater autonomous robot fixed depth control are respectively acquired; the decision network and the evaluation network are respectively established; through reinforcement learning, the decision network and the evaluation network can be continuously updated whenever an underwater autonomous robot moves forwards for each step during fixed depth control training till convergence; thefinal decision network for fixed depth control is acquired. The method is advantaged in that fixed depth control on the underwater autonomous robot under the condition of a completely-unknown underwater autonomous robot dynamic model is realized, and the practical value is high.

The invention discloses a face image restoration method based on a multi-discriminator generative adversarial network, and the method comprises the following steps: (1) carrying out the preprocessingof images in a disclosed face image database, and inputting the images into a generator, so as to obtain a generated image; (2) inputting the real image and the generated image into a plurality of discriminators to obtain a feedback value; (3) taking the feedback values of the plurality of discriminators as confrontation loss, and performing confrontation training on the generative adversarial network by combining the perception loss and the reconstruction loss; and (4) inputting the missing face image into the trained generator to obtain a repaired face image. Aiming at the problem of repairing the shielded or damaged face image, a generative adversarial network structure with multiple discriminators is adopted, so that the problem of low authenticity of the repaired image is solved, andthe repaired image is more natural and more real.

The invention provides a finite-time full-order sliding mode control method of a quadrotor unmanned aircraft. For the quadrotor unmanned aircraft whose model parameters are coupled in a non-linear manner, a full-order sliding mode control method is utilized and a one-order filter is adopted to design the finite-time full-order sliding mode control method of the quadrotor unmanned aircraft. The design of full-order sliding mode surfaces ensures the rapid and stable convergence of the system. In a practical control system, two full-order sliding mode surfaces including the design position and attitude angle are degisned as inner and outer control rings to realize rapid tracking of the aircraft, and the filter is added to improve the buffeting problem. The invention provides the finite-time full-order sliding mode control method of the quadrotor unmanned aircraft, based on the condition that the system model parameters are coupled in a non-linear manner, rapid convergence to a balance point in finite time of the system is ensured, the response speed of the system is increased, the buffeting of sliding mode control input is improved, and the rapid and stable tracking performance of the system is effectively realized.

The invention provides a method for performing joint inversion on the P-wave and S-wave velocity ratio of a reservoir by utilizing P-wave and converted wave data. According to the method, an inversion target function is constructed by adopting a concept that a gamma change rate acquired from gather seismic reflection information and a gamma change rate acquired from travel time information are infinitely approximate, and key parameters of the target function comprise a P-wave intercept, a P-wave gradient, a converted wave gradient and the P-wave and S-wave velocity ratio calculated according to the travel time information, and can be acquired from the actually input amplitude versus offset (AVO) information of a P-wave and converted wave gather and the travel time information, so that the target function can be rationally constructed, and a finally inverted P-wave and S-wave velocity ratio elastic parameter is accurate.

The invention discloses a three-phase load flow calculation method of a weak looped distribution network comprising multi-type distributed power sources. Firstly, the effective three-phase load flow calculation method of a weak looped distribution network is put forward in a distribution sequencing network based on a sequence component calculation method and in combination with characteristics of the weak looped distribution network and a loop analysis of a road. Secondly, detailed formula derivation of iterative computation of distributed generation (DG) of a PQ node type, a PQ (V) node type, an OV node type and a PI node type is conducted and the easy formula is introduced into the load flow calculation method. The three-phase load flow calculation method of the weak looped distribution network comprising the multi-type distributed power sources can effectively process multi-type DG and the weak looped distribution network and is distinct in calculation procedure, simple in programming, and easy to achieve. In addition, the three-phase load flow calculation method of the weak looped distribution network comprising the multi-type distributed power sources retains strong points of rapid calculation and stable convergence of a forward-back substitution method catering to branch circuits. Finally, effectiveness and good convergence of the three-phase load flow calculation method of the weak looped distribution network comprising the multi-type distributed power sources is tested and verified by means of a IEEE (institute of electrical and electronic engineers)-37 generatrix test calculation case, and the three-phase load flow calculation method of the weak looped distribution network comprising the multi-type distributed power sources has good ability of processing the multi-type DG and the looped network.

An accelerator beam track control method and system based on depth reinforcement learning aim to control the accelerator beam track at a target state; the method comprises the following steps: using adepth reinforcement learning method and the training data to pre-train a depth nerve network, and storing weight parameters of the trained depth nerve network and experience data of a track control strategy; using a beam position monitor to obtain the state data of the beam track online, returning the state data into the depth nerve network, and coupling the output of the depth nerve network to abeam track corrector; loading the weight parameters of the trained depth nerve network and the experience data of the track control strategy, and allowing the depth nerve network to adjust the control parameters through prediction control and on-line reinforcement learning, thus adaptively stably controlling the beam track at the target state.

The invention discloses a neural network full order slip form control method for a mechanical arm servo system, which, aiming at the mechanical arm servo system containing a dynamic execution mechanism, is designed by utilizing the full-order slip form control method and combining with the neural network. The full order slip form is designed to guarantee the finite time convergence of the system and can eliminate buffeting and singular problems by avoiding the differential item in a practical control system. Besides, the neural network is used for approaching unknown non-linearity of the system and non-determinacy of internal-external disturbance. The neural network full order slip form control method for a mechanical arm servo system realizes provides a method to eliminate the buffeting and the singularity problem of the surface of the slip form, effectively complements the unknown linearity and the control method of the internal-external disturbance, and realizes fast and stable control of the system.

The invention discloses a novel neural network model for simulating biological bidirectional cognition capability, and a training method. The model consists of a positive neural network and a negative neural network. The positive neural network completes the simulation of a positive cognition process from the input to the output, and the negative neural network completes the simulation of a positive cognition process from the output to the input, wherein the two neural network structures are symmetric, and share a weight. The corresponding connection weight matrixes of the positive and negative neural networks are in a transposition relation. According to the invention, a coordinative structure of the positive and negative neural networks which are symmetric in structure and shares the weight is built, thereby achieving the simulation of the biological bidirectional cognition capability. A mode of negative learning process is introduced into a process of a standard BP algorithm, and a novel neural network training method is proposed.

The invention discloses an enhanced-type constant-speed reaching law sliding-mode control method for a four-rotor-wing unmanned plane system. The enhanced-type constant-speed reaching law sliding-modecontrol method is designed for the four-rotor-wing unmanned plane system through combining with a constant-speed reaching law sliding-mode control method. The design of an enhanced-type index reaching law aims at guaranteeing that the sliding mode of the system can reach a sliding mode plane more quickly while the vibration of the system is not increased, thereby achieving the quick and stable control of the system.

The invention discloses a text intention recognition method and system based on projection gradient descent and label smoothing, and relates to the field of natural language processing question-answering systems. The method comprises the following steps: (1) obtaining an initial vector code through an embedded layer; (2) adding disturbance meeting L2 constraints to an embedded layer by using a projection gradient descent algorithm to form an adversarial sample; (3) using a Transformer network to encode the context semantic information; (4) scaling the category of the real intention by using label smoothing; (5) inputting the encoder output features into a classifier, and calculating the cross entropy between the encoder output features and the smoothed labels; (6) optimizing a target function; and (7) after model training is completed, predicting and outputting intention categories. According to the model, in a classification task, sufficient semantic vector coding can be carried out on an input intention; meanwhile, disturbance is added to a text embedding layer to form an adversarial sample, label sliding is conducted on a final classification target, and the robustness and generalization capacity of the model can be remarkably improved.

The invention discloses an enhanced-type double-power reaching law sliding-mode control method for a four-rotor-wing unmanned plane system. The enhanced-type double-power reaching law sliding-mode control method is designed for the four-rotor-wing unmanned plane system through combining with a double-power reaching law sliding-mode control method. The design of an enhanced-type double-power reaching law aims at guaranteeing that the sliding mode of the system can reach a sliding mode plane more quickly while the vibration of the system is not increased, thereby achieving the quick and stable control of the system.

The invention provides a morphological analysis inverse iteration method for determining space structural lofting states of a cable-strut-beam system. The method includes: using a nonlinear finite element inverse iteration method with construction process considered, using an inactivating-activating unit method to positively simulate construction process influences based on assumption of the lofting similar state, using a morphological compensation method to inversely correct a lofting state geometrical configuration and pretension scheme, gradually astringing a positive tension analysis result of a lofting similar state to a preset design state through positive-inverse loop iteration, and finally obtaining a precise lofting state geometrical configuration and pretension control scheme.

The invention relates to a quad-rotor aircraft nonlinear sliding mode pose control method based on an error exponential function. The invention design a quad-rotor unmanned aerial vehicle nonlinear sliding mode pose control method by using a nonlinear sliding mode control method in allusion to a quad-rotor unmanned aerial vehicle system containing a dynamic execution mechanism. The design of a sliding mode surface is to ensure quick and stable convergence of the system. In addition, the robustness and the tracking accuracy of the system can be improved by adopting a nonlinear function to design the sliding mode surface. The quad-rotor aircraft nonlinear sliding mode pose control method based on the error exponential function realize quick and stable control of the system.

The invention discloses a distributed generation power distribution network three-phase load flow calculation method taking a phase sequence mixing method into consideration. The distributed generation power distribution network three-phase load flow calculation method comprises the following steps that firstly, a power distribution network three-phase load model, a network phase sequence parameter model and a distributed power supply access model are built based on the phase sequence mixing method, an effective three-phase unbalance power distribution power network improved load flow calculation method is disclosed in a power distribution sequence network, and a sequence component method is fully utilized to process the advantages that a three-phase unbalance system has access to symmetrical distributed generation (DG) and PV nodes; and then PQ, PQ(V), PV and PI node type DG calculation models are deduced in detail and are induced into the discloses load flow calculation method. The distributed generation power distribution network three-phase load flow calculation method taking the phase sequence mixing method into consideration has the advantages of being clear in calculation process, simple to program, quick in calculation speed and good in convergence. Finally, the effectiveness and generality of the distributed generation power distribution network three-phase load flow calculation method are tested and verified through diffset. The distributed generation power distribution network three-phase load flow calculation method has good convergence and strong capacity of processing DG nodes and generating reactive power boundary cross.

The invention discloses a self-adaptive adjustment method of an aeromagnetic interference compensation coefficient, relating to the aeromagnetic interference compensation technology field and particularly relating to a regulation method of the aeromagnetic interference compensation coefficient. The self-adaptive adjustment method of the aeromagnetic interference compensation coefficient is aimed to solve the problem that the in the prior art, the aeromagnetic interference compensation solving matrix is not accurate in the aeromagnetic interference compensation coefficient which is caused by morbidness. The self-adaptive adjustment method of the aeromagnetic interference compensation coefficient comprises steps of expressing the magnetic interference generated by an airplane as HI, performing extension on a T-L model, performing fusion on a geomagnetic field HE and a magnetic interference HI, setting an initial value Theta 0,P0, calculating a final compensation coefficient vector Theta 1 of first collection data according to a formula of Theta m=Theta m-1+Km (ym-Delta m Theta m-1), obtaining a final compensation coefficient vector Theta N until N times of collection data are finished, enabling the estimate value of a compensation coefficient vector to be Theta <^ >=Theta n, and using a Theta <^ > as a self-adaptive regulation result of the aeromagnetic interference compensation coefficient. The self-adaptive adjustment method of the aeromagnetic interference compensation coefficient is applicable to the adjustment of the magnetic interference compensation coefficient.

The invention relates to an extended-state-observer-based full-order sliding mode control method of a chaotic system of permanent magnet synchronous motor. The method comprises: a chaotic model, an initialization system state, and correlated control parameters of a permanent magnet synchronous motor system are established; on the basis of coordinate conversion, the permanent magnet synchronous motor chaotic model is converted into one based on a standard extended state observer design suitable for a non-linear extended state observer design; a full-order sliding mode controller is designed based on the extended state observer to eliminate a buffeting phenomenon in the sliding mode control and guarantee that the system state can be convergent to a zero point rapidly and stably. According to the full-order sliding mode control method, the buffeting phenomenon in sliding mode control can be prevented and the chaotic state of the system can be guaranteed to be convergent to the zero point rapidly and stably.

The invention discloses a multi-fighter cooperative fault diagnosis method under switching topology. The method includes the steps that firstly, a multi-fighter system connecting diagram of each topology structure is constructed and represented with a digraph to obtain the Laplacian matrix and self-cycle matrix of each topology structure; secondly, a state equation and an output equation of each node fighter are obtained, and a system model with an executor fault is set up for each node; then, a global augmented system state equation based on the digraphs is constructed for each node; finally, a multi-fighter cooperative fault diagnosis observer is set up, collected input data and output data of the all the nodes are sent into the observer to obtain the fault estimated value of each node, and therefore the executor faults of the flight control system are estimated. By means of the method, the fault of any node and the faults simultaneously appearing in multiple nodes can be diagnosed on line in the cooperative flight control system, and online fault diagnosis and real-time fault estimation of the flight control system can be carried out.

The invention relates to a small unmanned helicopter nonlinear control method. A small unmanned helicopter self-adaptive control method based on an immersion and invariant set method is provided, and realizes the effect that a small unmanned helicopter can maintain a control effect with stable posture under the situation of parameter uncertainty. Thus, the technical scheme adopted by the invention provides a posture nonlinear self-adaptive control method for the small unmanned helicopter, which applies the immersion and invariant set method to the control of a small unmanned helicopter posture system while the small unmanned helicopter is under the situation of parameter uncertainty. The posture nonlinear self-adaptive control method is mainly applied to small unmanned helicopter nonlinear control.

The invention discloses a gradient extremum search algorithm-based double closed-loop lighting energy-saving control method. The illumination requirement is kept through PID closed loop control; meanwhile, continuous iteration is carried out through a gradient extremum search algorithm to quickly search out the minimum value of energy consumption of a lighting system; and stable output of the minimum value is kept. The control has an important control role when applied to the lighting system. Aiming at the illumination energy consumption problem, the most ideal condition is to ensure the regional illumination demand and to reduce the energy consumption to the minimum value. According to the gradient extremum search algorithm-based double closed-loop lighting energy-saving control method, the grouping condition, namely the percentage of a lamp bank (each group of lamps) is adjusted through a Newton extremum search control algorithm, so that the condition that an illumination control region reaches the illumination requirement, i.e., the target illumination value is ensured; meanwhile, low-energy consumption stable output of the lighting system can be kept; the requirements of users are met; and the energy-saving effect is also achieved.

The invention provides an integrated least-mean-square (LMS) adaptive filter, which comprises a digital filter, wherein the digital filter integrates a step length 2 mu which is a decimal fraction into an integer of more than or equal to 1, and outputs an output signal y(n) which is integrated and amplified at the same time; an output end of the digital filter is connected with an error calculation module which is used for performing order-of-magnitude matching calculation on a desired signal d(n) and an error signal e(n); and by integrated processing, the decimal multiplication operation of the LMS adaptive filter is avoided, and the complexity of an implementation circuit is reduced. The invention also provides a method for acquiring an adaptive weight coefficient by using a variable step length. The invention has the advantages that: by arranging an error judgment module, the step length is variable; and a large signal can be stably and quickly converged during the guaranteeing of the convergence speed of a small signal.

The invention provides a four-rotor aircraft nonlinear sliding mode position and attitude control method based on a single exponential function. For a four-rotor UAV (unmanned aerial vehicle) system having a dynamic execution mechanism, the invention designs the four-rotor UAV system nonlinear sliding mode position and attitude control method by utilizing a nonlinear sliding mode control method. The design of a sliding mode surface is to ensure fast and steady convergence of the system; and besides, since the sliding mode surface is designed based on a nonlinear function, robustness and tracking precision of the system can be improved. The four-rotor aircraft nonlinear sliding mode position and attitude control method based on the single exponential function realizes fast and stable control of the system.

The invention relates to an enhanced fast power reaching law sliding mode control method of a quad-rotor UAV system. The enhanced fast power reaching law sliding mode control method is designed for the quad-rotor UAV system by combining with a fast power reaching law sliding mode control method. The design of an enhanced fast power reaching law is used for ensuring a sliding mode of the system canreach a sliding mode surface more quickly, does not increase the chattering phenomenon of the system, and achieves fast and stable control of the system.

In order to provide an X-ray CT apparatus and the like that reconstruct an image using an iterative approximation method which ensures stable convergence and can be executed at high speed, a computation device 5 of an X-ray CT apparatus 1 calculates matrices A, B, D, R, and R′ on the basis of the scanning conditions input through an input device 6 (step 1). Then, the computation device 5 calculates each element of a matrix I−α(SB^TDA+βSR) (step 2). Then, the computation device 5 calculates the operator norm ∥I−α(SB^TDA+βSR)∥ of the matrix I−α(SB^TDA+βSR) (step 3). Then, the computation device 5 determines a relaxation coefficient α such that a predetermined conditional expression is satisfied (step 4).