26 results about "Nonlinear approximation" patented technology

The invention relates to the technical field of intelligent sensing systems, in particular to an algorithm and system for avoiding water surface floating objects and improving monitoring precision based on radar water level monitoring, and the method comprises the following steps: collecting radar echoes, constructing spatial-temporal characteristics based on waveform asymmetry and coherent attenuation gradient, and calculating the spatial-temporal characteristics; according to the method, the central moments of the front edge and the rear edge of the echo are calculated, the waveform asymmetry is quantified, the dynamic stability is analyzed in combination with the attenuation gradient of the multi-time-lag coherence coefficient, the floating object interference is recognized from the double dimensions of the spatial form and the time memorability, and the actual water level is inverted. An effective scattering trailing interval is positioned, nonlinear approximation operation is performed on trailing data in combination with an index physical model, a shielded water surface energy distribution curve is reconstructed, a derivative zero point is solved, distance measurement deviation caused by floating objects is corrected, the real water level height is effectively restored, and the monitoring anti-interference capability and the data accuracy in a complex hydrological environment are improved.

The invention discloses a self-adaptive fuzzy PID flow field control method and device, and the method comprises the steps: generating an initial PID gain parameter in real time through a fuzzy reasoning module, introducing an RBF neural network module to carry out the dynamic optimization compensation of the initial parameter, and generating a correction gain. And a final PID parameter is dynamically synthesized by adopting a weighted fusion formula, and a weight coefficient is adjustable so as to balance the contribution of fuzzy rules and neural network learning. The nonlinear approximation capability of the RBF network is utilized, network parameters are updated in real time through a gradient descent method, a Jacobian matrix of a controlled system is output, and the sensitivity of the control quantity to input changes is accurately recognized. The parameter updating rate is dynamically adjusted through a momentum item and an exponential decay function in combination with environmental sensor data, flow feedback and historical deviation, and a closed-loop control loop of monitoring, fuzzy reasoning, RBF optimization, PID output and feedback is formed. The accuracy of flow field control is improved.

The application discloses a self-adaptive fuzzy PID flow field control method and device, and the method comprises the following steps: generating initial PID gain parameters in real time through a fuzzy reasoning module, introducing an RBF neural network module to dynamically optimize and compensate the initial parameters, and generating corrected gains. The final PID parameters are dynamically synthesized by using a weighted fusion formula, and the weight coefficient is adjustable to balance the contribution of fuzzy rules and neural network learning. By using the nonlinear approximation ability of the RBF network, the network parameters are updated in real time through the gradient descent method, the Jacobian matrix of the controlled system is output, and the sensitivity of the control quantity to the input change is accurately identified. In combination with the environmental sensor data, the flow feedback and the historical deviation, the parameter update rate is dynamically adjusted through the momentum term and the exponential decay function, and a closed-loop control loop of monitoring-fuzzy reasoning-RBF optimization-PID output-feedback is formed. The application improves the accuracy of flow field control.

The present invention discloses an active control method for a squeeze film damper based on RBF neural network PID optimization. The method comprises the following steps: obtaining rotor system state parameters through a data acquisition module; inputting the state parameters into an RBF neural network, and utilizing the RBF neural network to perform online optimization of the PID control parameters; and, based on the optimized PID parameters, driving a piezoelectric ceramic actuator via a piezoelectric ceramic controller to adjust the damping force of the squeeze film damper, thereby achieving active control of rotor system vibration. By integrating the nonlinear approximation of an RBF neural network with the stability of PID control, the method addresses the hysteresis and locking issues of traditional passive control, achieves active suppression of rotor system vibration, and significantly improves control accuracy and reliability.

The invention provides a drilling-expanding mixed machining material removal power prediction method based on SR-TELM, relates to the technical field of numerical control machine tool power prediction, and deeply excavates a physical mechanism between machining parameters and material removal power in drilling-expanding mixed machining material removal through a symbolic regression algorithm. By embedding a mechanism model into hidden layer neurons of the TELM, the advantage of analyzing a physical mechanism by a symbolic regression algorithm and the strong nonlinear approximation capability of the TELM are organically fused, the power prediction precision is effectively improved under a limited data condition, the physical interpretability of the model is ensured, and the power prediction efficiency is improved. And a new solution is provided for machine tool drilling-expanding mixed machining material removal power modeling.

Provided is an electronic device for controlling surface heart and a method of controlling the electronic device. The electronic device includes a speaker, a temperature sensor, a memory, and a processor electrically coupled to the speaker, the temperature sensor, and the memory. The processor obtains first temperature information based on impedance information measured in a coil included in the speaker; obtains second temperature information measured by the temperature sensor, the second temperature information based on a heat source disposed adjacent to the speaker; predicts a surface temperature of a surface area of the electronic device, opposite to an internal area in which the speaker is disposed, based on the first temperature information and the second temperature information using a nonlinear approximation function; and controls an audio signal input to the speaker based on the predicted surface temperature.

The invention provides a marine three-phase asynchronous motor neural network inversion sliding mode control method, which combines the advantages of nonlinear approximation capability and good real-time performance of an RBF neural network and good dynamic performance and strong anti-interference performance of sliding mode control, makes full use of the back-off decoupling characteristic of a back-stepping method, and improves the reliability of the control system. The neural network inversion sliding mode control method for the three-phase asynchronous motor for the ship is designed, and the purpose of improving the performance of a motor control system is achieved. For the problem that parameters cannot be determined in the modeling process of the marine three-phase asynchronous motor, a forward stability augmentation channel is designed. A backstepping method is adopted to perform decoupling control on the system, sliding mode control solves the disturbance problem of the control system, and the effect of an RBF neural network performs real-time estimation and compensates disturbance between system coupling of contact surfaces between control system frames and nonlinear friction force.

The invention discloses a load identification method and system. The method comprises the following steps: acquiring a first operation parameter of target equipment, and preprocessing the first operation parameter; performing feature extraction operation on the preprocessed first operation parameter to obtain a total feature vector; pre-training a first load identification model, and taking the total feature vector as the input of the first load identification model; and carrying out load identification according to the output of the first identification model. According to the method, the power feature, the waveform feature, the harmonic feature, the V-I track feature and the wavelet scattering feature are combined for identification, the features of the load data can be excavated from multiple angles and fully utilized, and the precision of load identification is improved; according to the method, the wavelet scattering network is constructed, the multi-scale analysis capability and the non-linear approximation capability are achieved, complex signals can be better processed, and the multi-scale features of the signals are effectively extracted.

The invention discloses a curve envelope fitting method based on a VGG16 network. The algorithm uses a data set sample with a label established by CCD acquisition to train a neural network, and applies the neural network algorithm to the acquisition data set to verify the accuracy and infer the microscopic morphological characteristics of the optical fiber surface; a reading data set is created through a tenserflow framework, and a gray value change sequence of the group of images at a certain pixel point (a, b) is recorded as X(a, b)(t); a cubic Hermite interpolation method is used to supplement the sequence X(a, b)(t) into a one-dimensional sequence X2(a, b)(t) with a size of 224*224, and then the sequence X2(a, b)(t) is converted into a two-dimensional image matrix X2(a, b)(m, n); the actual height of the pixel point predicted by the specially designed convolutional neural network is processed and output, and compared with the sample label to make the error within a set threshold range; the application of the neural network enables the algorithm to have better self-learning, self-organizing ability and fault tolerance and excellent nonlinear approximation ability, which can improve the accuracy and fault tolerance of the envelope algorithm and has certain reference significance.

The invention relates to a model-free adaptive temperature control method and system in an infrared heating process, and relates to the technical field of infrared heating control regulation, and the method comprises the steps: describing an infrared heating system as a nonlinear system of discrete time, and constructing a corresponding linearization data model; and then introducing an RBF neural network to estimate the dynamic characteristics of the infrared heating system, and estimating control law parameters on line. A rolling optimization prediction mechanism of model prediction control is used to process time lag characteristics in an infrared heating process, and an RBF-MFAPC controller is constructed. And finally, taking the data acquired in real time as the input of the controller, and dynamically adjusting the power of the infrared heater through real-time feedback, thereby realizing the accurate control of the temperature in the infrared heating process. The method does not need to depend on model parameters in the infrared heating process, achieves the high-precision control of the temperature in the infrared heating process through the nonlinear approximation system dynamic characteristics of the RBF neural network and a rolling optimization mechanism of model prediction control, and improves the adaptability of the system.

The invention discloses a neural network asymptotic tracking control method suitable for a robot system, and the method comprises the steps: introducing synaptic plasticity and internal plasticity into a conventional echo state network through simulating the adaptive characteristics of a biological nervous system, enabling the synaptic plasticity to dynamically adjust the connection weight of a neuron, enabling the internal plasticity to adaptively adjust the dynamic characteristics of the neuron, and enabling the neural network to continuously adjust the connection weight of the neuron; synchronous optimization learning of the neural network weight and the neuron state is realized, and an echo state network with a mixed learning mechanism is formed. The novel neural network is combined with a robust error symbol integral (RISE) controller, by means of the powerful nonlinear approximation capability of the neural network and the integral characteristic and the robust mechanism of the RISE controller, an unknown nonlinear part widely existing in a robot system is effectively processed, and asymptotic tracking control of the robot system on an expected trajectory is achieved. A simulation result verifies the effectiveness and stability of the method.

This invention provides a method for predicting material removal power in drilling-expansion hybrid machining based on SR-TELM, relating to the field of CNC machine tool power prediction technology. It delves into the physical mechanism between machining parameters and material removal power in drilling-expansion hybrid machining by using a symbolic regression algorithm. By embedding the mechanism model into the hidden layer neurons of TELM, it organically combines the advantages of symbolic regression in analyzing physical mechanisms with the powerful nonlinear approximation capability of TELM. This effectively improves power prediction accuracy under limited data conditions while ensuring the physical interpretability of the model, providing a new solution for modeling material removal power in drilling-expansion hybrid machining of machine tools.

The present invention belongs to the technical field of unknown radiation source signal analysis, and its purpose is to provide a method and system for identifying the modulation mode of radiation source signals. The present invention proposes a method for identifying the modulation mode of unknown radiation source using the characteristic parameters of the signal itself in conjunction with a global optimization neural network. During the implementation process, the characteristic parameters of the radiation source signal are first obtained, that is, the shallow characteristics of the radiation source signal itself are estimated, and then the neural network is constructed and the optimal network structure is obtained. Based on the superior performance of the deep neural network in nonlinear approximation, the characteristic parameters of the radiation source signal are input into the optimal network structure after multiple layers of individual coding changes, so as to mine and capture the potential characteristics of the radiation source signal, thereby making the modulation mode output result more stable and efficient. Compared with the recognition of ordinary neural networks, the noise resistance and recognition accuracy of the modulation mode recognition of the radiation source signal in a low signal-to-noise ratio environment of the present invention are significantly improved.

The present invention proposes a wind turbine model predictive control method and system based on Gaussian process error compensation. It does not rely on complex mechanism models, but predicts the model output compensation in real time through the Gaussian model, and dynamically corrects the model accordingly, effectively overcoming the high dependence of traditional methods on model accuracy and simplifying the modeling process. The Gaussian model is used to predict future wind speeds and output compensation. The Gaussian model has excellent nonlinear approximation capabilities and can accurately predict future complex and changeable working conditions and system states, significantly improving the effectiveness of the MPC control strategy and achieving more accurate wind turbine control.

The invention discloses an electric ship power adjusting device based on self-adaptive control. The electric ship power adjusting device comprises a sensor module, a data processing module, a self-adaptive control module, an actuator driving module and a power output module. The sensor module collects state parameters, operation parameters and environment disturbance information, and the state parameters, the operation parameters and the environment disturbance information are input into the self-adaptive control module after being standardized by the data processing module. The self-adaptive control module processes disturbance, non-linear terms and actuator faults through cooperative work of built-in disturbance observation, unknown non-linear approximation, fault-tolerant control and preset performance constraint units, generates self-adaptive control instructions, converts the instructions into driving signals through the actuator driving module, controls the power output module to adjust power, rotating speed and energy distribution, and controls the power output module to output power. Dynamic power adaptation of the electric ship is achieved, and navigation stability is guaranteed.

The present invention discloses a time-vertex joint graph filter group based on joint graph frequency domain sampling and a design method. First, sampling is defined based on the joint graph frequency domain; then, the time-vertex joint graph filter group is completely reconstructed based on joint frequency domain critical sampling; finally, joint orthogonal and joint biorthogonal graph filters are designed; the filter group has a symmetrical structure; the redundancy of subband coefficients is reduced, and the data transmission efficiency is improved. The method provided by the present invention designs a time-vertex joint frequency domain spline graph filter group by defining up- and down-sampling operations based on the joint graph frequency domain. The filtering operation and sampling operation of the filter group are both performed on the joint frequency domain, so that the vertex domain and time domain information of the time-varying graph signal can be effectively utilized. The simulation experiment part verifies the reconstruction performance, nonlinear approximation performance and denoising performance of the time-vertex joint frequency domain sampling graph filter group, which is significantly better than the existing frequency domain sampling graph filter group on static graphs.

The invention discloses an unmanned aerial vehicle interference observer design method based on an RBF neural network. According to the method, the nonlinear interference observer is combined with the RBF neural network, and the real-time online learning and nonlinear approximation capabilities of the RBF neural network are utilized to perform online approximation on an estimation error generated by the nonlinear interference observer in actual operation, so that the influence of model parameter drift on the interference estimation precision is effectively weakened, and the interference estimation accuracy is improved. And the problem of noise amplification caused by high gain is avoided. According to the method, on the premise that a huge offline sample library does not need to be constructed, the improved interference observer can adaptively track various uncertain interferences in the flight process of the unmanned aerial vehicle, interference estimation errors are remarkably reduced, the real-time performance and accuracy of interference estimation are improved, a more reliable interference compensation basis is provided for an unmanned aerial vehicle control system, and the interference compensation efficiency is improved. The flight stability and control precision of the unmanned aerial vehicle in a complex dynamic environment are guaranteed, and the requirement for efficient and safe operation of the unmanned aerial vehicle is met.

The application belongs to the technical field of intelligent control, and particularly relates to a consistency learning control method based on a neural network, which comprises the following steps: S1, constructing a linear parameterization model of consistency output of multiple intelligent agents; S2, designing a loss function by outputting relevant consistency output through a neural network, and obtaining a linear parameter update algorithm in the time and iteration directions through a gradient descent method; and S3, using the output of the neural network to design a controller, and constructing a consistency learning control scheme of the multiple intelligent agent system based on neural network approximation. The scheme updates parameters based on the time axis and the iteration axis, does not require a system model and strict matrix conditions, and realizes perfect tracking in the whole time period. The application breaks through the traditional cognition of taking error as an index parameter, directly fits equivalent parameters through a neural network, solves the problems of insufficient fitting capacity of a projection algorithm and nonlinear approximation error under a complex system, and significantly improves the convergence speed of the system and the adaptability of consistency control.

The invention belongs to the technical field of intelligent control, and particularly relates to a consistency learning control method based on a neural network, and the method comprises the steps: S1, constructing a linear parameterization model of multi-agent consistency output; s2, a loss function is designed by outputting related consistency output through a neural network, and a linear parameter updating algorithm in the time direction and the iteration direction is obtained through a gradient descent method; and S3, designing a controller by using the output of the neural network, and constructing a consistency learning control scheme of the multi-agent system based on neural network approximation. According to the scheme, the parameters are updated based on the time axis and the iteration axis, a system model and strict matrix conditions are not needed, and full-time-period perfect tracking is achieved. According to the method, the traditional cognition that errors serve as index parameters is broken through, equivalent parameters are directly fitted through the neural network, the problems of insufficient fitting capacity and nonlinear approximation errors of a projection algorithm under a complex system are solved, and the convergence speed of the system and the adaptability of consistency control are remarkably improved.

The present invention discloses a high-precision predetermined time attitude tracking control method for a flexible spacecraft subject to multiple disturbances, the method comprising the following steps: step one, establishing a mathematical model of the flexible spacecraft based on a rotation matrix; step two, establishing an error kinematics and dynamics model of the flexible spacecraft; step three, based on the good nonlinear approximation capability of a radial basis neural network RBFNN, approximate estimation and compensation of lumped disturbances including external disturbances, parameter uncertainty, actuator failure and structural vibration; step four, designing a controller based on predetermined time theory and sliding mode control theory, so that the maximum convergence time of the system is independent of its initial state. The beneficial results of the present invention are: the lumped disturbance can be compensated by using a neural network under the condition of unknown prior information, wherein actuator failure and saturation problems are also considered; the system stabilization time can be preset; and the accuracy and stability of the attitude tracking control system of the flexible spacecraft are improved.

This invention discloses a reinforcement learning-based attack detection method for quadcopter drones, comprising: estimating the state values ​​of the drone during flight using Kalman filtering; obtaining the corresponding residual correlation terms; dividing the entire detection space into several detection intervals; setting reward values ​​for two actions: continuing detection and detecting an attack; setting different simulated attack times; training Q-tables for different simulated attack times using the Saras algorithm; fusing the trained Q-tables for different simulated attack times using a weighted average; fitting the weighted average fused Q-table using a neural network; and detecting attacks on the drone online using the Q-table fitted by the neural network. This invention utilizes reinforcement learning technology and integrates the nonlinear approximation capability of neural networks to achieve the detection of subtle and covert attacks on quadcopter drones. It not only improves the recognition ability of subtle and covert attacks but also enables online detection of new attack patterns and reduces attack detection latency.

The application provides a neural network inversion sliding mode control method for a marine three-phase asynchronous motor, which combines the nonlinear approximation capability of an RBF neural network, good real-time performance, and the advantages of good dynamic performance and strong anti-interference of a sliding mode control, fully utilizes the back-off decoupling characteristics of a backstepping method, and designs a neural network inversion sliding mode control method for the marine three-phase asynchronous motor, so as to improve the performance of the motor control system. For the problem that the parameters cannot be determined in the modeling process of the marine three-phase asynchronous motor, a forward stability augmentation channel is designed. The backstepping method is used for decoupling control of the system, the sliding mode control solves the disturbance problem of the control system, and the RBF neural network is used for real-time estimation and compensation of the disturbance between the system coupling and the nonlinear friction force of the indirect contact surface of the control system framework.

The application discloses a kind of nonlinear system dynamic event triggering optimization control methods of reinforcement learning, it is related to triggering tracking control field, comprising the following steps: step S1, establish nonlinear multi-agent system model;Step S2, construct double-layer MLP nonlinear approximation model;Step S3, design consensus reinforcement learning control law;Step S4, design dynamic event triggering mechanism;Step S5, design weight update law and verify stability.The application adopts the nonlinear system dynamic event triggering optimization control method of reinforcement learning described above, while significantly saving communication and computing resources by using adaptive dynamic event triggering mechanism, the optimal control performance of the system is realized;Closed-loop system stability is strictly guaranteed, and Zeno behavior is effectively excluded, the effectiveness of the proposed strategy is verified by numerical testing on a multi-motor system.

The present invention proposes a piezoelectric drive robust control method based on a sliding mode observer and a radial basis function neural network. The method designs a sliding mode state observer with disturbance compensation to observe system disturbances and system states. Since the radial basis function neural network can also be used for interference suppression, the RBFNN has the characteristics and capabilities of parallel processing, self-learning and nonlinear approximation, and can generate reasonable outputs through a sufficient fitting process. Through the design of this control method, the influence of the system not being modeled and various disturbances can be compensated and the chattering effect caused by the sliding mode controller can be reduced, further improving the tracking performance of the piezoelectric drive system output.

The present application belongs to the technical field of hypersonic vehicle control, and relates to a reusable carrier flight-propulsion-missile coupling fault-tolerant control method. The purpose of the present application is to realize stable tracking control of the reusable carrier. The method comprises the following steps: constructing a longitudinal dynamics model of the reusable carrier; constructing a longitudinal dynamics simplified model of the reusable carrier flight-propulsion-missile; designing a sliding mode controller of the reusable carrier; and designing a RBF neural network of the reusable carrier. Through the powerful nonlinear approximation capability of the RBF neural network, the sliding mode fault-tolerant control is designed, the sensor and actuator fault-tolerant compensation is considered, and the stable tracking control of the reusable carrier is realized. The method is a reusable carrier flight-propulsion-missile coupling fault-tolerant control method, and has a wide application prospect.

The invention belongs to the technical field of hypersonic flight vehicle control, and relates to a flight-push-missile coupling fault-tolerant control method for a reusable vehicle. The invention aims to realize stable tracking control of a reusable vehicle. The method comprises the following steps: constructing a reusable vehicle longitudinal dynamic model; the invention relates to a vertical dynamics simplified model of a flying bomb of a reusable launch vehicle. Designing a reusable vehicle sliding mode controller; the invention relates to a vehicle RBF neural network design capable of being repeatedly used. The sliding-mode fault-tolerant control is designed through the powerful nonlinear approximation capability of the RBF neural network, and the stable tracking control of the reusable vehicle is realized by considering the fault-tolerant compensation of the sensor and the actuator. The method is a coupling fault-tolerant control method for the flying-pushing missile of the reusable launch vehicle, and is wide in application prospect.