57 results about "Nonlinear approximation" patented technology

A vehicle dynamics behavior reproduction system capable of describing accurately behavior of a motor vehicle in a lateral direction even for nonlinear driving situation includes a vertical wheel force arithmetic means (105), a lateral wheel force arithmetic means (110), a cornering stiffness adaptation means (115), a state space model/observer unit (120), a selector (130), a delay means (135), and a tire side slip angle arithmetic means (125). Vertical wheel forces (F_Zij) and tire side slip angles (α_ij) are determined by using sensor information and estimated values while lateral wheel forces (F_Yij) are determined in accordance with a relatively simple nonlinear approximation equation. The lateral wheel force (F_Yij) and the tire side slip angle (α_ij) provide bases for adaptation of cornering stiffnesses at individual wheels. Vehicle motion is accurately described to a marginal stability by using adapted cornering stiffnesses (C_ij) and other information.

System and method for modeling a nonlinear process. A combined model for predictive optimization or control of a nonlinear process includes a nonlinear approximator, coupled to a parameterized dynamic or static model, operable to model the nonlinear process. The nonlinear approximator receives process inputs, and generates parameters for the parameterized dynamic model. The parameterized dynamic model receives the parameters and process inputs, and generates predicted process outputs based on the parameters and process inputs, where the predicted process outputs are useable to analyze and/or control the nonlinear process. The combined model may be trained in an integrated manner, e.g., substantially concurrently, by identifying process inputs and outputs (I/O), collecting data for process I/O, determining constraints on model behavior from prior knowledge, formulating an optimization problem, executing an optimization algorithm to determine model parameters subject to the determined constraints, and verifying the compliance of the model with the constraints.

The invention relates to a heterogeneous cloud wireless access network resource allocation method based on deep reinforcement learning, and belongs to the technical field of mobile communication. Themethod comprises the following steps: 1) taking queue stability as a constraint, combining congestion control, user association, subcarrier allocation and power allocation, and establishing a random optimization model for maximizing the total throughput of the network; 2) considering the complexity of the scheduling problem, the state space and the action space of the system are high-dimensional,and the DRL algorithm uses a neural network as a nonlinear approximation function to efficiently solve the problem of dimensionality disasters; and 3) aiming at the complexity and the dynamic variability of the wireless network environment, introducing a transfer learning algorithm, and utilizing the small sample learning characteristics of transfer learning to enable the DRL algorithm to obtain an optimal resource allocation strategy under the condition of a small number of samples. According to the method, the total throughput of the whole network can be maximized, and meanwhile, the requirement of service queue stability is met. And the method has a very high application value in a mobile communication system.

The invention provides a motor-vehicle tail-gas concentration prediction method of an LSTM-based deep space-time residual-network. Powerful data nonlinear-approximation capability and self-learning capability of a deep learning method are utilized for establishing an LSTM-based depth-space residual-network structure, space-time prediction of urban motor-vehicle tail-gas concentration is realized,urban motor-vehicle tail-gas space-time data and external influence factor data are collected, the urban motor-vehicle tail-gas space-time data are used for storage, training and prediction, two-dimensional prediction analysis of time and space is directly carried out, external conditions of weather, holidays and the like are considered at the same time, and a fully-connected two-layer network isused for simulating external influence factor characteristics to enable a prediction result to be more comprehensive and accurate.

A constrained non-linear approximator for empirical process control is disclosed. The approximator constrains the behavior of the derivative of a subject empirical model without adversely affecting the ability of the model to represent generic non-linear relationships. There are three stages to developing the constrained non-linear approximator. The first stage is the specification of the general shape of the gain trajectory or base non-linear function which is specified graphically, algebraically or generically and is used as the basis for transfer functions used in the second stage. The second stage of the invention is the interconnection of the transfer functions to allow non-linear approximation. The final stage of the invention is the constrained optimization of the model coefficients such that the general shape of the input/output mappings (and their corresponding derivatives) are conserved.

The invention discloses a distribution prediction method of Mn and associated heavy metals, such as Cu, Ni, Zn, Pb, Cd and the like, in a soil-water interface contaminated flow in a manganese mine area. On the basis of runoff water sample actual detection data of a sampling point of a soil-water interface in the manganese mine area and along-river distribution status of heavy metal content in sediments, a BP prediction model of the manganese and associated heavy metal distribution in the soil-water interface contaminated flow in the manganese mine area is constructed on the basis of a GIS technology through a geostatistic method with the GIS technology with a BP neural network introduced. The predict model has excellent fault-tolerant capability and nonlinear approximating mapping capability. A predict result of contents and spatial distribution of the manganese and associated heavy metal of the soil-water interface contaminated flow in manganese mine area is more objective and accurate.

The invention discloses a calibrating method for an electronic compass. The calibrating method is used for promoting measuring accuracy of the electronic compass based on a self-adaption differential evolution algorithm and a Fourier neural network theory and is especially suitable for an orienting system with low cost and higher precision. The calibrating method comprises the following steps: utilizing a Fourier neural network to perform error modeling on the electronic compass and utilizing an improved self-adaption differential evolution algorithm to optimize a weight value of the Fourier neural network, so as to acquire an accurate error model to compensate a measuring value of the electronic compass. The error model which is established according to the calibrating method is capable of realizing accurate mapping of a sample space and has a higher nonlinear approaching capability. According to the calibrating method, a minimum local part is avoided, the defects of over-slow convergence rate and oscillation of the neural network are overcome and the influence of an outside magnetic field on the electronic compass is efficiently compensated, thereby greatly promoting the measuring accuracy of the electronic compass.

The invention discloses a linear pseudo-spectrum GNEM guidance and control method. According to the linear pseudo-spectrum GNEM guidance and control method, an original nonlinear optimal control problem is converted into a group of problems for solving a system of linear algebraic equations based on the concept of GNEM through linearization and Gauss pseudo-spectrum dispersion by combining nonlinear approximation model predictive control, linear quadratic optimal control and a Gauss pseudo-spectrum method. The linear pseudo-spectrum GNEM guidance and control method has the advantages that the calculation efficiency of solving the optimal control problem is quite high, and high calculation precision can be obtained just through several nodes; in addition, a final solution can be presented through a smooth function relevant to control over disperse nodes, and the method is quite suitable for online calculation. It is indicated through a simulation result that by applying the method to tail-section attack guidance with terminal angle constraint, compared with an MPSP method, the linear pseudo-spectrum GNEM guidance and control method has higher calculation efficiency and high calculation precision and can be completely applicable to a guidance framework for terminal guidance, and a smaller required overload command is generated compared with self-adaptive tail-section proportion guidance.

The invention discloses a fusing method of an atmospheric attack angle and an inertia attack angle in transonic flight stage. The fusing method comprises the following steps of: firstly, data acquisition; secondly, transonic flight stage judgment; thirdly, inertia attack angle solution; fourthly, complementary filtering fusing; and fifthly, BP neural network correction. Through a complementary filter, the method inhibits impact wave interference fluctuation of the atmospheric attack angle, and enables complementary fused attack angles to eliminate fierce fluctuation and become stable. Further, by utilizing a favorable learning ability and a non-linear approximation property of the neural network, the method ensures a better anastomosis degree between a real attack angle and a final fused modified attack angle which is treated by the neural network, and enables whole precision and reliability of the attack angle to be improved further. By replacing a primary atmospheric attack angle in the transonic flight stage with the fused modified attack angle, the invention can effectively improve whole precision of the attack angle during flight, and provides a beneficial foundation of accurate navigation and flight control.

Methods and apparatuses for measuring a light absorbance are provided. The method measures light absorbance of at least one detection chamber of a microfluidic device, including the detection chamber and at least one reference chamber. The detection chamber may accommodate a test subject. The method includes detecting a plurality of reference transmitted light intensities for the at least one reference chamber and estimating a value between the plurality of reference transmitted light intensities through nonlinear approximation. The estimated value is then applied to light absorbance measurement of the detection chamber to reduce a light absorbance error of the detection chamber.

A method of providing a low cost quantized nonlinear scaler for a continuous curve includes the steps of providing a quantized coefficient table representing an approximation of the nonlinear continuous curve; applying the table in hardware to locate coefficients in the table, and applying the coefficients to a scaling pipeline to get scaled data. The nonlinear approximation in the table is achieved by following the rules that the sampled data must be consistent, the quantization points must be consistent, the coefficient curve must be symmetric about its centerline, and the quantization values must be selected so that the sum of all coefficients values in the coefficient curve over all the source points for a given resultant point is always equal to a normalization value.

The invention provides a non-affine and nonlinear fault-tolerant controller design method for a non-affine and nonlinear system with disturbance and parameter uncertainty. The designed observer can be applied to the nonlinear system with non-affine parameters, and the observer has ideal robustness in the presence of the parameters of large-range variation. Fault information and disturbance information are hidden in the observer. A fault-tolerant controller is dynamically designed based on the observer. Due to the fact that the system is the non-affine and nonlinear system, the design of the controller is not easy. The invention provides the dynamic non-affine and nonlinear approximation method. The non-affine and nonlinear system is approximated into an affine nonlinear system with time-varying parameters. Parameters which need to be known are estimated online through a filter. A non-affine flight control system is used to verify the effectiveness of the provided method. The robust fault-tolerant control of the non-affine and nonlinear system can be realized.

The invention discloses an error compensation method for a photoelectric encoder. Based on improved particle swarm optimization and a Fourier neural network principle, the method is used for improving the measurement accuracy of the photoelectric encoder, and is particularly suitable for an angle measuring system requiring low cost and high accuracy. According to the method, a compass error is modeled by a Fourier neural network, and the weight of the neural network is optimized by the improved particle swarm optimization, so that an accurate error model is obtained to compensate a measured value of the photoelectric encoder. The error model established by the method can realize accurate mapping of a sample space and has high nonlinear approximation capability; and by the method, local minimum is avoided, a defect that the neural network has ultralow convergence rate, oscillates and the like is overcome, measurement errors of the photoelectric encoder are effectively reduced, and the measurement accuracy of the photoelectric encoder is greatly improved.

The invention discloses a remote sensing data-based high-precision agricultural region ground surface temperature retrieval method, which comprises the following steps: firstly, calculating relationship between thermal radiation intensity and temperature, and performing non-linear approximation on Planck function by utilizing exponential function; and on the basis, carrying out derivation of a high-precision retrieval method of the ground surface temperature of the agricultural region by combining ground surface thermal radiation transmission equation. According to the method, the ground surface emissivity and atmospheric transmittance are required: NDVI of the ground surface of the agricultural region can be extracted by utilizing the data of ASTER visible light waveband and near infrared waveband so as to further calculate the emissivity data; the content of atmospheric water vapor can be retrieved by utilizing the data of MODIS near infrared waveband, and accurate atmospheric transmittance data can be obtained by fitting the relationship between atmospheric transmittance and the content of atmospheric water vapor through utilizing segmental cubic polynomial fitting. The ground surface temperature of the agricultural region can be accurately obtained, the space-time distribution of the ground surface temperature can be accurately analyzed, the space difference of the regional temperature can be analyzed, thus providing basic data for agriculture, weather, hydrology, ecology, biogeochemistry and the like.

The invention discloses a convolutional-fuzzy neural network method for actively controlling global spatial noise of a vehicle. The method comprises the steps of arranging a plurality of secondary paths around a vehicle noise reduction area; acquiring a noise residual signal of each secondary path; adopting a convolutional-fuzzy neural network to firstly perform offline identification to obtain asecondary path model and then serve as an adaptive active noise control algorithm of the secondary path at the same time to correct controller parameters online; and finally, outputting multi-azimuthnoise cancellation signals. According to the invention, the convolutional-fuzzy neural network is used for identifying the inverse model of an object, a very effective method is provided for vehicle global space nonlinear noise identification, and the identification precision of the secondary path is improved by utilizing the nonlinear approximation capability of the convolutional-fuzzy neural network to a function; an active feedback noise elimination system is adopted, and a stable secondary path model is established; and the problems that vehicle global space noise is difficult to control and the frequency band is narrow are solved.

The invention discloses a composite orthogonal neural network prediction control-based intelligent ship tracking method. The method comprises the steps of: obtaining a predetermined trajectory in themovement process of a ship, and calculating optimization algorithm predicted thrust of each propeller through an optimization algorithm according to the predetermined trajectory and predicted outputs;carrying out weighted stacking on the optimization algorithm predicted thrust and neural network predicted thrust through the neural network predicted thrust so as to output thrust to be generated byeach propeller; predicting the position, heading and speed of the ship through a prediction model; and correcting predicted values of the position, heading and speed of the ship and taking the corrected predicted values as the predicted outputs. According to the method, a composite orthogonal neural network is combined to put forward a new model prediction strategy; the neural network is simple in algorithm and high in learning convergence speed, and has the excellent characteristics of high linear and nonlinear approximation accuracy and the like; and the learning algorithm of the neural network can be offline completed, so that the online calculation time is greatly shortened.

The present invention discloses a method for improving interpolation position accuracy of a cutting bed based on an improved BP neural network. Parameters of a current node are input into a trained BP neural network interpolation model to obtain predicted values of the node parameters output by the BP neural network interpolation model. The predicted values of the node parameters output by the BP neural network interpolation model are substituted into a B-spline curve to obtain a predicted interpolation node. The deviation between the predicted interpolation node and the actual interpolation node and the corresponding feedback compensation output are calculated. The feedback compensation output is compared with an original machining trajectory curve; the node parameters for the next interpolation are optimized and predicted with the Newton path searching method; the predicted node parameters are substituted into the fitted B-spline curve to obtain the actual interpolation node; and the interpolation operation is performed. The method of the invention has the advantages of good nonlinear approximation ability, high interpolation precision and fast convergence speed.

The invention discloses a lead-acid storage battery SOH estimation method based on SA and ANN algorithms. The method comprises the steps: S1, carrying out cyclic charging and discharging experiment ona lead-acid storage battery, and recording the one-to-one correspondence between the actual capacity and the number of cycles of the cyclic charging and discharging experiment, and the time of constant-voltage charging and constant-current charging in a charging stage of each experiment; S2, based on the test result of the step S1, determining an influence factor with the highest SOH associationdegree with the lead-acid storage battery, and establishing an artificial neural network regression model; and S3, training weights and bias values in the artificial neural network regression model through an improved simulated annealing algorithm, establishing a new regression model, and estimating the battery capacity data points of the SOH of the lead-acid storage battery by using the new regression model. According to the invention, the artificial neural network model ANN is adopted, the method has the advantages of being good in nonlinear approximation capacity and generalization performance, small in number of training samples and high in fitting precision, the number of model parameters is small, and the calculation speed is high.

The invention discloses an electrocardiogram identity recognition method. The method comprises the following steps: acquiring an ECG (Electrocardiogram) signal of a human body; preprocessing the collected ECG signal to obtain a clean short-period ECG signal; generalized S transformation is carried out on the short-period ECG signals, and a phase domain feature vector Y1, a time domain feature vector Y2 and a frequency domain feature vector Y3 are extracted; inputting the phase domain feature vector Y1, the time domain feature vector Y2 and the frequency domain feature vector Y3 into a sparsity-constrained nonlinear approximation model to obtain an optimal dictionary and a corresponding optimal sparse coefficient matrix, and performing lightweight processing on the optimal sparse coefficient matrix to obtain a sparse coefficient vector; and inputting the sparse coefficient vector into a trained deep neural network model based on a bidirectional long-short-term memory network for identity recognition. According to the invention, the multi-modal feature vector is extracted from the original ECG signal to serve as the input vector of the deep neural network, so that the recognition precision is improved, and the recognition rate is increased.