55results about How to "Strong nonlinear fitting ability" patented technology

The invention discloses a wind power combined predication method based on ensemble average empirical mode decomposition and an improved Elman neural network, and belongs to the technical field of wind power prediction. The method comprises the following steps: step 1, extracting wind speed sequence historical data and normalizing the data; step 2, making sequence analysis of the extracted wind speed sequence historical data through empirical mode decomposition; step 3, reconstructing a phase space for the sequences obtained through empirical mode decomposition; step 4, cyclically selecting the number of nodes in a hidden layer to train an Elman neural network, and superposing the prediction results of the sequences to obtain a wind speed prediction result; and step 5, making error analysis of the wind speed prediction result. The modeling process of the method is simple and practical. By adopting the method, the wind power can be predicted quickly and effectively. The method is of great significance to the security and stability and dispatching operation of a power system under the condition of wind power grid-connection.

The invention discloses an anti-attack and defense method based on an LSTM (Long Short Term Memory) detector. The method comprises the following steps that: 1) generating a candidate detector of an LSTM-FNN (Fuzzy Neural Network) structure, and asking the detector to detect abnormal samples as more as possible on a premise that the detector does not misjudge a normal sample; 2) storing the candidate detector into a register queue, detecting a training dataset by an abnormality detector taken out of the register queue, and deleting the detected abnormal sample to enable different abnormality detectors to cover different abnormal areas; and 3) in a detection stage, detecting a detected sample by all abnormality detectors, and comprehensively judging whether the sample is abnormal or not. Byuse of the method, a neural network is combined with a negative selection algorithm, each LSTM detector guarantees that the normal sample can not be misjudged, the abnormality detector set guaranteesthat the abnormal situation may be covered, and an algorithm detection effect is improved.

The invention relates to a forecasting method of burning and explosion characteristics of organic compounds based on a support vector machine; the method utilizes various structural parameters which reflect the structural characteristics of molecules to describe the burning and explosion characteristics of organic matters according to the principle that various burning and explosion characteristics are decided by the molecular structure; the introduction of the support vector machine of a powerful machine learning algorithm can carry out training and forecasting to the burning and explosion characteristics of the organic matter and the nonlinearity, nondeterminacy and complexity which exist in the molecular structure, thus establishing a forecasting module with stability and effectiveness; the established forecasting module is applied in the forecasting of the burning and explosion characteristics of other unknown compounds with the advantages of high forecasting precision, fastness and convenience.

The invention discloses an excellent driver lane changing imitation model establishment method based on a GRU network, belonging to the field of smart car automatic driving. The method comprises the following steps: performing real vehicle experiment under a lane changing condition, acquiring an excellent driver steering characteristic parameter, a vehicle dynamics parameter and a track parameter,and forming a lane changing behavior data set; performing training learning for the lane changing data set through the GRU network, and obtaining the excellent driver lane changing imitation model based on the GRU network. In the method provided by the invention, based on strong nonlinear fitting ability of the GRU network on a long-time sequence, one simple and efficient excellent driver lane changing imitation model is realized, further improvement of accuracy of learning can be guaranteed based on fast learning, imitating a lane changing behavior of an excellent driver can be realized better, and in the future, the method has certain referential significance in such field and other prediction problems related to driver models.

The invention discloses an electric vehicle power battery SOC intelligent detection device which is characterized in that the intelligent detection device includes a battery parameter acquisition platform and a battery SOC estimation system, the battery parameter acquisition platform collects real-time parameters of voltage, current and temperature of an electric vehicle power battery group, the battery SOC estimation system estimates an SOC value through the collected parameters, and a battery SOC system is a nonlinear, time-delay and multivariable coupling complex real-time system with a high requirement. According to the device, a problem that a conventional detection device can not obtain an ideal effect of the intelligent detection of the electric vehicle power battery SOC.

Provided is a ranging method for overhead line power distribution network single-phase earth faults. The method comprises: using an off line method to inject three-phase symmetric high-voltage pulse to an overhead line power distribution network on a bus end, and sampling zero-module voltage in a preset time window from zero time, to obtain a zero-module voltage waveform; performing energy normalizing processing on the zero-module voltage waveform to obtain an energy normalizing waveform; extracting fault point reflected wave arrival time, front time, front steepness, and wave front energy from the energy normalizing waveform; using the arrival time, front time, front steepness, and wave front energy as input variables of a neural network, and using fault distances as output of the neural network to establish a ranging model. The errors of fault distances obtained by the ranging method for overhead line power distribution network single-phase earth faults are low.

The invention relates to a river flow forecasting method, comprising the following steps: 1: data preprocessing; 2: VMD model decomposition; 3: component reconstruction of VMD decomposition results; 4: integration of a VMD-BP model and river flow forecasting. Variational mode decomposition is an effective method to process non-stationary signals. A river flow forecasting method based on a VMD-BP model is proposed and established in combination with the advantages of a BP neural network in processing nonlinear function fitting. The method decomposes original data into a plurality of intrinsic mode functions reflecting data characteristics, and smooths the data, thereby solving the problems of nonlinearity and undulation of water level flow data, and improving the forecasting accuracy.

The invention relates to an air knife pressure real-time optimization control method in the galvanizing process. On the basis of a plating thickness neural network prediction model, the corresponding error correction based on variable lag time and the real-time optimization technology based on an increment PID algorithm are adopted, and a good anti-interference and follow-up control effect is adopted. When the plating thickness deviates from a set value because of the outside interference, the air knife pressure is optimized in real time based on the difference between a plating thickness predicted value after being subjected to the error correction, and the set value, and the plating thickness is made to be kept nearby the set value; and during product switching, iterative optimization is performed on the air knife pressure based on a plating thickness predicted value not subjected to the error correction, and the plating thickness is made to fast complete the switching process closely along with a changing curve of the set value. According to the pressure real-time optimization control method in the galvanizing process, bad influences caused to the plating thickness by the outside interference are effectively overcome due to the above technology, fast switching between products in different plating thicknesses is achieved, the plating quality fluctuation can be obviously reduced, the excessive zinc consumption is reduced, and the percent of pass of galvanized products is increased.

The invention discloses an RBF adaptive neural network repetitive controller suitable for a repetitive servo system. According to the controller, the self-adaptive weight adjustment of an RBF neural network is utilized to approximate a servo motor input and output differential equation with unknown parameters; according to a repetitive control method, a control quantity at the current moment is corrected by utilizing the operation information of a previous period, so that periodic interference is overcome, and the tracking of a given periodic reference signal by an output quantity is realized.According to the RBF adaptive neural network repetitive controller suitable for the repetitive servo system, on the basis of the periodic operation characteristics of the repetitive servo system, onone hand, the RBF neural network is utilized to approximate the system model of unknown parameters, and on the other hand, the repetitive control method is introduced to eliminate common periodic interference in an repetitive operation process.

The invention discloses a traffic environment pedestrian multi-dimensional motion feature visual extraction method comprising a step 1 of constructing a pedestrian motion database; a step 2 of extracting a pedestrian detection frame image of the same pedestrian in consecutive image frames; a step 3 of extracting the HOG feature of the same pedestrian motion energy map; a step 4 of constructing a pedestrian motion pose recognition model based on an Elman neural network; a step 5 of determining a pedestrian pose in a current video by using the pedestrian motion pose recognition model based on the Elman neural network; a step 6 of calculating the instantaneous speed sequences of the pedestrian in the X-axis and Y-axis directions to obtain the real-time speed of the pedestrian; and a step 7 ofaccording to a three-dimensional scene in an intersection environment, obtaining the position information of the pedestrian in the image in real time, and obtaining the real-time motion feature of the pedestrian in combination with the pedestrian pose and the real-time speed. The method has high recognition accuracy and good robustness, is convenient to use, and has a good application and promotion space.

The invention relates to a video image Chinese wolfberry branch detection method for a Chinese wolfberry harvesting and clamping device. Compared with the prior art, the defect that the position of aChinese wolfberry branch is difficult to identify is overcome. The method comprises the following steps: obtaining and preprocessing a training sample; constructing and training a Chinese wolfberry leaf and key point detection network; constructing and training a Chinese wolfberry branch positioning model; collecting and preprocessing video images of the Chinese wolfberry branches to be detected;and obtaining a Chinese wolfberry branch detection result. According to the invention, the selection method of the ROIAlign input feature map in the traditional Mask-RCNN is optimized; meanwhile, description of a target area is improved according to the shape of a Chinese wolfberry leaf, and an inclined frame with a direction is used for replacing a common rectangular frame, so that on the basis of ensuring the data reliability, the memory space is reduced, and the branch identification and positioning accuracy and efficiency are improved.

The invention discloses a GRNN motorized spindle thermal error modeling method based on genetic algorithm optimization, and the method comprises the steps: building a four-layer generalized regression neural network (GRNN) structure, carrying out the global search of a generalized regression neural network smooth factor sigma through a genetic algorithm, and simply and accurately finding a global minimum value; initializing a generalized regression neural network smooth factor sigma by adopting a random population initialization mode, constructing a fitness function of a genetic algorithm (GA), calculating individual fitness, executing natural operation on a population, and selecting, crossing and inheriting individuals; establishing a GRNN framework, training samples, so population evolution gradually reaches the training precision, and finally, verifying the generalization of a generalized regression neural network (GA-GRNN) optimized by a genetic algorithm by adopting experimental data of different rotating speeds. According to the method, global optimal search is carried out on the smooth factor sigma of the GRNN by using the genetic algorithm, and the prediction precision and generalization ability of the GRNN are improved.

The invention provides a short-term photovoltaic power prediction method based on similar day wavelet transform and a multi-layer perceptron, a power prediction model is established by adopting an artificial intelligence technology, and a short-term photovoltaic power prediction model based on similar day wavelet transform and the multi-layer perceptron is established. According to the model provided by the invention, the advantages of various algorithms are combined, and the prediction accuracy can be effectively improved. The main innovation point of the method is that similar daily wavelet transform and a multilayer perceptron are combined to be used for the research of predicting the output power of the photovoltaic power station.

The invention discloses a personalized HRTF rapid modeling acquisition method, and the method comprises the following steps: S1, performing correlation analysis on a physiological appearance parameter database, and screening out characteristic physiological appearance parameters in combination with physical analysis; S2, performing multiple data dimension reduction on the personalized HRTF database, and extracting a feature matrix corresponding to a sample; S3, establishing, training and fitting a neural network of a relationship between the collected sample feature physiological shape parameters in the personalized HRTF database and the corresponding HRTF feature matrix; S4, inputting feature physiological appearance parameters of the user, predicting an HRTF feature matrix of the user by using the trained neural network, and reconstructing the feature matrix into a complete HRTF database. The method has the advantages that the personalized HRTF corresponding to the user can be rapidly and accurately obtained through a small number of feature physiological shape parameters, and therefore the positioning precision and user experience of the virtual auditory display technology can be improved to a certain degree.

The invention relates to a neural network fault arc detection system based on generalized S transformation. The system comprises a training sample generation module, a neural network training module and a fault arc identification module, the training sample generation module comprises a fault arc experiment and simulation data acquisition module and an arc feature extraction module based on generalized S transformation, and the fault arc identification module comprises a user real-time total load data acquisition and processing module, a neural network model module and a fault identification result module. A neural network fault arc identification method is provided on the basis of the system, an S transformation feature extraction method and a neural network mode identification method are fused, features of fault arc current signals can be accurately captured through S transformation, time-frequency features are grasped, and the problems that in the prior art, feature discrimination is not high, and confusion is prone to occurring are solved. Through verification, the load identification effect of the method has high accuracy, and a technical basis can be provided for a series of advanced applications of a non-intrusive load identification technology.