13 results about "Relief algorithm" patented technology

The invention belongs to the technical field of fault judgment, and discloses a wind turbine generator system fault judgment method based on a Relief algorithm and a related device. The wind turbine generator system fault judgment method based on the Relief algorithm comprises the steps of selecting characteristic quantities and preprocessing the characteristic quantities, iteratively calculating the weight of the preprocessed characteristic quantities according to the Relief algorithm to obtain a characteristic variable set, establishing a regression model between the average wind speed and the characteristic variable set and calculating the regression model, obtaining a normal reference characteristic quantity state of the wind turbine generator system under the real-time state data, and judging whether the wind turbine generator system has a fault or not through a state judgment algorithm according to the real-time state data and the normal reference characteristic quantity state of the wind turbine generator system under the real-time state data; the abnormal condition of the wind turbine generator system can be found in time, an early warning mechanism is triggered, the early warning accuracy is remarkably improved, and occurrence or expansion of faults is avoided.

The application provides a ReliefF feature selection method for random multi-subspace of big data, and comprises the following steps: S1, dividing an original feature space to generate a plurality of feature partitions containing a plurality of random subspaces of the same size and no intersection; S2, using a ReliefF or Relief algorithm in each random subspace to obtain a local weight of a feature, and then combining the local weight vectors of the random subspaces in each feature partition to obtain a total weight vector; and S3, integrating the total weight vectors of the plurality of feature partitions into a maximum weight vector of each feature, that is, obtaining the maximum weight vector of all features by averaging the weight scores of the plurality of feature divisions. The application fully considers the diversity of the subspaces and the contribution information of the samples to the features in the feature selection process, and has the ability to explore each subspace in a high-dimensional space.

The invention discloses a wind turbine generator blade icing fault detection method, which comprises the following steps of: performing feature selection on multi-dimensional feature data collected by a wind turbine generator SCADA (Supervisory Control And Data Acquisition) system by adopting a Relief algorithm, evaluating a feature weight according to contribution of features to distinguishing blade icing and a normal state, and screening out key features; constructing a new wind speed-power feature X1 and a new environment temperature-cabin temperature feature X2 on the basis of the screened feature data and a visual graph thereof in combination with an icing rule; constructing a cross LSTM prediction model, predicting one group of data according to the characteristics of multiple groups of data, and generating a residual signal; relates to the technical field of wind turbine generator fault detection, and through feature selection and an improved LSTM algorithm, the time sequence change features in the blade icing process can be more accurately captured, the fault detection accuracy is improved, and the detection accuracy is effectively improved.

A kind of computing offloading and resource allocation method based on deep reinforcement learning, aiming at the problem of high task computing delay in multi-user multi-edge server scenario, with the goal of minimizing the total task processing delay of users. By selecting the deep reinforcement learning method based on value function, the D3QN algorithm combining Double-DQN algorithm and Dueling-DQN algorithm based on DQN algorithm with GRU network is proposed for edge computing computing offloading and resource allocation processing method, so as to improve the resource utilization of edge server and effectively alleviate the overestimation and Q value uniqueness problem in DQN algorithm, and obtain a better offloading scheme. Finally, the algorithm is simulated and compared with various baseline algorithms, which proves that the proposed algorithm can achieve the purpose of optimizing the total task processing delay of users and reducing the task drop rate.

The present invention relates to a non-line-of-sight signal recognition method based on a channel feature weighted model. The method comprises the following steps: after acquiring channel impulse response (CIR) data, nine features are constructed on the CIR data. First, seven basic channel features are extracted from a CIR waveform diagram. Two new variables, an attenuation factor (Mr) and a peak time factor (IDiff), are further constructed based on signal propagation characteristics. The attenuation factor (Mr) quantifies the attenuation degree of a first path, and the peak time factor (IDiff) measures the difference between the first path and the peak path position under different channels, fully reflecting the differences between the two variables between different channels. The method further calculates the weight of each feature according to a relief algorithm, explores the ability of the feature to distinguish close-range samples, introduces weight coefficients, and performs support vector machine (SVM) classification on the weighted features to improve the ability to recognize non-line-of-sight signals.

The invention provides a rigidity parameter optimization method and device for a vibration isolator, and the method comprises the steps: obtaining data used for representing the working state of a vibrating screen and vibration isolation performance data to form a sample set, and determining an optimization target and a constraint condition; evaluating factor weights of the candidate factors by adopting a Relief algorithm, and determining an optimization factor set according to the factor weights; the candidate factors comprise at least two of rubber hardness, rubber diameter, rubber rod height, rubber pre-pressing amount and initial angle of the vibration isolator; determining an optimization factor combination meeting an optimization target in the optimization factor set by adopting a particle swarm optimization algorithm; and verifying the optimization factor combination and outputting a target parameter. According to the method, the Relief algorithm is utilized to perform importance evaluation and screening on the candidate factors, and the particle swarm optimization algorithm is combined to search the optimal combination in the optimization factor set, so that the technical effect of efficiently determining the optimal solution of the stiffness parameter of the vibration isolator in a complex parameter space is achieved.

The invention discloses a tumor hyperspectral image classification method based on multi-modal feature fusion, and the method comprises the following steps: S1, obtaining brain tumor hyperspectral image data, and carrying out the reflectivity calibration and normalization preprocessing; s2, screening an optimal wavelength combination from the hyperspectral image through a DGWCR algorithm to obtain spectral features; s3, screening five spectral indexes sensitive to tumor tissues from preset 38 spectral indexes by using a Relief algorithm, and calculating to obtain spectral index characteristics; s4, using an IAPS algorithm to extract rotation invariant spatial texture features of the image; s5, splicing and fusing the spectral features, the spectral index features and the spatial texture features along channel dimensions; and S6, inputting the fused features into a machine learning classifier, and outputting a classification result. According to the method, through combination of spectral features, tumor sensitive spectral indexes and spatial invariant features, the classification accuracy OA under small samples is remarkably improved to be larger than 95%.

This invention provides a method and intelligent equipment for continuous non-explosive mining of hard rock, applicable to the field of mining machinery and equipment. This method includes sensor deployment, data processing, automatic adjustment of operating parameters, and self-learning. By installing vibration sensors, torque sensors, speed sensors, and temperature sensors on the mining equipment and collecting time series data, the system uses feature extraction based on wavelet packet singular value decomposition, feature dimensionality reduction using the Relief algorithm, and an improved XGBoost model to identify rock formation hardness, achieving highly accurate hardness detection. Based on the identification results, the control center automatically adjusts operating parameters to ensure a high degree of match between the operating parameters and the rock formation hardness, achieving an efficient and stable mining process, reducing energy consumption while minimizing wear caused by vibration. By recording and analyzing data from actual operations, the XGBoost model is continuously optimized to enhance the system's adaptability and accuracy. By introducing intelligent, data-driven technologies, the system achieves adaptive capabilities in mining environments with complex and changing rock formation hardness.

The application provides a vehicle license plate detection method based on spatial clustering and time sequence exposure, realizes fast and efficient detection and identification of vehicle license plate information of small development boards and other equipment through a traditional image recognition algorithm, and achieves high detection efficiency; a machine learning method is the best choice for vehicle license plate detection, but in order to adapt to small development board equipment with low computing power, a traditional image recognition algorithm is better; in order to solve the problems of poor precision, generalization ability and robustness of the traditional recognition method, a spatial license plate fast strip-shaped recognition area and a time sequence license plate efficient recognition area of image data of a video detection device are extracted, and a license plate spatial distribution and a time sequence exposure degree rule are obtained; on the basis of the traditional image recognition algorithm, the picture is optimized according to the rule, the algorithm pressure is relieved, the calculation time is reduced, the delay is reduced, the detection efficiency is improved, and the false detection rate and the missed detection rate are reduced.

The application discloses a kind of fusion feature screening and dynamic optimization's BP neural network tobacco moisture prediction method, it is related to cigarette production intelligent control technical field.The method includes: collecting the multiple-source production data related to moisture regaining process in silk production line;Using Relief algorithm carries out feature screening, selects the key feature variable that influences moisture regaining export moisture;The normalized processing is carried out to the feature data screened out, and is divided into training set, test set and verification set;BP neural network prediction model with specific structure is constructed and trained, the model adopts the structure of linear layer and activation function alternation, and introduces one-dimensional convolution layer to reduce parameter quantity, uses Mini-batch gradient descent method and cosine annealing learning rate scheduling strategy;Using the model trained, according to real-time production data, the moisture value of moisture regaining export is predicted, and according to this, water addition control is guided.The application can realize the accurate prediction of tobacco moisture, effectively improve the stability of moisture regaining export moisture, to improve product quality.

The invention relates to the technical field of power system operation and control, and discloses a method and system for extracting transient stability key response characteristics of a sending-end system.The method comprises the steps that negative impedance equivalent simplification is conducted on a new energy unit, and a mathematical model for new energy grid-connected sending is constructed; based on the mathematical model, analyzing the influence of new energy access on the transient characteristics of the system from the perspective of transient energy, and extracting key factors influencing the power angle stability level of the system; and constructing an original feature set based on the key factors, and performing feature dimension reduction by adopting a Relief algorithm to obtain an optimal feature subset. According to the method, the technical problem of realizing the integrity of information extraction for complex characteristics of a large power grid is solved.

The application discloses a kind of based on HRRP feature's bistatic radar passive jamming identification method;The invention is first to HRRP feature is normalized, alignment and determines support area range etc. Preprocessing operation, then extract a total of 18 kinds of features including peak point length, normalized variance and front end structure ratio, and use the feature selection algorithm based on Relief algorithm and specificity coefficient to select features with high discrimination and low redundancy to construct feature vectors, and finally use SVM to complete passive jamming identification. The application proves that the combination of bistatic radar and selected features can effectively improve the performance of jamming identification.

The invention discloses a prediction model screening method and device based on a weighted information criterion method, and relates to the technical field of data analysis. By introducing a multi-stage feature screening mechanism and combining a partial autocorrelation function (PACF) and a Relief algorithm, interference of redundant and invalid features on a model evaluation result is reduced, and scientificity and stability of prediction model screening are improved from the source. Different evaluation indexes (including error indexes, information criterion indexes and generalization performance indexes) can be endowed with differentiated weights through a weighted information criterion function, so that multi-dimensional performance expression is considered in a model screening process, and single-index deviation is avoided.