63 results about "Kernel extreme learning machine" patented technology

The invention relates to the technical field of synchronous reluctance motor structure optimization, in particular to a permanent magnet auxiliary synchronous reluctance motor structure parameter multi-objective optimization method. The method comprises the steps that motor structure optimization parameters are selected, and the change range of the parameters is determined; carrying out comprehensive sensitivity analysis according to the selected optimization target; performing hierarchical processing on the optimization variables according to the comprehensive sensitivity index, dividing the optimization variables into a strong sensitive layer and a weak sensitive layer, and taking parameters in the strong sensitive layer as to-be-optimized parameters of the next algorithm; comparing the prediction precision of the back propagation neural network, the radial basis, the extreme learning machine, the support vector machine and the kernel extreme learning machine, and selecting an agent model with an optimal prediction effect; combining the established high-precision agent model with a fast non-dominated sorting genetic algorithm, and searching an optimal combination of to-be-optimized parameters; and finally, carrying out single parameterization scanning determination on parameters in the weak sensitive layer by utilizing finite element simulation so as to obtain optimal structural parameters of the permanent magnet auxiliary synchronous reluctance motor.

The invention relates to a transformer fault diagnosis method. The invention discloses an oil-immersed transformer fault diagnosis method fusing a digital twinborn model, and the method specifically comprises the four steps: 1, achieving the equalization of a fault sample through employing the I NNOS; 2, extracting 9-dimensional candidate ratio signs, and performing fusion and dimension reduction on the ratio signs by using sparse principal component analysis (SPCA); 3, a kernel extreme learning machine (KELM) model is built, an AO algorithm is introduced to optimize parameters of the KELM model, and an AO-KELM optimal diagnosis model is built; and 4, establishing a digital twinborn five-dimensional model to achieve real-time interactive mapping between a physical entity and a virtual space and realize online diagnosis of the transformer fault. The transformer operation state can be simulated in real time, the fault diagnosis precision is improved, and intelligent application of transformer digital twinning is achieved.

The invention belongs to the technical field of electric power information physical system security, and discloses an urban power grid information physical system security situation early warning method based on cross-space risk propagation. A power grid physical layer and information layer coupling model and a cellular space false data injection attack model are constructed, and a fault cross-space propagation mechanism is simulated through an event-driven cellular automaton theory; an integrated kernel extreme learning machine model is adopted to predict the operation state of the power grid, multi-dimensional data fusion is realized through a radial basis kernel function, and an integrated structure is adopted to fuse a plurality of model prediction results so as to improve the model prediction precision; and establishing an early warning system including voltage out-of-limit, line overload and load loss indexes, and dynamically distributing weights and dividing early warning grades in combination with an entropy weight method. According to the method, the problem of low precision of urban power grid security situation early warning under multivariate disturbance is solved, and the accuracy and robustness of security situation early warning can be effectively improved.

The invention relates to the technical field of power prediction, in particular to a photovoltaic power generation power prediction method, device and equipment and a computer storage medium. According to the photovoltaic power generation power prediction method provided by the invention, in order to make up the defects of a heuristic algorithm, a VMD-KELM photovoltaic power generation power prediction model based on BWO optimization is provided; the model combines the advantages of VMD (variational mode decomposition), KELM (kernel extreme learning machine) and BWO (white whale optimization algorithm), and the photovoltaic power environment can be quickly and accurately predicted according to the nonlinear data mining requirement and the high-uncertainty photovoltaic power environment.

The invention discloses a lithium battery health state estimation method and system. Belongs to the technical field of lithium batteries, and comprises the following steps: firstly, collecting single lithium battery data under the condition of continuous charging and discharging, then introducing an adaptive factor to improve Kalman filtering to obtain adaptive Kalman filtering, and using the improved adaptive Kalman filtering to carry out missing value filling on the collected data to improve data integrity. Secondly, improving an information acquisition optimization algorithm by using a quadratic interpolation method and an elite random reverse learning strategy to obtain an IIAO, optimizing hyper-parameters of a time sequence analysis model LightTS model by using the IIAO, inputting the processed lithium battery data into the optimized LightTS model for prediction to obtain a preliminary prediction result, and carrying out prediction on the preliminary prediction result; and finally, performing error correction on an error between the preliminary prediction result and the true value by using an online sequential reduced order kernel extreme learning machine (OS-RKELM) to obtain an error prediction value, and finally, superposing the preliminary prediction value and the error prediction value to obtain a final lithium battery health state prediction value.

The invention discloses a circuit breaker fault diagnosis method based on global-local feature extraction. Comprising the following steps: 1, collecting vibration signals of a circuit breaker in a normal working condition and a fault state as original data; 2, extracting original time domains and frequency domains of the vibration signals of the circuit breaker operating mechanism in different states and time-frequency characteristics of a generalized S transformation matrix, and regarding the original time domains and frequency domains as global characteristics; 3, extracting local features of the vibration signals by using a texture feature extraction mode of matching a local binary pattern with a gray-level co-occurrence matrix; 4, further screening the features by using maximum correlation and minimum redundancy to obtain a feature set; and 5, in combination with a kernel extreme learning machine classifier and GL-mRMR feature extraction, obtaining a GL-mRMR-KELM model applied to circuit breaker vibration data classification. The method is used for fault diagnosis of a high-voltage circuit breaker, and solves the technical problems of low universality and poor accuracy under small samples in a current operating mechanism state diagnosis algorithm.

The invention discloses a battery health state estimation method and system, and the method comprises the steps: extracting the strong correlation features of a battery health state through fusing a physical model and relaxation time based on electrochemical impedance spectroscopy data, and carrying out the feature screening and reconstruction through a random forest algorithm, and then optimizing network parameters of the multi-scale kernel extreme learning machine by using a sparrow algorithm so as to estimate the state of health of the battery. The estimation method provided by the invention can capture data features from multiple angles, has good generalization ability, and has significant advantages in the aspects of precision and efficiency compared with an existing estimation method.

The invention belongs to the technical field of power transmission line monitoring and dynamic capacity increasing. A power transmission line dynamic current-carrying capacity prediction and capacity increase method based on double-model adaptive fusion comprises the following steps: data acquisition and preprocessing: acquiring historical weather and current-carrying capacity data, and cleaning and standardizing the historical weather and current-carrying capacity data; weather-driven prediction: optimizing a kernel extreme learning machine through an artificial egret optimization algorithm to predict future meteorological parameters, and substituting the future meteorological parameters into a heat balance equation to calculate a first current-carrying capacity predicted value; performing data-driven prediction, optimizing variational mode decomposition by adopting a dynamic perception and accurate capture algorithm, and directly predicting a second current-carrying capacity predicted value in combination with a long-short-term memory network; in the fusion step, weights are dynamically distributed according to errors of the two models on the verification set, and self-adaptive weighted fusion is carried out to obtain a final point prediction value; and a probability modeling step: fitting probability distribution based on historical errors, and constructing a confidence interval. According to the invention, the precision and reliability of current-carrying capacity prediction are improved, and dynamic capacity increasing and safe operation of the power transmission line are realized.

The invention discloses a fault diagnosis and health management prediction method based on a laser gyroscope, and belongs to the technical field of inertial navigation equipment health management. The method comprises the following steps: acquiring and preprocessing a time sequence data set of operation core parameters of the laser gyroscope; performing wavelet packet decomposition on the time sequence data to extract high-frequency fault features, extracting time sequence statistical features by a sliding window method, and combining principal component analysis to perform dimensionality reduction and fusion to obtain a multi-dimensional feature vector; inputting the feature vectors into an improved kernel extreme learning machine model, optimizing parameters through a particle swarm optimization algorithm, and then outputting fault types and grades; on the basis of a fault diagnosis result, an analytic hierarchy process is adopted to endow parameter weights, and a health factor calculation model is constructed to quantify a health state; taking the health factor sequential sequence and the key parameter degradation trend as input, and combining a bidirectional long-short-term memory neural network with health factor sequence constraint to predict residual life and a confidence interval; new data are regularly brought in, and model parameters are updated through transfer learning to realize dynamic iteration. The method solves the problems of early fault recognition lag and low life prediction precision of a traditional method, is suitable for the fields of aerospace, precision navigation and the like, and has remarkable engineering application value.

The invention discloses an SVG voltage ride-through control parameter identification method based on a kernel extreme learning machine in a double-solution space, and the method comprises the following steps: S1, dividing SVG fault stages, building an SVG steady-state control model and a fault control model, and according to the difference of the influence of control parameters on steady-state characteristics and fault characteristics, carrying out the recognition of SVG voltage ride-through control parameters; dividing a solution space of SVG control parameters into a double-solution space composed of a steady-state parameter solution space and a dynamic parameter solution space; s2, extracting a corresponding observed quantity characteristic value, constructing an input-output sample, and establishing a kernel extreme learning machine model in a double-solution space; s3, training a kernel extreme learning machine by using the input-output samples, and inputting RT-LAB hardware in-loop test data into the trained kernel extreme learning machine to identify SVG control parameters; according to the method, the adaptability and the accuracy of identification are improved, so that accurate modeling of the SVG is realized.

The invention provides a resin composition for a halogen-free flame-retardant cable sheath and a preparation method of the resin composition, and belongs to the technical field of polymer composites for cables, the preparation method comprises the following steps: S1, raw material grading pretreatment and core parameter accurate determination; s2, gradient pre-dispersion and uniformity feedback regulation and control; s3, step-by-step grafting compatibility and grafting rate feedback regulation and control; s4, carrying out dynamic flame-retardant regulation and control melt blending; and S5, gradient cooling granulation and finished product performance verification. In two key links of gradient pre-dispersion and dynamic melt blending, an improved fruit fly optimization algorithm-adaptive fuzzy neural network algorithm and an improved sparrow search algorithm-kernel extreme learning machine algorithm are respectively embedded, and the two algorithms are bidirectionally interactively fused, so that the problems that traditional process parameters are set according to experience and the performance fluctuation is large are solved, and the dynamic melt blending method is suitable for industrial production. The synergistic improvement of the uniformity, the flame retardant property and the mechanical property of the resin composition is realized.

The invention discloses a lithium battery health state estimation method and system, electronic equipment and a storage medium, and belongs to the field of battery state prediction.The method comprises the steps that probability modeling is conducted on a small amount of lithium battery charging and discharging time sequence data through a variational self-encoding network, an expansion sample with degradation consistency is generated, and an expansion data set is formed; key health indexes reflecting battery aging characteristics in the extended data set are constructed into a characteristic matrix as input; a kernel extreme learning machine is adopted as a nonlinear regression prediction model, global optimization is carried out on kernel function parameters and regularization coefficients of the nonlinear regression prediction model through an artificial bee colony optimization algorithm, and finally an SOH prediction model which is obtained through training and represents the relation between the battery health state and the characteristics is obtained; and outputting a battery health state estimation result based on the SOH prediction model. According to the method, the health state prediction precision and efficiency can be improved.

The present invention provides a method and a device for training a bearing fault diagnosis model. The training method includes the following steps: performing data segmentation on the fault vibration signals of different types of bearing faults to obtain a training set, where the training set includes a number of training samples; calculating the silhouette coefficients of the training samples in the training set under different parameters to select the optimal parameters of the enhanced composite multi-scale slope entropy; using the enhanced composite multi-scale slope entropy after parameter selection to extract fault features to obtain a training feature set; and using the training feature set to train the honey badger algorithm optimized kernel extreme learning machine to obtain a trained composite multi-scale fault diagnosis model. The present invention can effectively identify bearing faults of different types and damage degrees, and the average recognition rate reaches 99.9%.

This application provides a method, apparatus, and electronic device for detecting outliers in satellite telemetry data. The outlier detection model for telemetry data is trained using a kernel extreme learning machine, which can effectively detect outliers in satellite telemetry data, greatly improving the efficiency and accuracy of outlier detection.

The application relates to a fan blade icing fault diagnosis method based on a GWO-REKELM, and the method comprises the following steps: 1) pre-processing and normalizing all fan operation data; 2) using a random forest to perform feature screening on all fan operation data; 3) using an oversampling technology to balance the data set of the screened fan operation data; and 4) using a grey wolf optimization kernel extreme learning machine model with an improved objective function to train and test the data set, and completing fan blade icing fault diagnosis. Compared with the prior art, the application has the advantages of improving the correctness of fan blade icing fault diagnosis and the like.

The invention provides a pumped storage unit vibration trend prediction method and system based on a combined model, and belongs to the technical field of pumped storage unit detection. Comprising the following steps: acquiring an original vibration signal and performing variational mode decomposition by adopting a self-adaptive signal decomposition method to generate a plurality of subsequences; a singular spectrum analysis method is adopted to extract dominant components of all the subsequences, and superposition reconstruction is carried out on residual components; the processed subsequences are converted into a training set and a test set of a prediction model through phase-space reconstruction, and the training set and the test set serve as prediction model input for model training; and on the basis of a kernel extreme learning machine model and an IBiGRU-KAN model, predicting the high-frequency sub-sequence and the low-frequency sub-sequence in each sub-sequence, and superposing the prediction results of each sub-sequence to determine the vibration trend of the pumped storage unit. According to the method, complex data processing and model calculation tasks can be efficiently executed, and the vibration trend of the pumped storage unit is accurately predicted.

The invention relates to a tobacco shred moisture early warning method, a tobacco shred moisture early warning device and a storage medium, and the tobacco shred moisture early warning method comprises the following steps: obtaining a to-be-detected tobacco shred sample; inputting the to-be-detected tobacco shred sample into a pre-trained tobacco shred moisture prediction model to obtain a tobacco shred moisture prediction result; wherein the pre-trained cut tobacco moisture prediction model is obtained by training a preset kernel extreme learning machine prediction model by adopting an improved hunter prey optimization algorithm; judging whether the tobacco shred moisture prediction result exceeds a preset moisture threshold value or not; and if yes, sending out an early warning prompt. According to the invention, the problem of inaccurate moisture early warning in the tobacco shred production process is solved, and the accuracy of moisture early warning in the tobacco shred production process is improved.

The present application relates to a kind of short-term load prediction method based on ISSA-HKELM, the method includes: first, in view of the defect of kernel extreme learning machine KELM, combined with Gaussian kernel function and polynomial kernel function, construct the hybrid kernel extreme learning machine HKELM with stronger generalization ability;Second, in view of the problem that sparrow search algorithm is easy to fall into local extremum, adaptive t distribution strategy and dynamic adaptive weight are introduced to improve sparrow search algorithm;Third, the improved sparrow search algorithm ISSA is used to optimize the parameter of hybrid kernel extreme learning machine HKELM and establish ISSA-HKELM prediction model;Finally, short-term load prediction is carried out using the established ISSA-HKELM model.Compared with prior art, the present application has good prediction accuracy and robustness and the like advantages.

The present invention discloses a temperature extraction method for a Brillouin optical time-domain analysis system (BOTDA) based on a nuclear extreme learning machine, comprising: using the Brillouin optical time-domain analysis system to collect Brillouin gain spectrum parameters of a test optical fiber; using the nuclear extreme learning machine to analyze the parameters collected by the Brillouin optical time-domain analysis system; using the obtained real matrix as training data for the nuclear extreme learning machine; training the nuclear extreme learning machine using the training data to extract accurate temperature information; and utilizing a faster processing speed to improve system performance. The present invention introduces a nuclear extreme learning machine algorithm to improve the temperature extraction accuracy and efficiency of the Brillouin optical time-domain analysis system, which is beneficial to the application of the Brillouin optical time-domain analysis system in actual detection.

The invention belongs to the field of industrial network security, and particularly relates to an industrial control network traffic intrusion detection method combining deep learning and multi-kernel learning, which comprises the following steps of: firstly, acquiring an industrial control traffic data sample and performing feature cleaning, and then establishing a multilayer deep neural network DNN model; and the optimal parameter configuration of the model is obtained through iterative training. Then extracting a middle layer representation result of the DNN as a kernel matrix of mapping; the kernel matrixes obtain respective weights through a multi-kernel learning MKL process, and linear combination is carried out. And finally, the combined kernel matrix replaces the original shallow kernel function of the kernel extreme learning machine KELM, and a multi-depth kernel extreme learning machine MDKELM model for detecting the traffic sample type is formed. The model is deployed at the edge of a device and a network, external access traffic is detected, whether the external access traffic belongs to normal traffic is judged, and if not, early warning is given out in time.

The invention discloses a sailboat motion dynamic prediction method based on grey wolf optimization and a kernel extreme learning machine. The method comprises the following steps: establishing a three-degree-of-freedom motion prediction model based on a grey wolf optimization algorithm and a kernel extreme learning machine algorithm; determining input characteristics and three-degree-of-freedom output response of the three-degree-of-freedom motion prediction model, and dividing the model into a training set and a verification set according to a proportion; performing preliminary training on the three-degree-of-freedom motion prediction model based on the training data set to obtain a preliminarily trained three-degree-of-freedom motion prediction model; testing the preliminarily trained three-degree-of-freedom motion prediction on the verification set, and iteratively searching a parameter position for reducing the fitting fitness of the three-degree-of-freedom motion model by adopting a grey wolf optimization algorithm to obtain a trained three-degree-of-freedom motion prediction model; and based on the trained three-degree-of-freedom motion prediction model, dynamic prediction of ship control motion is realized.

The application discloses a street lamp line ground fault intelligent diagnosis system and method, belongs to the technical field of power equipment operation monitoring, and comprises a data acquisition and preprocessing unit, a feature extraction unit and a fault diagnosis unit. The data acquisition and preprocessing unit collects original residual current data of a street lamp line through a residual current transformer and an oscilloscope. The application adopts a parrot optimization algorithm to adaptively determine optimal decomposition parameters of VMD, overcomes the limitations of artificial parameter setting, ensures the quality of data decomposition, lays a foundation for extracting high-quality fault features, solves the problems of lack of reliability of existing fault protection, great influence of human factors, and low accuracy of existing street lamp ground fault diagnosis systems and methods, calculates sample entropy values, constructs a multi-dimensional feature vector set representing line states, inputs the multi-dimensional feature vector set into a fault diagnosis unit, and the fault diagnosis unit trains a kernel extreme learning machine classifier by using the feature vector set.

The invention discloses a porous scaffold SLM preparation method and device based on machine learning, and belongs to the technical field of additive manufacturing. The method comprises the following steps: firstly, designing a composite lattice porous structure consisting of inner-layer simple cubic cells and outer-layer body-centered cubic cells, wherein the inner-layer simple cubic cells and the outer-layer body-centered cubic cells are directly connected through nodes; then, a Latin hypercube sampling design test scheme is adopted, and initial small sample data of the selective laser melting process are obtained; then, a regression synthesis minority class oversampling technology is introduced to carry out enhancement processing on the data set; on the basis, establishing a kernel extreme learning machine prediction model optimized by a swarm intelligence algorithm; and finally, reversely solving optimal process parameters based on the target performance and completing preparation. Through the composite structure design, the stress shielding effect is effectively relieved, the problems of high experiment cost, less sample data and low prediction precision in traditional process optimization are solved, and efficient and accurate customized manufacturing of the high-performance medical CoCrMo implant is achieved.

The invention discloses a target identification method for a nonlinear aperture inverse imaging simulation dynamic intersection process. The method comprises the following steps: S1, generating a knowledge base containing nonlinear aperture inverse imaging simulation data; s2, performing multimode feature extraction and data fusion based on the knowledge base to obtain a fusion feature vector; s3, further fusing the fusion feature vectors to obtain depth features; and S4, training a kernel extreme learning machine by using the depth features after dimension reduction, and performing target recognition by taking the trained kernel extreme learning machine as a kernel classifier. According to the method, the dynamic rendezvous imaging data knowledge base can be established to the greatest extent through analog simulation under the condition that actual measurement data is deficient, early-stage training data is provided for an identification algorithm, and the algorithm debugging efficiency and accuracy during actual measurement application are improved.

The invention discloses a distributed power supply output interval pseudo measurement generation method considering uncertainty, comprising the following steps: for a wind generating set, constructing and training a double-layer kernel extreme learning machine model, the first layer of the model is used for correcting the predicted wind speed, and the second layer of the model is used for correcting the predicted wind speed; the second layer is used for predicting uncertainty upper and lower limit intervals of fan output under the corrected wind speed; for a photovoltaic generator set, a dual-output feedforward neural network model is constructed and trained, and the model directly outputs an output upper and lower limit interval of a photovoltaic system under a predicted meteorological condition; and taking the predicted output upper and lower limit intervals as interval pseudo-measurement data of the corresponding distributed power supply nodes, and applying the interval pseudo-measurement data to state estimation or operation optimization of the power distribution network. According to the method, the output fluctuation range of wind power and photovoltaic power at the future moment can be effectively described, reliable interval pseudo measurement data is generated, and the defect of insufficient monitoring of distributed power supply nodes is overcome.

The present disclosure provides a method for determining the phreatic line of a stepped reservoir bank slope group, a program product and an electronic device, and relates to the technical field of engineering data processing. The method comprises: obtaining first monitoring data of a target slope and second monitoring data of a related associated slope; fusing the first monitoring data and the second monitoring data according to a time-varying weight of hydraulic association to obtain a hydraulic association fusion feature; processing the hydraulic association fusion feature of the target slope and the hydraulic association fusion feature of a plurality of sample data by using a kernel extreme learning machine to obtain a phreatic line elevation inversion value; constructing a hydraulic association strength field based on the distance between slopes and the fluctuation rate of the reservoir water level; fusing the phreatic line elevation inversion value and the key parameters in the first monitoring data to obtain a group perception fusion feature; and obtaining a phreatic line elevation optimization value according to the group perception fusion feature, the hydraulic association strength field and the phreatic line elevation inversion value. The present disclosure improves the accuracy and stability of determining the phreatic line of the stepped reservoir bank slope group.

The present invention provides an effective wind speed soft measurement method, comprising a wind turbine SCADA data acquisition module, an offline optimization calculation module based on an improved gray wolf optimized IGWO algorithm, an effective wind speed soft measurement modeling module based on a kernel extreme learning machine (KELM) algorithm, and the like. The signal output end of the wind turbine SCADA data acquisition module sequentially passes through a SCADA data preprocessing and normalization module, an offline optimization calculation module based on an improved gray wolf optimized IGWO algorithm, an effective wind speed soft measurement modeling module based on a kernel extreme learning machine (KELM) algorithm, an effective wind speed soft measurement model parameter optimization module based on the IGWO algorithm, and an effective wind speed soft measurement model performance evaluation module, and finally outputs the effective value of the wind speed. The present invention establishes an effective wind speed soft measurement model, which does not require wind tunnel experiments or virtual simulations to obtain the data required for effective wind speed soft measurement modeling, but only requires SCADA data of the wind turbines actually operating in the wind farm. The model has the characteristics of low cost, high prediction accuracy, and ease of implementation.

The application provides an abnormal load identification method and system of a subway direct-current power supply network and a readable storage medium, belongs to the technical field of urban rail transit direct-current traction power supply systems, obtains feeder line current data based on a single-end traction substation feeder line current collection current loop, and performs sampling window management; performs variational mode decomposition on the same window feeder line current, calculates energy weight, energy entropy and sample entropy, and constructs a multi-dimensional entropy feature matrix; then generates state identification through energy entropy first-level discrimination and correlation dimension second-level screening, and inputs the state identification into a kernel extreme learning machine diagnosis model after fusion with the entropy features, and outputs normal load, system oscillation, locomotive oscillation or short-circuit fault classification results to generate corresponding protection disposal.

The present application proposes a vocal singing practice quality grade evaluation method of fractional order spectrogram deep learning, including the following steps; Step S1, collect the audio signal of the singer's vocal vowel singing practice, and according to the singing index, label the corresponding quality grade, construct the sample data set, and use it for model training, testing and verification; Step S2, convert the audio signal into a series of fractional order spectrograms; Step S3, construct the fractional order spectrogram deep feature extraction network based on DenseNet and channel attention mechanism in the model, input the extracted fractional order spectrogram deep features into the BiLSTM network, and extract the time sequence features of the singing practice signal; Step S4, the quantum fireworks algorithm is used to optimize the hyperparameters of the kernel extreme learning machine of the model, and the extracted time sequence features are mapped to a high-dimensional space for quality grade decision, forming the evaluation result; Step S5, train the model; The present application can better adapt to the characteristics of non-stationary signals and provide more accurate spectrum analysis.

The invention relates to the technical field of cable detection, and discloses a cable insulation state diagnosis model construction method, a cable insulation state diagnosis method and a cable insulation state diagnosis device. Combining a plurality of base learners including a gradient boosting decision tree algorithm model, a lightweight gradient elevator algorithm model, an extreme gradient boosting tree algorithm model and a support vector machine algorithm model to construct a target base learner combination model, and using an improved frost ice optimization algorithm to optimize a mixed kernel extreme learning machine as a meta learner; therefore, the finally constructed cable insulation state diagnosis model can give consideration to different feature selection logics and model advantages, the diagnosis precision and generalization ability are effectively improved, information redundancy is reduced, and the problem of performance limitation of a single model is solved.