43results about How to "Reduce input dimensionality" patented technology

The invention relates to a stacked SAE (Sparse Autoencoder) deep neural network-based bearing fault diagnosis method. The first layer of a network is applied to the qualitative judgment of a bearing fault, that is, the first layer of the network is applied to the fault type judgment of the bearing fault; and the second layer of the network is applied to the quantitative judgment of the bearing fault, that is, the second layer of the network is applied to the severity judgment of the bearing fault. According to the method of the invention, empirical mode decomposition (EMD) and an autoregressive (AR) model are combined together to perform pre-processing on original bearing signals, the parameters of the AR model are extracted and are adopted as the input of the network, and therefore, the input dimensions of the network can be greatly reduced, the simplification of calculation can be facilitated, and the training and testing of the network can be accelerated; a deep neural network on which the method of the invention is based can further automatically extract features of the input and qualitatively and quantitatively determine the bearing fault automatically, and therefore, the diagnostic accuracy of the method of the present invention can be ensured, and at the same time, dependence on signal processing expertise can be decreased, manual judgment is not required, the consumption of manpower can be decreased; and thus, the method has a higher practical value in the era of big data.

The invention discloses a neural network and particle swarm optimization algorithm-based building energy consumption predicting method. The method comprises the following four main steps of collecting related data of building energy consumption , and pre-processing the data; determining an input output item and a network structure of a multi-layer feedforward neural network model, wherein the multi-layer feedforward neural network model has an error back-propagation learning function; optimizing the connection weigh value and threshold value of a BP network by using a particle swarm algorithm; performing short-term prediction on the building energy consumption by using the neural network model which is obtained through optimization. According to the predicting method, pre-input variables are subject to main component analysis by using a statistical product and a service resolve scheme software, and the variable according with a main component extraction requirement is selected, so that the input dimension is reduced; the structure and parameters of the neural network model are optimized through the overall optimization ability of the particle swarm algorithm, so that compared with the current building energy consumption predicting method, the predicting model provided by the invention has the advantages of simple structure, high predicting precision and the like.

The invention discloses a power load SOM-FCM hierarchical clustering method based on load fluctuation feature extraction. The power load SOM-FCM hierarchical clustering method comprises the followingsteps: S1, taking power load active time series data as an input, preprocessing the load data, and obtaining a volatility feature vector; S2, performing first layer coarse clustering on the volatilityfeature vector obtained in the step S1 by using a self-organizing map neural network (SOM) clustering algorithm, and obtaining the clustering result and a weight matrix; and S3, using the coarse clustering result in the step S2 as an input of a self-adaptive fuzzy C-means algorithm (FCM), adding effective function judgment, and finally obtaining the clustering result with the optimal clustering number. The invention adaptively clusters the power load by extracting the load fluctuation feature parameters, solves the clustering problem in the case that the electric load composition is complicated and the number is large, clusters the load from the load curve volatility itself, can easily obtain the analysis data, is simple in calculation, and is easy to transplant.

The invention discloses a fault diagnosis method for an induction motor of an electric automobile based on broad learning. According to the method, an incremental broad learning (IBL) method is adopted to diagnose a fault of the induction motor, the structure of IBL is simple, and the network can be effectively trained and retrained. Feature extraction (empirical mode decomposition and sample entropy), incremental broad learning and singular value decomposition are combined to form an intelligent diagnosis method for the induction motor of the electric automobile. Experimental results show that the method is superior to other algorithms in fault diagnosis for the induction motor of the electric automobile. In addition, the IBL simplified by singular value decomposition (SVD) is smaller inerror and higher in system stability than the previous IBL.

The invention discloses an energy spectrum analysis method based on Hilbert curve transformation and depth learning, belonging to the field of radiation environment monitoring and image recognition, which has the characteristics of high recognition rate, good stability and strong adaptability. The invention comprises the following steps: (1) acquiring a detection energy spectrum and a simulation energy spectrum and performing pretreatment; (2) Transforming the traditional one-dimensional energy spectrum analysis into two-dimensional image recognition, training and testing the full spectrum input depth learning; (3) A depth learning algorithm for fast nuclide recognition is constructed, and the effect of the depth learning classifier is analyzed by determining the classification threshold and ROC curve.

The invention discloses a WT-KPCA-SVR coupling model based gas emission quantity prediction method. The method comprises the following steps: firstly, performing data preparation, namely collecting gas emission quantity monitoring data and corresponding factors, extracting gas emission quantity subsequences according to wavelet transform, and separating out a trend item subsequence and a fluctuation subsequence; determining the influence factors of each subsequence according to a gray relative analysis method, performing kernel principal component dimensionality reduction on the influence factors of each subsequence to reconstitute the principal component of each subsequence; combining the reconstituted principal components of all the subsequences and values of all gas emission quantity subsequences into a sample set; respectively establishing support vector machine regression models of the trend item subsequence and the fluctuation subsequence according to a training sample; synthesizing the two models to obtain a final gas emission quantity prediction model; performing model precision detection by using a detection sample, wherein the model can be used if passing the detection. The model is reliable in design principle, simple in prediction method, high in prediction accuracy and friendly in prediction environment.

The invention relates to a traffic sign recognition method and system with a depth learning model based on a compact neural network, wherein the method and system are suitable for traffic sign detection and recognition in an image or video. The system comprises an image acquisition module, a picture preprocessing module, a traffic sign recognition module, and a voice prompting module. The picture preprocessing module carries out three steps of color positioning, shape detection, and picture zooming to obtain pictures with the same sizes. The traffic sign recognition module is used for classifying and identifying the pre-processed pictures, obtaining a traffic sign classification and recognition result by the trained depth learning model based on a compact neural network, and then transmitting the result to the voice prompting module for voice prompting. According to the system provided by the invention, the compact-neural-network-based depth learning model having advantages of small modules, high precision, low computing consumption, and high transportability to a mobile phone platform and the like serves as a core; and the system has advantages of many type of identified traffic signs, high precision, and good real-time performance.

The invention discloses a joint modulation recognition method based on RBM networks and a BP neural network. The joint modulation recognition method comprises the steps that 1, a modulation signal ispreprocessed; 2, feature parameters are extracted; 3, a training sample and a test sample of each modulation mode are randomly generated; 4, an input layer and a first hidden layer of the BP neural network are taken as one RBM network to be trained; 5, a collection of parameters of the RBM network is initialized; 6, the RBM network is trained to obtain the collection of the parameters and output features of the hidden layer; 7, the first hidden layer and a next layer of the BP neural network are taken as a visible layer and a hidden layer of a second RBM network to be trained, the output of the first RBM network is taken as the input of the second RBM network, and the step 5, the step 6 and the step 7 are repeatedly conducted until collections of parameters of all RBM networks are acquired; 8, the BP neural network is retrained until an optimal solution state is achieved; and 9, test data is normalized, the trained BP neural network is input, and the modulation mode recognition rate iscalculated. The joint modulation recognition method has the advantages that the input dimension is reduced, and the modulation recognition rate is increased.

The present invention provides a multi-technology fusion aero-engine fault intelligent diagnosis method. A data analysis method is employed to perform identification of fault data samples to match faults which has the similarity with samples to be subjected to analysis in a fault database being not lower than a determination value as suspected faults. According to the typical fault factor decision table, all the suspected faults are subjected to multi-time screening to obtain the limited number of mainly suspected faults with the maximum possibility. A pattern identification algorithm is employed to perform pattern identification of the mainly suspected faults, learning training samples when identification are from fault data features in a fault sample data feature library, identification features when identification are features of samples to be identified, and identification results are subjected to one or more than one kinetics feature inspection. The aero-engine structure-type fault intelligent diagnosis method can selectively determine the learning object identified by the mode to reduce the learning range, the recall ratio [Eta] is reduced from 1 to 33%, and the learning time is reduced from 90s to 19s. Through the inspection link, the identification result is inspected, and the false alarm probability of interference signals is reduced from 33% to 0.

The invention discloses a method for evaluating and calculating electric transmission line ice coating environment influence factors under a micrometeorological condition. The method comprises the following steps that first, on the basis of historical data provided by an overhead line structure micrometeorological station, the part with the insulator tension rapidly increasing in the micrometeorological station is sought, and corresponding micrometeorological parameters are acquired; the increasing speed of the insulator tension is calculated through a data processing formula; then, a normalization method is used for normalizing micrometeorological data with different dimensions and insulator tension data; after the above steps are completed, the correlation degree of each micrometeorological parameter and the insulator tension is calculated through a Deng grey correlation calculation formula; then, the existence time of the micrometeorological parameters is accumulated, and the accumulative time coefficient is calculated; after the above steps are completed, the calculated results are combined, and at last the method for evaluating and calculating the electric transmission line ice coating environment influence factors under the micrometeorological condition is obtained.

The invention provides a shield tunneling machine cutterhead torque prediction method and system based on a parallel neural network. The method comprises the following steps: selecting working parameters having great influence on shield tunneling machine cutterhead torque from a shield tunneling machine actual working database; preprocessing the selected working parameters; establishing a parallelneural network cutterhead torque prediction model according to the preprocessed working parameters, and training the parallel neural network cutterhead torque prediction model; and evaluating the prediction precision of the trained parallel neural network cutterhead torque prediction model on the cutterhead torque on different data sets. High-precision real-time prediction of the cutterhead torque is achieved, a driver can be guided to adjust operation parameters of the shield tunneling machine in advance, efficient and safe propelling of the shield tunneling machine is guaranteed, and therefore the intelligent level of the shield tunneling machine is improved.

The invention relates to a deep foundation pit deformation prediction method and system based on a neural network and rough set classification. Quantitative analysis can be carried out by utilizing arough set theory, so that the characteristics of the relationship between the data are reasoned and explained, a large number of index attributes influencing deep foundation pit deformation are reduced, the input dimension of a neural network is reduced, and the reduced index attributes are input into the neural network for prediction, so that the efficiency and accuracy of deep foundation pit risk prediction are improved.

The invention discloses an NOx online prediction method and system based on an intelligent method, and the method comprises the steps: analyzing a flue gas generation mechanism of a coal-fired unit and an SCR system mechanism, determining an input variable of an NOx emission prediction model, collecting operation data related to NOx emission, and constructing training set data; reducing an input dimension by adopting a PCA algorithm based on the collected training set data, and reconstructing a sample space; establishing an SCR inlet NOx dynamic prediction model by adopting an LSTM recurrent neural network model; and obtaining operation data related to the NOx emission amount at the current moment, and determining the optimal SCR inlet NOx prediction value at the current moment. By means of the method, the SCR inlet NOx concentration is predicted in advance, the ammonia spraying amount is adjusted in time, the SCR inlet NOx prediction precision is improved, and the method has guiding significance in pollutant emission and cost of a coal-fired unit.

The invention belongs to the technical field of face recognition, and particularly relates to a glasses removal method for fine-grained face recognition, which comprises the following steps: segmenting an initial glasses-worn face image into three image blocks, respectively identifying the three image blocks by Part1, Part2 and Part3, wherein the Part2 comprises a complete glasses part; establishing an ERCNN (Eyeglass Removing Deep Convolutional Neural Network) model; taking the Part2 as the input of the convolution layer of the ERCNN network model; performing feature selection and maximum element operation through an MFM unit in the network, and reconstructing the Part2 by utilizing operations of deconvolution, average pooling and weighted summation of elements one by one, thereby obtaining a new image block Part2_new after the glasses are removed; and combining the output Part2_new with the original Part1 and the original Part3 to obtain a complete face image of which the glasses areremoved.

The invention discloses a building energy consumption prediction method based on a longicorn beard optimization algorithm and a neural network, and the method comprises the following four main steps:1), collecting building energy consumption related data, and carrying out the principal component analysis and normalization preprocessing of the data; 2) determining input and output items and a network structure of a multilayer feedforward neural network model with an error back propagation learning function; 3) optimizing the connection weight and threshold of the BP network by using a longicorn beard algorithm; and 4) performing short-term prediction on the building power consumption by utilizing the optimized BAS-BP prediction model. According to the method, principal component analysis is carried out on the pre-input variables by utilizing a principal component analysis algorithm, and the variables meeting the principal component extraction requirement are selected, so that the inputdimension is reduced; and the structure and parameters of the neural network model are optimized by using the global optimization capability of the longicorn beard algorithm. Compared with an existing building energy consumption prediction method, the prediction model has the advantages of being simple in structure, high in prediction precision, short in operation time and the like.

The invention discloses a robot skill learning method and system based on reinforcement learning and unsupervised learning, and the method comprises the steps: enabling a robot controller to control an execution mechanism to act when a specific task is executed, and enabling an image collection subsystem to collect a real-time state image of the execution mechanism in a task environment. An encoder of the unsupervised learning subsystem converts the real-time state image into a real-time state code, takes a state image when a set execution mechanism executes and completes a specified task as a target image, and converts the target image into a target code; the deep reinforcement learning subsystem outputs an action value of the execution mechanism at the next moment according to the real-time state code and the target code; and the robot controller controls the execution mechanism to make corresponding actions in real time according to the action value of the execution mechanism until the execution mechanism executes and completes the task. Compared with a common end-to-end network, the system provided by the invention is more universal, can be applied to different targets and tasks, and can also compress the training time.

The invention discloses a PCA-FIG-SVM (Principal Component Analysis-Fuzzy Information Granulation-Support Vector Machine)-based absolute gas emission prediction method, belonging to an absolute gas emission prediction method for a coal mine stope face. The method comprises the following steps: (1) collecting absolute gas emission monitoring data and influence factors; (2) carrying out principal component modeling on the influence factor data and reconstructing principle components; (3) carrying out FIG on a time sequence composed of the absolute gas emission monitoring data; (4) establishing an SVM regression model of granulation data; (5) predicting the absolute gas emission. According to the method disclosed by the invention, a novel prediction model is provided for an absolute gas emission change trend, the change trend and the change space range are focused and researched, the principle components are extracted by utilizing PCA, the influence of redundant information is effectively reduced, the input dimensionality of the SVM model is reduced, and the design precision is reliable; the prediction method is simple, high in prediction precision and environment-friendly in prediction environment.

The invention provides a home scene layout method based on artificial intelligence. The home scene layout method comprises the following steps: acquiring a house type segment code sequence and a housetype segment classification label sequence of each functional room; training the neural network to obtain a house type segment sequence function allocation neural network; performing matching operation on the house type segment code sequence to be identified by using a house type segment sequence function allocation neural network to obtain a corresponding house type segment classification labelsequence; the obtained house type section classification label sequence is utilized to place materials layer by layer according to a home scene constraint tree to complete home scene layout of a single functional room; and repeating the steps until the layout of all the functional rooms forming the whole house type is completed. According to the home scene layout method, the processing efficiencyof a computer is effectively improved, the error of house type information is reduced, the neural network is used for identifying the house type segment code sequence, a function is allocated to the house type segment, the home layout information can be processed in real time, and the time for matching the collocation information is shortened.

The invention discloses a multivariate load prediction method for a user-level comprehensive energy system. The method comprises steps of carrying out ntrinsic mode decomposition on the electrical load, the cold load and the thermal load through complete ensemble empirical mode decomposition of self-adaptive noise, and decomposing strong non-stationary components obtained through decomposition again through variational mode decomposition; extracting principal components from the weather calendar rule feature set based on kernel principal component analysis to realize data dimension reduction;and forming an input set by using the decomposed non-stationary and stationary sequence components and principal components extracted from the weather calendar rule feature set respectively, performing prediction by using a DBiLSTM neural network and an MLR respectively, and finally reconstructing a prediction result to obtain a final prediction result. According to the method, the coupling characteristic of the multi-element load in the user-level comprehensive energy system can be deeply mined, the multi-element load prediction precision of the user-level comprehensive energy system is effectively improved, and a very good prediction effect can also be achieved for the multi-element load with randomness and large volatility.