109 results about "Backpropagation neural nets" patented technology

In a data storage system, a cache is managed by a predictive cache management engine that evaluates cache contents and purges entries unlikely to receive sufficient future cache hits. The engine includes a single output back propagation neural network that is trained in response to various event triggers. Accesses to stored datasets are logged in a data access log; conversely, log entries are removed according to a predefined expiration criteria. In response to access of a cached dataset or expiration of its log entry, the cache management engine prepares training data. This is achieved by determining characteristics of the dataset at various past times between the time of the access / expiration and a time of last access, and providing these characteristics and the times of access as input to train the neural network. As another part of training, the cache management engine provides the neural network with output representing the expiration or access of the dataset. According to a predefined schedule, the cache management engine operates the trained neural network to generate scores for cached datasets, these scores ranking the datasets relative to each other. According to this or a different schedule, the cache management engine reviews the scores, identifies one or more datasets with the least scores, and purges the identified datasets from the cache.

The invention discloses an on-line transmission line lightning shielding failure trip early-warning method. The method comprises the following steps of: performing statistics on historical lightning shielding failure trip information to obtain transmission line lightning shielding failure trip probability distribution by a two-dimensional information diffusion theory and a conditional probability method; selecting radar forecast data, such as echo intensity, echo tops and vertical accumulated liquid water content; establishing a lightning current magnitude prediction model based on a back propagation neural network; and sending real-time early warning and an early warning grade of transmission line lightning shielding failure trip probability according to the predicted lightning current magnitude and the side distance to lightning stroke and by virtue of the transmission line lightning shielding failure trip probability distribution model. According to the real-time forecast data of a meteorological radar, the method provided by the invention can predict the trip probability of a transmission line and send an early warning signal, thereby providing reference for decision-making analysis of grid dispatching operators, making a transmission line dispatching strategy in time, improving power supply reliability, lowering economic loss of a grid, and improving the reliable running ability of the grid.

The invention provides a short-term traffic flow prediction method based on a nerve network combination model. The method is used to construct a counterpropagation nerve network combination prediction model and the short-term traffic flow prediction method is provided based on the model. Aiming at a characteristic of a traffic flow, a fuzzy C mean value clustering algorithm is used to cluster the traffic flow. For bunch generated through clustering, a counterpropagation nerve network prediction model is constructed. According to grade of membership, a weighted sum of prediction model prediction results is calculated and is taken as a final prediction result. In order to increase prediction precision, a taguchi method is used to carry out test designing so as to test influences of different structure parameters on prediction model prediction precision, and an optimum structure parameter is used as an initial structure of the prediction model. By using the method in the invention, the prediction precision of the short-term traffic flow can be effectively increased, an influence of a noise on the prediction precision in training data is reduced and operation time is reasonable.

The invention belongs to the technical field of network security, in particular to a security situation intelligent prediction method, device and system based on a deep neural network, and the methodcomprises the steps: taking an automatic encoder as a basic unit, combining with an error back propagation BP neural network, and constructing a deep self-encoding network model for network security situation unsupervised training learning; sequentially carrying out unsupervised layer-by-layer pre-training and supervised model parameter fine adjustment by combining expert knowledge and the hierarchical evaluation deep self-encoding network model to obtain a trained network model; and predicting the target network security situation based on the trained network model. According to the invention, the deep auto-encoder is used as a basic structure, an unsupervised layer-by-layer algorithm is used for pre-training, and a supervised algorithm is used for parameter fine tuning, so that the problem of dependence on a network security data label is solved, automatic monitoring and intelligent early warning of a security situation are realized, and the accuracy and timeliness of situation prediction are improved.

The invention discloses a method for recovering a signal based on a BPNN (Back Propagation Neural Network), comprising the following steps: S1. acquiring insertion pilot frequency information of a signal transmitter and receiving pilot frequency information of a signal receiver in a unknown channel, and accordingly constructing a training sample set; S2. building a BPNN model consisting of an input layer, a hidden layer and an output layer; S3. successively inputting each group of sample information in the training sample set into the BPNN model to perform training, so as to obtain a well trained BPNN model; and S4. receiving a signal from the unknown channel and inputting the signal into the well trained BPNN model by the signal receiver, so as to recover an original signal transmitted bythe signal transmitter. Through adoption of the method of the invention, the original signal transmitted by the signal transmitter can be recovered according to the signal received from the unknown channel by the signal receiver, thereby avoiding signal distortion caused by the unknown channel, and improving accuracy and stability of signal transmission.

The present invention discloses a method and system of scheduling and linking virtual network functions. The method comprises the following steps of constructing a Markov initial model according to the network environment information and the network request information; carrying out the deep reinforcement learning training on the Markov initial model to obtain a Markov training model by the randomgeneration action and in a back-propagation neural network manner; real-timely obtaining the network environment information and the network request information, and obtaining the virtual network function placement nodes and the service chains of the network requests according to the obtained network environment information, the network request information and the Markov training model. Accordingto the present invention, by carrying out the deep reinforcement learning training on the Markov initial model to obtain the Markov training model via the random generation action and in the back-propagation neural network manner, the optimal deployment of the network functions and the service chains can be realized furthest, and the purposes of reducing the total delay of the network requests and improving the network resource utilization rate are achieved.

The invention discloses an image super-resolution reconstruction method based on a back-propagation neural network. The contents of the method mainly include image input, preprocessing, image denoising, image-border maintaining and the back-propagation neural network. The method includes the steps that first, a synthetic aperture radar image is preprocessed, multiplicative noise of the image is converted into additive noise, an improved non-local mean value is adopted to conduct the denoising, and an exponentiation operation is adopted to restore the image; then, a kernel function is adopted to maintain clear borders of a reduced image; finally, through the treatment of the back-propagation neural network, a super-resolution reconstruction result is obtained. According to the method, a neural network model is adopted, and a large number of computing resources and a large amount of calculation time are saved; according to the improved non-local mean value method, a low-resolution image is reconstructed into a high-resolution image, speckle noise of the synthetic aperture radar image is denoised, and the combination of the modified non-local mean value method and the back-propagation neural network greatly improves the reconstruction effect.

An embodiment of the invention discloses a defect classifying method for a data center monitoring system. The method includes: constructing a network input matrix according to monitoring states of monitoring items in monitoring resources, and according to the input matrix, classifying the defects of the monitoring resources through a BP (back-propagation) neural network which is trained in advance. The embodiment of the invention further discloses a defect classifying device for the data center monitoring system. According to the defect classifying and the defect classifying device, stability and safety in operation of equipment in the data center can be guaranteed, and normal operation of various businesses can be guaranteed.

The invention belongs to the technical field of human body characteristic parameter identification, and discloses a human body characteristic parameter extraction method and system based on image analysis, and the method comprises the steps: extracting a user contour, a key characteristic region and characteristic points through the photo image processing of a user, and obtaining a key part of user characteristic parameter information; scanning enough human body scanning data in a human body database to obtain a neural network model capable of reflecting human body characteristics; directly generating a neural network model according to the input parameter information of the key parts of the height, the chest circumference, the waistline and the hip circumference, and generating a human body characteristic curve matched with the real human body shape; starting from the photo information of the user, searching a similar human body three-dimensional model matched with the body of the user in a 3D scanning human body database; The back propagation neural network (BP neural network) is used for drawing a three-dimensional human characteristic curve, and human characteristic parameterscan be well extracted through the application of the method.

A DDoS attack detection system based on a back-propagation neural network algorithm comprises a data acquisition module, a data preprocessing module, a resource scheduling module, a rule base module,data analysis modules and a response module, wherein the data acquisition module is used to collect network traffics; the data preprocessing module is used to preprocess original network traffic data;the resource scheduling module is used to allocate the appropriate data analysis module for data analysis; the rule base module is used to filter the data with intrusion features; the data analysis modules are used to real-timely update rules of the rule base module; and the response module is used to detect intrusion behaviors and give responses and alarms. The DDoS attack detection system disclosed by the invention has the beneficial effects as follows: the accuracy of attack detection can be improved; the situation that a traditional detection method based on a feature detection algorithmcan only detect known attack modes and can helplessly process unknown attacks can be improved; and the shortcomings of slow convergence speed and low accuracy of a traditional back-propagation neuralnetwork algorithm can be improved.

The invention provides a wireless intelligent propagation method based on a Cost231-Hata model. The model is based on application of a 5G network and assistance of wireless LET network data, and provides a wireless propagation model which is established in different environments and under the influence of various external factors and is high in stability. According to the model, engineering parameters, map parameters and self-searched features in LET data are analyzed; the average signal received power (RSRP) of the model is predicted; detailed division of loss of a transmission path is considered; a more comprehensive transmission path loss is adopted and is composed of a free space loss Ls and a diffraction and scattering loss Lms of a roof to a street, a back propagation neural networkis constructed to predict RSRP at different geographic positions, and the coverage intensity of wireless signals in a new environment can be accurately predicted through the model, so that the networkconstruction cost is reduced, and meanwhile the network construction efficiency is improved.

The present invention discloses an identification method of a crank call. The method comprises the steps of: establishing a crank call identification model based on a back propagation neural network;designing input and output variables of the back propagation neural network; designing a particle swarm optimization to calculate an initial back propagation neural network parameter so as to improvethe precision and the convergence speed of the back propagation neural network; providing a method for establishing a training data set and a verification data set to improve the training efficiency;and calculating an identification effect assessment value of the back propagation neural network, and identifying whether a calling number is a crank call, a normal call or a suspected crank call or not by employing a back propagation neural network with the maximum identification effect assessment value. The identification method of a crank call is accurate and rapid, and good in anti-crank calleffect.

The invention discloses an immune feature recognition method based on a neural network. Variable region sequences (CDR3 sequences) of B cell receptors (BCR) or T cell receptors (TCR) of subjects are obtained according to high-throughput sequencing, and are compared with BCR or TCR variable region sequence (CDR3 sequence) set of a control group to obtain immune feature sequences, different from thesequences of the control group, of the individual subjects or a subject group; and immune feature models of the subjects and the control group are constructed by utilizing a feedforward back propagation (BP) neural network algorithm, so the immune features of samples can be identified at a molecular level.

The invention relates to a turboshaft engine steady-state model identification method based on PSO-NARXPSO-NARX-based turboshaft engine steady-state model identification method, and discloses a PSO-NARX network-based turboshaft engine steady-state model identification method. The invention discloses a PSO-NARX network-based turboshaft engine steady-state model identification method. The method ischaracterized in that a particle swarm optimization algorithm (PSO) is used for optimizing characteristic parameters of the NARX network, and a mean square error of model prediction output and targetoutput is used as a fitness function of particles, so that the optimization effect on the NARX network is improved. A PSO-NARX network is applied to identification of a steady-state model of a certaintype of turboshaft engine; c. Compared with a back propagation (BP) neural network and an NARX network, the method has the advantages that the precision of the steady-state model of the turboshaft engine identified by the PSO-NARX network is higher, the precision requirement of practical application can be met, and a better convergence effect is shown. According to the method, tThe problems thata turboshaft engine steady-state building model is complex, difficult and not high in precision can be well solved.

The invention discloses a single channel abdominal recording fetal electrocardiogram extraction method. The single channel abdominal recording fetal electrocardiogram extraction method includes the following steps that (1) a signal is collected in an abdomen of a mother body, the signal is preprocessed to remove baseline drift, and the denoised signal is taken as a network target signal; (2) an electrocardiogram component of the mother body is estimated through singular value decomposition (SVD) and a smooth window (SW), and the estimated electrocardiogram component of a chest of the mother body is taken as a network input signal; (3) then a back propagation (BP) neural network is constructed, a hidden layer is arranged as 15 neurons, the number of iterations is 500, the learning rate is arranged as 0.1, and the target error is 0.000001; and (4) the BP neural network method is used for training a network, a network template is obtained, and then a fetal electrocardiogram signal is extracted. According to the single channel abdominal recording fetal electrocardiogram extraction method, the clear fetal electrocardiogram signal can be obtained by collecting an abdomen leading mixed signal, the inconvenience of chest electrocardiogram collection to pregnant women is avoided, and certain value in practical application is achieved.

The invention discloses a water quality prediction method and device, electronic equipment and a computer readable storage medium. The water quality prediction method comprises the following steps: establishing a water quality prediction model based on a neural network model, wherein the neural network model is one or a combination of more of a long-short term memory neural network model (LSTM), a recurrent neural network model (RNN), a back propagation neural network model (BP) and a convolutional neural network model (CNN); training the water quality prediction model based on historical water quality monitoring target data and potential characteristic factor data; and predicting the water quality based on the water quality prediction model. According to the technical scheme provided by the invention, the influence of the potential characteristic factor data related to the water quality monitoring target data is fully considered, and the data is flexibly applied, so that the characteristics which can be provided by the data are reasonably found, and the simulation accuracy of the model is improved.

The invention discloses an expiratory air detection device and an establishment method of an expiratory air marker of the expiratory air detection device. Two modeling methods are provided, wherein the one modeling method comprises: respectively screening VOCs with significant differences in expiratory air by using three characteristic variable screening methods including unrelated variable elimination method (UVE), competitive adaptive reweighting algorithm (CARS) and continuous projection algorithm (SPA), and training training-set data by applying a machine learning algorithm back propagation neural network (BPNN) to establish an analysis model so as to obtain an expiratory air compound fingerprint spectrum, and the other modeling method comprises: taking all expiratory air components and concentrations as input, establishing an analysis model by adopting a machine learning algorithm to obtain an expiratory air compound fingerprint spectrum, and selecting the first ten important VOCsestablished by the analysis model by utilizing a feature classification method.

The invention discloses a time-varying channel estimation method and system based on deep learning. A network input sample is reasonably constructed; the method is based on a single hidden layer neural network, and comprises the following steps: firstly, fully utilizing channel change characteristics in historical channel information and other characteristics in a received pilot signal; and further improving the performance of channel estimation by using the advantages of least square estimation; secondly, carrying out offline training on a back propagation neural network by using the constructed sample, and then obtaining time-varying channel information in real time in an online mode. In order to reduce the calculation complexity, only the received pilot signals and the information of the pilot sub-channels are adopted, and the pilot sub-channels are modeled by adopting a polynomial basis expansion model to reduce to-be-estimated parameters so as to carry out time-varying channel estimation. According to the method, the channel estimation precision can be remarkably improved, the calculation complexity is low, and the method is suitable for efficient acquisition of time-varying channel information in a high-speed moving scene.

The invention belongs to the technical field of automobiles, and particularly relates to a driving style identification algorithm based on factor analysis and machine learning. According to the driving style identification algorithm, firstly, data strongly related to a driving style are selected as driving style characteristic parameters, dimensionality reduction is conducted on the characteristicparameters through factor analysis to obtain public factors, redundancy between driving data is reduced, and corresponding physical significance is given to the public factors; the public factors aretaken as an input, and labels of corresponding driving styles are marked for different drivers by adopting a Gaussian mixture model clustering algorithm; and a driving style identification model is trained by using a back propagation neural network optimized by a genetic algorithm. By fusing unsupervised learning and supervised learning, the identification cost can be effectively reduced. The genetic algorithm is used for optimizing the initial weight of the back propagation neural network, so that the identification precision of the model can be effectively improved, and the blank that the driving style cannot be identified statically is filled.

The invention provides a training method of a water quality parameter inversion model, and a water quality monitoring method and device. The training method comprises the following steps: acquiring water quality parameters and hyperspectral data in a target time period; performing waveband processing on the hyperspectral data to obtain multi-waveband spectral data; conducting correlation analysison the water quality parameters and the multiband spectral data, and taking the multiband spectral data corresponding to the multiple bands with the maximum correlation as target spectral data; and based on a training set constructed by the water quality parameters and the target spectral data, training a pre-constructed back propagation neural network until the back propagation neural network reaches convergence, thereby obtaining a water quality parameter inversion model. The method breaks through the limitation that an existing method is influenced by the time-phase characteristics of remote sensing images and models are difficult to unify, and reduces the analysis difficulty between the water quality parameters and the hyperspectral data.

The invention provides a fused partial discharge type identification method based on a DS evidence theory. The invention relates to the fields of power systems, deep learning technologies, image processing technologies and the like. According to the method, firstly, a convolutional neural network is utilized to input partial discharge PRPD map image features for recognition to obtain a recognition rate, then statistical features of PD signals are extracted and input into an SVM classifier to obtain classification probabilities, and finally, a DS evidence theory is utilized to perform fusion judgment of partial discharge types on the two probabilities. Compared with a traditional support vector machine (SVM) and a back propagation neural network (BPNN) algorithm, the method provided by the invention is advantaged in that the correct recognition rate is remarkably improved, the effect of improving the recognition rate of two defects with relatively high similarity is particularly obvious, and robustness is relatively good.

The invention discloses an island microgrid hierarchical control method considering communication data disturbance under CPS concept. The method is based on the CPS concept, and divides a hierarchicalcontrol structure into two layers, that is, a network layer and a physical layer, by utilizing a communication network in the micro grid. In the network layer, influence of a CDD on system control effect is analyzed, and a strategy of adopting a back-propagation neural network to finish data compensation and eliminate the influence of the CDD is proposed; and in the physical layer, P-omega / Q-U droop control is adopted as primary control of voltage and frequency of the microgrid, and meanwhile, communication data in the network layer and a consistency protocol with virtual leaders are utilizedto finish secondary control of DER output voltage and angular frequency, thereby ensuring that the voltage and angular frequency of the microgrid is reliably controlled.

The invention discloses an integrated detection method of an edge side cloned node based on a counterpropagation neural network. The method comprises the following steps: S1. collecting, by an edge side computing node, a data set; S2. training and testing a BPNN by using the data set; S3. waiting for a new information packet; S4. extracting channel information from the information packet, storingreference channel information, and calculating a channel difference value; S5. accumulating the credibility of each node; S6. inputting the credibility of each node into the BPNN to judge whether a clone attack exists; and S7. if no clone attack exists, updating the reference channel information, and if the clone attack exists, sending a clone attack alarm. According to the integrated detection method disclosed by the invention, integrated detection is performed on a data source node on the edge side node by using the BPNN, meanwhile whether the clone attack exists in multiple nodes, thus improving the detection efficiency of the cloned node and reducing the network transmission load and the load of a central network. According to the integrated detection method disclosed by the invention,the influence of random noise of a channel is also reduced by accumulating the credibility and classifying the BPNN, and the accuracy of the detection of the cloned node is improved.

The invention provides a micro-service state detection method, a model training method, equipment and a storage medium, and the method comprises the steps: obtaining a plurality of index parameters of a micro-service for each micro-service under a micro-service architecture, inputting the plurality of index parameters to a micro-service state detection model, obtaining a detection result used for indicating whether the micro-service is abnormal, wherein the micro-service state detection model is obtained by training a back-propagation neural network model. The micro-service state detection model of the embodiment is used for predicting the health state of the current micro-service, and the prediction function of the health state of the micro-service is realized by analyzing multiple index parameters of the current micro-service, so that abnormal micro-services are isolated in advance, the stability and continuity of the whole application or service are ensured, and the service quality of the system is effectively improved.

The embodiment of the invention discloses a shale brittleness index prediction method based on logging data. The shale brittleness index prediction method comprises the steps of: collecting a rock core, extracting an initial logging curve of a collection well corresponding to the rock core, performing a triaxial compression test on the collected rock core to obtain a stress-strain curve, and calculating the brittleness index measured through the test; identifying and selecting an effective logging curve from the initial logging curve by adopting linear regression and sensitivity analysis, carrying out standardization processing on the effective logging curve, and then establishing a prediction model through adoption of principal component analysis and a back propagation neural network method; and performing prediction performance evaluation on the prediction model, and training and correcting the prediction model again based on an evaluation result. According to the method, the conventional logging data and the brittleness index measured in a laboratory are utilized, and the model is established by using the principal component analysis and the back propagation neural network to predict the brittleness index, so that the purposes of reducing the calculation amount and improving the prediction accuracy are achieved.

The invention discloses an ozone concentration estimation method fusing satellite remote sensing and ground monitoring data, and belongs to the technical field of environment monitoring. The method comprises the steps of 1, collecting, preprocessing and fusing multi-source sample data to obtain input parameters; 2, based on a multilayer mapping back propagation neural network, establishing an ozone concentration estimation operation basic model; and 3, based on the influence factors, the forward time and the spatial range, searching an optimal input parameter combination of the obtained ozone concentration estimation operation basic model, and accurately estimating the ground ozone concentration according to the obtained optimal input parameter combination to obtain the spatial continuous distribution condition of the ozone concentration, and realize the ozone concentration estimation method fusing satellite remote sensing and ground monitoring data. The method has the advantages of high accuracy, high reliability and simplicity in operation, the used multi-source sample data are free and open-source, the universality is enhanced, the ozone concentration can be quickly estimated, and a continuous distribution diagram of the ozone concentration in a target area can be drawn.

The invention provides a fusion positioning method. The method is characterized in that the information acquired by a wireless radio frequency technology and a vehicle-mounted sensor is used as the observation information, a longitudinal motion model and a transverse motion model of the vehicle are respectively established as state equations, extended Kalman filtering recursion is carried out on the two models respectively, the real-time vehicle motion state information is used as input, a BP (Back Propagation) neural network model is established to determine the model probabilities of the twomodels, and multi-model estimation combination is realized according to the model probability, and an intelligent multi-model fusion method is formed. The method is advantaged in that real-time switching between the longitudinal motion model and the transverse motion model is achieved according to a motion state of a vehicle, that the state equation can accord with the actual motion state of thevehicle is guaranteed, and accurate and reliable positioning of the running vehicle in the tunnel and other shielded environments is achieved.

The embodiment of the invention discloses a method and device for establishing a net present value prediction model,, a storage medium and electronic equipment, and the method comprises the steps: carrying out the training of a back propagation neural network model based on a dispersed fault block oilfield development database, generating an initial net present value prediction model, and determining the network parameters and MAPE of the initial net present value prediction model; repeating the operation of training the back propagation neural network model based on the dispersed fault block oilfield development database for N times to obtain corresponding network parameters and MAPE corresponding to N initial net present value prediction models; and optimizing the N groups of network parameters and the MAPE based on an empire competition algorithm until a group of optimal network parameters remain, and returning the optimal network parameters to the back propagation neural network model to generate a target net present value prediction model. Through the scheme, the prediction precision and efficiency of the net present value of the dispersed fault block oil field group are improved, and objective and accurate evaluation of the dispersed fault block oil field group is realized.

The invention relates to the technical field of signal processing, in particular to an atmospheric delay estimation method, device and system based on a back propagation neural network and a storage medium, and the method comprises the steps: building the neural network through employing the related information of known atmospheric delay value pixel points of a to-be-observed region as training data and employing an error back propagation principle, training the neural network by using training data, inputting related information of unknown atmospheric delay value pixel points of the to-be-observed area into a trained network, and continuously solving new unknown atmospheric delay value pixel points on the basis of updating known information until atmospheric delay values of all pixel points are solved. The method can be applied to computer end software and is matched with corresponding hardware equipment. According to the method, the problem of low estimation precision caused by the fact that an atmospheric delay value model extracted in the prior art is not representative can be effectively solved, the influence of other errors of data on an estimation result in the prior art isovercome, and the atmospheric delay value estimation precision is improved.