57 results about "Time series modeling" patented technology

The invention discloses a method for predicting service life of a product by an accelerated degradation testing based on degradation distribution non-stationary time series analysis. The method comprises the following steps of: 1, acquisition and preprocessing of test data; 2, parameter estimation of degradation distribution; 3, time series modeling of degradation distribution parameters; 4, accelerated degradation modeling based on the degradation distribution; and 5, service life prediction based on degradation distribution. The method can macroscopically describe degradation statistical regulation of all samples under accelerated stress, overall analyze the time series of the degradation distribution in an accelerated degradation process, can extrapolate the degradation distribution under the accelerated stress to normal stress to obtain product reliability reflecting the product accelerated degradation random process volatility regulation and service life relationship prediction, improve the credibility on estimation result of service life prediction and reliability, and save more time and is more efficient compared with performance degradation prediction under normal stress level.

The invention discloses a power system abnormal data identifying and correcting method based on time series analysis. The power system abnormal data identifying and correcting method includes data preprocessing, time series modeling, abnormal data identifying and abnormal data correcting. Data preprocessing includes the step of identifying and correcting missing data in data to be detected and data suddenly changing to be zero. Time series modeling comprises the steps of conducting time series analyzing on the preprocessed data to be detected and establishing a model according to the time series, and a difference autoregression moving average model is used for modeling the data to be detected. According to abnormal data identifying, the fitting residual series of the established difference autoregression moving average model is analyzed, an error confidence interval is set, and abnormal data are identified. According to abnormal data correcting, a neural network method is used for establishing a prediction model for correcting the abnormal data, the data value of the moment when the abnormal data exist is predicted, and the abnormal data are corrected. The power system abnormal data identifying and correcting method is easy to implement and high in accuracy.

The invention discloses a product order prediction method and device with time series characteristics. The product order prediction method comprises obtaining statistics of order data of enterprises at every time point according to stored historical order data; selecting a prediction model according to the time series characteristics of the order data and determining a prediction output equation of the prediction model; enabling the statistics of the historical order data to be processed as a prediction input table according to the requirements of the prediction model and training a corresponding prediction network model; and utilizing the prediction input table of the prediction order quantity to calculate to obtain the prediction order quantity of orders according to the prediction model which is well-trained through prediction orders and the prediction output equation. The invention also provides the product order prediction device according to the product order prediction method. The product order prediction device mainly comprises a data acquisition module, a data preprocessing module, a time series modeling module and an order prediction module. The product order prediction method and device with the time series characteristics have the advantages of solving the nonlinear problem of product order prediction, meeting the requirements of system real-time performance and improving the prediction accuracy.

The invention discloses a method for optimizing the new energy capacity ratio in layers in a power grid. The method comprises the steps that iteration calculation is conducted on the inner layer and the outer layer, a calculation model is that time series modeling is conducted on output force of the inner layer on the basis that the characteristics of new energy of a region are considered, the optimal energy-saving emission reduction benefit of the power grid serves as a goal, the factors such as the load characteristic, the unit peak shaving characteristic and the thermoelectricity coupling characteristics of heat supply units of different types are comprehensively considered, and therefore an annual time series production analog simulation model related to new energy power generation is established; the outer layer is provided with a capacity ratio optimizing model, the energy-saving emission reduction benefit of the model of the inner layer serves as a fitness function, so that the individual optimizing direction is updated, the new energy power generation ratio capacity is determined, the blindness of random generation of new energy installed capacity is reduced, and the optimizing efficiency and the accuracy are improved. The method for optimizing the new energy capacity ratio in layers in the power grid can be applied to new energy capacity optimization of the province-level power grid and has important guiding significance in grid source planning and practical power system dispatch of the province (region) power grid with the requirements for new energy installed capacity planning and low carbon electric power.

The invention relates to a video action classification method based on CNN-LSTM (Convolutional Neural Network-Long Short Term Memory) and action., and belongs to the field of computer vision. The method comprises the following steps: S1, exporting multi-layer depth features from a pre-trained convolutional neural network CNN to represent video actions, capturing context relationship information between different video frames by using Conv-LSTM and FC-LSTM, and performing time sequence modeling on the video actions; s2, through TAM and JSTAM, time significance and space-time significance in action representation are enhanced; s3, two attention models are adopted to obtain video action global representation containing key information, and a PCA dimension reduction algorithm is utilized to perform dimension reduction and decorrelation on a high-dimensional action representation vector; and S4, allocating different weights to the outputs of the two independent networks of the time attention network TAN and the joint time attention network JSTAN, and integrating the plurality of representation vectors into a final classification vector.

The invention discloses a lake and reservoir algal bloom predicating method based on multielement nonstationary time series analysis and neural network and support vector machine compensation, and belongs to the technical field of water quality monitoring. The method comprises the steps of characteristic factor nonstationary time series modeling, error influence factor kernel principal component analysis, neural network error modeling according to the situation of large sample data, support vector machine error modeling according to the situation of small sample data, final error compensation and predicating result obtaining. The problems that existing algal bloom predication precision is not high, and predication is hard to carry out according to the small sample data are solved, the description of the algal bloom forming process corresponds to reality better, and the result of algal bloom modeling predication is more accurate. The advantage compensation of a time series analysis method suitable for linear system modeling and a statistical learning method suitable for nonlinear system modeling is achieved, and the algal bloom predication accuracy is improved.

The invention discloses a residual network and attention mechanism-based drug relationship extraction method. The method comprises the following steps of: S1, carrying out vector representation on words in a drug entity relationship data set; S2, carrying out time-series modeling on a drug relationship statement by utilizing a two-layer bidirectional long-short term memory model neural network; S3, importing residual connection into the constructed two-layer bidirectional long-short term memory model neural network; S4, decomposing a deep semantic meaning automatically obtained by the two-layer bidirectional long-short term memory model neural network into a memory space and an attention space, fusing memory information and attention information, and inputting the fused information into aSoftmax classifier to extract a drug relationship. According to the drug relationship extraction method disclosed by the invention, dependency relationships between long-distance words are effectivelysolved, gradient diffusion is overcome, model overfitting is prevented, the model robustness is good and the classification effect is good.

The invention provides a wind power output time series modeling method based on fluctuation characteristics. The method includes the following steps that historical wind power output data are collected and systemized, and a wind fluctuation curve variation trend is quantitatively described; multi-dimensional joint probability distribution of statistical parameters of all kinds of wind fluctuations is counted according to natural months, and transition probabilities of all the kinds of wind fluctuations are calculated; random sampling is carried out by means of the multi-dimensional joint probability distribution and the transition probabilities according to natural months, output values of output data points of the wind fluctuations are figured out, and a simulated wind power output time series is obtained. The historical wind power output data are directly applied to the method, statistical analysis and a random sampling theory are adopted, the random fluctuation characteristics of wind power output are simulated, and future wind power output scenes corresponding to reality can be constructed.

The invention discloses a gas path fault diagnosis method and system for aeroengine dynamic process. The method includes the steps of establishing a feature extraction model based on a multi-layer kernel extreme learning machine; adopting a hidden Markov model based on time series modeling for fault identification. The method solves the problem that the conventional data-based engine fault diagnosis uses time series measurement data to diagnose the fault with low accuracy in the existing aeroengine gas path fault diagnosis in the dynamic process, is suitable for the engine dynamic fault diagnosis in consideration of the degradation of gas path components and the redundancy of sensor parameters, and has a positive promotion effect on engine health management and maintenance cost reduction.

The invention discloses a lithium ion battery residual life prediction method based on a time convolutional neural network, and belongs to the field of lithium ion battery fault prediction and health management. The battery equipment degradation process is a highly nonlinear and complex multi-dimensional system and is high in time-varying property, the prediction process of an existing algorithm needs expert knowledge and priori knowledge, time and labor are wasted, and the prediction process is difficult and low in precision. According to the method, a hidden model in a time sequence is mined through the powerful time sequence modeling capability of the neural network, and the nonlinear mapping relation between measured parameters and the service life is automatically established. Due to a parallelism mechanism of convolution calculation, accelerated training can be carried out by using graphic calculation, and the calculation is faster. The invention provides the calculation method of a parameter filter, when invalid parameters and redundant parameters are too many, a part of parameters can be automatically screened, the prediction workload is reduced, and the training efficiency is improved.

The invention provides an analysis method of an output characteristic of a wind power plant. The method comprises the following steps of 1, collecting, analyzing and arranging historical wind power output data of different wind power plant groups, 2, normalizing the analyzed and arranged historical wind power output data, and 3, inspecting the historical wind power output data by a visual method. The method can serve as an early foundation of medium and long term wind power output time series modeling, and can also be applied to planning and operation analysis of a power system containing wind power.

The invention discloses a discrete wavelet multiscale analysis based random error compensation method for an MEMS (Micro Electro Mechanical system) gyroscope, and the method is used for improving the data measurement accuracy of the MEMS gyroscope. The method comprises the following steps: decomposing a signal of the MEMS gyroscope by using a binary orthogonal discrete wavelet Mallat algorithm step by step, wherein the decomposing scale comprises 3 levels and a decomposed approaching signal and a decomposed detail signal are subjected to time series modeling and kalman filtering after the first level is decomposed; performing the second-level decomposing on the filtered approaching signal and then gradually decomposing and filtering in the same manner. The filtered approaching signal at the final layer and the various detail signals are subjected to signal reconstruction. According to the discrete wavelet multiscale analysis based random error compensation method for the MEMS gyroscope, the random error compensation effect of the MEMS gyroscope is improved, and the method has an important role for a vehicle-mounted or ship-based measurement occasion requiring high accuracy, high efficiency and high stability.

The invention discloses a behavior recognition method based on a space-time attention enhancement feature fusion network, and belongs to the field of machine vision. According to the method, a networkarchitecture based on an appearance flow and motion flow double-flow network is adopted, and is called as a space-time attention enhancement feature fusion network. Aiming at a traditional double-flow network, simple feature or score fusion is adopted for different branches, an attention-enhanced multi-layer feature fusion flow is constructed to serve as a third branch to supplement a double-flowstructure. Meanwhile, aiming at the problem that the traditional deep network neglects modeling of the channel characteristics and cannot fully utilize the mutual relation between the channels, the channel attention modules of different levels are introduced to establish the mutual relation between the channels to enhance the expression capability of the channel characteristics. In addition, thetime sequence information plays an important role in segmentation fusion, and the representativeness of important time sequence features is enhanced by performing time sequence modeling on the frame sequence. Finally, the classification scores of different branches are subjected to weighted fusion.

The invention relates to a power distribution network optimization scheduling method and system considering dynamic reconstruction of a network frame. Firstly, carrying out active power distribution network element time sequence modeling is carried out; then, establishing an active double-layer optimization scheduling model considering network dynamic reconstruction with the economy of power distribution network dispatching and the stability of fast voltage as objectives; and performing decimal encoding on branches in each basic loop, setting an iterative condition of a network frame population, adding a radial constraint condition of a power distribution network frame into an iterative strategy, and solving the active double-layer optimization scheduling model. The method can effectivelyreduce the operation cost of the power distribution network, smooth the voltage level of the power distribution system and improve the system voltage stability.

The invention provides a wind power output time series modeling method and system. The method comprises the steps of obtaining historical wind power output data of a target wind power plant and an adjacent wind power plant; according to the historical wind power output data, calculating correlation coefficients of the target wind power plant and the adjacent wind power plant to obtain a correlation coefficient set; and according to the correlation coefficient set, generating a simulated wind power output time series. According to the technical scheme provided by the method and the system, a sample wind power plant required for modeling is determined by calculating the correlation coefficients of the target wind power plant and the adjacent wind power plant; and the time series and fluctuation of wind power output are considered, and the spatial correlation of the wind power output is reflected, so that the wind power prediction precision is improved.

The invention discloses an MEMS gyro random drift error processing method which is characterized in that based on an existing MEMS gyro and at a signal processing back-end, measured data is analyzed by an Allan variance method, various main random error sources are effectively separated and magnitude of each error coefficient is determined; time-series modeling is conducted on the basis of checking and preprocessing the data; and finally, interference noise is effectively inhibited by multiple times of Kalman filtering processing so as to raise precision of the gyro.

The invention provides a video action detection method based on time sequence convolution modeling. The method comprises the steps of firstly, adopting an action proposal generation technology to generate proposal fragments containing complete actions as much as possible; secondly, screening out a complete proposal with high overlapping degree by using non-maximum suppression in cooperation with weighted IoU, and then performing sparse sampling on the proposal to obtain a specified number of video frames; extracting space-time features of the video frames by adopting a deep network model; dividing the obtained frame-level spatial-temporal features into three stages according to an evolution mode, and performing sequential modeling on the features of each stage by using sequential convolution; and finally, predicting an action category and detecting an action occurrence time interval by using a classifier and a regression device. By applying the method, the incompleteness of the proposal can be overcome, and the time sequence information of the video stream is reserved to the maximum extent, so that the action in the video can be detected more accurately.

The invention discloses an examination room cheating behavior analysis method based on motion feature enhancement and long time sequence modeling, and belongs to the video behavior recognition field and the deep learning field. The method comprises the steps: collecting a data set, carrying out the behavior type marking of the data, extracting a video stream as an image frame, improving the capturing capability of a model on a moving target based on a motion feature enhancement method, carrying out information fusion between frames through a feature spectrum shift mode, carrying out modeling on a long-time sequential relationship based on a sequential pyramid method, and completing the construction of an identification model; secondly, initializing a behavior recognition classification model by adopting a Xavier method according to an image obtained by the data set, obtaining a sampling sequence of video frames by adopting a segmented extraction mode, carrying out iteration to a preset number of iterations based on a loss function of the classification model, and completing training of the model; and finally, using a video frame sequence obtained through sampling for reasoning testing, and obtaining a specific behavior category result.

The invention provides a video-based behavior recognition method, a video-based behavior recognition device, terminal equipment and a computer readable storage medium. The behavior recognition methodcomprises the steps of extracting a feature of a first full connection layer and a feature of a last convolution layer from an RGB frame of a video through a feature extractor; inputting the feature of the first full connection layer into a full connection long short-term memory network LSTM to carry out time sequence modeling, and inputting the feature of the last convolution layer into a convolution LSTM to carry out time sequence modeling; inputting the features subjected to time sequence modeling through the full-connection LSTM and the features subjected to time sequence modeling throughthe convolution LSTM into a joint optimization layer; and performing behavior identification on the RGB frame of the video through the joint optimization layer. The method can solve the problems thatan existing behavior recognition method is complex in background information, not high enough in time sequence modeling capacity and the like.

The embodiment of the invention discloses an anomaly detection method and device for transformer monitoring data. The method comprises the following steps: acquiring a to-be-detected sequence of on-line monitoring data of a transformer; constructing an abnormal data identification model by adopting time sequence modeling and an isolated forest algorithm; constructing an abnormal type recognition mode based on an improved multi-dimensional SAX vector representation method; adopting the abnormal data identification model to identify abnormal data of the to-be-detected sequence; determining the abnormal type of the abnormal data by adopting the abnormal type identification mode, wherein the abnormal type comprises an invalid abnormal mode and an effective abnormal mode; and when the exception type is the invalid exception mode, carrying out relevance verification on the exception type by adopting sequence relevance analysis. According to the scheme, on the basis of effectively recognizing abnormal data information, abnormal modes can be deeply analyzed, and effective abnormal points and invalid abnormal points can be accurately distinguished.

The invention relates to the field of network security, in particular to a malicious attack detection method, system and device based on cloud WAF and a medium. The malicious attack detection method based on the cloud WAF comprises the steps: after an access request for a service line source station is received, performing parameter extraction on the access request; judging whether the access request is abnormal or not based on an HMM model; if the access request is abnormal, judging whether the access request is an attack based on an SVM classifier; if so, intercepting the access request. According to the invention, the HMM model is suitable for processing continuous dynamic signals, and the excellent time sequence modeling capability of the HMM model is organically combined with the extremely strong dichotomy capability of the SVM classifier under the condition of small samples; a re-training mechanism is added, so that model attenuation is effectively slowed down, the robustness isgood, the behavior of bypassing detection can be effectively prevented, the high accuracy is ensured, and the method has the advantages of lower missing report rate and false alarm rate and cost saving.

The present invention provides a method for dividing photovoltaic output fluctuation types, comprising: I, collecting and processing historical data; II, determining feature vectors according to feature indexes reflecting photovoltaic output features; III, performing clustering analysis by using a self-organizing feature map neural network; and IV, dividing photovoltaic output fluctuation types. The method of the invention can serve as an early basis to be applied in medium and long term output time series modeling of photovoltaic power generation, and provides early technical means for generation of photovoltaic simulation data required by time series production simulation containing large-scale new energy sources, and annual new energy source consumption ability analysis.

The invention discloses a method for improving the overall classification accuracy of long-tail distribution speech based on transfer learning, and the method comprises the steps: firstly building an R-CNN model composed of a CNN and an RNN network through the training of a data set presenting long-tail distribution, enabling the CNN network to be used for extracting speech features, and enabling the RNN network to carry out the time sequence modeling of the speech features extracted by the CNN network; further mining speech information, and extracting inter-class separable features for subsequent speech classification; then, training the R-CNN model twice, in the first model training, using data of long tail distribution for model training, and obtaining preliminary model parameters; in the secondary model training, using the data in balanced distribution for model training, and fixing and migrating the CNN network shallow parameters obtained in the primary model training to the secondary model training; and carrying out speech classification prediction by using the model after secondary training, thereby improving the overall classification effect of the speech classification model.

The invention provides a PMF-based microblog user interest prediction method, which comprises the following steps: S1) obtaining original data of microblog user posting behavior, social circle information and social relationship; S2) carrying out automatic text marking on the original data of users, microblog user posting behavior and social circle information, establishing a user interest theme matrix and carrying out social relationship mining on the original data of the social relationship to obtain a social trust relationship matrix between users; S3) carrying out corresponding time-series modeling on the data obtained in the S2) to form a user posting behavior time-series model and a social circle information time-series model; and S4) substituting the user posting behavior time-series model, the social circle information time-series model and the social trust relationship matrix obtained in the S2) and the S3) into an SC-PMF prediction model to obtain a microblog user interest prediction result. The method integrates factors of user history behavior, user social trust relationship and user social circle blog information and the like, and solves the problem of cold start.

A computing device for time series modeling and forecasting includes a processor, and a memory coupled to the processor. The memory stores instructions to cause the processor to perform acts including encoding an input of a multivariate time series data, and performing a non-linear mapping of the encoded multivariate time series data to a lower-dimensional latent space. The next values in time of the encoded multivariate time series data in the lower dimensional latent space are predicted. The predicted next values and a random noise are mapped back to an input space to provide a predictive distribution sample for a next time points of the multivariate time series data. One or more time series forecasts based on the predictive distribution sample are output.

The present invention discloses a low-voltage early warning method and device based on a time series. The method comprises a step of acquiring a plurality of voltage monitoring data at a voltage monitoring point of a distribution network terminal, a step of carrying out differential operation on the plurality of voltage monitoring data to obtain a stable voltage monitoring sequence, a step of carrying out time series modeling on the stable voltage monitoring sequence of the voltage monitoring point to obtain a time series model, and a step of carrying out low-voltage early warning according tothe time series model. According to the low-voltage early warning method and device based on a time series, the low-voltage early warning is effectively realized.

The invention discloses an automatic lie detection method and system based on domain adversarial training, and the method comprises the steps: S1, multi-modal feature extraction: extracting text feature representation, audio feature representation and facial feature representation; S2, performing multi-modal feature fusion, and obtaining multi-modal feature representation by using an adaptive attention mechanism; S3, performing time sequence modeling, and capturing context information in the dialogue by using a bidirectional recurrent neural network to assist lie detection of the current sentence; S4, performing domain adversarial training, extracting lie feature representation irrelevant to speakers by using a domain adversarial network, and reducing the influence of speaker difference onautomatic lie detection performance; S5, predicting a lie level, and inputting to-be-tested data into the lie classifier subjected to domain adversarial training for predicting the lie level of the individual. The system comprises a multi-modal feature extraction module, a multi-modal feature fusion module, a time sequence modeling module, a domain adversarial training module and a lie level prediction module which are sequentially connected from top to bottom.