94 results about "Decomposition of time series" patented technology

A signal processing system and method for breaking a time series into piece-wise discrete trends and determining whether new data represents the continuation of a trend. The method identifies and utilizes at least one set of trend determination parameters, which have favorable trend fit characteristics relative to other possible sets of parameters. In a semi-automated embodiment, the error and trend length characteristics are cross-plotted for multiple sets of possible parameters, and one or more of the parameter sets is selected from the graph. In an automated embodiment, an objective function is formulated from the characteristics, and an optimization technique is applied to identify one or more good parameter sets.

The invention discloses a time sequence analysis based optical transmission network trend prediction method, which specifically comprises the steps that 1) network management performance parameters are screened, and network management performance parameters capable of reflecting the network running state are selected; 2) network management performance data is acquired, and specified network management performance data is acquired through a northbound interface of the optical transmission network for network management; 3) a time sequence is formed, the network management performance data is acquired in an uninterrupted manner in a sampling period, and the network management performance data is arranged into a time sequence of a certain performance parameter feature value according to a certain time interval; 4) the time sequence is decomposed, and a trend term, a periodic term and a stochastic term in the time sequence are decomposed through analyzing a time sequence sample; 5) predicted values of the decomposed terms are calculated, and the predicted values are estimated according to respective prediction models in allusion to the three different types of decomposed terms; and 6) a final predicted value is calculated, and the final predicted value is calculated according to a time sequence addition model and performs cross validation with an actual value.

The invention relates to an ultrashort-term slide prediction method for wind power. An atomic sparse decomposition method with quite high non-stable signal tracking and prediction capacity is used as a front decomposition method of a neural network. A wind power time sequence is decomposed into an atomic component and a residual error component, the atomic component is automatically predicted, the residual error component is predicted by the neural network, atomic decomposition results are updated by adding the latest wind power real-time data, and further the wind power of a next moment is slidably predicted. Actual wind field data prove that the model can effectively avoid non-stability of the wind power, sparser decomposition effects are achieved, and statistical intervals of absolute average error and root mean square error computation values can be remarkably reduced. Therefore, the ultrashort-term slide prediction method has the advantages that non-stability of the wind power can be effectively avoided, the sparser decomposition effects are achieved, and the statistical intervals of the absolute average error and root mean square error computation values can be remarkably reduced.

The invention discloses an intelligent power grid false data injection attack detection method which combines a variational mode decomposition technology and a machine learning technology. The detection method is added behind a power system state estimation part, and comprises the following steps: firstly, decomposing a system state time sequence into a set of a plurality of sub-sequences with different center frequencies by using a variational mode decomposition technology; secondly, in order to compress redundant data to facilitate model training, expressing data segment characteristics by using characteristic quantities based on statistical indexes; and finally, constructing a false data injection attack detector under the smart power grid by taking the calculated feature set as a training set of an online sequential extreme learning machine. In the process, the credible authenticated data can be used as an extended training set to continuously optimize the performance of the detector with the online learning capability. The invention aims to establish an efficient and accurate false data injection attack detection method for the smart power grid to ensure safe control and stable operation of a power system.

The present invention relates to a system (100) and method (800) capable of indirectly monitoring respiratory and cardiac variables. A decomposition technique on time-series of features extracted from the chest audio signal α(t) is proposed. The proposed monitoring system (100) may acquire the acoustic signal on the chest of a subject (140) by means of a wearable transducer (150). The proposed system may estimate a number of physiological variables such as flow estimates, respiration rate, inspiration and expiration markers and cardiac related variables. Cough and apnea detection, adventitious sound recognition, activities performed, and information about energy estimation and the status of a monitored subject can be derived as well.

The invention discloses a landslide displacement dynamic prediction method based on multiple influence factors. The landslide displacement dynamic prediction method is specifically implemented according to the following steps of 1, decomposing a landslide cumulative deformation displacement time curve according to a time sequence addition model; 2, extracting trend term displacement from landslidecumulative deformation displacement time curve decomposition by adopting a moving average method; 3, predicting trend term displacement by adopting a cubic polynomial; 4, selecting a main influence factor from the predicted trend term displacement by adopting a grey correlation degree sieve and taking the main influence factor as an initial input vector of a deep learning LSTM neural network model to predict the landslide periodic term displacement; and 5, according to a time sequence decomposition principle, superposing the predicted values of the displacement sub-sequences to obtain a finalpredicted value of the displacement, thereby finishing the dynamic prediction of the landslide displacement, and solving the problem that information at a long ago moment cannot be reserved in the prior art.

The present invention discloses a GNSS position time series periodic characteristic mining method, and provides an accurate GNSS position time series periodic signal analysis and characteristic analysis method by considering the limitation of a conventional harmonic function, and introducing the methods, such as the wavelet analysis and FAMOUS time-frequency analysis methods, a maximum likelihood estimation method, etc. The method is characterized by decomposing an original time series into new time series of different frequency bands according to the different inherent frequencies of the different signals in the original series, and decomposing and reconstructing the signals, processing each layer, thereby obtaining a needed part. The method of the present invention overcomes the limitation of a conventional GNSS time series model, facilitates reflecting the periodical change of a coordinate series really, enables the precision and the reliability of the GNSS topocentric coordinates to be improved further, obtains the high-precision position and speed parameters, and provides the bases for knowing the influence mechanisms and the change rules of the relevant geophysical phenomena deeply.

The invention discloses a financial time series combination prediction method based on adaptive noise complete ensemble empirical mode decomposition (CEEMDAN). According to the method, firstly, a financial time sequence is decomposed into eigenmode function (IMF) sub-sequences and remainders through CEEMDAN; long-term short-term memory (LSTM) is combined with support vector regression (SVR) to construct prediction models which are respectively used for predicting IMF sub-sequences and remainders; and finally, each prediction sequence is reconstructed to obtain a prediction result. Experimentalresults show that the method is suitable for industrial production, single LSTM model, sVR model, compared with other mixed models, the prediction model shows better performance in short-term prediction of the financial time series, and the problems that an existing financial time series prediction method is tedious in process, difficult to directly predict nonlinear and non-stable financial dataand low in prediction accuracy are solved.

The invention discloses a landslide displacement prediction method based on the PSO-SVR and DES combination. The method comprises the steps of acquiring a displacement monitoring time sequence and a plurality of initial influence factors of a displacement monitoring point at a landslide, removing random noise of the displacement monitoring time sequence by adopting a wavelet denoising method, anddecomposing the displacement monitoring time sequence into two components, namely a periodic term and a trend term, by using an HP filter. On the basis, the invention further discloses a preparation method, the water level and the rainfall are used as influence factors. The main influence factor characteristics of landslide displacement are extracted by adopting a principal component analysis method, a hybrid prediction optimization model combining a PSO-SVR (particle swarm optimization support vector regression) machine and a DES (double exponential smoothing) machine is established. The landslide displacement prediction is realized by constructing periodic term and trend term training sample components respectively. Finally, the trend term displacement prediction value and the season term displacement prediction value are superposed to obtain a landslide total displacement prediction value.

A method for discover network failure, an electronic device and a computer-readable storage medium, wherein, that method for discovering network failure comprise the following steps: generating a failure data table according to data recording XDR signaling; extracting periodic fluctuation data according to the failure data table, wherein the periodic fluctuation data is used as a failure signalingdata source of a time series decomposition model; regularizing the periodic fluctuation data according to the data requirements of the time series decomposition model; performing time series decomposition on the regularized periodic fluctuation data based on the time series decomposition model to obtain an early warning threshold for prompting a network fault, the early warning threshold being used for network fault monitoring.

The invention discloses a wavelet transform-based fine-granularity self-learning integration prediction method which comprises the following steps: by adopting time sequence decomposition based on wavelet transform, predicting time sequences of different variable coefficients with different granularities, so as to relatively precisely reveal the variation rules of the time sequences; with the combination of the time sequence decomposition, extracting characteristics of a plurality of related factors, sufficiently capturing main influence factors, and predicting future trend through rule statistics, thereby being rapid, convenient and simple; applying a model-based aggregation algorithm frame to the regression process, thereby enabling the model to have robustness which is relatively good when being compared with that of a single based learner; predicting with the combination of composite models based on wavelet transform, SVR and Ensemble, thereby obtaining prediction performance which is relatively precise when being compared with that of a conventional single model. The wavelet transform-based fine-granularity self-learning integration prediction method can be widely applied to the technical field of big-data mining and machine learning.

Described is a system for predicting system trajectories toward critical transitions. The system transforms a set of multivariate time series of observables of a complex system into a set of symbolic multivariate time series. Then pair-wise time series of a transfer entropy (TE) measure are determined, wherein the TE measure quantifies the amount of information transfer from a source to a destination in the complex system. An associative transfer entropy (ATE) measure is determined which decomposes the pair-wise time series of TE to associative states of asymmetric, directional information flows, wherein the ATE measure is comprised of an ATE+ influence class and a ATE− influence class. The system estimates ATE+, TE, and ATE− trajectories over time, and at least one of the ATE+, TE, and ATE− trajectories is used to predict a critical transition in the complex system.

The invention discloses a singular spectrum analysis-based landslide mass displacement prediction method. The method is specifically implemented according to the following steps of performing data preprocessing on a time sequence by utilizing a spectral decomposition theory and an embedded reconstruction theory of singular spectrum analysis to obtain the accumulated landslide displacement data; removing the trend term displacement from the accumulated displacement to obtain the periodic term displacement; adopting Gaussian fitting to perform fitting prediction on the trend term displacement; selecting influence factors from the predicted trend term displacement by adopting a rapid multi-principal-component parallel extraction algorithm, and selecting the LSTM model related parameters by utilizing a Bayesian optimization algorithm; constructing a training set, a verification set and a prediction set, and establishing an LSTM network model to predict the periodic item displacement; and according to a time sequence decomposition principle, superposing the predicted values of the displacement sub-sequences to obtain a final predicted value of the displacement, thereby finishing the landslide body displacement prediction method. According to the present invention, the problem that multi-source heterogeneous influence factors are difficult to fuse for collaborative and dynamic forecasting in the prior art, is solved.

Embodiments of the present invention provide techniques for decomposing a signal using a sawtooth, or triangle wave, transform. Specific embodiments transform an input signal into a piecewise-linear sawtooth or triangle wave function, and construct upper and lower envelopes for the input signal in sawtooth space. A component function for the input signal (e.g., an Intrinsic Mode Function (IMF)) is then generated based on the sawtooth or triangle wave function and the upper and lower envelopes. In various embodiments, the results generated in sawtooth space are reversely transformed into the original data space of the input signal. In this manner, an input signal may be decomposed into one or more component functions without the time-consuming and relatively unpredictable sifting process of the Empirical Mode Decomposition (EMD) method.

The invention provides a self-adaptive multi-source InSAR ground subsidence monitoring time series data splicing method and system. According to the method, the problems in three aspects of imaging geometry, space coverage and time matching degree in land subsidence time sequence splicing are considered, and in the imaging geometry, the precision of an InSAR monitoring result is improved by estimating vertical deformation of InSAR data; a method for constructing a spatial grid is provided for different spatial coverage ranges and spatial resolutions of a multi-source InSAR data set; considering the difference of spatial references, the external data is proposed for the first time to unify the spatial references so as to reduce system errors; and a self-adaptive time sequence model is constructed by utilizing a thought of time sequence decomposition, so that the introduction of errors of a common fitting model is avoided, and the precision of spliced data is improved. According to the method, the applicability of large-scale splicing of the land subsidence time series data of the multi-source InSAR monitoring result is high, and the precision of the spliced data is obviously improved.

The invention discloses an industrial steam quantity prediction method based on machine learning and relates to the technical field of industrial thermal power generation. The prediction method is based on historical working condition data collected by a boiler sensor and the actually discharged steam amount. The method comprises the following steps of carrying out time series decomposition on historical working condition data to obtain two characteristic data including an internal trend and periodic information of the historical working condition data; learning the internal trend and period information of the historical working condition data and the actual discharge amount corresponding to the historical working condition data through an LSTM algorithm; finally, training and constructingan LSTM algorithm prediction model, inputting working condition data newly collected by a boiler sensor into the LSTM algorithm prediction model after being subjected to time sequence decomposition,and outputting the predicted steam amount of the boiler. The invention further provides an industrial steam quantity prediction device based on machine learning. The prediction device is combined withthe prediction method, and a steam quantity prediction result with higher accuracy can be output.

The invention discloses a data prediction method based on an FEEMD decomposition time sequence. The data prediction method comprises the steps: building an ELM prediction model, and carrying out the optimization of the ELM prediction model through the fusion of a rhododendron search algorithm; decomposing the historical data time sequence into an intrinsic mode function component and a residual component by adopting an FEEMD algorithm; decomposing real numerical signals in the intrinsic mode function component and the residual component by adopting a variational mode decomposition method; taking a result obtained by variational mode decomposition as input of an ELM prediction model to solve prediction values of an intrinsic mode function component and a residual component; and superposingthe prediction values of the eigenmode function component and the residual component to obtain a final prediction value. According to the data prediction method, the ELM prediction model is optimizedby adopting the rhododendron search algorithm, so that rapid and accurate data prediction is realized.

The invention belongs to the technical field of music prediction model, and discloses a time series solving method and system for music traffic prediction. The method comprises the following steps that: carrying out model construction for a music user, and using a weighted fuzzy kernel clustering model to carry out a music use clustering operation; on the basis of artist dataset obtained by user clustering, carrying out the construction of the music traffic prediction model, and using algorithm based on feasible coefficient space segmentation for prediction; and verifying the efficiency and the accuracy of the artist music traffic prediction in ta music playing traffic prediction model. The method assists a player manufacturer to reasonably arrange a player background and a proper networkbandwidth so as to effectively lower the probability of occurrence of crash, meanwhile, through the research of the integral classification of music users and a classification prediction result of music, the analysis of the player manufacture for the development tendency of the music traffic is improved, and a result generated by prediction reduces loss for enterprises.

The invention relates to a city large-scale road network traffic speed prediction method based on deep integrated learning. The method comprises the following main steps: acquiring traffic flow detection data of all detection points in a road network; decomposing a speed time sequence into a plurality of intrinsic mode functions and residual sequences; adding an external variable to establish a three-dimensional space-time depth input tensor, stacking a detector on a third dimension depth dimension; calibrating parameters of a convolutional neural network model, performing prediction for a matrix composed of the intrinsic mode functions and the residual sequences through the calibrated model; reestablishing a predicted speed time subsequence, restoring the subsequence to be a predicted speed time sequence of all the detection points of a road network level. In the method provided by the invention, a complicated nonlinear and non-steady speed time sequence is decomposed into a pluralityof subsequences with higher periodicity, one-time multi-step prediction of city large-scale road network traffic speed is realized, meanwhile, prediction precision and prediction efficiency are improved, and the method has good spatial extensibility.

The invention provides a low-frequency geomagnetic signal denoising method based on shift-invariant sparse coding. The method comprises the following steps: firstly, decomposing a noise-containing magnetotelluric time sequence into a low-frequency effective signal and a noise-containing high-frequency signal by using mathematical morphology filtering; and then dividing the high-frequency signal containing noise into a plurality of segments, autonomously learning a feature structure of human noise from the high-frequency signal containing noise by using shift invariant sparse coding to construct a learning type redundant dictionary, and performing signal-noise separation on the high-frequency effective signal containing noise by using the learned redundant dictionary to obtain a denoised high-frequency effective signal; and adding the low-frequency effective signal and the high-frequency effective signal to obtain a full-band magnetotelluric effective signal. Strong human noise in low-frequency magnetotelluric data can be effectively removed, the quality of the magnetotelluric data is remarkably improved, and the deep detection effect of a magnetotelluric method is improved.

The invention discloses an electric energy meter fault number prediction method, and relates to a fault prediction method. In the prior art, an electric energy meter demand or alternating method is adopted, and the number of faults occurring in batches of electric energy meters every month is not predicted. The method comprises the following steps of: solving a moving average number sequence of asequence for imported data, establishing an ARIMA model for the moving average sequence, and predicting the future of the moving average sequence by using the model; and finally, restoring the seasonality of the predicted data, and finally obtaining a prediction result. According to the method, the time sequence decomposition model, the exponential smoothing model and the time sequence ARIMA modelare combined for use, and advantages and disadvantages can be made up; the accuracy and economy of predicting the fault number of the electric energy meter are improved, the quality risk and propertyloss caused by overdue and overstock of the electric energy meter are avoided, the cost input returned by the overdue meter is reduced, and the overstock of the inventory electric energy meter is avoided.

The invention discloses a time sequence prediction method combined with multiple models. The method comprises the following steps: decomposing a time sequence into three time sequences of long-term trend, seasonal fluctuation and irregular fluctuation by using an STL decomposition algorithm; the method comprises the following specific steps: step 1, data preprocessing; obtaining a training set, a verification set and a test set; step 2, according to the obtained time sequence of the long-term trend, adopting a traditional prediction method ARIMA model and Prophet model for prediction, and selecting a model with the best precision; step 3, according to the obtained time sequence of seasonal fluctuation, seasonal prediction is carried out by adopting an adaptive filtering method, an ARIMA model and a Prophet model, and a model with the best accuracy is selected; step 4, according to the obtained time sequence of irregular fluctuation, predicting the time sequence of irregular fluctuation by using a neural network model; and step 5, adding the prediction results from the step 2 to the step 4 to obtain a final prediction result.

The invention discloses a short-term demand forecasting method for an electric energy metering device of a business expansion project based on conditional probability adjustment, and belongs to the technical field of metering device demand forecasting. The method of the present invention takes the demand prediction of the electric energy metering device of the business expansion project as the object, first combines the historical data of the past years in the marketing business system, uses the time series decomposition model to predict the overall demand, and comprehensively considers the business expansion on the way from "measurement demand determination" to " According to the conditional probability distribution theory, the conditional probability adjustment model of the in-transit project is established, and the obtained preliminary forecast results are dynamically adjusted in real time, so as to obtain more accurate demand forecast results. The short-term demand forecasting method for electric energy metering devices in industrial expansion projects based on conditional probability adjustment can adjust, optimize and improve the original demand forecast results in a timely manner according to the changes in demand data of recent engineering projects, and provide more accurate information for procurement, inspection, production, and warehousing and distribution. demand plan.