45results about How to "Achieve high-precision forecasting" patented technology

The invention relates to an expressway traffic flow parameter prediction method and a system based on a gated neural network GRU. The method comprises the following steps: according to the high-speedroute information of the collected data and the longitude and latitude information of a toll station of a section, the research section data is initially screened, then the abnormal data is cleaned according to the manifestation of the abnormal data, and then the velocity time series is calculated in a certain time period, and then the missing data is filled in the missing time series, the filledspeed time series data is divided into training data and test data, and the traffic flow prediction model is obtained by training the training data, finally, the error analysis is carried out by usingthe predicted data and test data. The invention utilizes the advantage of GRU long-time memory data characteristics to obtain higher prediction accuracy, relatively less prediction model parameters and good portability, and can provide technical support for traffic guidance and traffic accident management and dispatching of traffic management departments.

The invention provides a complex equipment maintenance decision-making method based on fault prediction. The method comprises the following steps: A, determining a feature factor related to an equipment fault, setting a fault threshold of the feature factor, and collecting the historical data of the feature factor; B, predicting the numerical value of the feature factor through a gray model and aBP neural network model respectively; C, determining weights of the gray model and the BP neural network model; and D, carrying out numerical prediction on the equipment feature factor based on the combination model determined by the weight, taking the moment when a fault threshold value is reached as a prediction fault moment, and determining the optimal maintenance opportunity. The method has the following advantages: the advantages that the gray model has low requirements for sample data volume and the BP neural network has strong autonomous learning capability are combined, high-precisionprediction of the feature factor is effectively realized, the maintenance time is determined in time according to comparison with a fault threshold, preventive maintenance can be carried out in time,and normal operation of equipment is ensured.

The invention provides a neural network wind power short-term forecasting method based on a fuzzy partition theory. The neural network wind power short-term forecasting method based on the fuzzy partition theory adopts the mode of combining a fuzzy theory, artificial intelligence and a statistical theory through analyzing the important features of wind velocity variation and the relationship between wind velocity and power. When wind power forecasting is conducted, wind scale fuzzy partition processing is conducted on wind velocity data obtained from weather forecasting according to periods of time, BP neural network partition forecasting is conducted, a forecast power value is obtained through multiplying a partition forecast value by a membership degree value of the partition forecast value and adding all partition values, a probability statistics modified algorithm is conducted, and the forecast power is obtained. The neural network wind power short-term forecasting method based on the fuzzy partition theory improves the accuracy of a forecasting model effectively.

The invention discloses a residual life analysis method for a corroded oil and gas pipeline based on improved adaptive GEV distribution. The method comprises the following steps: 1) acquiring a maximum corrosion depth data sequence X(i)=(x1,x2,lambda,xG) of the corroded oil and gas pipeline; 2) inputting a maximum corrosion depth data sequence of the corroded oil and gas pipeline into an improvedGEV distribution model, and simulating and predicting parameters of the improved GEV distribution model by an MCMC method to obtain the statistical parameter values of a threshold parameter eta, a position parameter mu and a scale parameter sigma; 3) judging an extreme value distribution type to which the maximum corrosion depth of the corroded oil and gas pipeline belongs according to the threshold parameter eta, and then analyzing the residual life of the corroded oil and gas pipeline according to the extreme value distribution type to which the maximum corrosion depth of the corroded oil and gas pipeline belongs. By adopting the method, the limitation problem of single distribution of the maximum corrosion depth of the oil and gas pipeline is solved, and high precision prediction of theresidual life of the corroded oil and gas pipeline is realized.

The invention discloses a tool wear state detection method for industrial unbalanced data. The method comprises the steps of: preprocessing the historical monitoring data of the numerically-controlledmachine tool cutter obtained by the sensor to form source training data with the cutter wear state label; carrying out the same preprocessing of the historical monitoring data of the to-be-detected tool, and forming auxiliary training data together with a few types of data obtained through the oversampling synthesis of the source training data; training a cutter wear state prediction model by using the training data through a transfer learning method; and preprocessing and inputting the real-time sensor data of the to-be-detected cutter into the prediction model, and obtaining the abrasion state of the cutter in real time. According to the method, the balance of the training data is fully ensured, and the distribution consistency of the training data and the test data is fully ensured, sothat the problem of high-precision detection of the tool wear state under the condition of less and unbalanced industrial data volume is solved.

The invention belongs to the field of power transformation engineering cost prediction, and particularly relates to an improved firefly algorithm-based power transformation engineering cost predictionmethod for SVM optimization. For improving optimization performance of an FA to optimize parameters of an SVM prediction model, the invention provides the improved firefly algorithm-based power transformation engineering cost prediction method for the SVM optimization. The method mainly comprises three parts including data processing, parameter determination and cost prediction; specially, in theparameter determination part, a position updating formula of the FA is improved by adopting a Gauss disturbance technology based on a conventional FA to search for optimal parameters; and the methodenhances the capability of fireflies in escaping from local optimum and improves the optimization performance of the FA so as to optimize the parameters of the SVM prediction model. Through Schaffer function testing, the proposed Gauss disturbance FA has the advantages of high convergence speed, good search capability and the like, and can realize high-precision prediction of power transformationengineering cost level.

The invention relates to an adaptive vehicle state prediction system and prediction method based on model and parameter dynamic adjustment. During working, process noise parameters in a robust volumeKalman filtering unit are dynamically updated through a fuzzy inference system unit, and model parameters in the robust volume Kalman filtering unit are dynamically updated through a model parameter prediction unit; and high-precision prediction of the vehicle state is completed based on the sensor information collected by a vehicle-mounted sensor signal measurement unit and the robust volume Kalman filtering unit. According to the method, the model has a dynamic updating capability while the vehicle state is predicted, the prediction precision is continuously improved based on continuous self-adjustment of an algorithm, and the development of a vehicle active safety control technology is promoted.

The invention discloses a controlled airspace strategic flow prediction method based on a gray long-short-term memory network, and belongs to the technical field of air traffic flow management. The controlled airspace strategic flow prediction method comprises the following steps: step 1, reading data; step 2, preprocessing the data; step 3, preliminarily screening influence factors by utilizing grey correlation analysis; step 4, extracting main characteristics by utilizing a principal component analysis method; step 5, establishing a gray strategic flow prediction model; step 6, establishinga long short-term memory network strategic flow prediction model; and step 7, establishing a gray long-short-term memory network combination prediction model. For the controlled airspace strategic flow prediction method, scientific basis can be provided for airspace structure optimization such as sector division and route adjustment of the control area; effective utilization of airspace resourcesis achieved; and a basis is provided for resource demand distribution such as future personnel investment, financial investment and fixed asset investment of the controlled area.

A method for cluster prediction of extreme wind speed includes constructing target wind measuring station and time-lapse wind measuring station at target wind measuring point according to near-term wind speed condition, de-noising data of wind measuring station, clustering wind speed samples, utilizing LS-SVM trains the denoised wind speed clustering sample data, and constructing the wind speed prediction model of each anemometer station under various step sizes, choosing each model to carry on the best combination of many kinds of steps, realizing the multi-step iterative forecast, improvingthe forecast precision, reducing the random error disturbance. The wind speed prediction along the railway line can be realized, and the wind speed environment in the accident-prone area can be knownin advance, which can guide the train operation timely and effectively, and ensure the safety of train operation.

The invention discloses an image distortion method based on matrix inverse operation in a virtual reality (VR) mobile end. The method comprises the following steps: 1, reading a current frame and an equipment state; 2, converting coordinates under a screen coordinate system at current-frame time into coordinates under a standard equipment system at the current-frame time; 3, according to the coordinates under the standard equipment coordinate system at the current-frame time, obtaining coordinates in a world coordinate system; 4, according to the coordinates in the world coordinate system, obtaining coordinates under the standard equipment coordinate system at corresponding next-frame time; 5,performing linear transformation on the coordinates under the standard equipment coordinate system at the next-frame time so as to finally obtain coordinates under the screen coordinate system through the transformation; and 6, endowing the coordinates under the screen coordinate system at the corresponding next-frame time with pixel RGB values of the coordinates under the screen coordinate system at each current-frame time so as to obtain a final distortion image. The method is a method for generating intermediate frames in VR, can effectively reduce jittering in a VR game, and improves user experience.

The invention discloses an electric power consumption probability prediction method based on a neural network, and the method comprises the following steps: collecting historical data of electric power consumption, dividing the historical data into a training set and a test set, and carrying out the normalization processing of all variables; constructing a neural network model based on a convolutional architecture and a self-attention mechanism; training a neural network model by using the processed training set data, and selecting a model with the best prediction precision as a trained neural network model by using the test set; recent data of power consumption are selected and preprocessed, the preprocessed recent data are input into the model, and an output value of the model is subjected to inverse normalization processing to obtain a probability prediction result. Compared with a traditional power load prediction method, the method has the advantages that modeling of power consumption data of different users in a power grid is achieved at the same time by means of the constructed neural network model, short-term and long-term modes in a time sequence can be captured, high-precision prediction of the time sequence is achieved, and a point prediction result and a probability prediction result are output.

The invention discloses an urban rail transit emergency on-line passenger flow prediction and simulation system. The system comprises a real-time passenger flow monitoring module and an emergency passenger flow prediction module, wherein the real-time passenger flow monitoring module is used for predicting a passenger flow evolution state in a future preset duration under a normal condition, and generating a passenger flow index of real-time time granularity of urban rail transit, and visually displaying the predicted passenger flow evolution state to monitor the real-time passenger flow. Theemergency passenger flow prediction module is used for adjusting a train operation time table for emergency occurred in a road network, predicting the passenger flow distribution condition under the influence of the event according to the adjusted time table, and publishing the influence degree information of each terminal arrival station to passengers. The urban rail transit emergency on-line passenger flow prediction and simulation system can monitor passenger flow on line in real time and automatically adjust train operation when an accident occurs.

The invention provides a hot continuous rolling strip steel width prediction method based on principal component analysis in cooperation with a random forest, and relates to the technical field of hot continuous rolling process control. The method comprises the following steps: firstly, determining an arrangement form of hot continuous rolling production line equipment, determining a temperature system, rolling mill equipment parameters and rolling boundary conditions; then according to characteristics of a production line, determining actually measured data which needs to be acquired and is about steel grade change, specification change and width of a first steel block after roller change; carrying out standardization processing on the acquired actual measurement data; carrying out dimension reduction processing and feature selection on the standardized data set by adopting a principal component analysis method, and determining an input variable of a random forest width prediction model for strip steel width prediction; dividing the data set after dimension reduction processing and feature selection based on principal component analysis into a training set and a test set according to a certain proportion, and constructing and training a random forest width prediction model according to a random forest algorithm; finally evaluating the prediction precision of the random forest width prediction model.

The invention discloses a power load prediction method and device based on a highway neural network, and the method comprises the steps: obtaining power load data and weather data of a prediction place in a historical time period, and solving a correlation coefficient of the power load data and the weather data; obtaining a feature matrix according to the correlation coefficient, and classifying the running days of the historical load by adopting a clustering analysis method; for each type of operation days, taking the operation day type, daily power load data and related weather data as input, and respectively establishing a corresponding load prediction model based on an artificial intelligence expressway neural network; and determining the type of the current operation day, and carrying out power load prediction by adopting the load prediction model corresponding to the type of the current operation day. According to the scheme, the accuracy of power load prediction is effectively ensured.

The invention discloses a method for forecasting the commodity barcode registration quantity based on BP neural network. The method comprises the following steps: analyzing the autocorrelation analysis chart of the commodity barcode registration quantity and generating the training sample; establishing a BP neural network, and using the generated training sample for training, so as to obtain a forecasting model of commodity barcode registration quantity; forecasting the commodity barcode registration quantity by using the forecasting model. The method takes into account the non-linear characteristics of the data of commodity barcode registration quantity, and enables high-precision prediction of the commodity barcode registration quantity. The method adopts the single hidden layer BP neural network under the premise of ensuring high prediction accuracy, which greatly reduces the computational difficulty and complexity compared with the multi-hidden layer neural network. The method is characterized by simple operation.

The invention discloses a method for predicting the porosity of a tight sandstone reservoir. The method comprises the following steps: determining a longitudinal wave impedance three-dimensional data volume of sandstone; determining a longitudinal wave impedance threshold value for dividing the thin reservoir and the thick reservoir; obtaining a prediction result of the porosity of the thick reservoir; obtaining a prediction result of the porosity of the thin reservoir; and combining the prediction results of the porosity of the thick reservoir and the porosity of the thin reservoir into a data volume to obtain a prediction result of the porosity of the whole reservoir. According to the method for predicting the porosity of the tight sandstone reservoir, the porosity distribution of the thick reservoir can be effectively predicted, the porosity distribution of the thin reservoir is considered, the method has a high coincidence rate through actual drilling verification, and high-precision prediction of the porosity of the tight sandstone reservoir is achieved. And through a classification prediction mode, the depiction of the porosity of the reservoir is more accurate, and an effective result is provided for efficient development of the compact sandstone reservoir.

The invention provides a disease risk prediction method and system based on multi-source graph neural network fusion, and belongs to the technical field of medical treatment. In the method and the system, patient features are constructed based on disease duration information in combination with historical diagnosis information of a patient, and a disease risk prediction data set is formed; constructing a multi-source disease relation network, proposing disease network feature extraction based on a graph neural network, and performing patient disease feature matrix completion; a disease risk prediction model based on multi-source disease relation network fusion is provided, and high-precision prediction of disease risks is achieved.

The invention relates to a method for predicting penetration performance of high-speed impact concrete of a projectile body considering an erosion effect, and belongs to the field of impact dynamics.By combining two mechanisms of projectile body surface melting and aggregate cutting, the projectile body high-speed impact concrete penetration performance prediction method coupled with two projectile body erosion mechanisms is provided, and erosion and projectile body motion parameter evolution in the projectile body high-speed penetration process are effectively predicted; according to the method, the projectile body surface melting caused by projectile target friction heat and the cutting effect of hard particles on the projectile body surface in the projectile body penetration process are considered, the combined action of multiple erosion mechanisms obtained through experimental observation is fully considered, and the bottleneck of a traditional prediction method considering a single erosion mechanism is broken through. By means of the prediction method, high-precision prediction of the penetration performance of the high-speed impact concrete of the projectile body can be achieved, and key technical support is provided for structural design of the high-speed penetration drilling projectile and evaluation of the concrete protection performance.

The invention relates to a pressure drop prediction method for a horizontal oil-water two-phase flow based on a dynamic contact angle, which comprises the steps of respectively inputting the density [Rho]o and [Rho]w of the oil phase and the water phase, the motion viscosity [Nu]o and [Nu]w of the oil phase and the water phase, the interfacial tension [Gamma]ow and the pipeline inner diameter D; inputting apparent velocity parameters of the oil phase and the water phase; equally dividing the pipeline inner diameter D into N parts; calculating the pressure drop of the oil phase and the water phase; and obtaining pressure drop parameters.

The invention provides a short-term power load prediction method based on multiple factors and an improved feature screening strategy. The short-term power load prediction method comprises the following steps of S1, importing an original multi-factor data set and performing data preprocessing; S2, constructing a candidate feature variable set; s3, performing hour granularity feature screening based on data set reconstruction and an RReliefF algorithm; S4, introducing a k-means clustering label based on cosine similarity; S5, determining a final input variable set; and S6, carrying out model training and prediction. The method pays attention to front-end data processing of short-term power load prediction, can be used in combination with various current mainstream prediction models, can remarkably improve the prediction precision of the model, and has wide universality. According to the method, the problem of characteristic variable selection rules based on the hour granularity can be effectively solved, the shape and mode information of the load curve is added into the characteristic variables, and the prediction performance of the short-term power load is remarkably improved by improving the quality of front-end input data.

The invention discloses a long-wave ground wave propagation delay time-varying characteristic prediction method applied to a long distance, and the method comprises the steps: firstly building a long-term monitoring system of long-wave ground wave propagation delay, and obtaining long-wave ground wave propagation delay data; selecting proper position points near the propagation path, and acquiringthree kinds of meteorological data of temperature, humidity and atmospheric pressure of the position points; preprocessing the acquired propagation time delay data and meteorological data; and finally, according to the preprocessed propagation time delay data and meteorological data, using a generalized regression neural network (GRNN) for prediction, and obtaining long-distance long-wave groundwave propagation time delay time-varying characteristics. According to the invention, high-precision prediction of long-distance long-wave ground wave propagation delay time-varying characteristics isrealized, and the precision of a land-based long-wave navigation / time service system is improved.

The invention discloses a vehicle trajectory prediction method based on global attention and state sharing. The solution method comprises the following steps that a GAS-LED trajectory prediction model of a codec LSTM model with a global attention mechanism and state sharing is used; in the GAS-LED track prediction model, a state sharing mechanism with an encoder and a decoder is adopted to reduce the calculation workload, and meanwhile, two parallel calculation GAS-LED track prediction models are adopted to output prediction of the transverse lane changing behavior and the longitudinal driving distance of the vehicle in parallel; in the track prediction task of the lane level, the lane where the vehicle is located is focused on, and the GAS-LED track prediction model outputs corresponding prediction results for the transverse lane change and the longitudinal driving distance; and historical information of the current vehicle and the surrounding vehicles is used as the input of the GAS-LED trajectory prediction model 2, and then the two GAS-LED trajectory prediction models are used in parallel to obtain more output results convenient to predict. Through the scheme, the purpose of high-precision prediction is achieved, and the method has very high practical value and popularization value.

The invention discloses a high-precision weld joint morphology prediction method suitable for myriawatt-level laser welding, and aims to solve the problem that the weld joint morphology prediction precision of myriawatt-level laser welding is low due to the fact that a weld pool and plume coupling behavior cannot be considered in an existing weld joint morphology prediction method. According to the method, a compressible two-phase flow numerical calculation method based on pressure is adopted to solve the coupling behavior of the molten pool and plume, and therefore high-precision prediction of the myriawatt-level laser welding seam morphology is achieved. Firstly, a welding seam morphology function and welding parameters at the initial moment are input; secondly, a compressible two-phase flow numerical calculation method based on pressure is adopted to obtain a welding seam morphology function at the next moment; and drawing a welding seam morphology function at the next moment, and extracting welding seam morphology and welding seam morphology characteristics. Compared with an existing weld joint morphology prediction method, the weld pool and plume coupling behavior in myriawatt-level laser welding can be accurately calculated, the algorithm is simple and easy to implement, the calculation efficiency is high, the physical conservation is good, and high-precision prediction of the myriawatt-level laser welding weld joint morphology can be achieved.

The invention discloses a GRU-based harmonic residual segmented tide level prediction method. According to the method, firstly, a site hourly astronomical tide sequence is calculated by using a tide harmonic analysis method according to long-term actually-measured tide level data of a site to be measured, the actually-measured hourly tide level sequence of the site is aligned according to time, the astronomical tide sequence is subtracted, and the hourly tide level reconciliation residual sequence is obtained; the harmonic residual error sequence is divided into two sections of samples as input variables by comprehensively considering the influence action time periods of characteristic factors such as monsoon and typhoon, two tide level residual error GRU prediction models are formed through training respectively, and a residual error prediction result sequence is obtained through calculation; and finally, the residual prediction result sequence is added to the astronomical tide sequence of the corresponding time sequence, and the tide level prediction result is obtained. According to the method, tide level prediction can be achieved only by using single-station tide long-time sequence data, participation of other factors is not needed, high-precision prediction of station tide level data is achieved, and the efficiency of the tide level prediction process is improved.

The invention relates to a solar radiation prediction method based on TCN-Attention. The working process of the solar radiation prediction method comprises the following steps: S1, carrying out a series of data preprocessing such as necessary cleaning, conversion and specification on historical solar radiation data collected by a meteorological station, and carrying out normalization processing on the data so as to enable each index to be in the same order of magnitude; S2, dividing the data set into a training set and a test set; S3, establishing a TCN network training model based on an Attention mechanism; and S4, inputting historical solar radiation data to the trained prediction model, and outputting an ultra-short-term solar radiation value in the future 1h. The Attention mechanism module is introduced through the hidden layer of the TCN model to highlight key information, and the accurate prediction effect is further improved.

The invention discloses a region tail gas migration prediction method and system based on domain adaptation and a storage medium. The method comprises the steps: obtaining and processing historical tail gas data and external factor data of a source region and a target region, taking monitoring points as nodes for the source region data and the target region data, and enabling the source region data and the target region data to be connected pairwise; constructing graph structure data by taking the weight as the reciprocal of the monitoring point distance, and dividing a time sequence set according to the tail gas concentration change characteristics of the source region and the target region; constructing a tail gas spatio-temporal feature extraction module, and performing shallow feature extraction and fusion on the time sequence data of the source region and the target region; constructing an automatic encoder, and utilizing the encoder to non-linearly map shallow spatio-temporal features of the source domain and the target domain belonging to different feature spaces to the same feature space; performing depth extraction on the shallow layer features, and outputting a prediction result. According to the method, efficient utilization of the source domain data is realized by utilizing the domain adaptation method, so that higher-precision regional tail gas prediction of the target domain lacking data is realized.

The invention discloses a polypeptide detection method based on deep learning. The method comprises the following steps: acquiring mass spectrometry data of a training sample; acquiring a training set according to the mass spectrum combined data; and training a deep learning-based target detection model by using the training set, and detecting the polypeptide in the to-be-detected sample by using the trained target detection model. According to the polypeptide detection method based on deep learning, the deep learning method has high feature extraction capacity, 2D distribution features of polypeptides can be effectively captured, and high-robustness detection and high-sensitivity detection of different polypeptides are achieved; meanwhile, the classification error function used in the constraint function is designed based on the cross entropy loss function, and high-precision prediction of the polypeptide target probability can be achieved. The method is based on an artificial intelligence technology, and detection of dense polypeptide targets in a complex sample can be realized.

The invention provides a taxi passenger flow prediction method, which comprises the following steps of: firstly, splicing get-on passenger flow and get-off passenger flow into the same dimension, performing spatial node serialization, normalizing data to obtain original data, and dividing the original data into training data and test data; secondly, performing spatial attention feature extraction on the training data by adopting a spatial multi-head attention algorithm to obtain global spatial features; flattening the global spatial features and the training data into the same dimension, and performing time feature extraction by adopting a first multi-layer perceptron to obtain global time features; and inputting the global time features and the training data into a second multi-layer perceptron for iterative training to obtain a trained flow prediction model, and inputting the test data into the prediction model to obtain a prediction result. The method can effectively extract the global spatial features and the global time features, and can achieve the high-precision prediction of the taxi passenger flow.

The invention discloses a component quantitative analysis method, a test system and a storage medium. The method comprises the following steps: acquiring spectral data of a plurality of samples after laser excitation and the real content of components in each sample; training a partial least squares regression+support vector machine regression model by utilizing the spectral data of each sample, and during training of the partial least squares regression model, taking the spectral data with the maximum contribution degree corresponding to each sample in the training set as input data and taking the real content of each component of the corresponding sample as a training label; when a support vector machine regression model is trained, acquiring the predicted content of each component of a sample in a prediction set by using the trained partial least squares regression model, using the residual error between the predicted content and the real content as a training label, and using the spectral data of each sample in the prediction set as input data. According to the invention, quantitative measurement of components is realized based on the constructed model, and a set of test system for quantitative analysis of components by using laser-induced breakdown spectroscopy is constructed.