38results about How to "Reduce the difficulty of forecasting" patented technology

The invention discloses an image data multi-label classification method. The method comprises the following steps: decomposing an input image, extracting high-order correlation of features by utilizing a neural network, decomposing tag data, extracting high-order correlation of tags by utilizing the neural network, and decoding a feature code of the input image from an input space to a tag space by adopting the neural network comprising multiple layers of full connection layers; constructing a loss function, initializing a training parameter, adopting a random gradient descent method to minimize a final loss function as a target, and training and solving to obtain an optimal training parameter; and inputting to-be-tested image data into the trained model for prediction, and outputting to obtain a label result to realize multi-label classification. According to the method, the problem that the secondary correlation and the multi-correlation of the labels cannot be extracted at the sametime when a person works in front of the image data is solved, the prediction difficulty caused by too sparse image data is reduced, and the accuracy of multi-label classification is improved.

The invention discloses an anchor-free remote sensing image target detection method and system based on scene enhancement, and the method comprises the following steps: 1, carrying out the linear enhancement of an obtained remote sensing image data set in a balance coefficient hybrid enhancement mode, and obtaining an enhanced training set; 2, constructing a target detection model based on scene enhancement, training the target detection model through the training set obtained in the step 1 until a preset stop condition is met, and obtaining a trained target detection model, wherein the trained target detection model is used for remote sensing image target detection. According to the method, a more convenient and robust balance coefficient hybrid enhanced data augmentation mode is provided, the feature extraction capability and the category prediction capability of the network are enhanced by using scene information, and the detection precision is improved.

The invention relates to a power load prediction method and device, computer equipment and a storage medium; and the method comprises the steps: in response to a power load prediction request, obtaining power load related feature data corresponding to the power load prediction request, the power load related feature data comprising historical power load values; further dividing the feature data into life power load related feature data and production power load related feature data; and then inputting a pre-constructed power load difference prediction model to obtain a life power load prediction difference and a production power load prediction difference corresponding to the prediction time information, and performing calculation processing on the historical power load values to obtain anpower load prediction result. According to the invention, through the life power load related feature data and the production power load related feature data, respective prediction is carried out byusing the model, and the fine degree of power load prediction is improved; the variance in the prediction process is reduced by predicting the power load difference value, and the accuracy of power load prediction is improved.

The invention relates to an electric energy measuring error estimation method. The method comprises the following steps: preprocessing metering data measured by an electric energy metering device to obtain a metering error sequence; performing wavelet conversion decomposition on the metering error sequence to obtain multiple groups of error sequence components; constructing a robust extreme learning machine model, regarding each group of error sequence component and the measured metering data as the input quantity to add in the robust extreme learning machine model as the input quantity, thereby obtaining an error prediction value of each group of error sequence component; adding the obtained error prediction values to predict the next generation of error value; preprocessing the error value to obtain the new metering error sequence, updating the measured metering data, performing iterative updating and cyclically predicting. Through the electric energy metering error estimation methodprovided by the invention, the rule in the original signal is extracted through the application of the wavelet conversion decomposition and the robust extreme learning machine model, the prediction difficulty is greatly reduced, the prediction precision and operation speed under the existence of the noise condition are improved.

According to the wind speed prediction method based on the LMD and the LSTM, the actual wind speed is decomposed into a plurality of components, the prediction difficulty is reduced, and a time sequence model is established in combination with the LSTM for wind speed prediction; comprising the following steps: (1) extracting steady-state data; (2) wind speed feature extraction based on LMD; (3) establishing a wind speed prediction model based on LSTM; the LMD and the LSTM are combined, and the wind speed prediction precision is improved by utilizing the characteristic that the combined prediction has the advantages of the two algorithms; the local mean decomposition method is used for decomposing data, so that modal aliasing can be effectively eliminated, and the model prediction precision is improved; by using the excellent time sequence feature extraction capability of the LSTM, the wind speed prediction method can be effectively popularized to wind speed prediction of different stations with complex geographic features, and the accuracy of wind speed prediction is improved in time and space.

The invention discloses a selection and trend prediction method of underground coal mine electromagnetic radiation intensity time series data. The method comprises the following steps: sampling time series data of underground coal mine electromagnetic radiation intensity to obtain a sample data set; performing clustering granulation processing on the sample data set to obtain multiple data subsets; determining the significance of sample points in each data subset based on Hausdorff distance, and removing or reserving each sample point according to the significance so as to obtain a predictiondata set; and predicting the change trend of the underground coal main electromagnetic radiation intensity according to the prediction data set. According to the selection and trend prediction methodof underground coal mine electromagnetic radiation intensity time series data disclosed by the invention, the computation complexity of the trend prediction is reduced, and the robustness of the prediction result is enhanced.

The invention discloses an open channel gate front floating ice state prediction model and a construction method and application thereof, and the method comprises the steps: exploring the relation between gate front ice accumulation and transportation data based on open channel gate front floating ice monitoring data in ice-water two-phase water flow, building an ice transportation and accumulatedice discrimination model based on a support vector machine, and therefore, the ice accumulation and transportation state in front of the open channel bulkhead gate in the ice-water two-phase water flow is judged. According to the invention, the support vector machine is used for obtaining the optimal classification hyperplane of the gate front floating ice accumulation and transportation state according to the gate front ice-water two-phase water flow incoming flow condition and the boundary condition, accurate prediction of the gate front floating ice state is achieved, and therefore effective data support is provided for open channel gate front ice accumulation and transportation research and cold region water conservancy projects.

The invention relates to an overheat early warning method for a supercritical boiler heating surface pipe wall, in particular to an overheat early warning method for the supercritical boiler heating surface pipe wall using discretization conversion. In order to solve the problems that the generation technological requirements are high, overheat cannot be eliminated completely, over-frequent replacement or upgrade of pipelines can bring huge economical burdens to use units and pure temperature prediction cannot be popularized in practical application easily, the overheat early warning method for the supercritical boiler heating surface pipe wall using discretization conversion is provided. The method is implemented through the following steps that first, a HistoryTable is established; second, the variables from dc1 to dc17 are output; third, the variables from gc1 to gc17 are output; fourth, the dmark variable is output; fifth, a disaggregated Model is acquired; and sixth, whether the supercritical boiler heating surface pipe wall is to be overheated or not is subjected to early warning. The overheat early warning method is applied to the overheat early warning field of the supercritical boiler heating surface pipe wall.

The invention discloses a time series data prediction method based on time series decomposition and LSTM, and the method comprises the steps: 1, collecting time series data, carrying out the preprocessing of the time series data, obtaining a time series sample set which meets the data demands of a prediction model, carrying out the division of a training set and a test set, and obtaining a first training set and a first test set; step 2, establishing a first neural network for trend component and remainder prediction based on LSTM, performing training and parameter adjustment through the first training set, predicting the first training set by using a trained first neural network model to obtain a trend component and remainder prediction result of the first training set, and further processing the prediction result into a second training set; step 3, establishing a second neural network based on ANN, and performing training and parameter adjustment through a second training set; and 4, performing joint prediction on the test set by using the trained first neural network model and the trained second neural network model to obtain a fitted time sequence data prediction result.

The invention relates to a dynamic prediction method for movement deformation of mining overburden strata, in particular to a quantitative dynamic prediction method for determining any predicted level and rock stratum sinking, inclination, curvature, horizontal movement, horizontal deformation and vertical deformation values at any moment in the mining overburden strata. According to the dynamic prediction method for the movement deformation of the mining overlying strata, the mining overlying strata is layered according to the lithology, and according to the lithology evaluation coefficient of the overlying strata and the sinking coefficient model determined on the basis of the lithology evaluation coefficient, the sinking coefficient influence factor of each layered strata in the overlying strata is solved, so that the prediction accuracy of the separation of the overlying strata is improved; and the new model omits a very tedious process of judging the key layer of the overlying strata, and the calculation amount and the prediction difficulty are greatly reduced.

The invention discloses a short-term power load prediction method and system based on a hybrid model. According to the method, the time sequence characteristics of the high-frequency component subsequences are extracted through the LSTM prediction model, the short-term power load is predicted in cooperation with the ELM-CATBOOST mixed prediction model composed of the CATBOOST prediction model and the first ELM prediction model, original power load data are decomposed into a plurality of intrinsic mode function components through the CEEMDAN decomposition algorithm, the model prediction difficulty is reduced, and the prediction efficiency is improved. The prediction accuracy is improved; besides, an LSTM prediction model is utilized to extract time sequence features of the high-frequency component subsequences, historical power load data and original power load data of the high-frequency component subsequences are combined to jointly serve as input features of an ELM-CATBOOST hybrid prediction model, input feature dimension information is greatly enriched, the advantages of a single model are integrated by using the ELM-CATBOOST hybrid prediction model, and the prediction accuracy is improved. The method has higher robustness and accuracy, and different input features and prediction models are adopted for high and low frequency component subsequences, so that the model complexity can be reduced.

The embodiment of the invention provides an image recognition method and device and computing equipment. The method comprises the steps: acquiring a to-be-processed first image; inputting the first image into an optical character recognition network, and obtaining character information in the first image, wherein the optical character recognition network is trained in an auxiliary mode through a semantic segmentation network, and the semantic segmentation network is used for predicting position information of characters in the image; and sharing the features extracted by the intermediate layer of the optical character recognition network with the optical character recognition network. According to the invention, the semantic segmentation network assists in training the optical character recognition network, so that the optical character recognition network automatically notices the character region in the first image, thereby reducing the prediction difficulty of the optical character recognition network, improving the precision of the optical character recognition network, and realizing the accurate recognition of a complex scene, such as seriously bent, rotated and vertical characters.

The invention discloses a wind power prediction method based on secondary modal decomposition and cascade deep learning, and the method comprises the steps: firstly collecting and obtaining original wind power data and wind speed data, carrying out the signal preprocessing of the collected data, and decomposing a wind power and wind speed time sequence into a series of relatively stable sub-sequences through secondary modal decomposition, so that effective modal decomposition can be carried out on wind power signals, then a convolutional neural network gated cycle unit prediction model (CNN-GRU) is used for extracting implicit features of a coupling relation between time sub-sequences generated by decomposition and wind speed, time correlation between the time sub-sequences is further extracted, finally a prediction value of wind power is output, the wind power can be more effectively predicted, and the optimal prediction performance is achieved, and the method has a good application prospect.

The invention discloses a severe convective weather prediction method and system for improving a three-dimensional generative adversarial neural network based on a hybrid evolutionary algorithm, and the method comprises the steps: reading radar echo data from an original Doppler weather radar, generating a training data set, and dividing the training data set into a plurality of groups of input data; constructing a hybrid evolutionary algorithm by using a genetic algorithm and a cross entropy algorithm; establishing an improved three-dimensional confrontation generation neural network model based on a hybrid evolutionary algorithm; training the three-dimensional generative adversarial neural network model through the training data set to obtain a trained three-dimensional generative adversarial neural network model; the number N of radar echo data needing to be input is obtained according to the to-be-predicted time range, and the latest N pieces of preprocessed radar echo data are input into the trained three-dimensional confrontation generation neural network model for severe convective weather prediction. The severe convective weather prediction accuracy can be effectively improved.

The invention discloses a prediction method and device for generating capacity of a tubular turbine. The method comprises the steps: obtaining prediction parameters used for predicting the generatingcapacity of the through-flow turbine, wherein the prediction parameters comprise the water inlet amount H of a water storage device connected with the through-flow turbine, the water outlet amount F of the water storage device, the turbine efficiency E1, the generator efficiency E2 and the gravity constant g; based on a first preset formula, determining a plurality of first predicted generating capacities of the tubular turbine within a preset time according to the prediction parameters; according to the actual historical generating capacity curve of the tubular turbine, repairing each first predicted generating capacity, and obtaining a corresponding target prediction generating capacity. According to the method, the technical problems that the prediction difficulty of the generating capacity of the water turbine is increased and the accuracy of the predicted generating capacity is relatively low due to many parameters involved in the generation prediction of the water turbine and strong randomness of the parameters when the generating capacity of the through-flow water turbine is predicted in the prior art are solved.

The invention relates to the field of three-dimensional seismic inversion, in particular to a method for converting horizontal well data into plumb shaft data and a method for seismic inversion. According to the method, the parameters are quantified, the horizontal well data is effectively converted into the plumb shaft data, the converted plumb shaft data is combined with a depth domain inversiontechnology, so that the prediction precision and the longitudinal resolution of the three-dimensional seismic reservoir are effectively improved, and the difficulty of reservoir prediction and development is reduced.

The invention discloses a lottery user activity prediction method, which comprises the following steps: acquiring original user data; extracting and converting the original user data; classifying andloading the original user data into a database in a specified format; and loading the original user data into a database in a specified format. Preprocessing the original user data stored in the database to obtain multi-dimensional user data; Obtaining a prediction feature set related to user activity according to the multi-dimensional user data; inputting The prediction feature set into a pre-trained GBDT algorithm-based activity prediction model to predict user activity. Correspondingly, the invention also discloses a lottery user activity prediction system, a terminal device and a computer-readable storage medium. The technical proposal of the invention can reduce the prediction difficulty of the lottery user activity and improve the prediction accuracy.

The invention discloses a project budget balance prediction method and device. The method comprises the steps of obtaining project data related to project end or balance according to service content of a to-be-predicted project, then obtaining key features according to the service data in the project data, and inputting the key features into a prediction model to predict to obtain project budget balance information of the to-be-predicted project. According to the method, a plurality of service factors influencing project budget balance are considered; according to the method, the service datacorresponding to various service factors are acquired, and the service data are combined and analyzed by using the machine learning algorithm to obtain the project budget balance prediction result, sothat the accuracy of budget balance judgment is improved, and the prediction difficulty and complexity are reduced compared with a traditional project budget balance clearing mode through manual operation.

The invention discloses a lottery user product participation prediction method, which comprises the following steps: obtaining original user data, extracting and converting the original user data, andloading the original user data into a database in a specified format according to classification; preprocessing the original user data stored in the database to obtain multi-dimensional user data; obtaining a prediction feature set related to a user product participation degree according to the multi-dimensional user data; inputting the prediction feature set into a pre-trained fusion predictionmodel to predict the user product participation degree, wherein the fusion prediction model is at least generated by fusion of a Bayesian classifier, a stochastic forest classifier and an iterative decision tree classifier. Correspondingly, the invention also discloses a lottery user product participation prediction system, a terminal device and a computer-readable storage medium. The technical proposal of the invention can reduce the prediction difficulty of the participation degree of the lottery user product and improve the prediction accuracy rate.

The invention relates to the technical field of image processing, and provides an image super-resolution system based on empirical mode decomposition, and the system comprises an input module which is used for obtaining a first image; the first image is a low-resolution image; the feature extraction module is used for extracting features of the first image; the IMF prediction module is used for predicting a plurality of IMF feature maps according to the features of the first image; the plurality of IMF feature maps are located at different frequencies; the IMF prediction module comprises a plurality of parallel branches, each branch is a CNN filter bank, and the number of the branches is the same as that of the IMF feature maps; the reconstruction module is used for converting each IMF feature map into a new IMF according to a set amplification proportion to obtain a plurality of new IMFs; and superposing the plurality of new IMFs to obtain a second image, wherein the second image is a super-resolution image. According to the technical scheme, the problem of low iris recognition precision caused by low image resolution in the prior art is solved.

The present application relates to a power load forecasting method, device, computer equipment, and storage medium. The method includes: responding to a power load forecast request, acquiring power load-related characteristic data corresponding to the power load forecast request, including historical power load and further divided into characteristic data related to domestic electricity load and characteristic data related to production electricity load; then input the pre-built electricity load difference prediction model to obtain the predicted difference of domestic electricity load corresponding to the forecast time information and The production power load forecast difference, and the historical power load value is calculated and processed to obtain the power load forecast result. In this application, through the characteristic data related to the domestic electricity load and the characteristic data related to the production electricity load, the use of the model is used to perform separate predictions, which improves the precision of the electricity load prediction; variance, which improves the accuracy of electric load forecasting.

The invention relates to an electric energy measuring error estimation method. The method comprises the following steps: preprocessing metering data measured by an electric energy metering device to obtain a metering error sequence; performing wavelet conversion decomposition on the metering error sequence to obtain multiple groups of error sequence components; constructing a robust extreme learning machine model, regarding each group of error sequence component and the measured metering data as the input quantity to add in the robust extreme learning machine model as the input quantity, thereby obtaining an error prediction value of each group of error sequence component; adding the obtained error prediction values to predict the next generation of error value; preprocessing the error value to obtain the new metering error sequence, updating the measured metering data, performing iterative updating and cyclically predicting. Through the electric energy metering error estimation methodprovided by the invention, the rule in the original signal is extracted through the application of the wavelet conversion decomposition and the robust extreme learning machine model, the prediction difficulty is greatly reduced, the prediction precision and operation speed under the existence of the noise condition are improved.

The invention discloses a discretized transformation supercritical boiler heating surface tube wall overtemperature early warning method, and the invention relates to a supercritical boiler heating surface tube wall overtemperature early warning method. The present invention aims to solve the problems of high production process requirements, inability to prevent over-temperature, excessively frequent replacement or upgrading of pipelines, which will bring huge economic burden to the user unit, and difficult to popularize in practical application by simply predicting the temperature, and proposes a Pre-warning method for over-temperature of supercritical boiler heating surface tube wall based on discretized transformation. The method is through 1. Establishing the historical data table HistoryTable; 2. Outputting variables from dc1 to dc17; 3. Outputting variables from gc1 to gc17; 4. Outputting dmark variables; 5. Obtaining a classification model Model; Whether the wall is about to overheat and carry out early warning and other steps to achieve. The invention is applied to the field of overtemperature early warning of the tube wall of the heating surface of the supercritical boiler.

The invention discloses a method for selecting time-series data of electromagnetic radiation intensity underground in a coal mine and predicting a trend, comprising the following steps: sampling the time-series data of electromagnetic radiation intensity underground in a coal mine to obtain a sample data set; Obtain multiple data subsets; determine the importance of sample points in each data subset based on the Hausdorff distance, and remove or retain each sample point according to the importance to obtain a forecast data set; predict the coal mine according to the forecast data set The change trend of downhole electromagnetic radiation intensity is predicted. According to the selection and trend prediction method of time-series data of electromagnetic radiation intensity in underground coal mines of the present invention, the computational complexity of trend prediction is reduced, and the robustness of prediction results is enhanced.

The invention relates to the field of three-dimensional seismic inversion, in particular to a method for converting horizontal well data into plumb shaft data and a method for seismic inversion. According to the method, the parameters are quantified, the horizontal well data is effectively converted into the plumb shaft data, the converted plumb shaft data is combined with a depth domain inversiontechnology, so that the prediction precision and the longitudinal resolution of the three-dimensional seismic reservoir are effectively improved, and the difficulty of reservoir prediction and development is reduced.

The tail cover separation mechanism comprises a first separation tail cover and a second separation tail cover, and the first separation tail cover comprises a first cover body, a first limiting ring, a first annular supporting piece and a first connecting lifting lug; the first cover body is semi-cylindrical; the first limiting ring is positioned at the front end of the first cover body and is integrally connected with the first cover body; the first limiting ring is in transition fit with a limiting clamping groove in the tail of the navigation body, and the first limiting ring is connected with the limiting clamping groove through a first explosive bolt; the first annular supporting piece is installed in the first cover body. The first connecting lifting lug is fixed at the tail end of the first cover body; the second separation tail cover and the first separation tail cover are the same in structure, the second separation tail cover and the first separation tail cover are installed at the tail of the navigation body in a matched mode, and key components such as a tail vane and a propeller of the navigation body can be effectively protected when the navigation body is impacted by tail beating force in the water entering process.

The invention relates to the field of intelligent driving, specifically discloses a driving assistance system-oriented dangerous target detection method and device, and aims at solving the problem oflow dangerous target detection precision under field complicated traffic scenes. For the aim, the dangerous target detection method comprises the following steps of: identifying a detection frame position of a dangerous target in an obtained vehicle body external image and a Cartesian product of a target type and a distance type according to a preset dangerous target detection model; obtaining a danger level of the target according to the Cartesian product and a danger level matching table; and marking a detection frame of the target in the image according to a color corresponding to the danger level. Meanwhile, the dangerous target detection device, a dangerous target storage device and a dangerous target processing device can execute the steps in the dangerous target detection method. Through above technical scheme, the accuracy of image-based dangerous target detection can be enhanced under real complicated traffic scenes, so as to prevent traffic accidents.

The invention discloses a multi-label classification method for image data. Decompose the input image, use the neural network to extract the high-order correlation of the features, decompose the label data, use the neural network to extract the high-order correlation of the label, and use the neural network containing multiple fully connected layers to convert the feature code of the input image Decode from the input space to the label space; construct a loss function, initialize the training parameters, use the stochastic gradient descent method to minimize the final loss function as the goal, and train and solve to obtain the optimal training parameters; then input the image data to be tested into the trained The prediction is made in the model, and the label result is output to realize multi-label classification. The invention solves the problem that previous work on image data cannot simultaneously extract the secondary correlation and multiple correlation of tags, reduces the prediction difficulty caused by too sparse image data, and improves the accuracy of multi-label classification.

The invention relates to an image-based displacement measurement method and system, and belongs to the field of image processing. The method comprises the following steps: constructing a detection network based on a central anchor point; the basic network of the detection network is Faster-RCNN, and an anchor frame generation strategy based on a center anchor point, an area intersection-to-union ratio calculation mode based on the center anchor point and a regression loss function based on the center anchor point are adopted; the central anchor point refers to the centroid coordinate of the target, the anchor frame based on the central anchor point is a square position frame with the central anchor point as the center, and the marking information of the anchor frame comprises the coordinate of the central anchor point and one half of the side length of the square position frame; training the detection network based on a regression loss function to obtain a marker detection model; and inputting the marker images collected at the two moments into a marker detection model to obtain a displacement measurement result of the marker. The precision of displacement measurement can be improved.

The invention discloses a multi-layer cooperative real-time classification early warning method for a wind power climbing event. The method comprises the following steps: 1, obtaining wind power historical data for preprocessing; 2, formulating a wind power climbing event classification strategy; 3, establishing a multi-layer collaborative prediction model of a decomposition layer, a prediction layer, a correction layer and a feedback layer; and 4, identifying different types of wind power climbing events to carry out real-time early warning. According to the method, EMD decomposition, GRU prediction, SVR correction and multi-step rolling prediction of actual measurement information feedback are utilized, real-time early warning is carried out after different types of wind power climbing events are recognized according to the wind power climbing event classification criterion, corresponding measures can be taken in time before the climbing events happen, and therefore safe and stable operation of a power system is guaranteed.