117results about How to "Optimize forecast results" patented technology

The invention relates to a short-term load prediction method based on a clustering and sliding window. The method comprises the following steps of: preprocessing electric power load data; clustering historical data of a prediction user by utilizing a clustering algorithm, and adjusting clustering parameters; selecting k data from near to far of the prediction time in a category, containing most data, in clustering results to form a sliding window k; predicting the k selected data by utilizing a combination model based on the sliding window, and acquiring a primary prediction result; and correcting the primary prediction result of the combination model according to meteorological factors to obtain a final load prediction result. Compared with the prior art, the method has the advantages of high prediction precision, good adaptability and the like.

The invention discloses a wind power station power projection method based on error statistics modification. The procedures adopted by the invention is that firstly, data pretreatment is performed, the projection and measured data and the statistical data of the wind power station are divided into a plurality of sample sets according to different terrain heights; secondly, the sample sets are used for the training of a wind power projection model so as to form a 24-hour output short-term forecast model of the wind power station, the error distribution condition of the power can be projected through analyzing the projected power value and measured power value so as to obtain the expected value of the power projection error of the wind power station; thirdly, the power projection value of the wind power station can be obtained according to the numerical weather prediction data, weather forecast data and the trained model; and finally, the final power modification value of the wind power station can be obtained through modifying the power projection value according to the projection error statistical calculation result of the wind power station.

The present invention belongs to the field of power prediction of wind power generation and particularly relates to a method for predicting a wind power interval based on a particle swarm optimization nucleus limit learning machine model. The method comprises: carrying out data preprocessing, i.e. preprocessing historical data in SCADA according to correlation between a wind speed and power; initializing a KELM model parameter and carrying out calculation to obtain an initial output weight betaint; initializing a particle swarm parameter; constructing an optimization criterion F according to an evaluation index and carrying out particle swarm optimization searching to obtain a model optimal output weight betabest; and bringing test data into a KELM model formed by betabest to obtain a wind power prediction interval and evaluating each index of the prediction interval. The method is easy for engineering realization; a good prediction result can be obtained; not only can a future wind power possible fluctuation range be described, but also reliability of the prediction interval is effectively evaluated, possible fluctuation intervals of wind power at different confidence levels are given out and reference is better provided for a power system decision maker.

The invention discloses a method for completing and predicting an attribute missing data set based on a generative adversarial network. The method comprises the following steps: 1) normalizing data minmax and using one hot code for discrete attribute, wherein the missing value is marked as 0; 2) using that data set to establish a missing position coding vector with respect to the sample; 3) constructing a generative antagonistic network and an auxiliary prediction network for data filling and label prediction; 4) restoring that result before minmax normalization according to the maximum and minimum value in the attribute; 5) selecting a suitable super parameter through testing. The data distribution information and the label information in the data set are fully utilized, and at the same time, another auxiliary prediction network included in the method can directly input attribute missing data to give the prediction result of the label after the training is completed. The process is simple and has higher prediction accuracy.

The invention relates to a training method and system of a commodity personalized ranking model. The training method comprises the following steps: according to long-term interest characteristics in historical commodity data, carrying out off-line training on the commodity personalized ranking model, obtaining a parameter corresponding to each long-term interest characteristic, i.e. obtaining the commodity personalized ranking model with high precision, and eliminating short-term characteristics in the historical commodity data to reduce time consumption; and at a unit time interval, obtaining real-time commodity data, expanding the commodity personalized ranking model subjected to the off-line training, and carrying out on-line training on the expanded commodity personalized ranking model according to the long-term interest characteristics and the short-term characteristics in the real-time commodity data to obtain an updated parameter corresponding to each long-term interest characteristic and an updated parameter corresponding to each short-term interest characteristic. Therefore the expanded commodity personalized ranking model is updated for one time at the unit time interval to obtain the model with higher timeliness and realize the balance of the precision and the timeliness of the model so as to obtain a better prediction result.

The invention discloses a method for labeling and complementing gastric cancer pathological slices based on pseudo-label iterative annotation. The method comprises the steps that 1), pseudo-label samples are produced by using the original positive samples and the original negative samples of the gastric cancer pathological slices; 2), image segmentation is conducted on the pseudo-label samples, and the pseudo-label samples are used as training images and transmitted to U-Net to be trained; 3), data augmentation is conducted on the original positive samples and transmitted to the trained U-Netin step 2) to be tested, reduction is conducted based on an augmentation manner, and finally weighted averaging is conducted on all images and the images are integrated to obtain a gastric diseased probability graph; 4), the parts of which the gastric cancer diseased probability is higher than a threshold value are screened out, extracted and spliced to the original negative samples to generate the pseudo-label samples of the next iteration; iteration is constantly conducted on the above processes to finally obtain the gastric cancer pathological slices which are completely annotated. By meansof the method, human resources needed to be consumed by slice annotation are greatly reduced, the quantity and quality of a training data set are improved, and probability is provided for training amore accurate deep learning model.

The invention provides a welding defect detection method and device, electronic equipment and a storage medium, and relates to the field of defect detection. The method comprises the steps: constructing a sample data set, wherein the sample data set comprises a plurality of preprocessed images containing welding spot areas; constructing a multi-task learning network model, wherein the multi-task learning network model comprises at least two sub-networks; employing at least two sub-networks for detecting defect position information and predicting defect classification information at the same time; training a multi-task learning network model by utilizing the sample data set; inputting a prediction result output by the trained multi-task learning network model into a decision fusion model; and finally, collecting the welding spot area graph, inputting the collected image into the trained multi-task learning network model, and outputting a prediction result according to the weight determined by the decision fusion model. The method has the advantages of being convenient to test and high in precision.

The invention relates to a device for assisting judicial case judgment based on machine learning, which utilizes a large amount of document data and trains a model to learn the relationship between case fact description and the fine range and relevant legal provisions, and realizes the prediction of the fine range and the law label of any given case fact description text. The invention relates toa device for assisting judicial case judgment based on machine learning. Including: defining the proper nouns in the description of the facts of a given case and dealing with them; Extracting multiplesemantic features from the text to achieve a deeper level of semantic representation; Machine learning method based on multi-label classification is used to classify the law items and obtain the lawlabels related to the description text of the case facts. Single-label classification training model based on machine learning predicts the range of possible fines in related cases. The invention applies machine learning to the judicial field for the first time, realizes deeper semantic representation by multiple feature extraction modes, improves the accuracy and generalization ability of the training model well, has higher reference significance for the final judgment of a case, and is conducive to the realization of the same case and the same judgment.

The invention discloses a monocular image depth estimation method based on a pyramid pooling module. In a training stage, a neural network is firstly constructed, which comprises an input layer, a hidden layer and an output layer. The hidden layer includes a separate first convolution layer, a feature extraction network framework, a scale recovery network framework, a separate second convolution layer, a pyramid pooling module, and a separate connection layer. Each original monocular image in the training set is used as the original input image. The optimal weight vector and the optimal bias term of the trained neural network model are obtained by calculating the loss function value between the predicted depth image and the real depth image corresponding to each original monocular image inthe training set and inputting it into the neural network for training. In the testing phase, the monocular image to be predicted is input into the neural network model, and the predicted depth imageis obtained by using the optimal weight vector and the optimal bias term. The advantages are high prediction accuracy and low computational complexity.

The invention relates to a complex attack identification technology for a wireless intrusion detection system. The complex attack identification technology for a wireless intrusion detection system comprises the following steps: step 1, data acquisition; step 2, single step attack identification; step 3, complex attack identification; step 4, information display interface. The beneficial effect of the invention comprises: 1) the hybird architecture is applied to the identification of the single step attack, the identification accuracy of the single step attack is enhanced, and the false positives and false negatives are reduced; 2) the evaluation mechanism is used for identifying the complex attack, and the simple yes or no is not used for judging the single step attack behaviors, so that the attack sequence is generated, therefore, the distortion of the final result due to the information distortion among the modules can be effectively reduced; 3) an algorithm for identifying the complex attacks and predicting the final intention of the attackers is designed, which has better prediction results for complex attack intention with obvious characteristic behaviors.

The invention belongs to the technical field of image processing, and discloses a character image automatic segmentation method based on deep learning and an information data processing terminal. Themethod comprises: collecting character pictures to form a training data set; constructing a deep neural network model of first-level image semantic segmentation; inputting the collected training dataset into a first-stage deep neural network to generate a trimap; constructing a second-level deep neural network model; inputting the collected training data set and the obtained trimap into a second-level deep neural network to generate a segmented character mask image; and synthesizing the character mask image and the figure original image to obtain a segmented character image. According to thecharacteristics of the character image, the character and the image background of the character image are automatically segmented, characters in the image are automatically screened, and the characters from the background picture are separated in combination with character features. The method can be used for automatic character matting, and can also be used for character photo background replacement and background processing through background blurring.

The invention provides a landslide displacement prediction method based on wavelet transform-rough set-support vector regression (WT-RS-SVR) combination. By means of the method, according to the characteristics of the influence factors of landslide displacement, the complex displacement process and landslide displacement monitor data measured in real time, accumulative displacement of a typical monitor point is decomposed into trend-term displacement and periodic-term displacement through WT, and a trend-term displacement prediction function is obtained through curve fitting; screening is conducted on the influence factors of the landslide displacement through an RS algorithm, and selected factor sets are used as input factor sets of an SVR machine, accordingly a landslide displacement optimization prediction model based on WT-RS-SVR combination is established, and the precision of a prediction result is analyzed and evaluated. The prediction result of the landslide displacement prediction method can well embody the development and change tendency of the landslide displacement. The landslide displacement prediction method has high prediction capacity, and is accurate, effective and practical.

The invention discloses an urban traffic road section speed prediction method and a system based on multi-road-section space-time correlation. The method comprises the steps of acquiring the speeds ofthe latest p historical time points of all road sections in an optimal feature subset corresponding to a to-be-predicted road section; wherein p is a positive integer; and inputting the acquired speeds of the latest p historical time points into a pre-trained GRU neural network, and outputting the prediction speed of the (p + 1) th time point of the to-be-predicted road section. The prediction model quantitatively and dynamically considers space-time correlation among all road sections in a road network from the aspects of traffic parameters, road section connectivity, road grades and the like, and can select a road section subset which is beneficial to speed prediction of a road section to be predicted from the road network. The prediction model can realize accurate prediction of the speed of the urban traffic road section.

The invention discloses a residual service life prediction method for large-scale equipment based on multi-parameter feature fusion. The method comprises the steps: obtaining multiple sensor time sequence parameters of large-scale equipment in a laboratory through a large-scale online monitoring system; performing regression analysis on the multi-parameter continuous values by using a ReliefF algorithm, and obtaining a parameter type with relatively large correlation with the equipment state through feature weight screening; performing data dimension reduction and feature extraction on the screened parameters based on a principal component analysis method, and obtaining a health index representing the operation state of the large-scale equipment through weight fusion; constructing an HMM model based on an expectation maximization algorithm, taking the health indexes as a training set for model training, and finding a grading model used for evaluating the equipment health state corresponding to the current health index; calculating an exponential likelihood value through a Viterbi algorithm to obtain a health index nearest to the likelihood value, and predicting an exponential difference by using a weighted average method to obtain a health state fitting curve; and calculating a residual service life prediction value of the large-scale equipment.

The invention provides a three-dimensional human body posture estimation method and a computer readable storage medium. The method comprises the following steps: acquiring a single-person image from an original image by adopting a human body detection network and carrying out standardization processing on the single-person image; predicting two-dimensional coordinates of the key points from the single-person image by using a two-dimensional attitude estimation method; generating a three-dimensional coordinate from the two-dimensional coordinate, including: predicting a first three-dimensional coordinate of the key point by using a three-dimensional attitude generator; performing symmetric processing on the two-dimensional coordinates according to a symmetric structure of a human body joint, and predicting second three-dimensional coordinates of the key points by using a three-dimensional attitude generator; enabling the first three-dimensional coordinate and the second three-dimensional coordinate to respectively calculate difference values with the corresponding labels, the sum of the results bing used for back propagation, and obtaining three-dimensional human body posture estimation. A connection relation and a symmetric relation between key points of a human body are fully utilized, and the purpose of optimizing a prediction result can be achieved; and meanwhile, on the basis of an original data set, the expansion of training data is realized, and the robustness of the model is enhanced.

The invention discloses a wind speed prediction method along a railway line considering wind direction and a confidence interval. A plurality of low-correlation wind speed prediction models are established by using historical wind speed data, the intelligent integrated optimization prediction results are optimized, the prediction accuracy is improved, and the wind direction is established at the same time. Wind speed prediction error joint probability distribution, combined with the wind direction prediction value of the target anemometry point, to obtain high confidence wind speed predictioninterval; the wind speed prediction error of each simulation and the wind direction real value at the corresponding time are taken as the first observation value of two-dimensional discrete random variable to construct the wind direction.The joint probability distribution of wind speed prediction error is used to establish the mapping relationship between wind direction and prediction error. Basedon wind direction, the high confidence interval of wind speed prediction error is obtained, which significantly improves the robustness of wind speed prediction, avoids the singularity of absolute wind speed prediction, and provides more accurate and effective prediction information for train operation decision-making.

The invention discloses a method for predicating interval probability of short-term wind power. The method comprises the following steps: acquiring a number of historical wind power from a wind power plant as a sample set; establishing optimization criteria according to the prediction interval coverage probability, the prediction interval bandwidth mean square root and the prediction interval average offset; establishing the interval predicating model of the short-term wind power based on an artificial bee colony nerve network, optimizing and updating a nerve network weight threshold to the optimization criteria through an artificial bee colony algorithm; according to the optimal weight threshold, establishing a nerve network and performing interval predication to the wind power to be predicated; performing state division to the historical wind power, establishing a Markov chain prediction model, and calculating the transition probability of each status; predicating the wind power interval according to the Markov chain status transition probability and the interval predication, and calculating the probability of the numerical point in the predication interval. When the short-term wind power interval predication is executed, the probability distribution of the numerical point in the interval is considered, thus the method can provide the basis to an optimization system.

The invention discloses a nuclear extreme learning machine quantile regression-based wind power interval prediction method. The method comprises the steps that a wind power plant output power and windspeed data is collected; the data is processed simply, and unreasonable data is deleted; a nuclear extreme learning machine quantile regression model is built; by means of a particle swarm algorithm,nuclear extreme learning machine quantile regression parameters are optimized, and a regression module is determined; test data is put, and a wind power prediction interval is obtained. Accordingly,the nuclear extreme learning machine quantile regression principle and a nuclear extreme learning machine model are effectively combined, and the optimal model parameter is obtained by conducting search and optimization through the particle swarm algorithm, uncertain information in wind power can be effectively grasped, then a better prediction result is obtained, and the basis can be provided forsafe and stable running of wind power integration.

The invention discloses a language model pre-training method. The language model pre-training method comprises the following steps of carrying out word segmentation on the corpora in a model accordingto characters and sub-words; extracting 15% of the generated segmented words immediately for position covering, and calculating the covered semantic distribution; controlling the sub-word mixing in the model by using an independent door control unit; and performing synchronous training on the semantic distribution and the prediction of the masking words. According to the method, the prediction result of the model after the BERT pre-training can be obviously improved.

The invention discloses a reservoir parameter logging interpretation method based on a regression committee machine. The method includes the following steps of 1, selecting logging data sensitive to to-be-predicted parameters as input; 2, normalizing each input attribute value; 3, obtaining reservoir parameters according to results of rock physical experiments; 4, combining the logging data of each layer with experimental data of the reservoir parameters to form a data set; 5, randomly dividing the data set into a training data set body and a test data set body; 6, selecting several intelligent regression algorithms to serve as front-mounted regression predictors; 7, using the training data set body as input and training each intelligent regression algorithm to obtain a corresponding regression prediction model; 8, using the test data set body as input and setting a prediction value by each prediction model; 9, for each set of input data, combining the prediction values set by the regression prediction models, and adopting a committee decision mechanism for setting final prediction values.

The invention discloses a narrow-band discrete distribution parabolic equation method for forecasting ASF with high precision. The method specifically comprises the following steps: firstly, sampling a loran-C current time-domain signal; performing discrete fourier transform on the sampled signal; decomposing the sampled signal into a plurality of frequency current components; adopting a flat ground formula for calculating a magnetic field irradiated by each frequency current component in a near zone; calculating a far zone magnetic field by regarding a magnetic field result at a boundary of a near-far zone as an initial field for a discrete distribution parabolic equation method for an uneven grid part, thereby acquiring the magnetic field generated by each frequency current component on the earth surface; and lastly, adopting fourier inversion on the basis of sliding window thought, thereby acquiring a time-domain magnetic field signal recovered by the frequency domain magnetic fields irradiated by the frequency current components on the earth surface. The method provided by the invention can overcome the defect that the present theory is difficult to forecast practical long-distance loran-C signal ASF distribution. Compared with the existing frequency domain method, the method has the advantage that the forecast precision is obviously increased and has the characteristic of high practicability.

The invention discloses a reservoir fluid identification method by using logging data based on a committee machine. The method comprises the following steps: 1) selecting logging data as input data; 2) normalizing the input data; 3) according to an oil testing result, obtaining a reservoir fluid type; 4) making the logging data and the fluid type to form a data set; 5) randomly dividing the data set into a training data set and a test data set; 6) selecting pre-classifiers; 7) training each pre-classifier to obtain a corresponding classification model; 8) using the test data set as an input togive a category through each classification model; 9) for each set of input data, combining the categories given by each classification model, and adopting a committee decision-making mechanism to give a final classification type. The method simulates the decision-making mechanism of the committee and combines a plurality of the pre-classifiers to reduce falling into the local minimum and make the decision of the committee more scientific and accurate.

The invention discloses a roller kiln temperature prediction integrated modeling method capable of combining a mechanism with data. The method comprises the following steps that: through the analysis of factors which affect temperature change, from a perspective of the temperature change and energy, establishing a mechanism model; then, considering situations that roller kiln sintering is a very complex process, a whole sintering process can not be described through a single mechanism model and the mechanism model has model errors through simplification, establishing a data model to predict model errors so as to make up mechanism output, i.e., utilizing errors to serve as a training sample to establish an error prediction model of a nonlinear time-varying process based on local weighted kernel principal component regression; and finally, combining the mechanism model with the data model to establish a roller kiln temperature prediction integration model. By use of the model, the state change of a process can be better tracked, and a good guidance function is provided for roller kiln temperature control so as to improve the production quality and the percent of pass of a product.

An over-the-air (OTA) mobility service platform (MSP) is disclosed that provides a variety of OTA services, including but not limited to: updating software OTA (SOTA), updating firmware OTA (FOTA), client connectivity, remote control and operation monitoring. In some exemplary embodiments, MSP is a distributed computing platform that delivers and / or updates one or more of configuration data, rules, scripts and other services to vehicles and IoT devices. In some exemplary embodiments, the MSP optionally provides data ingestion, storage and management, data analytics, real-time data processing,remote control of data retrieving, insurance fraud verification, predictive maintenance and social media support.

The invention discloses a prediction method and system based on a forest discrimination model. The method includes the steps that first, based on a random forest algorithm, modeling is performed through modeling data, the modeling data are binned, and the modeling result is obtained through solution of the discrimination model; second, according to the established model and the modeling result, data to be predicted are graded, and the prediction result is obtained. Through the prediction method and system based on the forest discrimination model, a simple, extended and well-founded binning method can be provided, evaluation of data is simplified, a good binning result can be obtained with the method while an advanced operational theory does not need to be mastered, and thus a prediction result with good effect is obtained. The modeling efficiency and the precision of the model are improved to a large extent.

The invention discloses a medical image focus detection modeling method, device and system based on federated learning, and the method comprises the steps: enabling a global server S to transmit a generated global parameter [omega]<0> to local focus recognition clients C<k>; generating a global parameter [omega]<theta+1> by using detection head network parameters returned by the K local focus identification clients C<k>; and sending the global parameter [omega]<theta+1> to each local focus identification client C<k> to obtain a corresponding medical image focus detection model. The training information of the intermediate model is transmitted to a data holder through codes, respective data information does not need to be shared, and the model is integrated through a corresponding strategy, so that better training and prediction results are returned.

The invention relates to a depth map encoding / decoding method based on difference sampling blocks, and belongs to the technical field of the depth map in the three-dimensional video encoding standard.The method comprises the edge detection, macro block segmentation, intra-frame prediction mode selection, intra-frame multi-resolution macro block mode encoding and decoding end; the block reconstruction is performed on the decoded code stream, and the upper sampling processing is performed on the reconstructed block according to an encoding rule. The depth map encoding / decoding method has the advantage that a depth map encoding method based on different sampling blocks for the feature that the depth map only contains object contour information, a good prediction result is acquired by performing self-adaptive geometric division on the depth macro block containing discontinuous motion fields, thereby improving the depth map encoding efficiency, and reducing the complexity of the depth mapencoding.

The invention provides an atmospheric pollutant concentration prediction method integrating machine learning with LSTM. The method comprises the following steps: obtaining atmospheric pollution monitoring data, selecting training data and test data, and completing the data preprocessing; for the prediction target and the training data, constructing an atmospheric pollutant concentration predictionmodel fusing machine learning and LSTM; inputting the training data into a prediction model, and training the prediction model; inputting the test data into the trained model to obtain a prediction result of the test data; analyzing the accuracy of a prediction result of the test data, if the accuracy meets the requirements, performing model fusion and prediction, and if the accuracy does not meet the requirements, adjusting model parameters, and then performing model training. The method has the advantages of simple data, fast calculation speed, full consideration of atmospheric pollutant data, extraction of the time and space distribution characteristics of the data, and high prediction precision.

The invention discloses a network coupling time sequence information flow prediction method based on causal logic and graph convolution feature extraction. The method comprises the following steps: establishing a causal logic network by using transfer entropy based on node time sequence data, performing feature extraction on logic network node data by using a graph convolutional network (GCN), performing flow prediction by using a gated cycle unit (GRU) on the basis of a graph information feature h, and performing training optimization on parameters by using a back propagation algorithm. By adding the logic information of the causal network and combining the graph convolution network and the gating cycle unit, the information flow prediction precision is improved.