113results about How to "Improve forecast results" patented technology

The invention discloses an establishing method of an uncertainty mid-term and long-term hydrological forecasting model, comprising the following steps: using a wavelet analysis (WA), an artificial neural network (ANN) and a hydrological frequency analysis (HFA) in combination to establish the uncertainty mid-term and long-term hydrological forecasting model; dividing the original sequence into two sections of a main sequence and a random sequence on the basis that WA is used to reveal multiple time dimension variation characteristics of the hydrological sequences, adopting ANN for analogue forecasting on the main sequence and hydrological frequency analysis on the random sequence and overlapping the results of the two sections to be a final forecasting value. The model is used for the mid-term and long-term hydrological forecasting in the Yellow River estuary area, and compared with the traditional method, the results show that the model can reveal the time and frequency structures and variation characteristics of the sequences, has high forecast value result precision and acceptance rate, can quantitatively analyze and describe the impact of hydrological uncertain factors on the forecasting result and can obtain the analogue forecasting value of different frequencies to corresponding hydrological sequences.

The invention discloses a behavior recognition method for learning a human skeleton of a neural network based on an end-to-end space-time diagram. The method is used for behavior recognition of a human 3D skeleton. The method specifically comprises the following steps: obtaining a human body 3D skeleton key point position data set for training, and defining an algorithm target; performing clustering expression on each frame based on the spatial position to obtain a spatial node relation; calculating a time track of each joint point, and performing relation measurement according to the time track to obtain a time node relation; establishing a joint learning framework of the space-time diagram learning and the diagram convolutional neural network; and estimating the behavior category of thecontinuous human body 3D skeleton by using the learning framework. The method is suitable for human body action analysis in a real video, and has a good effect and robustness for various complex conditions.

The invention discloses a method for protecting privacy in a federated learning prediction stage based on a PSI technology. The method comprises the following steps that firstly, a prediction serviceparty calculates a prediction result of a model of the prediction service party, then the prediction service party and the prediction service party execute an improved PSI protocol, a part of model calculation results of the prediction service party are encrypted, a prediction demand party decrypts the calculation result of a data provider in combination with data of the prediction demand party, and finally a prediction result of a keyword id shared by the prediction service party and the prediction demand party is obtained. According to the method, the PSI technology is utilized, the data areencrypted through the key derivation function, the privacy protection requirement in the federated learning prediction stage is met, the last link of federated learning privacy security protection isbroken through, and landing of more application scenes of federated learning is promoted.

The invention discloses a friend relationship mining method combining network topology characteristics and user behavior characteristics. The method comprises the following steps of (1) establishing a friend relationship network diagram and randomly selecting 90% of friend relationship connection edge data as a training set and the rest of 10% as a test set; (2) constructing two weighted undirected graphs of a friend relationship network based on topological similarity; (3) constructing two weighted undirected graphs of a friend relationship network based on user behavior characteristic similarity; and (4) utilizing a community detection algorithm (CNM algorithm) based on weighting modularity to respectively carry out community division on the four weighted undirected graphs, if any two users are classified as one community in a division process of the four communities for at least three or more times, considering that the two users are friends. The method introduces the topology characteristics and the behavior characteristics are introduced into the user friend relationship networks and mines whether the two users are the friends through community division.

The invention discloses a coauthor recommending method under a scientific and technical literature heterogeneous network. Probability that a pair of authors establish a cooperation relationship in the future is in direct proportion to willingness that two nodes establish the cooperation relationship with each other, so that the coauthor recommending based on cooperator willingness is provided, and cooperating authors are recommended to authors by calculating probability that two authors cooperate in the future. The coauthor recommending method includes defining attention degree of the nodes according to future influence increase degree and rate of the nodes and ages of the nodes; designing cooperation willingness of two authors due to different relationships on the basis of the attention degree; designing topological feature attributes under different relationships on the basis of the willingness of different relationships; taking the topological feature attributes as an independent variable of a logic regression model, utilizing parameters of a real data training model, and using an acquired function model to calculate probability that the two authors cooperate in the future. The coauthor recommending method takes willingness of two authors which cooperate into consideration, and the willingness is based on node influence and age, so that prediction results of cooperation relationships are improved, and recommending quality of coauthors of authors is improved.

The invention discloses a glomerular cell image recognition method based on a deep neural network, and the method comprises the steps: obtaining a to-be-detected pathological image based on an artificial intelligence and deep learning technology; preprocessing the pathological image to obtain a plurality of slice images; inputting each slice image into a preset neural network model for identification and segmentation to obtain a glomerular region map; carrying out cell counting on the glomerular region map; Sub-images of the glomerulus in the kidney in the pathological image can be quickly andaccurately segmented, and cells in the glomerulus are counted by using a traditional method and a deep learning fusion model, so that the problems of large workload, low efficiency and high misdiagnosis rate of artificial identification of the glomerulus in the pathological image are solved; the invention optimizes the algorithm of glomerular sub-image segmentation and glomerular cell counting inpathological images, uses more data training algorithms, improves the accuracy of segmentation and counting, and relates to the field of biomedical image processing.

The invention discloses a method of recommending hit songs and singers in a music on-demand network. Considered are the fact that future heat of each node of the music network needs to meet the change law of historical heat of the node and the fact that the heats of different types of nodes are kept in certain relations; future heats of songs and singers are acquired through a function model prediction phase and a coordinated adjusting phase, and on such basis, the top songs and singers are recommended; in the function model prediction phase, a function model of historical heat time sequences of the nodes is designed and used for primarily predicting the future heats of the nodes; in the coordinated adjusting phase, average edge betweenness of different relational edges of the music on-demand network is counted, and adjusting factors are calculated to adjust the primary future heats obtained in the function model prediction phase. The method has the advantages that considered are both the historical heat information of the nodes and the heat relation of the different types of nodes, ranking prediction results of the future heats of the nodes is improved, and singers and songs are recommended with better quality.

Aiming at improving the prediction accuracy of the summer precipitation in northeast China and further improving the climate service capacity, the northeast summer precipitation multi-mode combined downscaling prediction method is developed. The method takes advanced climate mode forecast information and early stage actual condition factor information both at home and abroad into consideration, combines a mode error correction technology, and adopts a singular value decomposition method to respectively establish a coupling type relation between a small-scale northeast summer precipitation field, a large-scale mode summer circulation forecast field and a large-scale early stage external forced actual condition field, thereby establishing a northeast summer precipitation forecast model. Theoptimal northeast summer precipitation multi-mode combined downscaling forecasting model with regional characteristics is obtained through the comparison and inspection of multi-mode and multi-schemeforecasting effects. The northeast summer precipitation multi-mode combined downscaling prediction method can effectively improve the northeast summer precipitation prediction accuracy rate of climatemodes both at home and abroad, and better provide technical support for government disaster prevention and reduction decisions.

The invention relates to relates to a BP neural network model based fault prediction method and system, and belongs to the technical field of fault prediction of the radar system. The fault predictionmethod comprises that state parameter normalized sample data of a transmitting-receiving assembly is input to a BP neural network model, a predicted value of a state parameter is calculated, comparison analysis is carried out on predicted and reference values of the state parameter, and a fault prediction result is output. The BP neural network model is used to predict the state parameter of a digital transmitting-receiving assembly, the state parameter value in next time can be predicted on the basis of existing sample data, the fault prediction result is obtained conveniently and rapidly, amaintenance staff can know the work state of the digital transmitting-receiving assembly timely according to the fault prediction result, support is provided for realizing predicative maintenance ofthe assembly, and the reliability of the radar system is improved.

The invention discloses a tunnel advance geology forecast method based on blast hole drilling information. The method comprises the following steps of A, establishing a standard database [Ki] of drillability indexes before the implementation of forecast; and B, implementing advance geology forecast by the following specific steps of: (1) numbering advance blast holes on a tunnel face; (2) recording acquired drilling rate v, drilling pressure p, drilling rod rotating speed w, rock slag, return water amount and return water color one by one; (3) calculating the drillability index K of each group of parameters according to a drillability index formula (7); (4) comparing the calculated drillability index k with [Ki], if K belongs to (min[Ki], max[Ki]), the basic surrounding rock grade of a drilling segment is grade i; (5) comprehensively judging the surrounding rock grade in combination with rock slag ingredients, water return condition and the surrounding rock condition of the tunnel face according to a table 2; (6) comparing and verifying in combination with a forecasted surrounding rock grade and a practical surrounding rock grade determined after tunnel excavation; and (7) expanding corrected data into [Ki]. The method is convenient to operate, is low in cost, is accurate and efficient, and can be used for performing real-time advance geology forecast.

The invention discloses a traffic prediction method based on a multi-scale graph convolutional network model, and the method comprises the steps: employing a multi-scale graph convolutional network and a GRU, and capturing the time dependence of a traffic network, i.e., the local time change trend of a traffic flow, and the spatial dependence, i.e., a topological space structure at the same time, and predicting the traffic flow of each road section in the future time step through the traffic flow of the historical time step, thereby accurately predicting the traffic flow of the road network. The method can effectively predict the temporal and spatial change characteristics and rules of the traffic flow, is high in prediction precision, and improves the traffic flow prediction effect.

The invention discloses an application method of a machine learning classification model in adolescent autism auxiliary diagnosis. The method is characterized by being implemented according to the following mode. The method comprises the steps of 1, establishing a model training method; 2, constructing a model evaluation index; 3, performing characteristic engineering of the autism auxiliary diagnosis system; 4, performing data dimension reduction processing; 5, carrying out feature selection; and 6, carrying out model training and result analysis. According to the invention, a machine learning method is introduced into the field of autism research; the high efficiency and the reliability brought by the method are greatly helpful to the auxiliary diagnosis of autism. The application fieldof the invention can be embodied in: (1) disease diagnosis and treatment, (2) smoking addiction, network addiction and network game addiction, (3) cognition and other health fields, and the like.

The invention relates to the technical field of artificial intelligence. The invention discloses a game formation intensity prediction method and device, electronic equipment and a storage medium, andthe method comprises the steps: obtaining formation information of a to-be-predicted formation, wherein the formation information comprises all game roles in the to-be-predicted formation and the position information of all game roles in a game scene; based on each game role, obtaining a role feature corresponding to the to-be-predicted formation; based on the position information of each game role in the game scene, obtaining role position features corresponding to the to-be-predicted formation, wherein the role position features comprise position features of each game role in the to-be-predicted formation and relative position features among the game roles; splicing the role features and the role position features to obtain spliced features; and based on the spliced features, obtainingthe formation intensity of the to-be-predicted formation. By means of the prediction method provided by the invention, the formation intensity of the to-be-predicted formation can be quickly and accurately obtained.

The invention discloses a pedestrian attribute recognition system and method based on multi-layer feature learning, the system comprises a feature bottom-to-top extraction module, a bottom-to-top feature fusion module, a feature prediction module, a multi-layer prediction fusion module and a test module, and the method comprises the following specific steps: processing pictures layer by layer frombottom to top to obtain multi-layer features; fusing the features of the adjacent layers layer by layer from top to bottom, compressing the channel by the feature map obtained by the higher layer, carrying out feature fusion and channel dimension reduction on the compressed channel and the feature map sampled by the upper layer, and outputting the feature of the current layer; obtaining preliminary prediction results of different levels through a maximum pooling layer and a full connection layer according to the fused features and the extracted uppermost features; overlapping the preliminaryprediction results of different levels, and correspondingly endowing each attribute predicted by each level with a weight value to obtain a final prediction result; and extracting a prediction resultcorresponding to the picture, and calculating a result of each index. According to the method, a group of specific weights are learned for each attribute according to the predicted values obtained bythe fused features, so that each attribute can better utilize multi-layer features to obtain a better recognition effect.

The invention discloses a sales prediction method and system, and relates to the field of sales prediction, and the method comprises the steps: obtaining a predicted category and corresponding time information; obtaining corresponding feature data according to the predicted category and the corresponding time information; and inputting the feature data into a preset model combination correspondingto the predicted category, and predicting the sales volume of the predicted category in the time corresponding to the time information. According to the invention, the system is directly adopted to replace manpower to predict the sales volume, so that the capability requirement on purchasing personnel is reduced; and a computer system is adopted for prediction, so that the dimensions and the datavolume considered during prediction are greatly increased, and a foundation is laid for accurate prediction. In addition, the sales volume is predicted by using a multi-model combination, and the precision is greatly improved compared with prediction by using only one model. And each category has a corresponding preset model combination, so that the matching degree is better, and the prediction result of each category is further improved.

The invention discloses a load prediction method based on industry-classified power utilization characteristic analysis, and the method comprises the following steps: S1, dividing power users into industry power utilization users and resident power utilization users, dividing the resident power utilization users into residence community users and small-capacity public transformers; S2, identifyingsaturated users and unsaturated users in industrial power utilization users and residential district users by using a load saturation identification technology; S3, taking the average load value of the saturated users as a future prediction value in recent three years; carrying out typical growth pattern analysis on unsaturated users and installation users; S4, summarizing and counting the natural growth rate of the small-capacity public transformer; S5, calculating the coincidence rate in the industry and the coincidence rate between the industries respectively; and S6, summarizing all typesof loads to obtain a whole-region load prediction result. Classification prediction is carried out on the basis of power utilization growth characteristic analysis of mass user load data, and then the simultaneous rates of all levels are applied to summarize and calculate the total load prediction result of the area.

The invention provides a method for manufacturing a seismic slice with an overthrust fault, which comprises the following steps of: carrying out explanation on a geologic horizon (with an overthrust fault) and a fault; connecting the geologic horizon with the fault so as to form a new horizon according to the explanation results of the geologic horizon and the fault; carrying out up-and-down horizon connection on the obtained new horizon at the position of the fault so as to form a closed fault block; and according to the formed fault block and the connected horizon, executing an operation ofslicing.

The invention discloses an equipment fault defect diagnosis method based on an improved U-Net neural network. The method comprises the following steps: constructing an improved U-Net network model which contains five down-sampling network layers and five up-sampling network layers and has triple constraints; wherein the triple constraints include three-level loss constraints on a fifth upsamplingnetwork layer located at the bottommost layer, two-level loss constraints on a fourth upsampling network layer adjacent to the fifth upsampling network layer and one-level loss constraints on a firstupsampling network layer located at the topmost layer. According to the invention, defect crack position and shape prediction and generation are carried out on an original image by using an improved U-Net neural network model; 3D display interaction is carried out through a WebGL method, the precision and correctness of equipment appearance crack pixel-level prediction are further improved, meanwhile, the picture experience feeling of man-machine interaction is remarkably improved, effective recognition and precise positioning of the equipment appearance defect crack position are facilitated,and the maintenance cost is reduced.

The invention discloses a credit score card model training method and a taxpayer abnormal risk assessment method. The risk assessment method provided by the invention comprises the following steps of:s1, obtaining a training sample; s2, obtaining initial features; s3, serializing the category features; s4, selecting continuous feature initial binning points; s5, selecting a continuous feature optimal binning point; s6, ensuring the monotonicity of a continuous feature binning result, s7, selecting features with high predictability for feature selection, s8, training the risk model by using alogistic regression model, and s9, predicting the abnormal risk of the taxpayer by using the learned taxpayer abnormal risk model. According to the method, the prediction result of the model is improved, potential risk taxpayers can be found in advance, tax authorities are helped to control invoice receiving and invoicing behaviors of the risk taxpayers in advance, and the cost of false invoicingof enterprises made by criminals is increased.

The invention provides a CAD prediction model building method and device and electronic equipment, and relates to the technical field of modeling, and the method comprises the steps: obtaining genotype data and physical condition data of a sample object; calculating by utilizing the genotype data to obtain a CAD multi-gene risk value; and establishing a CAD prediction model based on the CAD multi-gene risk value and the physical condition data. The technical problem of relatively low accuracy of data obtained by performing CAD prediction at present in the prior art is solved.

The invention discloses a deep learning network-based expressway freight volume prediction method and system. The method comprises the steps of obtaining input data of a model; constructing a highway network diagram, calculating a Dijkstra matrix, calculating a Pearson coefficient matrix of the entrance truck flow and the exit truck flow, and combining the Dijkstra matrix and the Pearson coefficient matrix to form a composite adjacent matrix; inputting the input data and the composite adjacent matrix into a graph convolutional layer of the model, fusing a Laplacian matrix and a spatial attention weight matrix, and aggregating spatial information by adopting a graph convolutional neural network; fusing the input data with the time attention weight matrix, and learning time features by using a long-short term memory network; and carrying out inverse normalization on the output of the connection layer to generate a final prediction result. The method fully considers the space information aggregation capability of the graph convolutional neural network and the time sequence learning capability of the long and short term memory network, can obtain a high prediction result, and can be widely applied to the field of intelligent traffic.

The invention discloses a vehicle operation parameter prediction method and system containing space-time characteristics, electronic equipment and a readable storage medium, and the method comprises the steps: S1, constructing a multi-view space-time diagram of a research region, taking an AOI region in the research region as a vertex, and taking the region feature quantities of two AOI regions assides; S2, inputting the information of the multi-view space-time diagram and historical data of a research period into a constructed MGCAN network to extract space-time features; wherein the historical data is historical vehicle operation parameters of each AOI area in the research time period; and S3, converting a vehicle operation parameter prediction result of each AOI area in the research time period by utilizing the space-time features. According to the method, the multi-view space-time diagram is constructed through the diagram structure, the multi-view space-time diagram and the space-time characteristics in the historical data are extracted through the MGCAN, vehicle operation parameter prediction is achieved through a brand-new means, and the method can be particularly applied to private car travel flow prediction.

The invention relates to a statistical forecast method and an apparatus for the urban heat island strength, which considers the influence of at least three weather factors such as horizontal total wind speed, wind direction and air temperature to the measuring result of the urban heat island strength, applies the historical observation data of the city to be researched for establishing a data analysis process apparatus, and obtains the numerical value forecast result of the specific weather factors such as horizontal total wind speed, wind direction and air temperature of the city to be researched by a numerical value mode so as to correctly perform the statistical forecast for the urban heat island strength and provide a certain technology support for the weather study and atmosphere science.

The invention discloses an assimilation method for inversion of the sea fog humidity by a meteorological satellite. The assimilation method comprises an all-weather sea fog monitoring module, a sea fog humidity construction module, a sea fog temperature construction module and a temperature and humidity data assimilation module. The all-weather sea fog monitoring module is used for carrying out all-weather monitoring on yellow sea fog based on MTSAT satellite data to obtain a three-dimensional distribution state of the sea fog; the sea fog humidity construction module and the sea fog temperature construction module are used for constructing humidity and temperature observation in a fog area based on sea fog information; and the temperature and humidity data assimilation module is used forassimilating by utilizing cycling-3DVAR. The method has the beneficial effects that the forecast initial field can be improved according to the inversion fog area and the change characteristics of thetemperature and humidity structure in the sea fog development process, so that the sea fog short-time proximity forecast result is further improved.

The invention discloses an intelligent chart generation method and device, a computer system and a readable storage medium, which relate to the technical field of big data. The method comprises the steps of receiving a user request sent by a user side, and generating a target object based on the user request, matching historical data corresponding to the target object in a database based on the target object to obtain first processing data, establishing a prediction model according to the data type of the first processing data, and predicting the target object by adopting the prediction modelto obtain target data associated with the target object, and obtaining a preset template based on the user request, generating a target chart according to the target data, the first processing data and the preset template, and sending the target chart to a user side, thereby solving the problems that in the prior art, a BI chart lacks of autonomously realizing prediction of future data trend basedon historical data, and the manual analysis cost is high, and the efficiency is low.

The invention provides a health condition risk prediction method and device, computer equipment and a storage medium, and belongs to the field of intelligent medical treatment in an artificial intelligence technology. The method comprises the steps of obtaining target prediction information of a target user in a first preset time period; determining a risk level of the target user on at least oneprediction dimension according to the target prediction information and reference prediction information of the specified user; and performing risk prediction on the health condition of the target user according to the risk level of the target user on the at least one prediction dimension. According to the invention, risk prediction is carried out on the health condition of the target user in at least one prediction dimension according to the target prediction information of the target user and the reference prediction information of the specified user. Since the specified user is a user witha risk in a health condition, the risk prediction is carried out based on the reference prediction information of the specified user, the prediction result is more accurate and reliable, and the prediction result is more comprehensive according to the risk level of the target user in at least one dimension.

The invention provides an urban area crowd flow prediction method and system, and the method comprises the steps: downloading a data set which comprises trajectory data and external influence factor data; preprocessing all data, including crowd flow calculation, external influence factor data one-hot coding and normalization processing; constructing a network structure based on ResNet and LSTM, wherein the network structure comprises a ResNet sub-network for simulating urban area crowd flow spatial features, an LSTM sub-network for simulating urban area crowd flow time features, an external factor neural network for simulating influence of external factors on crowd flow, and a fusion module; constructing a training data set and a test data set, taking a network structure based on ResNet and LSTM after network training and test as a model of urban regional crowd flow prediction, and inputting the preprocessed crowd flow state of each region of a target city and external influence factors into the network structure, and finally, obtaining a crowd flow prediction result of each region of the city in a certain time period in the future.

The invention provides a robot behavior teaching method based on meta-learning. The robot behavior teaching method is characterized by comprising the steps: acquiring a teaching video; and learning the teaching video by using the trained neural network model. The training process of the neural network model comprises the following steps: collecting training content; preprocessing the training content to obtain a preprocessed comparison video, a preprocessed teaching video and a preprocessed motion video; constructing an initial neural network model; taking the preprocessed teaching video as aninput, obtaining a demonstration action, and calculating the loss of the demonstration action; updating the initial neural network model according to the demonstration action loss to obtain an updated model; taking the preprocessed motion video and the track action as input to obtain a predicted track action, demonstrating semantics, motion semantics and comparison semantics, and calculating target action loss and semantic loss so as to construct total loss; updating the updated model based on the total loss; and until the total loss is stably converged to a total loss threshold, obtaining atrained neural network model.

The invention discloses a comprehensive energy load prediction method and system based on a user energy consumption label. The method comprises the following steps: S1, carrying out the analysis of auser energy consumption behavior through combining an energy consumption scene, and constructing a user energy consumption label system; s2, selecting an energy consumption label of a key user, and embedding user labels for short-term and medium-term and long-term load prediction; s3, constructing a CNNLSTM load prediction model based on the user label; and S4, carrying out comprehensive energy short-term and medium-term and long-term load prediction by utilizing the CNNLSTM load prediction model. According to the method, differentiated energy consumption characteristics of different typical regions are analyzed from static and dynamic dimensions, the user tags are determined by adopting a fishbone diagram analysis method and an importance ICCS judgment method, short-term and long-term medium-term load prediction of the sub-user tags is carried out by applying a convolutional neural network and long-term and short-term memory, and a prediction result is more comprehensive and accurate.

The invention discloses a human skeleton motion prediction method based on multi-task non-autoregressive decoding, which is used for solving the problem of motion prediction of a human 3D skeleton. The method specifically comprises the following steps of obtaining a human body 3D skeleton key point data set for training, and defining an algorithm target; establishing a graph convolution encoder, and performing feature learning on the input human body 3D skeleton to obtain features of the input skeleton; establishing a classifier, and performing behavior recognition on the input human body 3D skeleton input; establishing a non-autoregressive decoder, and predicting a human body 3D skeleton at a future moment; performing behavior recognition on the predicted human body 3D skeleton by using ashared graph convolution encoder and a classifier; and using the joint learning framework to carry out human body action prediction at a future moment. The method is used for human body action prediction analysis in a real video, and has good effect and robustness for various complex conditions.