86results about How to "Good predictive accuracy" patented technology

The invention provides a Dubins path based obstacle avoidance control device and method for a driverless car. The device comprises a camera arranged outside the left A pillar of the driverless car, acamera arranged outside the right A pillar of the driverless car as well as a laser radar arranged on the roof of the driverless car. The method comprises the following steps: the radar performs positioning and speed measurement on obstacles entering a camera monitoring range; a sequence of an obstacle to collide with the smart car is determined with an obstacle avoidance algorithm; the smart carreplans the current obstacle avoidance path according to the new obstacle sequence to finish the obstacle avoidance operation. By arranging the cameras in the middles of the two A pillars of the driverless car, the problem about bind area of the driverless car is solved effectively, and perfect and accurate degree of information acquisition is increased; different obstacle avoidance judgment algorithms are adopted for different obstacles in driverless operation, and obstacle avoidance accuracy and precision of the driverless car are improved.

The invention discloses a power consumer load interval prediction method based on deep learning. The method comprises the following steps of (1) establishing a large consumer historical load data preprocessing model; (2) establishing a load point prediction model based on an LSTM time recurrent neural network; and (3) adopting a load interval prediction algorithm of a point prediction value scaling coefficient. In this way, according to the method, a user load preprocessing model based on a state vector machine method is established to carry out preprocessing analysis on the single user historical data; and according to the processed historical data, an LSTM machine learning method is adopted to find a prediction model for reducing a user load prediction error to the maximum extent, and the load interval prediction of a single user is carried out by using a point prediction value scaling coefficient load interval prediction algorithm, so that the accurate load interval prediction can be carried out on the load of the single power user with strong random fluctuation, and the prediction accuracy of the user load is obviously superior to that of a traditional method.

The embodiment of the invention provides a training method and device of a question-answer pair classification model. The method comprises the steps of obtaining question-answer pair data; extracting question-answer pair characteristics from the question-answer pair data; labeling a classification tag on the question-answer pair data according to the quality of the question-answer pair data; training the question-answer pair classification model through the question-answer pair characteristics and the classification tag. The quality of the question-answer pair data is adopted to automatically label a large quantity of training sets, the question-answer pair classification model is classified, that is, to predict the quality score, the artificial strategy is avoided, thus the problems that feature information adopted in the artificial strategy is few, the active feedback rate of a user is low, the subjective judgment of a quizzer is depended on, the cheating phenomenon in the advertisement is serious, and the strategy is unstable caused when the feedback information of the user to new question-answer pair data and historical question-answer pair data is imbalanced are solved, and good forecast accuracy rates are obtained in both the historical question-answer pair data and the new question-answer pair data.

The invention relates to the maintenance field of operation and maintenance equipment, in particular to an intelligent operation and maintenance alarm filtering method and system based on multi-platform autonomous prediction. The method comprises the following steps: (1) data acquisition integration; (2) data quality inspection; (3) data cleaning; (4) feature engineering; (5) sample collection; (6) model training and parameter optimization; ) model release (8) model use; (9) model feedback and optimization. The system includes a data acquisition and integration module, a data quality inspection module, a data cleaning module, a feature engineering module, a sample sampling module, a model training and parameter optimization module, a model release module, a model alarm filtering module, and a model feedback and optimization module. The invention guarantees the real-time processability of low-level alarm events, and avoids the possibility of potential sudden serious alarm events due to subjective judgment errors of experts and inability to work 24 hours a day.

The invention discloses a network flow type prediction method based on deep learning. A multistage prediction scheme of ''edge pre-classification + center fine classification'' is adopted, that is, pre-classification is performed at first and then fine classification is performed, and deep learning models of pre-classification and fine classification are respectively constructed on an SDN switch and an SDN controller of a network edge, wherein the network function virtualization NFV technology is adopted, computing resources of switches in the SDN network and a distributed deep learning network constructed by links are used as hardware resources required for the pre-classification model, and the SDN controller is used as the hardware resource required for the fine classification model; andthe pre-classification model uses four joint features, and the fine classification model uses ten joint features. By adoption of the multistage prediction scheme in the network flow type prediction method disclosed by the invention, the communication overhead of the switch to the controller can be reduced, and the load of the controller can also be alleviated; the prediction is achieved by usinga capsule network method as early as possible; and meanwhile, the deep learning model is periodically trained by using an autonomously updated training data set to improve the prediction accuracy.

The invention relates to an electronic device, a litigation data processing method and a storage medium. The method comprises: obtaining a judgment document of a predetermined type of lawsuit, performing analyzing to obtain the focus of the dispute; based on the relationship between the disputed focus in the pre-established evidence list and the name of evidence and the disputed focus obtained ineach decision document, analyzing the judgment document to obtain the valid evidence information matched with the obtained disputed focus in the decision document; based on the focus sentences obtained from each judgment document, analyzing the corresponding judgment documents to obtain the judgment sentences; based on the judgment sentences of the judgment document, obtaining the corresponding judgment results, and establishing the relationship among the focus sentences, the evidence information, the judgment sentences and the judgment results; training the Bayesian model to get the pre-judgment model corresponding to the predetermined type of litigation. The invention can improve the accuracy rate of litigation pre-judgment and effectively guide relevant parties to make corresponding operation.

The invention provides a construction method and a construction device as well as a sorting method and a sorting device for a support vector machine sorter, wherein the method comprises the steps of determining a non-linear weighted kernel function; determining a non-convex Lp fraction norm punishment target function based on the weighted kernel function; constructing the support vector machine sorter by utilizing the non-convex Lp fraction norm punishment target function. Compared with the technical scheme in the prior art, in which when high-dimensional small sample data are sorted, all characteristic-dimensional combinations need to be traversed to find the needed characteristics, the method has the advantage that the constructed support vector machine sorter can realize the characteristic selecting function of a sample original space after the non-linear kernel mapping; the method can be used for sorting the high-dimensional data to generate a more sparse model, realize more accurate characteristic selecting and obtain a better predication accuracy, so the calculation complexity is greatly reduced, and a data disaster is avoided.

The invention provides a medicament module pharmacokinetic property and toxicity predicting method based on a capsule network. After a comprehensive module fingerprint and a module descriptor are constructed and early-period preparing operation for establishing model is performed, a low-grade characteristic content of a molecule is extracted from an upper-layer low-grade characterized through convolutional or restricted Boltzmann machine operation; then a capsule network method is used for abstracting the high-grade characteristic of the molecule in a lower-layer high-grade characteristic; a relation between the high-grade characteristic and an active label is fit through a dynamic routing algorithm, thereby predicting the pharmacokinetic property and the toxicity class of an unknown smallmolecule. The method does not require collection of large scale datasets for training, optimization is performed on input through end-to-end and furthermore automatic dimension reduction is realized.A coupling coefficient is updated through iterating a dynamic routing process. The dynamic routing conveys all characteristics of an upper-layer capsule to a random lower-layer capsule, thereby greatly reserving a hierarchical position relation. The method realizes a better predicting effect than that of a traditional machine learning method.

The invention discloses a surrounding vehicle behavior adaptive correction prediction method based on a driving prediction field, which comprises the steps of: S1: carrying out surrounding vehicle behavior discretization and data set preprocessing, i.e., partitioning surrounding vehicle behaviors into N typical behaviors according to a transverse direction and a longitudinal direction; S2: acquiring traffic environment participation vehicle time series data, i.e., enabling each traffic environment participation vehicle to acquire a position, a speed and an acceleration of the vehicle at each moment in real time by using a positioning system; S3: establishing the driving prediction field, i.e., establishing the driving prediction field EP based on three elements of safety, efficiency and driving comfort, wherein EP=ES+EE+EC; S4: establishing a surrounding vehicle behavior prediction model on the basis of a maximum likelihood estimation method; and S5: carrying out surrounding vehicle behavior real-time prediction and model adaptive correction. According to the invention, safety, efficiency and driving comfort which influence driver behaviors are comprehensively considered; the driving prediction field is established in a driving region of a target vehicle and qualitative and quantitative analysis is carried out; and a new idea is proposed for surrounding vehicle behavior prediction.

The invention provides a customer segmentation-based method for controlling churn rate prediction. The method comprises the following steps of: a, collecting original data; b, carrying out pretreatment on the original data; c, extracting a client value parameter set; d, selecting a user segmentation algorithm; e, generating a user segmentation group; f, selecting an appropriate churn rate prediction algorithm; g, calculating and predicting the churn rate; h, feeding back the prediction result; and I, outputting the prediction result. The method comprises carrying out segmentation according to selected users, and furthermore predicting by using an appropriate churn rate predication algorithm, and the method has the advantages that the user attribute characteristics of the user are mastered effectively, the user is segmented accurately, and the user churn rate is predicted accurately.

The invention provides a construction method and device, a classification method and device of support vector machine. The construction method and device, the classification method and device of support vector machine comprises: make sure non-linear weighted kernel function of a single variable; make sure nonconvex Lpfraction norm penalty object function on the base of the weighted kernel function of a single variable; make use of nonconvex Lpfraction norm penalty object function to construct support vector machine. Compared with the technical proposal which needs to traverse all characteristic combination of dimension to look for the desired characteristics when high-dimensional data of small sample is classified in the existing technology, the invention constructs the support vector machine and the support vector machine is used to classify the high-dimensional data of small sample so as to produce more sparse model, to achieve feature selection of any structure more accurately, to obtain better prediction accuracy, to reduce computation complexity largely and to avoid data disaster.

The invention relates to a single-model multi-output age and gender identification method and system based on a face. The method is characterized by cascading age and gender models; through one network model, extracting the characteristic data of an age and a gender; merging the characteristic data of the age and the gender into one piece of characteristic data used for age prediction; and then, using the different output layers to generate one model so as to outputting age and gender identification results in parallel. By using the method, the identification speed and the accuracy of the ageand the gender are increased.

The invention relates to the field of web service quality prediction, and discloses a service neighborhood based Web Service quality prediction method. The method is characterized by including the following specific steps of similarity calculation, neighbor selection, model establishment, model solution and prediction. A prediction model obtained by neighbor user based feature vector learning is established, and is solved through a gradient iteration descent method, and finally a predicted value of web service quality of a target user is obtained. Meanwhile, the invention discloses a device applying the Web Service quality prediction method. The method and the device have the advantages that accuracy is high, the problem about prediction of web service quality in the cold boot process is solved by separating feature vectors of target services and feature vectors of neighbor services, and application value is high.

The invention discloses a full traffic prediction method based on a dual graph framework, and the method comprises the steps: (1) expressing a road network structure as a topological graph, taking anintersection as a node, and taking a road segment connected with the intersection as an edge; preparing historical edge and node data and future edge and node data; (2) constructing a historical information encoder, inputting historical data into the encoder, realizing information transmission between edges and nodes through multi-layer dual mapping, and splicing outputs of the multi-layer dual mapping into a historical feature tensor; (3) constructing a future prediction decoder, decoding the historical feature tensor into future spatial-temporal features, and outputting a future prediction result; (4) taking an error between the prediction result and the actual data as a loss function to perform model training until the loss function converges; and (5) carrying out model test by using the trained model, and carrying out application after the test is finished. According to the invention, the prediction result can obtain the total complete description of the future traffic condition, and the prediction accuracy is high.

The invention discloses an on-line monitoring early-warning device for an SPD, and the device comprises an operation statistics module, an inrush current sensing module, a voltage sensor, a current leakage sensing module, a temperature collection module, and a processing module. The processing module is connected with the operation statistics module, the inrush current sensing module, the voltage sensor, the current leakage sensing module and the temperature collection module. The device provided by the invention collects the total number of operation times of the SPD, the inrush current of the SPD, the loading voltage of the SPD, the leaked current of the SPD and the surface temperature through the operation statistics module, the inrush current sensing module, the voltage sensor, the current leakage sensing module and the temperature collection module, thereby achieving the comprehensive collection of all parameters of the SPD, and laying a foundation for the estimate prejudgment of the SPD. According to the current and voltage values collected by the inrush current sensing module and the voltage sensor, the device calculates the actual impedance of the SPD, and prejudges the electrical performances of the SPD through comparing the actual impedance with a preset impedance threshold value and combining the comparison results with the leaked current of the SPD.

The invention discloses a pheochromocytoma metastasis prediction system based on a molecular marker. The system is characterized in that the system comprises a variable input submodule, an analysis module and an output module; the variable input submodule comprises a tumour primary diameter input submodule, a primary tumour part input submodule, a catecholamine secretion type input submodule, a blood vessel invasion state input submodule, an ERBB-2 overexpression state input submodule and an SDHB mutation state input submodule; the analysis module can build a metastasis probability alignment chart and calculate a total risk value based on variables input by the variable input submodule and can calculate a pheochromocytoma metastasis predicted value of a pheochromocytoma patient according to the total risk value; the output module is used for outputting the pheochromocytoma metastasis predicted value of the pheochromocytoma patient. According to the pheochromocytoma metastasis prediction system based on the molecular marker, SDHB germ-line gene mutation and primary tumour ERBB-2 protein high-expression, the diameter and position of a primary tumour, blood vessel invasion and the catecholamine secretion type are combined, and accordingly the pheochromocytoma metastasis prediction system is built and shows more excellent prediction accuracy compared with separately used clinical risk factors.

The invention provides a link level to system level simulation interface method based on LTE-A. According to the effective signal to noise ratio mapping algorithm based on mutual information of the method, through use of the feature that in the algorithm, parameter compression is carried out on each subcarrier in an averaging linear mode, the idea that compression is carried out through an exponential function is provided. According to the method, firstly, the signal to noise ratio of each subcarrier is calculated; then mutual information corresponding to each signal to noise ratio is obtained in a table look-up mode, and data compression of the exponential function is carried out on the mutual information, thereby obtaining equivalent mutual information; and finally, a predicted block error rate can be obtained through the equivalent mutual information. The problem that the precision accuracy of the interface algorithm in a high-order encoding modulation scheme is low is solved. Through improvement of the interface algorithm, the prediction accuracy of the interface method in the high-order encoding modulation scheme is improved.

The invention discloses a device, a method and a system for monitoring a state of a hydraulic system. The method comprises the following steps that supervised learning model training is provided, andspecifically, a training set and a test set are divided according to an existing sample data set, and training, verification and optimization processes of the supervised learning model is completed; and the system is initialized, and specifically, the system is started. According to the device, the method and the system for monitoring the state of the hydraulic system, the method comprises the following steps that firstly, data dimensionality reduction is conducted on the large-scale hydraulic measurement original data by using an unsupervised PCA algorithm, the data processing amount is greatly reduced, the training and prediction speed are remarkably improved, the overfitting risk is reduced, the training and prediction speed are remarkably improved, the generalization ability of a modelis improved, and good prediction accuracy is achieved; and the requirement for real-time accurate evaluation of the hydraulic system is met, finally, the prediction accuracy of the hydraulic state can be remarkably improved, and a better application prospect is brought.

The invention discloses a sales prediction and neural network construction method and device, equipment and a storage medium, and the method comprises the steps: obtaining the historical sales data ofa target commodity in at least one time window and the current product information of the target commodity; performing hole convolution processing on the historical sales data by using a neural network model, and extracting trend features of the historical sales data, wherein the trend features represent the influence degree of the commodity information of the target commodity on the historical sales volume of the target commodity; and training the current product information by using the neural network model according to the trend characteristics, and obtaining sales prediction data of the target commodity. According to the method, tThe subsequent sales volume of the commodity can be accurately and effectively predicted by obtaining the historical sales data of the commodity, and meanwhile, the neural network of the method has better model characterization capability.

The invention discloses a web service recommendation method based on improved deep learning. The web service recommendation method comprises the following steps: extracting free text label data from aWSDL document of a web service; obtaining a user-scoring matrix from the free text label data of the web service by utilizing the LFM based on the SDAE; constructing a functional feature preference matrix of the user-service according to the user-scoring matrix, and constructing a mixed preference matrix in combination with the QoS data set of the web service; and predicting a missing QoS value of the mixed preference matrix by using a collaborative filtering algorithm based on the mixed preference matrix to realize intelligent recommendation. According to the invention, the web service recommendation method based on improved deep learning can be realized.

The invention relates to an automatic rhythm extracting method, a system and an application thereof in natural language processing; the method includes steps of applying an automatic text-voice alignment technology to generate a large-scale rhythm data set, and applying a circular neural network to perform modeling on the rhythm of a sentence, adding a bidirectional expanding mechanism; applying the automatically structured text rhythm data to a natural language processing task based on the circular neural network. The method fully uses the isomorphism properties of common sequence data in the text rhythm sequence and the natural language processing task; through an alternative training method under the multi-task study, the natural language processing task is promoted without the assistance of artificially explicit marked semantic information. The practice of the method can overcome shortcomings of low efficiency, different standards, incapability of large-scale application of the artificial rhythm marking; meanwhile, the method can transfer semantics and pragmatics in massive voice data to the other tasks.

The invention discloses a method for predicting fall risk of elderly people. The method comprises the following steps: dividing a plantar pressure area into a hallux area, a second-fifth toe area, a forefoot area, a middle foot area and a heel area; dividing a support phase into an initial contact section, an initial metatarsal bone grounding section, an initial forefoot flat section, a heel off-ground section and a final contact section; then, based on the plantar pressure areas and the supporting phase, carrying out plantar pressure testing on a subject by using a Footscan plantar pressure flat plate testing system so as to obtain pressure change curves of the different plantar pressure areas in all supporting phase periods; constructing a deep neural network model by using a convolutional neural network and a recurrent neural network, training the prediction model, and selecting an optimal prediction model as a foot pressure prediction model; and finally, inputting the pressure change curves into the foot pressure prediction model to obtain a prediction value. The method has the characteristics of high data measurement precision, various characteristic indexes and good prediction accuracy.

The invention belongs to the technical field of news software development, and discloses a network news summary system and method for automatically generating abstracts by adopting multiple strategies, which are used for carrying out data acquisition in multiple fields of multiple news websites, automatically generating article abstracts, generating news short messages based on news abstracts andcarrying out reading analysis. Multiple strategy automatic abstract generation algorithms serve as the core, multiple news short messages are correspondingly generated for each original news report grabbed from the network, news short messages with large text content are removed, a user is helped to quickly preview and browse interested news reports, and the click rate and the reading rate of a news website can be increased. In addition, the use value of efficient and standard domain classification after aggregation of multiple news websites can be improved; and rapid reading of a large amountof news is achieved, and a large amount of time and energy for reading news original texts of users are saved.

A movie recommendation system based on cross-modal fusion comprises an input module, where the input information comprises feature information of each user in a user set, feature information of each movie in a movie set, interaction information of each user in the user set to each movie in the movie set, and poster pictures of each movie in the movie set; the system also comprises a preprocessing module, a single-mode coding module, a cross-mode fusion module and an output module. Visual information of movie poster pictures is extracted based on a deep neural network model, and useful information which can be used in the recommendation process is enriched; on the other hand, interaction and fusion of text and visual information are realized based on a cross-modal fusion algorithm, so that movie recommendation is better carried out on the user; compared with a traditional recommendation algorithm which only uses single-mode information, does not interact two-mode feature information and does not distinguish single-mode and multi-mode interaction modes, the invention has a better recommendation effect.

The invention provides a telecom customer loss probability prediction method and system based on an end-to-end model, relates to the technical field of data security, and the telecom customer loss probability prediction method based on the end-to-end model comprises the following steps: S1, obtaining telecom customer data, and marking the obtained data; S2, preprocessing the data, processing abnormal values and missing values, standardizing the data, and training a customer probability prediction sub-model; S3, fusing results of the sub-models, training a fusion model, and obtaining a customerloss probability prediction model; and S4, obtaining a customer loss probability value. According to the telecom customer loss probability prediction method and system based on the end-to-end model,big data and artificial intelligence technologies are applied to predict the customer loss probability of the telecom industry; and the method based on ensemble learning is applied to probability prediction, has good prediction accuracy, reduces the cost and time of manual judgment, and provides important reference data for enterprise maintenance customers.

The invention discloses a three-dimensional point cloud single-stage target detection method for decoupling classification and regression tasks, and the method comprises the steps: (1) carrying out the voxelization processing of a point cloud, enabling the data to become an ordered grid structure from an unnecessary point cloud, (2) carrying out the feature extraction of a network through three-dimensional sparse convolution, and obtaining a high-order feature map, and (3) in the feature map, using a double-head detector to aggregate the features and predict the classification, regression frame and direction of the target. In order to solve the problem of feature entanglement between target detection subtasks, a double-head detection network structure is designed, features concerned by classification and regression tasks can be extracted from high-dimensional features, and the subtasks are predicted. On the basis of decoupling, related information in the two tasks is combined by using a joint detection method to jointly predict a target category. According to the method, the accuracy of three-dimensional target detection is improved, and the method can be easily migrated to other methods.

The invention relates to the technical field of computer networks and especially relates to a transactional and QoS (Quality of Service) combination based service quality performance predication method and device. The method includes enabling Web service to be transactional and analyzing a combination problem of the service with transactional demands; analyzing a performance problem of service with transactional attribute combination; and according to performance comparison of the transactional combination service, proposing a transactional and QoS combination based service selection method. The invention also includes a service quality performance predication device employing the above method. The device also includes an extraction module, a processing module, an extraction module and a determining module.

The invention provides a risk prediction model and a risk prediction system for endometriosis-related ovarian cancer. The risk prediction model comprises the following regression equation: Logit(P)= ln(P / 1-P)=-11.149+0.113*age+0.170*physique index+0.002*mass volume+2.420*blood flow signal. Compared with the prior art, the prediction model has the advantages that the prediction accuracy of the ovarian cancer risk is greatly improved, and a treatment scheme can be better formulated clinically.

The invention particularly relates to a lithium battery SOC prediction method for improving an ant colony algorithm and optimizing particle filtering, and the method comprises the following steps: carrying out the discharge test of a lithium battery under different working condition currents, and preprocessing the test data; performing parameter identification according to the preprocessed experimental data, and constructing a state equation according to an ampere-hour integral method in combination with SOC prediction influence factors; establishing a measurement equation of a battery theoretical prediction model according to the second-order Thevenin equivalent model; using an improved ant colony algorithm to optimize particle filtering; and predicting the SOC change of the battery through optimized particle filtering. According to the prediction method provided by the invention, the situation that a traditional ant colony algorithm is easy to fall into a local optimal solution is improved; the improved ant colony algorithm is utilized to optimize particle filtering, the problems of low particle diversity and poor particle appearing when the SOC is estimated through the particle filtering algorithm are solved, the problems that a lithium battery SOC estimation method is complex and low in accuracy are solved, and the estimation precision is effectively improved.

The invention relates to a production fluid water content measurement method for a high-water-cut oil well based on time-frequency characteristics. The method adopts a double-ring high-frequency capacitive sensor, a water content fluctuation signal time-frequency transformation module, and a wellhead water content prediction network based on artificial intelligent. The method comprises the following steps: wellhead water content fluctuation information is firstly acquired through the double-ring high-frequency capacitive sensor provided with a specific structure and applicable for an oil wellhead; time-frequency transformation is then carried out on acquired water content fluctuation signals, so that a time-frequency spectrum of the water content fluctuation signals is obtained; and the time-frequency simultaneous distribution spectrum obtained through the transformation is used as an input of a convolutional neural network, flow characteristics of the detected signals are extracted layer by layer through multilayer convolution-pooling operation, the extracted characteristics are finally output to softmax for water content measurement, and a water content tag is obtained through wellhead testing.