129 results about "Decision level" patented technology

The invention discloses a bearing fault diagnosis method based on an LS-SVM and the D-S evidence theory. Through a multi-level information fusion method, the least squares support vector machine (LS-SVM) is adopted in a feature layer, and the D-S evidence theory is adopted in a decision-making layer to solve the problems that the single sensor fault diagnosis accuracy is low and the sensitive features are difficult to extract. Firstly, the wavelet noise reduction technology is used to improve the signal-to-noise ratio of a rolling bearing vibration signal, and eight parameters in the time domain and the frequency domain are introduced as feature parameters of the bearing vibration. Secondly, the bearing is fault identified by the LS-SVM. Thirdly, the LS-SVM feature output is used as the D-S evidence theory input, and the D-S evidence theory is used for fault decision making. The method can effectively improve the accuracy of rolling bearing fault diagnosis. The invention has certain significance for improving the fault diagnosis accuracy of the rolling bearing and the reliability of the diagnostic system.

The invention provides an intelligent electrical-network fault diagnosis method based on multilevel feedback adjustment. The method comprises the steps that historical switching information and electrical quantity attribute information of different devices in an electrical network when the electrical network has faults are stored in a historical fault information base; electrical quantity information of the devices in the electrical network is obtained when the electrical network has faults; diagnosis at the rough identification level is carried out according to the obtained electrical quantity information of the devices in the electrical network, and suspected fault elements are determined in a minimal breaking zone defining method based on the equivalent network to form a candidate set of the suspected fault elements; diagnosis at the fuzzy decision making level is carried out; diagnosis at the accurate positioning level is carried out; and according to results of three levels of fault diagnosis, intersection elements are obtained from suspected fault element sets E1 and E2 to finally determine a diagnosis result of fault elements. The electrical network with faults is analyzed at different levels according to the different sources of the multi-source fault information and the obtaining and processing difficulty of different type of information, and different types of fault information is fully utilized and information is complementary to improve the accuracy of fault diagnosis.

The invention relates to a multi-source information fusion fire prediction method based on a dynamic integrated neural network. Innovative logic design, establishing a fire prediction model based on amulti-source information fusion method; various fire characteristic signals pass through an information layer, a characteristic layer and a decision-making layer in sequence; in a feature layer, LSTMand RBF-BP neural networks in deep learning are used as sub-networks to carry out adaptive learning on multi-source fire feature signals; according to the fire prediction method, the output result issubjected to integrated analysis, and fire prediction is completed through the decision-making layer, so that the problems of time-varying characteristics and nonlinear characteristics of the fire signal and high missing report rate and false alarm rate of the single-characteristic signal fire prediction method are solved, and the recognition accuracy of the fire prediction system is effectivelyimproved. The method has high expandability, a complete prediction model can be established only by providing a data set again after a detection environment is changed, and the method has high adaptability.

The invention provides an intelligent network connecting test system based on a cross-layer collaborative framework for vehicles. An integrated verification and evaluation system which is comprehensive, multimodal, full synchronizing and large scale for the vehicle intelligent network connecting tests is formed by the cross-layer collaborative framework of a sensing layer, a communication layer, an interaction layer, a computation layer and a decision-making layer. The system of the invention overcomes the defects that safety guarantee and large-scale verification are difficult to realize or form in the prior art of testing.

The invention provides a DEMATEL-ANP-VIKOR mixed selection method of power system security risk reduction schemes. The method comprises the steps of 1) establishing an evaluation index system, 2) establishing the influence relation structure matrix of target layer elements by use of a DEMATEL method, 3) determining each index weight of a criterion layer by use of an ANP method, 4) ranking the security risks of a decision layer by use of a VIKOR method to obtain compromise power system security risk reduction schemes, and 5) selecting the schemes. The DEMATEL-ANP-VIKOR mixed selection method of the power system security risk reduction schemes has the advantages that an interactional restrictive relation between influence factors is taken into full account, and a compromise planning method is taken as the core to provide the optimal rank of the maximum community benefit and minimum individual regret compromise alternative schemes, and consequently, the compromise feasible risk settling schemes closest to the ideal solution are obtained.

The invention discloses a traffic monitoring method for vehicle locus tracking and collision early warning. The method is realized by use of an intelligent monitoring information hierarchical structure model, and the model comprises an environment perceiving layer, an object layer, a feature layer, a semantic layer and a decision layer. The method comprises the following steps: obtaining a traffic image in real time, and storing the traffic image in the environment perceiving layer; extracting a vehicle image in the traffic image and storing the vehicle image in the object layer; processing the vehicle image in real time by use of a model obtained from a convolutional neural network and an onsite acquisition device with an online learning capability so as to obtain identity information and motion parameters of a vehicle for storing in the feature layer; processing feature parameters to obtain a motion trend of the vehicle and storing the motion trend in the semantic layer; determining whether a vehicle is about to have a collision through a conclusion of the semantic layer; and if it is determined that the vehicle is about to have a collision, an early warning is generated. According to the invention, through preprocessing, learning and identifying the traffic image, whether an accident takes place can be analyzed; and the early warning is generated and correlation personnel is notified. Further disclosed is a traffic monitoring apparatus.

The invention relates to a risk assessment method for a thermal pipeline system. The method comprises the following steps of: 1, collecting and surveying materials required by pipeline system risk factors; 2, performing macroscopic test and professional inspection on pipelines; 3, performing professional analysis; 4, determining numerical values of the risk factors according to the technical standard, operating parameters and expert assessment; 5, inputting the numerical values of the risk factors to a computer, and performing risk calculation by using a risk assessment model; 6, outputting a risk assessment result, so that a decision maker decides whether to accept according to the risk assessment result; 7, if the risk assessment result cannot be accepted by the decision maker, finding and processing the major factor influencing the value at risk and performing risk calculation again; and 8, if the risk assessment result is accepted by the decision maker, finishing the assessment procedure. According to the risk assessment method, risk assessment is performed by using the risk model, so that a decision-making basis is supplied to the decision-maker, accidents such as breakage, collapsing or burst are reduced or avoided, the service life of the pipelines is prolonged, and human life and property are prevented from being lost.

The invention discloses a multi-modal sentiment classification method based on a heterogeneous fusion network, and belongs to the technical field of opinion mining and sentiment analysis. The method comprises the steps of 1) preprocessing video data; 2) constructing a text feature vector and identifying a text emotion category; 3) constructing a picture feature vector and identifying a picture emotion category; 4) constructing an audio feature vector and identifying an audio emotion category; 5) constructing a multi-modal global feature vector and identifying a multi-modal global emotion category; 6) constructing a multi-modal local feature vector and identifying a multi-modal local emotion category; and 7) adopting a voting strategy to obtain a final sentiment classification result. The heterogeneous fusion network adopts two fusion forms of intra-modal fusion and inter-modal fusion, two fusion angles of macroscopic fusion and microscopic fusion, and two fusion strategies of feature layer fusion and decision-making layer fusion. According to the method, the implied associated information among the multi-modal data can be deeply mined, and mutual complementation and fusion among the multi-modal data are realized, so that the accuracy of multi-modal sentiment classification is improved.

The invention provides a driverless automobile dynamic lane changing track planning method based on a Fraenet coordinate system. The method comprises the steps of establishing a path generation model,sensing the environment, sending a lane changing instruction by an upper behavior decision layer, planning an alternative motion track set by the path generation model, selecting alternative tracks and the like. According to the method, a discrete global trajectory is used as a target path, a cubic polynomial is adopted, the algorithm complexity is low, and the feasibility is good.

The invention relates to an architecture of an unmanned intelligent lifting system. The architecture comprises a perception layer, a decision making layer and a control layer; the perception layer obtains the on-site position of an assembled building through the positioning function of a GPS, meanwhile, an embedded sensing system collects environment information and tower crane information in realtime, collected data are processed through a kalman filtering algorithm, feedback is sent to the decision making layer in real time, and the various informaiton collecting function of the perceptionlayer is achieved; the decision making layer receives data information collected by the sensing layer, a decision is made through a neural network expert system, and a specific operation order of a tower crane is output; and the main function of the control layer is to receive a control order output by the decision making layer, and actual operation on hardware is completed through a vehicle control system according to the control order. According to the architecture of the unmanned intelligent lifting system, automatic lifting of the tower crane is achieved, positioning is precise, the production efficiency is high, the operation risk is low, and the manpower cost of the construction process is reduced.

The invention discloses a virtual experiment system and method based on multi-modal interaction. The experiment system mainly comprises an input layer, a sensing and recognition layer, a fusion layerand an application layer. The method comprises the steps: acquiring corresponding visual information, sensing signals and voice signals through a visual channel, a tactile channel and an auditory channel respectively; respectively identifying the information of different channels; fusing corresponding vector sets constructed by modal information of the visual channel, the tactile channel and the auditory channel on the AR platform by adopting a multi-modal fusion method based on a decision level according to an identification result; and according to the fusion result, presenting an experimentprocess and an experiment result by adopting voice navigation, visual display and tactile feedback modes. Through the application, multiple channels can be fully utilized, and an experimental processis achieved by adopting a multi-mode fusion method, so that the operation load of a user is reduced, the immersion of an experiment is improved, and the efficiency of virtual interaction is improved.

The invention provides an artificial potential field method for automatic driving vehicle decision-making layer path planning, and the method comprises the steps: S1, building a road environment through an artificial potential field method, and enabling a vehicle to drive in a road boundary; S2, acquiring surrounding information through a vehicle sensing device, and establishing different potential field functions for different types of obstacles, so the vehicle has different processing modes for different obstacles; S3, planning a set of safety paths according to the established potential field environment, and optimizing by using an improved genetic algorithm to find the shortest path. According to the method, the driving environment of an automatic driving vehicle is established by using an artificial potential field method, so that the vehicle performs different driving operations on different types of obstacles, and the improved genetic algorithm is used for optimizing the path, so that the vehicle is driven safely, and the energy consumption is minimum.

The invention discloses a gearbox bearing fault detection method and system, and the method comprises the steps: processing missing data and noise data in temperature feature information data, pressure feature information data and unbalance loading feature information data through employing feature preprocessing technologies, such as missing value filling, feature redundancy removal and outlier correction; establishing a first fault classification model algorithm Y1 by using an extreme value gradient lifting algorithm, and after the temperature, pressure and unbalance loading feature information of the bearing is processed, classifying the faults of the bearing; establishing a second fault classification model algorithm Y2 by adopting a closely connected convolutional network structure capable of processing a one-dimensional frequency spectrum sequence to carry out fault diagnosis classification on the vibration characteristic information of the rolling bearing; and establishing a fault diagnosis model fusing multi-sensor features, performing decision-making layer result fusion on two single-source fault diagnosis models including a first fault classification model algorithm Y1 and a second fault classification model algorithm Y2 through a D-S evidence theory, and then identifying the fault type of the gear box bearing.

The invention discloses a cross-domain migration electronic nose drift suppression method based on migration samples. The method comprises steps of projecting the source domain data and the target data to a subspace, performing edge maximum mean difference minimization processing, condition maximum mean difference minimization processing and separability maximization processing on sets of different domain data, and performing maximization processing on discrimination information to obtain a conversion basis P, a corresponding projection source domain data set and a projection target domain data set; calculating an unknown output weight of the adaptive extreme learning machine according to the projection source domain data set and the projection target domain data set to obtain a final adaptive extreme learning machine; and performing a drift suppression test on the target domain data of the unknown label. The method has the beneficial effects that the discrimination information of thesource domain and the target domain is stored while drift is inhibited. The edge distribution difference and the condition distribution difference are minimized, and the robustness and the classification accuracy of the model are improved. Knowledge migration is realized in a feature layer and a decision layer, and migration samples are fully utilized.

The invention relates to a processing method and device for enterprise intelligent decision analysis, computer equipment and a storage medium. The method comprises the steps of obtaining to-be-processed data corresponding to an enterprise identifier; performing feature extraction on the to-be-processed data to obtain a feature vector of the to-be-processed data; determining a logic relationship between the feature vectors through a logic layer of a pre-trained machine learning model; through a data analysis layer of the machine learning model, analyzing the feature vector by utilizing the logic relationship to obtain an evaluation value corresponding to the enterprise identifier; and processing the evaluation value through a decision-making layer of the machine learning model to obtain corresponding decision-making data. By adopting the method, the processing accuracy for enterprise intelligent decision analysis can be improved.

The invention provides a driving motor fault diagnosis model construction method based on intra-class feature transfer learning and multi-source information fusion, and the method comprises the steps:firstly proposing an improved hierarchical transfer learning method MSTL, which considers the neighbor relation between intra-class samples, maintains the local manifold structure of intra-class data, also can improve the separability of the domain data subjected to transfer learning to different categories, so that the adaptability of the fault diagnosis model to different distribution domain samples is improved; meanwhile, the feature set dimension can be reduced, and the fault diagnosis performance of the fault diagnosis model under variable working conditions is improved. Besides, aimingat the problem that a certain uncertain factor exists in a signal acquired by a single sensor, the D-S evidence theory is adopted to carry out driving motor multi-source information decision-making layer fusion, and secondary D-S evidence fusion is carried out on diagnosis results of vibration and current signals on a model. According to the feature transfer learning method MSTL and the multi-source information fusion diagnosis model provided by the invention, the fault diagnosis accuracy can be improved, and the method has a certain practical value.

The invention discloses a garbage classification method and system based on transfer learning and model fusion and a storage medium. In the method, the method includes classifying the junk image datathrough a classifier, and outputting a corresponding classification result; moreover, the generation process of the classifier adopted by the invention comprises the step of constructing two differentclassification networks into two strong classifiers by using an Adaboost algorithm, so that the generalization error rate is low, over-fitting is avoided, and the high accuracy of industrial requirements can be achieved. And the two strong classifiers are fused in a decision-making layer fusion mode, so that the feature diversity is fully utilized, and the classification accuracy is further improved.

The invention discloses a sudden respiratory infectious disease risk prediction model for regional epidemic prevention and control. The model comprises the following steps: S1, obtaining original data; s2, carrying out data management; s3, establishing a knowledge base; s4, carrying out decision-making layer fusion; and S5, establishing a risk prediction model. The invention has the following beneficial effects: (1) the system has advantages in multi-dimensional input data and output data processing; (2) any complex nonlinear problem can be processed, so that the nonlinear mapping capability is high; (3) information or data can be processed in parallel, and the data processing capability and performance are improved; (4) the self-adaptability is relatively high; and (5) the data fusion aspect is good.

A multi-task hierarchical machine learning model is configured to perform both a decision task to predict a decision result and an explanation task to predict a plurality of semantic concepts for explainability associated with the decision task, wherein a semantic layer of the machine learning model associated with the explanation task is utilized as an input to a subsequent decision layer of the machine learning model associated with the decision task. Training data is received. The multi-task hierarchical machine learning model is trained using the received training data.

The invention discloses a visual grading method and grading production line for the pear appearance quality. According to the method, information such as the fruit shape, the color, the defect type and the defect area of a sample can be reflected through characteristic values, and an appropriate machine learning model is selected as a classifier according to the features of the characteristics for accurately prediction, finally, the information is input into a decision-making tree for a decision-making level, and a decision-making classification process of a person is simulated to obtain a classification result. A multi-model fusion method is used in the whole grading process, various kinds of information is processed in a targeted mode, the over-fitting phenomenon is avoided, the processing result is more accurate, the detection efficiency and the detection precision are improved, and the requirement for rapid and accurate detection on an assembly line is met. The method is used for rapid grading of pears on the annular production line, the pears are graded through characteristic fusion, and the identification rate can reach 95% or above. The visual grading method and grading production line are suitable for appearance quality grading of different kinds of pears.

The invention relates to a fault diagnosis method and system for industrial equipment data, and the method comprises the steps: obtaining the detection data of a plurality of sensors under various types of faults, and enabling the number of the detection data of each sensor to be more than one; constructing a deep belief network for each sensor; optimizing each deep belief network, and determining a network hyper-parameter of each deep belief network; training each optimized deep belief network by taking the detection data of the sensor as input and taking the probability of the fault type corresponding to the detection data of the sensor as output; inputting to-be-diagnosed detection data into the trained deep belief networks, and outputting probabilities of a plurality of fault types; and adopting an improved DS evidence theory based on a Pearson coefficient to carry out decision-making layer fusion on the probability values of the fault types, and outputting the fault types. According to the invention, the accuracy of fault diagnosis is improved.

The invention discloses a rotating mechanical equipment virtual assembly model and method based on context awareness. A virtual assembly task is decomposed into an assembly knowledge layer, an assembly decision-making layer and a user control layer. The layers are endowed with different responsibilities. The virtual assembly model is defined for the assembly knowledge layer for containing assembly knowledge and assembly semantics to achieve contextualization of virtual assembly. Each part of the decision-making layer is regarded as an independent Agent, communication templates and decision-making rules between the Agents are defined according to the characteristics of the virtual assembly process, and user guide regularization is achieved. Through an offline clustering algorithm, the position and orientation of a user under a specific assembly situation are predicted, and time spent on adjusting a camera by the user is shortened. Compared with an existing assembly sensing technology, the model and method have the advantages that the assembly model is low in weight, interaction intelligence is achieved, and the interaction efficiency is high, and the assembly process can be customized to guide user operation.

The invention discloses a decision-making layer fusion method for multi-modal emotion classification. The method comprises the following steps: dividing samples in a multi-modal emotion data set into a training set and a test set; respectively constructing sentiment classification models of various modalities, and respectively training the sentiment classification models of various modalities by using samples of corresponding modalities in the training set; using the trained sentiment classification models of various modes to perform sentiment classification on samples of corresponding modes in the test set, and counting classification results to obtain sentiment classification confusion matrixes of various modes; performing sentiment classification on the corresponding modes of the tested sample by using the trained sentiment classification models of various modes; and performing decision-making layer fusion on sentiment classification results of various modes of the tested sample by using the classification confusion matrix to obtain a sentiment category of the tested sample. According to the method, the prior knowledge of information differences of different modes and the complementarity between the modes are fully utilized, and the accuracy and robustness of multi-mode sentiment classification can be effectively improved.

The invention discloses a decision-making layer data fusion method based on a multi-source sensor, and the method comprises the following steps: enabling the output of each sensor to be target-level data, carrying out the testing and analysis of the performances of each sensor before use, and obtaining some prior information and failure scenes of an output signal of each sensor; then, carrying outdata preprocessing on a target output by a single sensor, filtering a false detection target preliminarily, and smoothing abnormal jump data; and finally, fusing the preprocessed target-level data ofeach sensor according to priori knowledge obtained by early-stage testing, thereby obtaining stable and reliable output. According to the method, the interference of ghost introduced by false detection of a single sensor on the system is reduced, and the omission ratio of the system under specific complex working conditions can be reduced by complementary target information.

A text sentiment classification method based on a hybrid model belongs to the field of computer software. The method mainly comprises a data processing layer, a machine learning processing layer, a CNN processing layer, an LSTM-Attention processing layer and an adaptive decision layer. To-be-classified data is respectively processed by the machine learning processing layer, the CNN processing layer and the LSTM-Attention processing layer; and finally, the processing results are input into an adaptive decision-making layer together, and the adaptive decision-making layer adaptively adjusts theweight of each layer of result according to previous classification results of different processing layers, so that a final classification result is obtained. Compared with a single machine learning method and a single deep learning method, the method has the advantages that the processing effect is obvious, the result has good interpretability, and the application prospect is wide.

The invention discloses a model training and scene recognition method and device, equipment and a medium. The method comprises the following steps: training a first scene label and standard cross entropy loss of a sample image during the training of a scene recognition model, obtaining the parameters of a core feature extraction layer and a global information feature extraction layer through the training, then, according to a feature map output by the LCS module of each level and a loss value obtained by pixel-by-pixel calculation of the first scene label of the sample image, training a weight parameter of the LCS module of each level, and finally, training to obtain a parameter of a full-connection decision layer of the scene recognition model. Therefore, the scene recognition model has the high-richness feature extraction capability, scene recognition is performed based on the scene recognition model, and the accuracy of scene recognition is greatly improved.

The invention discloses a multi-target flexible job shop energy-saving scheduling method based on an improved grey wolf algorithm. The method comprises the following steps: firstly, constructing a flexible job shop energy-saving scheduling problem model; two-segment coding based on natural numbers is adopted; discretizing a continuity problem by adopting a mode based on an LOV rule; introducing an aggregation rate among individuals to obtain an initial population with relatively high quality; evaluating individuals in the initial population, determining an alpha wolf set, a beta wolf set and a delta wolf set of individuals of a decision-making layer according to a proportion, and adding a non-dominated solution into an external file; a dual-mode parallel search mode is used, tracking and searching capabilities are dynamically adjusted in the search process, improved tracking operation is introduced to improve the problem solving precision, variable domain search is adopted in the search mode, the evolution speed is improved, and a local optimal solution is broken through. According to the method, the machining sequence of the workpieces on the machines is reasonably arranged, and a better scheduling scheme is provided for a production enterprise from the three aspects of minimizing the maximum completion time, the total delay duration and the total energy consumption of the system.

The invention discloses a coal seam roof water and sand inrush risk evaluation method based on multi-source information fusion. The method comprises the following steps: comprehensively considering hydrogeological conditions, geological conditions, mining activity influence and other factors; taking the sand bed thickness, the base mining ratio, the sand burst fracture zone development height, thebed rock weathering index, the unit water inflow, the water saturation sand bed thickness, the permeability coefficient and the mining influence index as water and sand burst risk evaluation indexes,and on the basis, through a series of specific implementation steps of establishing a mathematical analysis model, establishing a decision-making level main control factor thematic map, constructinga judgment matrix to determine index weights, superposing the decision-making level main control factor thematic map and the like, finally, displaying the quantified result in the form of a graph by utilizing data storage, spatial data processing and analysis functions and the like of a geographic information system, and outputting a result graph through a graph output system, and achieving waterand sand bursting risk multi-source information partitioning according with the actual situation of a mine.

The invention discloses a financial data processing system. The system is constructed by being divided into four layers, and the four layers are respectively a data layer, a data management layer, ananalysis layer and a presentation layer. The data layer, the data management layer, the analysis layer and the presentation layer transmit financial data progressively in sequence. According to the processing system, financial operation monitoring, financial statement and accounting information analysis and evaluation, financial operation key index early warning and group company financial prediction are carried out, the system realizes short-term and long-term financial analysis, financial prediction and deep data mining, provides auxiliary decision support for a decision-making level, provides financial management information service for a business level, and provides an analysis and research system platform for financial analysis personnel.

The invention provides an expression recognition method based on classifier selective integration. The expression recognition method comprises the following steps: (1) generating a classifier pool; (2) evaluating the ability of an individual learner; (3) selecting a classifier sequence; (4) carrying out decision layer fusion. According to the invention, capability evaluation is carried out on prediction accuracy of individual learners and differences among different individual learners, and the capability evaluation is used as an evaluation standard of a selective integration algorithm for thecapability of the individual learners; an individual learner set with a better classification effect and better diversity can be selected according to the specific judgment condition of the verification sample, and the generalization ability of a classification system can be effectively improved when the expression data of batch unknown category labels is predicted.