197 results about "Fuzzy membership function" patented technology

The invention discloses a speech emotion recognition method. The method includes the steps that firstly, a speech signal is converted into a spectrogram to serve as initial input; secondly, a deep convolutional neural network is trained to automatically extract emotion features; thirdly, a stack type auto-encoder is trained for each kind of emotions, and all the stack type auto-encoders are fused to automatically construct membership functions of an emotion fuzzy set; fourthly, the features obtained in the second step are subjected to feature optimization by means of the fuzzy optimal theory in the third step; fifthly, emotion classification recognition is conducted by means of a Softmax classifier. The method takes abstract fuzzy properties of speech emotion information into consideration, the extracted emotion features are subjected to selective fuzzy optimization to improve the significance of the features, fuzzy membership functions in the fuzzy theory are automatically constructed by means of the concept of deep neural network layer-by-layer training, and the problem that the proper membership functions in the fuzzy theory are difficult to select and determine is solved.

Material selection systems and methods for quickly identifying which existing material best matches a desired set of properties are described so that product development time can be minimized. Users may input the properties they desire in a material, which properties they want searched and scored, the acceptable values of those properties, and a priority value for each property. Preliminary matching materials may be retrieved from a materials database, and an index value for each property value may be calculated. One of four fuzzy membership functions may then be utilized to calculate a scored property value for each property of each material. The scored property value may then be weighted to account for the priority value assigned to each property. The results may then be sorted in descending order based on their overall match scores, and output to the user so the best matching material(s) is readily identifiable by the user.

Disclosed herein are a system and method for evaluating the performance of infrastructure, in which a condition rating according to a result of a survey targeting experts is expressed as a sequentially distributed probability distribution function using a fuzzy membership function and a comprehensive performance measure score is more accurately expressed as a probability distribution function, thereby accurately evaluating a comprehensive performance measure rating. Also, a utility value is derived using a utility function and usable and functional evaluation measures for each facility of the infrastructure are collected, thereby comprehensively determining subjective performance evaluation measures of the infrastructure.

The invention discloses a fault tree and fuzzy neural network based automobile crane fault diagnosis method. The method includes: (1) establishing a top event fault tree of an automobile crane by a deductive method; (2) determining the numbers of input and output nodes of a fuzzy neural network according to fault tree branches and experiential knowledge, and establishing a structural model of the fuzzy neural network; (3) extracting a training sample according to knowledge contained in each branch of the fault tree, training the neural network, and establishing a network weight and a threshold matrix needed for neural network reasoning and calculation; (4) monitoring data on a platform by the aid of an existing automobile crane state, and applying a 3sigma criteria method in a statistical parameter method for determining a fuzzy membership function needed for fuzzy preprocessing; (5) inputting measured data into the fuzzy neural network for calculation, and outputting a fault mode. By the method, blindness and complexity during detection are avoided, and accuracy rate of diagnosis is increased.

The invention discloses a method for optimizing demand-side time-of-use power price based on photovoltaic grid-connected uncertainty. The method comprises the following steps: obtaining the daily load data after distributed energy is connected, and determining load peak-valley periods according to a fuzzy membership function; determining the objective function and the constraint condition of the optimization method of the demand-side peak-valley time-of-use power price in an operation cycle, and creating a peak-valley time-of-use power price optimizing model; by using a chance constraint theory, converting power balance constraint into deterministic equality constraint; and obtaining optimal peak-valley time-of-use power price by using a particle swarm optimization algorithm. The method provided by the present invention, based on peak shaving and load shifting adjustment of a power distribution network demand side according to the peak-valley time-of-use price optimization model, takes account of the connection of distributed power. At the same time, the optimization method adopts the chance constraint theory to solve the problem of prediction uncertainty of distributed power grid-connected power, reduces the decision risk caused by uncertainty and improves the pricing rationality of peak-valley time-of-use price.

The invention provides a product recommendation method based on deep learning. The method comprises the following steps: S1, crawling comment data of a product by means of a crawler; S2, conducting data preprocessing of the comment data; S3, conducting feature extraction of the data; S4, carrying out fine grain analysis of product comments; S5, giving quantitative marks to the product comments; and S6, with collaborative filtering combined, performing product recommendation. Meanwhile, the invention also provides a product recommendation system based on deep learning. Compared with the prior art, the method and the system combine the deep learning method to refine texts and quantifies the texts through a fuzzy membership function, can convert user comments into the marks of attributes of a product, then integrates a collaborative filtering method for recommendation, and can achieve better recommendation results.

The invention discloses a speech-emotion recognition method based on improved fuzzy vector quantization. The method extends the sum of fuzzy membership function from one to N so as to reduce the influence of sample wild-point on an iteration-training process to a certain extent, and adopts a clustering method based on similarity threshold and a minimum distance principle in the iteration-training process so as to avoid the problem that a clustering center is sensitive to initial values and easy to fall into local minimum values to a certain extent. Experimental results prove that the method can effectively improve the emotion recognition rate of the prior fuzzy vector quantization method.

The invention provides a case knowledge base representation and case similarity obtaining method. The case knowledge base expression and case similarity obtaining method comprises the steps of improving semantic representation capacity of semantic web standard description languages, establishing a case knowledge base and obtaining case similarity, wherein the step of improving the semantic representation capacity of semantic web standard description languages is that an N-element relation model, a fuzzy membership function and relation weight are introduced on the basis of semantic web standard description languages to improve the semantic representation capacity of the semantic web standard description languages; establishing a case knowledge base based on the strengthened semantic web standard description languages, wherein the case knowledge base include case knowledge representation models, the case base, a semantic web ruler base and an inference mechanism based on the semantic web standard description languages; obtaining case similarity. By means of the case knowledge base representation and case similarity obtaining method, the semantic representation capacity of the semantic web standard description languages can be improved so as to accurately represent case knowledge, facilitate case knowledge management and share and improve case recommendation speed and recommendation accuracy.

The invention provides a speech separation method based on a fuzzy membership function, and belongs to speech separation methods. The fuzzy membership function is combined in the speech separation method, so that more accurate definition of a membership degree of speech time frequency units to a target signal is obtained. An auditory oscillation model is built through human ear auditory system simulation, and speech pitch characteristics are extracted. The speech time frequency units are marked according to pitch cycle characteristics to form foreground streams and background streams. Whether the corresponding time frequency units are targets or noise is judged according to different marks. In the synthesis stage, a target unit multiplies a high weight, a noise unit multiplies a low weight, and resynthetized speech is obtained. By means of the speech separation method, the pitch cycle can be estimated more precisely, the time frequency units can be marked more accurately on the basis of characteristic clues, and the more complete target speech can be obtained. Due to the fact that the method is based on the pitch characteristics of the speech, good separation effects in complex and non-stationary noise are achieved, and the application range is wide.

The invention relates to an infrared image enhancement method for electrical equipment based on non-downsampling shear wave transform. The method comprises the following steps: step S100, establishinga mapping relationship between RGB and gray value of an infrared image; Step S200, performing non-downsampling shear wave transform on the infrared image of the electric equipment obtained in the experiment; S300, calculating the maximum inter-class difference of the infrared low-frequency image of the electrical equipment, obtaining a segmentation threshold between the background and the foreground, and segmenting the low-frequency image into two classes of the environmental background and the electrical equipment; S400, enhancing the low-frequency image, linearly enhancing the low-frequencyforeground electrical equipment, and performing histogram equalization and enhancement on the low-frequency background; Step S500, designing an optimized fuzzy membership function, and carrying out fuzzy enhancement on the high-frequency image of the electrical equipment; Step S600, synthesizing the enhanced low-frequency coefficient and high-frequency coefficient according to the NSST inverse transform into a new electrical equipment infrared image. Compared with the prior art, the invention has the advantages of higher image quality, stronger practicability and the like.

The invention relates to a method for fault diagnosis of rotating mechanical equipment based on interval-type evidence fusion, belonging to the technical field of fault monitoring and diagnosis of rotating mechanical equipment. The method is based on statistical analysis of typical fault data, comprising the following steps: building a set of fuzzy membership functions, and modeling each fault model mode in a fault archival repository by using the set of functions; modeling a fault inspection pending mode extracted during on-line monitoring by using a single membership function; matching the inspection pending mode with each model mode to obtain a reliability match interval of each fault supported by the inspection pending mode; providing an interval-type diagnostic evidence method obtained from the reliability interval by using the Monte Carlo Latin Hypercube sampling method; and fusing evidences, and then under the decision criterion, making the fault decision according to the fusion result. The decision based on the multi-evidence fusion result is more accurate than the decision made by the single-evidence fusion.

The invention discloses an admission control method and an admission control system in a heterogeneous wireless network environment. The method comprises the following steps of: when an access request is received, collecting the multi-domain information of each network and acquiring a target quality of experience (QoE) vector of a user in the access request by using an admission control entity; performing granularity partition on the information of each attribute domain in the multi-domain information according to a predefined performance index to form information granularity vectors of each attribute domain; according to the target QoE vector of a user terminal and a predefined fuzzy membership function, establishing fuzzy membership grade vectors of resource grains in each attribute domain; based on fuzzy evaluation vectors and inference rules of the resource grains in each attribute domain, establishing mapping relationships between the fuzzy evaluation vectors of the resource grains in each attribute domain and a network comprehensive evaluation index so as to acquire a comprehensive evaluation vector; and according to comprehensive evaluation, selecting the network having the highest network grade of service as an access network. Through the method and the system, an effective admission control strategy is designed for the heterogeneous wireless network and the overall performance of the heterogeneous wireless network is improved.

The invention relates to a method for determining abnormal gait comprising the following steps: (a) measuring ground reaction force generated during walking by a plurality of sensors arranged on the left-foot and the right-foot respectively; (b) applying measurements from each of the plurality of sensors to a predetermined fuzzy membership function to transform the measurements into the first fuzzy values; (c) applying the first fuzzy values to a predetermined fuzzy logic to generate the second fuzzy values for a plurality of gait phases; and (d) comparing the second fuzzy values with pre-stored data of normal gait to determine whether it is abnormal gait or not. Therefore, it is possible to determine abnormal gait more accurately even with fewer sensors.

The invention relates to a multi-target differential grey wolf algorithm-based reactive power optimization method of a power distribution network. Photovoltaic and load time sequence fluctuation is considered, a DSTATCOM is introduced and used as a compensation connected to an active power distribution network, segmentation is performed by taking hour as a time segment, the dynamic reactive powerof the DSTATCOM is smoothly changed according to change of an equivalent load after photovoltaic and load, fluctuating according to a time sequence, connected to the power distribution network, and the active network loss and the voltage deviation are reduced to the maximum extent under the condition that the minimum reactive compensation capacity is output. In order to solve the problem of multiple targets in a reactive power optimization model, an original grey wolf algorithm is improved, variation and cross in a differential algorithm are introduced, and multiple targets are processed by rapid non-domination sequencing, congestion distance and fuzzy subjection function. By the multi-target differential grey wolf algorithm-based reactive power optimization method, the influence on systemnetwork loss and voltage after time sequence photovoltaic and load connected to the power distribution network is effectively solved; and with the adoption of the multi-target differential grey wolfalgorithm, the problem of multi-target non-linear reactive power optimization is processed, and the global and local searching capability is balanced.

The invention provides a cold-rolled strip steel plate shape prediction control method, comprising the following steps: for cold-rolled strip steel of the same specification, establishing a corresponding plate shape prediction control fuzzy reasoning model; setting fuzzy membership functions of various parameters in the plate shape prediction control fuzzy reasoning model by being combined with the characteristics of influence of rolling force change on the strip steel plate shape; establishing plate shape fuzzy prediction control models by utilizing a Takagi-Sugeno fuzzy model modeling rule; selecting a corresponding plate shape fuzzy prediction control model to carry out online adjustment on a working roller bending device. A dynamic relationship is established among rolling force variation, forward pull variation of a roll mill, backward pull variation of the roll mill, and online adjustment variable of the working roller bending device by using a fuzzy modeling method, the adverse effect of transmission time lag existing between the roll mill body and a plate shape instrument on the plate shape control at the outlet of the cold-rolled strip steel, and two familiar plate shape defects of intermediate waves and edge waves existing in the cold-rolled strip steel products are effectively overcome.

The invention discloses a D-S-evidence-theory-based distribution automation terminal state diagnosis method. The method comprises: statistics of designated state characteristic quantities is carried out from operating information uploaded to a main distribution station from a distribution automation terminal; distribution automation terminal states are graded and a fuzzy membership function, corresponding to different distribution automation terminal states, of the state characteristic quantities is constructed; various membership parameters of the fuzzy membership function are determined; with the distribution automation terminal state grading result as a single element proposition, a basic probability assignment function and an evidence fusion result are calculated for the single element proposition and an identification frame; and matching is carried out on the evidence fusion result and a preset terminal state diagnosis decision-making criterion, so that a current state of the distribution automation terminal is obtained. According to the invention, the state of the distribution automation terminal can be determined accurately based on the operating information; the detection result is accurate and reliable; the diagnosis becomes convenient and rapid; and the operability is high.

The invention relates to a method for designing an industrial alarm on basis of alarm evidence fusion, which belongs to the technical field of industrial alarm design. The method is based on a fuzzy subjection function to construct a fuzzy threshold, the measured values of a process variable at each moment are compared with the fuzzy threshold to calculate an alarm evidence, and the alarm evidence shows the uncertainty degree that the value of the process variable exceeds or is lower than the fuzzy threshold; a linear weighting evidence is utilized to update rules, the alarm evidence of the current moment is fused with the alarm evidence in the past time, so the whole alarm evidence of the current moment is obtained, the alarm is judged to be sent out or not according to related judgment standards, the fusion process can effectively reduce the influence of uncertainty, so the alarm accuracy is improved; and a test sample is extracted according to the measured data of the process variable. A method for finding out the optimal fuzzy threshold under the lowest false alarm rate and missing report rate standards is provided.

A technical method of diagnosing rationality of city spatial patterns includes the following steps: 1, constructing an index system comprising a master target layer, a sub-target layer, a factor layer and a divisor layer; 2, after standard data processing, using an analytic hierarchy process and an entropy weight process to determine a weight of the divisor; 3, constructing HL comprehensive diagnosis models for rationality of city developing patterns, the models comprising a scale pattern USR diagnosis model, a spatial pattern UKR diagnosis model and a function pattern UFR diagnosis model; 4, using a fuzzy membership grade function process and a linear weighting summation process to calculate a rationality index for each sub-model; and 5, dividing cities into a highly reasonable city, a relatively reasonable city, an intermediate reasonable city, a lowly reasonable city, and an unreasonable city according to HL comprehensive indexes. The comprehensive diagnosis index system provided by the present invention has universality and comprehensiveness, and has availability and effectiveness when evaluating the data. The completed system evaluating method enables the calculated result of the rationality indexes of the urban developing patterns to be more accurate, and achieves a more scientific evaluation of the urban developing patterns.

The invention discloses a network comprehensive performance evaluation method oriented to service. According to the method, at first, all current service types are analyzed to obtain proportions of service of the various types; then the service with the highest proportion is selected, for the requirement for the network performance of the type service, a set of network parameters most important for service experience of the type service are selected for the type service to serve as evaluation indexes; a low-complexity FAHP is adopted for processing the parameters, and weight vectors of the parameters are obtained through calculation; network parameter values of a current network are measured in real time, a mode of fuzzy membership functions is introduced, a score corresponding to the measurement value of each network parameter is calculated, and the scores are weighted to obtain a performance evaluation index result of the service; evaluation results of other service in the network can be obtained according to the similar steps, finally, the proportions of all kinds of service are utilized for weighting the evaluation scores of the service, and the final network comprehensive performance evaluation result is obtained.

The invention discloses a tea infrared spectrum classification method of fuzzy discrimination clustering. A linear discrimination analyzing method is employed to extract the identification information of 14-dimensional training sample data, and 14-dimensional test sample data is projected to a discrimination vector to obtain the two-dimensional test sample data. The two-dimensional test sample data are subjected to fuzzy C-means clustering. A fuzzy interclass scattering matrix is calculated according to an initial clustering center, and the fuzzy total scattering matrix is calculated. An eigenvector is calculated according to the fuzzy interclass scattering matrix and the fuzzy total scattering matrix. A clustering central value is calculated in a characteristic space through the fuzzy membership function value. The average value of each 14-dimensional training sample is calculated respectively, and the Euclidean distance of the average values of the clustering central value and the training samples of the test samples. If the Euclidean distance from the clustering central value to the training samples is minimal, the tea belonging to the clustering central value is of the same type with the tea of the training samples, thereby realizing correct classification of different tea types.

The invention relates to a light face recognition method, which relates to the field of the computer mode recognition. According to the method, the influence of the light variation on the face recognition is improved on the aspects of pretreatment, characteristic extraction and classifier, so that the influence of the light variation on the face recognition performance is effectively reduced. By utilizing a coarseness zoning algorithm, a continuous contrast range interval is divided into a plurality of discrete sub-intervals in a nonlinear way according to the information of the light variation, the local nonlinear multilayer contrast characteristics (LNMCP) are extracted as the face characteristics, and the defects of the local binary pattern (LBP) and the LMCP can be effectively compensated; and a fuzzy membership function is introduced to store the probability of a measured sample belonging to different categories, and finally the weight of each layer is calculated through the information entropy, so that each layer of classification sub-results can be precisely merged together. Due to the adoption of the light face recognition method, the face recognition rate is effectively increased.

The invention relates to a distribution network connection method and belongs to the distribution network method technical field. The distribution network connection method adopts that a primitive connection type is the basic unit of the distribution network transmission electric energy; a point-to-point connection type from a power supply to a load comprises an overhead primitive connection and a cable primitive connection. The distribution network connection method identifies the types and magnitudes of the primitive connection, forms the distribution network connection method in an inferring mode, and accesses the types of the distribution network connection. The appraisal types of the distribution network comprise a reliability access, an economic access, and a security access. The distribution network connection method chooses an optimized connection model, introduces obscure subordinating degree function to solve the problem of inconformity of all kinds of index dimensions, and achieves selecting the optimized connection type under different emphasis through dynamic weighting. The distribution network connection method has the advantages of avoiding a former inherent flaw of the connection type, improving the security index, an obscure subordinating degree and a weighting thought, providing novel synthetic evaluation module for the distribution network connection, and has a significant practical value and referential meaning on construction of the distribution network by introducing an obscure inference method to constitute the distribution network connection type.