35results about How to "Solving Adaptive Problems" patented technology

The invention discloses a circulating fluidized bed boiler device using low calorific value coal gangue for post ignition, and belongs to a combustion system of a coal fired boiler. The device mainly comprises a hearth, a cyclone separator, a riser, a circulating ash heat exchanger, a header, a fluidized bed slag cooler and the like. The device is characterized in that: heat exchange and combustion are separated; the combustion is performed in a heat insulation region; a heating surface is arranged in a fluidized bed with finer particles; the device comprises a fine particle fluidized bed, a circulating ash heat exchanger and a fluidized bed slag cooler; the bottom part of the hearth is divided into two fluidizing regions; the flow direction of the bed material between the two fluidizing regions is changed by changing the fluidizing air parameter and slag discharge of the two fluidizing regions so as to fulfill the aim of adjusting the bed temperature; and both the hearth and the circulating ash heat exchanger are communicated with the fluidized bed slag cooler, so that the ash and the slag of the fluidized bed slag cooler can be independently discharged to greatly improve the adaptability of the boiler to fuels. The device has the main advantages of burning the fuels such as coal gangue with a low calorific value and the like and solving the problem of heavy wearing of the heating surface well.

The invention discloses a maximum information entropy-based chemical data stream real-time abnormity detection method. According to the method, a real-time clustering algorithm DSC-Stream (The Dimension Space Cluster-Stream) is used; a data dimension space is used as a starting point to divide the data dimension space into a plurality of dimension clusters according to the maximum entropy principle; the data in the same dimension cluster forms a micro cluster so as to realize the real-time clustering of the data streams; the algorithm can be used for effectively decreasing the quantity of the invalid grids generated in the traditional density grid algorithm so as to greatly improve the operational efficiency; meanwhile, through a method for setting dimension cluster similarity, the algorithm can be used for effectively solving the problem that the traditional density algorithm is sensitive to the scanning radius and the density threshold to cause poor robustness and can be better adapted to the stream data.

A domain self-adaptive method and device based on comparative adversarial learning are provided. On source domain data, a loss function Lcls(xs, ys) is used for training a whole network model, parameters in a feature extractor are fixed, only classifiers C1 and C2 are updated, classification loss of the classifiers is minimized, target domain sample discrimination differences of the classifiers are maximized, the parameters in the classifiers C1 and C2 are fixed, Ldis is used for updating parameters in the feature extractor, and a self-adaptive loss item is reserved in the step. The classifier C1 and the classifier C2 respectively use features of different data enhancement modes, so that diversity of the classifiers is guaranteed, the double classifiers can more efficiently find samples at classification boundaries, features learned by the model contain more effective information, so that the problem of self-adaption in the unsupervised field is well solved. On the basis of a traditional double-classifier-based confrontation method, the disclosed method takes decision boundaries of classifiers on a target domain into account and also pays close attention to inter-domain differences.

The invention provides a three-dimensional data reconstruction method based on deep learning after three-dimensional optical measurement of an SMT surface mounted electronic component three-dimensional optical measurement. The three-dimensional data reconstruction method comprises the following steps of step that 1, restoring the surface of a measured object is restored through a phase displacement method; step 2, constructing a neural network, and obtaining a multi-angle 2D image of the detected object through multi-angle shooting to restore a 3D model of the detected object; step 3, synthesizing the final most perfect three-dimensional contour by using the two types of 3D model data obtained in the step 1 and the step 2; and step 4, performing cross-parallel ratio calculation on the model data pushed by the traditional method and the neural network, and when the score is relatively low, updating and training the neural network in the step 2 by using the data in the step 1.

The invention provides a text writing style processing method, device and equipment and a storage medium, and relates to the technical field of computers. The method comprises the steps of obtaining an input text of a first writing style of a target application scene; and processing the input text by adopting a pre-trained writing style migration model of the target application scene to obtain a second writing style text with the same semantics as the input text. The writing style migration model is a model obtained by training a corpus text of a target application scene and a pre-constructedpseudo-parallel data set; the pseudo-parallel data set comprises a plurality of data pairs, each data pair comprises a corpus text of a second writing style and a back-translation text of a first writing style, and the back-translation text is a text obtained by back-translating a corpus of the second writing style. According to the scheme, the self-adaption problem in the field of text writing style conversion can be effectively solved, and the accuracy of text writing style conversion is improved.

The invention, which is suitable for the field of electronic circuits, provides a simplex-duplex automatic adapted circuit and a system of portable navigation devices. The adapted circuit comprises a master control module, and a first electronic switch and a second electronic switch which are connected with the master control module, wherein, the second electronic switch is connected with a USB interface module, and the USB interface module is connected with a power-supply control module. According to the invention, the self-adaptation of different serial communication interfaces of portable navigation devices is solved, an effective method for connecting external serial devices with portable navigation devices is provided, thus the problem of adaptive line waste caused by using traditional methods is reduced.

Provided is an adaptive threshold multi-target falling detection method based on video processing. According to the technical scheme, the method includes following steps: step 1, acquiring image information of users, and recording an aspect ratio, an effective area ratio and a center change of a human body image in a minimum external rectangle of the user during normal activity and falling; step 2, giving different weights to different sensitive degrees of falling by employing the aspect ratio and effective area ratio of the human body image, obtaining new determination parameters, and realizing fusion of falling determination modes; step 3, setting optimal thresholds for users of different body types; and step 4, realizing target falling detection according to the acquired image information of the users in step 1 and the fusion of the falling determination modes in step 2 and with the combination of the set optimal falling thresholds of the users in step 3. According to the method, the defect that people easily forget and contradict wears in a wearable type falling detection scheme is overcome, and difficulties of high misjudgment rate, expensive price and inconvenient laying dueto an environment type falling detection system are solved.

The invention provides a variable coefficient and local entropy combined driving radar image non-local mean value method, the pixel value of a current pixel point is solved according to the current pixel weight value and the value of each pixel in a search window, and the current pixel weight value is determined according to the ratio distance and the attenuation coefficient corresponding to eachpixel point. Compared with an existing speckle suppression method, the method has better smoothness in a uniform region, can also have good edge retention characteristics in an edge region, and also has a better effect in the aspect of image visualization.

The invention provides a humanoid intelligent control method for an unmanned aerial vehicle. The method comprises steps of determining an unmanned aerial vehicle flight mathematical model; designing ahumanoid intelligent controller; the humanoid intelligent control comprising a three-layer structure of an operation control layer, a parameter correction layer and a task adaptation layer; and finally, optimizing a proportional differential coefficient by adopting particle swarm optimization. The humanoid intelligent controller can determine a change strategy according to the deviation of the adjusted quantity and a change trend of the deviation, and a PID control parameter self-adaption problem of the four-rotor aircraft can be effectively solved, however, the proportion, the differential coefficient and the attenuation coefficient in humanoid intelligent control can influence the final control effect, so the particle swarm optimization algorithm is adopted to perform parameter optimization. Therefore, the humanoid intelligent controller based on the particle swarm algorithm can realize quick response of the angle of the unmanned aerial vehicle, and stability and anti-interference performance are improved.

The invention provides a text style migration technology-based character beautifying method, which comprises the following steps of S1, acquiring a first input text of a user, extracting keywords in the first input text, and performing label classification on the keywords; S2, obtaining a second input text of the style requirement input by the user, and judging a theme to which the style requirement belongs through the language model; S3, extracting and describing a plurality of corpus texts conforming to the same style theme through the language model, and performing sorting processing; S4, according to the label type to which the description main body belongs, introducing a related description main body with close relevance, and according to a corpus text with the related description main body, extracting and taking the corpus text as a supplementary description text; and S5, outputting a target text of a second writing style having the same semantics as the first input text. According to the method, the conversion of the text writing style is realized, the self-adaption problem in the field of the text writing style is effectively solved, and the accuracy of the conversion of the text writing style is improved.

The invention discloses a voice adaptive recognition method, system and device and a storage medium, solves the problem of online speaker adaptation of voice recognition acoustic modeling, realizes adaptive learning of a main network (namely an acoustic model) by introducing an auxiliary network (namely a network formed by a plurality of attention-based gating scaling adaptive layers), and simultaneously realizes adaptive learning of the main network (namely the acoustic model). And the auxiliary network adopts a self-attention mechanism to learn more distinctive speaker personality features, so that the recognition rate is improved.

The present invention discloses a method and device for sending data. The method comprises: a sending node sends data to at least two receiving nodes at the same time; the sending node receives feedback information sent by the receiving nodes and calculates a modulation and coding strategy according to the feedback information; and the sending node sends data to the at least two receiving nodes according to the modulation and coding strategy obtained through calculation, so that a link adaptation problem of point-to-multipoint transmission is solved, and air interface efficiency of point-to-multipoint transmission is greatly improved.

The invention discloses a flange-free waveguide fixing support and an installation method, and the structure of the fixing support comprises a cylindrical support, a waveguide tube clamp, a first fastener, a first deformable clamping sleeve, a waveguide tube, a second deformable clamping sleeve, and a second fastener. According to the current structure, the space and weight needed by fixing the waveguide can be reduced greatly, the number of waveguide layers can be flexibly increased or decreased by increasing the number of the deformable clamping sleeves and the number of the waveguide tube clamps according to the actual waveguide channel layout condition, the problem of waveguide position matching deviation caused by the installation error influence of related equipment at the two ends during waveguide tube installation is solved, and good adaptability and flexibility are achieved.

The invention relates to a carrying tool and particularly discloses a carrying tool suitable for a staircase. The carrying tool suitable for the staircase comprises a carrying frame. The carrying tool suitable for the staircase is characterized in that a rotating shaft is arranged at the bottom end of the carrying frame, the two ends of the rotating shaft are respectively provided with three wheel shafts which are evenly distributed, rotating wheels are arranged at the tail ends of the wheel shafts, and a pull rod is arranged at the front end of the carrying frame. The carrying tool is simple in structure, convenient to use, reasonable in design and flexible in application, effectively achieves carrying of a heavy object in the staircase, saves labor and carrying time and is suitable for wide application and popularization.