30results about How to "Reduce data complexity" patented technology

The invention belongs to the technical field of electric energy quality classification and identification, and discloses a voltage sag disturbance classification method based on a Long Shore-Term Memory algorithm. According to the method, firstly, voltage sag disturbance data are processed through a data processing layer, then feature values are extracted through two layers of LSTM networks, feature dimension reduction processing is conducted through a full connection layer, and finally classification recognition is conducted through a sigmoid network layer. The electric energy quality sag disturbance correlation characteristics are extracted through the deep learning algorithm, and the limitation of a classification recognition modeling method based on time domain, frequency domain, transformation domain and other physical characteristics in the complex power grid environment in the aspects of adaptability, algorithm efficiency and accuracy is overcome. By using the model, the defectsof tedious and complex extraction steps, possible loss of original characteristics of part of signals and the like in the traditional method can be well overcome; The defect that gradient explosion is likely to happen to a traditional recurrent neural network (RNN) is overcome, and high voltage sag disturbance recognition accuracy is achieved.

A supply network for supplying consumers with a medium is provided. Based on a reduction of data and a subsequent reconstruction of at least one pressure and/or flow value, the state at least at one further location in the supply network can be determined without having to be measured. An improved actuation of the at least one pump and/or of the at least one valve is thus possible. A device for pressure control in a supply network is also provided.

The present invention relates to a method for aligning a camera sensor to significant data which is text or barcode data to be recognized comprising the steps of:—capturing an image of the significant data by means of the camera sensor; —detecting a predominant alignment line of the significant data and detecting an angle thereof in relation to a horizontal line of the captured image; —determining image sections within the edge and line enhanced image which contain most likely significant data lines; —selecting a representative image section out of the determined image sections which is aligned with the predominant alignment line; —capturing a following image of the significant data; tracking the representative image section and determining the predominant alignment line out of the representative image section to achieve a fast calculation and audio or tactile feedback of the alignment quality to the user.

The present invention provides a reception apparatus for performing security tunneling and data re-transmission and unidirectionally receiving data from a transmission apparatus and a transmission apparatus for unidirectionally transmitting data to the reception apparatus. When the reception apparatus checks that there is an error in received data, the reception apparatus performs a switching operation or transmits a switching request signal to the transmission apparatus to notify of the error of the received data to the transmission apparatus. When the transmission apparatus has detected the switching, the transmission apparatus re-transmits data, in which an error is generated, to the reception apparatus. The transmission apparatus encrypts the data and transmits the encrypted data to the reception apparatus. According to the present invention, reliability and security of one-unidirectional data communication are improved.

The invention discloses an industrial building concrete carbonization depth prediction method. The method comprises the steps: obtaining historical data of industrial building carbonization depth in anatural environment and dividing the historical data into a training set and a test set; performing adaptive decomposition on the training set and the test set by using a variational modal decomposition method to obtain modal component sub-sequences of the training set and the test set; training each modal component sub-sequence of a gating cycle unit by using the modal component sub-sequence ofthe training set as a training prediction model, and establishing a gating cycle network carbonization depth prediction sub-model of each modal component sub-sequence of the training set; and predicting the modal component sub-sequences of the test set by using the gated cycle network carbonization depth prediction sub-model of each modal component sub-sequence of the training set, accumulating prediction values, and performing comparative analysis on the prediction values and a real value to complete industrial building concrete carbonization depth prediction evaluation. According to the method, the characteristics of the industrial building carbonization depth time sequence are mastered more meticulously, and the accuracy of carbonization depth prediction is greatly improved.

A non-uniformly sampled three-dimensional (3D) information representation method is revealed. Firstly, use a computer to reconstruct a 3D model according to a 3D model data. Then set up a projection point according to the 3D model. The projection point is projected onto the 3D mode so as to get a plurality of sampling points uniformly. Then create at least one 2D array data according to a plurality of horizontal parameters and a plurality of vertical parameters of a 3D sample model. The 2D array, data corresponds to the sampling points. Take samples from non-uniform area of the 3D model so as to generate a non-uniform element. Thus, the 3D model data is simplified and this is beneficial to representation, operation, storage and transmission of 3D data.

The invention provides a pilot working state recognition model establishing method. The method comprises the following steps of: S1, preprocessing flight data; S2, extracting features of the preprocessed flight data; S3, carrying out data dimensionality reduction on the extracted features by utilizing a Trelets algorithm; and S4, carrying out training and modeling on the dimensionality reduced data by utilizing a Gaussian process classifier optimized on the basis of cuckoo algorithm. The method has favorable time domain localization ability and is immune to noise interference so that the feature values are more cautious. According to the method, original data is reduced to lower dimensionality, so that the complexity of the dimensionality reduced data is reduced and basis is laid for further more correct classification; the parameter optimization is relatively easy and easier to converge; and global or approximate global optimization results are searched for functions as far as possible, so that the local optimization of models is avoided.

A method of forming a video signal by a decoder device is described wherein the method comprises receiving a bit-stream and decoder information, the decoder information signaling the decoder device the presence of one or more resolution components in the bitstream for forming a video signal, a resolution component representing a spatially subsampled version, preferably a polyphase subsampled version, of a first video signal having a first resolution, the one or more resolution components being part of a resolution component group, the group comprising a plurality of resolution components on the basis of which the first video signal is reconstructable; and, the decoder device parsing the bitstream and decoding the one or more resolution components into video frames on the basis of the decoder information.

The invention discloses a diagnosis system and method for fault classification and positioning of a transformer winding. The system comprises a sweep generator, a transformer winding, a detection terminal, an inductor L1, a divider resistor R1, a divider resistor R2 and a capacitor C. The sweep generator outputs a sweep-frequency signal, the sweep-frequency signal is input to the transformer winding from a grounding point N through the inductor L1 and the divider resistor R1, flows through a high-voltage winding A in the transformer winding and is output, and after the sweep-frequency signal passes through a capacitor C and is divided by the divider resistor R2, sweep-frequency data is collected by the detection terminal; and the detection terminal diagnoses the fault type and position of the transformer winding through a constructed deep learning convolutional neural network model. According to the method, the fault positioning problem is solved on the basis of realizing high-precision fault type diagnosis, and the method has certain engineering significance. The method and the system have a good effect on a simulation sample constructed in the invention, the fault type diagnosis precision is up to 100%, and the fault positioning accuracy is up to 83.33%.

The invention provides a method of correcting supersonic wave partial discharge detection. The method comprises the following steps of extracting a plurality of partial discharge signal linearity frequency spectrums from historical data and then processing to acquire a coefficient matrix and a principal component characteristic of each partial discharge signal, and constructing a classification model of the partial discharge signals and each type of a linearity frequency spectrum mean value; acquiring a linearity frequency spectrum mean value of pure noise signals, and according to the coefficient matrix, acquiring a classification model of pure noise supersonic wave signals; adding each type of the linearity frequency spectrum mean value of the partial discharge signals to the linearity frequency spectrum mean value of the pure noise signals and correcting the classification model of the partial discharge signals; acquiring a signal to be detected, according to the coefficient matrix, acquiring a principal component characteristic of the signal to be detected and introducing into the classification model corrected by the pure noise and the partial discharge signals, and screening signal output with a minimum Euclidean distance in the corresponding classification model. By using the method, an artificial factor interference is avoided, an introduced characteristic dimension is low and training data is less so that purposes of labor saving and time saving, and an objective detection result are reached.