36results about How to "Improve decomposition accuracy" patented technology

The invention provides a method for recognizing and separating magnetotelluric signal and noise. Firstly, a magnetotelluric signal collected in a strong electromagnetic interference environment is equidistantly segmented, by optimizing inherent time-scale decomposition, each magnetotelluric signal segment is subjected to adaptive decomposition to obtain a rotational component, and a sample entropyand a fuzzy entropy of the rotational component are extracted. Then, the sample entropy and the fuzzy entropy are taken as joint feature parameters to carry out fuzzy C-means clustering, according tothe feature parameters and a clustering method, a measured magnetotelluric signal sequence is divided into two kinds including a useful signal and a strong interference signal. Finally, only a magnetotelluric signal segment which is recognized as the strong interference signal is subjected to noise suppression processing by using a wavelet threshold, and a reconstructed magnetotelluric signal isobtained after superposition. The method has high reliability.

The invention discloses a reciprocating compressor bearing fault diagnosis method based on improved local mean value decomposition, which solves the problem that the traditional method fault information is less prominent and the accuracy is not high. The reciprocating compressor bearing fault diagnosis method comprises steps of using a sensor and a data collection system to test the reciprocating compressor vibration signal, using the improved local mean value decomposition method to decompose the vibration signal of the reciprocating compressor to form a series of PF components and decouple the fault information, adopting a correlation coefficient method to screen out the PF component containing the main state information from the massive PF components, and finally, calculating the transient amplitude spectrum of the chosen PF component, extracting the fault characteristic frequency, and determining the fault state. Compared with the diagnosis method based on the traditional improved local mean value decomposition, the fault characteristic frequency extracted by the improved method is more prominent and more accurate diagnosis on the fault of the reciprocating compressor bearing is realized.

The present invention discloses a complementary integrated empirical mode decomposition method for adaptive determination of decomposition parameters. The method is characterized in that: complementary integrated empirical mode decomposition is performed on original signals by adopting a method of gradually increasing the integrated average number of times m and gradually increasing the amplituderatio coefficient k, so that optimal decomposition parameters of the complementary integrated empirical mode decomposition can be adaptively determined, modal aliasing can be effectively suppressed, different decomposition parameters can be adaptively determined for different original signals, decomposition accuracy and computational efficiency of complementary integrated empirical mode decomposition are ensured, and the intrinsic modal components obtained by decomposition are more accurate and can more effectively represent the physical meaning of the original signals.

The invention discloses a mixed type fiber-optic gyroscope signal filtering method based on EEMD and FIR. The method specifically comprises the following steps that 1, a fiber-optic gyroscope signal is decomposed through an EEMD algorithm to obtain IMF components and residual errors of all layers; 2, Hilbert transform is performed on the IMF components of all the layers to obtain amplitude values and instantaneous frequencies of all the layers, threshold values are reckoned according to the instantaneous frequencies, and weight values are calculated through the threshold values; 3, FIR filtering processing is performed on the IMF components and the residual errors of all the layers by adopting an FIR filter, and then new IMF components and new residual errors of all the layers are obtained; 4, weighting reconstruction is performed on the new IMF processed by the FIR filter, and finally a denoised result is formed. According to the mixed type fiber-optic gyroscope signal filtering method based on the EEMD and the FIR, the advantages of the two methods are integrated, the data decomposition precision is improved through the EEMD method, low-pass filtering is further performed through the FIR filter method, therefore, the frequency precision during EEMD signal processing is improved, and the filtering effect is significantly enhanced; meanwhile, it is guaranteed that the method is still based on data, modeling does not need to be performed on the data, and the application range is wide.

The invention relates to the technical field of power systems, in particular to a non-intrusive residential user load decomposition method based on a residual convolution module. The method comprises the following steps: acquiring training data and preprocessing the data; constructing and training a load decomposition model: inputting a total active power sequence in training data into a residual convolution module, learning active power features by taking a CNN model as a basis in the residual convolution module, adding original input data and feature data learned by the CNN through cross-layer connection, further inputting the obtained data into a GRU network to learn time sequence features, and outputting a predicted value of the active power of the target electric appliance; comparing the predicted value of the active power of the target electric appliance with a true value, and continuously adjusting network parameters of the load decomposition model to obtain a trained load decomposition model; and decomposing the total active power of the user to be decomposed through the trained load decomposition model to obtain an active power decomposition result of the target electric appliance. The method is high in decomposition precision.

The invention discloses a wavelet packet extraction method for grid-connected inverter network-side harmonic current information. The method is based on a wavelet tree optimized decomposition and reconstruction algorithm for db4 wavelet packets. A maximum harmonic current component real-time extraction technology based on wavelet packet transform is put forward. According to the method of the invention, the wavelet frequency band, the maintaining frequency band and the zero-setting frequency band are judged by the threshold criterion after each layer of decomposition, and only the wavelet frequency band is decomposed during next layer of decomposition. The method has the advantages of small amount of decomposition operation and high wavelet decomposition precision. Compared with the prior harmonic current information extraction method, the method of the invention takes into consideration the computation limit of a digital processor, the amount of computation of decomposition and reconstruction in the process of harmonic information extraction is reduced without sacrificing the amount of maximum harmonic current information, and the method is more practical in engineering and has the advantage of high harmonic information extraction precision.

The invention relates to a power quality disturbance classification method based on an EWT-MPE-PSO-BP neural network, and the method comprises the steps: carrying out the accurate mode decomposition with anti-noise performance of different types of disturbance signals through EWT, and obtaining the mode components of different frequencies; then, introducing a time scale concept, and optimizing the traditional permutation entropy so as to be better suitable for a complex system problem; thirdly, introducing PSO to optimize the BP neural network, and converting the problem of searching for the minimum error in BP into the problem of searching for the optimal position of PSO, so that the defect that the convergence speed in the BP neural network is low is overcome, and the working efficiency of the BP network is improved; and finally, taking the extracted characteristic quantity as the input of the optimized neural network, and obtaining a final power quality disturbance classification result through multiple times of training. The method provided by the invention effectively solves the problems of inaccurate disturbance signal detection and low classification process speed, and is high in accuracy and high in working efficiency.

The disclosure proposes a multi-path spectrum analysis method and system of a slot wave signal. Detection signals are transmitted at different angles to obtain a plurality of slot wave signals, thereby realizing multi-directional detection of the target geological interface, increasing the detection range, and reflecting geological information of a to-be-detected stratum. In terms of slot wave signal extraction, the slot wave signals are decomposed by means of non-orthogonal decomposition to decompose multiple paths in the slot wave signals and a plurality of echo signals are fused to obtain amulti-path spectrum of the slot wave signals. The decomposition mode has the high decomposition accuracy and high practicability; the decomposed signals can be preserve the characteristics of interface signals as many as possible and the obtained signals can be expressed simply, so that the signal-to-noise ratio of the signal is increased, the error rate of the signal is reduced, and the accuracyof the slot wave detection is improved.