215 results about "Wavelet basis functions" patented technology

The invention discloses a method for denoising acoustic testing data of porcelain insulator vibration based on wavelet decomposition threshold denoising. The method comprises the following steps: using a vibration acoustic method to detect a porcelain insulator to acquire a porcelain insulator vibration response signal containing noise; selecting a suitable wavelet basis function for the porcelain insulator vibration response signal containing noise and then carrying out multi-resolution wavelet decomposition, and transforming the vibration response from a time domain to a wavelet domain; rationally selecting a threshold function and a threshold, and machining a wavelet coefficient corresponding to the noise according to the threshold function; carrying out wavelet reconstruction, transforming the vibration response after denoising treatment to the time domain from the wavelet domain; storing the de-noised vibration response and sorting an acoustic testing result of the porcelain insulator vibration. According to the method, the noise in the acoustic testing data of the porcelain insulator vibration can be effectively filtered, so that sorting accuracy is improved.

Through sensor of stress wave, the invention picks up signal of stress wave of malfunction in early stage. The invention mainly solves issues of data acquisition for stress wave, noise suppression, feature extraction of signal, and fault recognition and position. Based on signal character, wavelet analysis selects suitable wavelet basis function, and carries decomposition of fault signal in multiple scales. Reconstructing waveform and spectrogram from each decomposition in different scale picks up minute character so as to determine type of fault and put forward scheme of treatment. Thus, the invention reduces economic loss to minimum.

The invention relates to a traffic flow prediction method based on a quick learning neural network with double optimal learning rates. The method comprises the following steps of: normalizing m continuous traffic flow historical data which serve as the input of a prediction network; initializing weights and stretch and shift factors of a wavelet basis function by using a three-layer neural network, wherein the shift factors and transfer factors of the wavelet basis function employ a first learning rate, and network weights employ a second learning rate; providing a learning rate array, and performing network training of the double optimal learning rates; and outputting values of a current moment to first (m-1) periods, which serve as the input of a trained network, performing reverse normalization, and thus obtaining a prediction value of a traffic flow at a next moment of the current moment. The method has the advantages that the first learning rate and the second learning rate employ the optimal learning rates during network training at each time, quick network training can be realized, and high-accuracy prediction of the traffic flow is realized.

This invention relates to a method for interfusing fuzzy images based on region characters including the following steps: 1, applying small wave frames to carry out multi-dimension resolutions to being interfused images to get a series of high frequency components and a lowest component, 2, carrying out K average clustering to the low frequency part of the infrared images and dividing it to three kinds, 3, expressing the three kinds to an important target region, a second important region and a background region to make a decision to the interfusion based on the fuzzy region character and measurement target of the HF part of a sensor multi-image, 4, sending the final multi-resolution images into a filter made up of a same small wave basic function to be filtered and summing up the filtered image signals, lowering the transformation layer number of the small wave frame by one, carrying out a process of the fall-sample to the filter then to transform the next layer inversely and repeating the process till finishing the inverse transformation to the entire wave frames to get a final fused image.

The invention discloses a rotary machine fault detection method of dual-tree complex wavelet transformation with adjustable quality factors. The rotary machine fault detection method of dual-tree complex wavelet transformation with the adjustable quality factors comprises the steps of (1) building a reasonable sampling parameter set, building dual-tree complex wavelet base functions with different quality factors, (2) using each built dual-tree complex wavelet base function to carry out time-scale analysis on a vibration response signal of a rotary machine, calculating kurtosis information entropy of wavelet coefficients of each layer with participation of each dual-tree complex wavelet base function, selecting a dual-tree complex wavelet base function corresponding to the maximum feature kurtosis information entropy as the dual-tree complex wavelet base function which is in optimal matching with an impact component of the vibration signal, and (3) analyzing the vibration signal through the optimal dual-tree complex wavelet base function, and carrying out fault diagnosis. According to the rotary machine fault detection method of dual-tree complex wavelet transformation with the adjustable quality factors, the dual-tree complex wavelet base functions with any frequency-band focusing performance and time-domain oscillation performance can be built, the base function with the optimal matching performance can be selected in a self-adaptation mode, and accurate detection of periodicity impact type fault features and information of the impact period of a rotary machine device can be achieved.

The invention discloses a fault diagnosis method of a variable-speed bearing, which comprises the following steps of: sampling the vibration signals of the bearing at equal time intervals through an acceleration sensor by a data acquisition module controlled by a fault diagnosis module to obtain a vibration signal sequence x(n); carrying out wavelet transformation on the acquired vibration signal sequence x(n) by adopting a plurality of Morlet wavelets as the wavelet basis functions of the wavelet transformation to obtain wavelet coefficient wt(m,n); carrying out modular operation on the wavelet coefficient wt(m,n) to acquire the envelope ewt(m,n)=||wt(m,n)|| of the wavelet coefficient wt(m,n); transforming the wavelet envelope coefficient at each size from an equal time interval sampling result to an equal angle sampling result; carrying out Fourier transformation on the wavelet envelope sequences ewt(m,t) at various sizes, which are subjected to equal angle sampling, to obtain the frequency spectra eswt (m,f)=FFT(ewt(m,t)) of the wavelet envelope sequences ewt(m,t); and making eswt(m,f) into a three-dimensional image.

The invention relates to a remotely sensed image small object super-resolution rebuilding method, which provides a remotely sensed image small object super-resolution rebuilding model; the space resolution factor adds 1.5 ploidy of the initial picture and effect depresses the extraneous wave; the linear material and the empirical distribution estimate the atmosphere disturb image parameter H; the redundancy wavelet distribution adopts mirror-image wavelet base function and uses the cross cut correlation to achieve morphology wavelet non-linear wavelet encoding and depresses the high frequency immediately noise; it dynamically evens the effect of high frequency noise signal and high frequency detail signal. It is applied in satellite image military target identifying, small target detecting and earth source remotely sensed image measuring.

The invention provides a potential field abnormal separation method based on a wavelet spectral analysis. The method, which is implemented on a computer, comprises the following steps that: gridding is carried out on a potential field abnormal signal observed by an irregular grid so as to form a potential field abnormal signal of a regular grid; according to a purpose of potential field separation abnormity and a wavelet basis characteristic, a wave basis function is determined; and a wavelet multiple dimensioned analysis is carried out on the potential field abnormal signal of the regular grid so as to obtain potential field abnormal signals under different dimensions; radial power spectrums of the potential field abnormal signals that are obtained under different dimensions are respectively calculated; and according to the radial power spectrums of the potential field abnormal signals obtained under the different dimensions, corresponding filed source body depths are respectively estimated. By employing the provided method in the invention, quantitative evaluation can be carried out on field source depths corresponding to separated potential field abnormal signals.

The invention aims at providing a target tracking method based on multi-scale dimensional decomposition, and the target tracking method comprises the following steps of selecting a wavelet-base function to decompose a target angle or track measurement data onto a scale, predicting and filtering the measurement data on a low-frequency subspace of each scale by utilizing extended kalman filtering (EKF) algorithm to obtain a rough tracking result of the target on different scales, and further eliminating the influence of noise and wildvalue by utilizing the wavelet threshold algorithm on the high-frequency subspace of different scales; and converging the tracking data on different scales through the wavelet reconstruction algorithm to obtain the precise tracking data of the target. The target tracking method can effectively, accurately, reliably and stably track the target in different complicated environment, the multi-scale EKF algorithm is realized by utilizing field programmable gate array (FPGA) parallel processing structure, the wavelet decomposition and rebuilding, the EKF algorithm of different scales and the wavelet threshold denoising are simultaneously implemented, and the real-time performance of the target tracking is ensured.

The invention discloses a method for extracting engineering machine running characteristic signals in the technical field of signal processing, which comprises the following steps of: determining an optimal decomposition layer number of wavelet decomposition according to a wavelet coefficient whitening inspection method by acquiring a machine running state data and selecting a wavelet function; then calculating a de-noising threshold value of a wavelet coefficient of each decomposition layer, and de-noising the wavelet coefficient by adopting a soft threshold function; and performing inverse wavelet transform on the de-noised wavelet coefficient to obtain state data of the de-noised engineering machine running characteristic signals. The method improves the adaptability and computing speed, improves the signal-to-noise ratio of the acquired signals, and can meet the requirements of remote real-time monitoring, fault diagnosis and performance prediction.

The invention discloses a wood defect detecting and sorting device and method based on a depth camera and deep learning, the wood defect detecting and sorting device comprises an industrial personal computer, a detecting mechanism and a sorting mechanism arranged on the rear side of the detecting mechanism, and the detecting mechanism comprises a detecting conveying belt and a depth image collecting mechanism. According to the method, RGB images and depth information of the wood surface are collected through a depth camera, and RGBD color depth information is reconstructed through combinationof a GAN network and wavelet transform. In wavelet transformation, wood cracks in training data are manually marked, and a self-adaptive crack wavelet basis function is formed. Wavelet reconstructionis carried out on the basis to improve the efficiency of a subsequent algorithm, and the defect type is obtained through analysis in combination with a deep learning algorithm. The algorithm can sortwoods with different defect types, so that the defect discrimination efficiency is improved, and the sorting efficiency is greatly improved.

The invention discloses a transformer noise prediction method based on a wavelet neural network and the wavelet technology. A neuronal hyperbolic tangent S-type excitation function of a hidden layer in the traditional BP (back propagation) neural network is replaced with a wavelet-based function, momentum factors are introduced when parameters of the neural system are adjusted, and accordingly, a prediction model is higher in convergence speed and higher in error precision. Vibration and noise digital signals are decomposed by means of the wavelet decomposition technology, wavelet low-frequency coefficients obtained are used as input-output pairs for the prediction model, the wavelet low-frequency coefficients obtained by prediction are reconstructed by means of the wavelet reconstruction technology after modeling, and predicted noise digital signals are obtained. The transformer noise prediction method based on the wavelet neural network and the wavelet technology has the advantages that fewer training samples are required, time of training neurons in the neural network is shortened, and the problem that poor prediction effect is caused by ambient high-frequency interference noise contained in actually-measured transformer noise data is further avoided.

The invention discloses a segment hidden crack identification method based on matching pursuit and wavelet transform, and a technology of combining orthogonal matching pursuit and wavelet transformation is adopted to process a shield tunnel lining hidden water-containing micro-crack geological radar detection signal, the influence of a strong impedance interface can be effectively weakened, and micro-weak reflection signals of a target detection object are enhanced, so that the purpose of accurately detecting and identifying the shield tunnel lining hidden quality defects is achieved. Firstly,according to the sparse representation theory, a surface strong reflection and abnormal strong reflection stripping method for layer and wavelet constraint matching pursuit is provided, and by combining a matching pursuit algorithm and a strong reflection forming mechanism, a sparse dictionary matched with the characteristics of the strong reflection signals is selected, and two-time matching decomposition is carried out on each signal, so that the micro-weak target reflected signals submerged in the strong reflection can be well displayed. Secondly, a wavelet basis function matched with thesignal and a proper wavelet transformation scale are selected, and the image profile is processed and enhanced again by adopting a continuous wavelet transform method, so that the hidden water-containing micro-crack signals are effectively highlighted.

The invention discloses a method for calculating the characteristic peak of a derivative detection spectrum through using a wavelet. The method comprises the following steps: respectively carrying out continuous wavelet transformation on spectra through using the first derivative and the fourth derivative of a Gaussian function as wavelet basis functions, extracting a maximum point position from the wavelet calculation fourth derivative spectrum of gaus4 to obtain the position of an original spectrum peak, and extracting extreme point positions from the wavelet calculating first derivative spectrum of gaus1 to obtain the inflection point positions of the original spectrum peak. Characteristic points are obtained through detecting extreme points, so the method is more convenient and accurate than zero-crossing point calculation.

The invention discloses an intelligent fault diagnosis method for a pump station main device, and relates to the field of signal processing. The method comprises the steps of collecting a data signalof a to-be-tested key part of a to-be-tested pump station main device in a working state, and using the data signal as a to-be-tested sample; and inputting the to-be-tested sample into a VMD-gcForestdiagnostic model, sequentially drawing a modal function spectrum diagram and performing a multi-granularity cascade forest diagnosis to obtain a working status label corresponding to the to-be-testedsample, and obtaining the working state of the to-be-tested key part according to the working status label. The intelligent fault diagnosis method of the invention solves the problem that the waveletbasis function and the filtering threshold existing in an original vibration signal extraction process cannot be determined in the prior art; the problems of a lack of theoretical basis, the end pointeffect and the modal aliasing of EMD; and the problems that the existing fault diagnosis method based on the original vibration signal is extremely complicated in parameter adjustment, time-consumingin calculation, and low in diagnostic accuracy.

The invention discloses a wavelet denoising method for adaptively determining wavelet hierarchical series. The method comprises the following steps: acquiring a wavelet basis function according to anoriginal signal; obtaining a decomposition layer number according to the original signal and a wavelet basis function; decomposing the original signal according to the number of decomposition layers to obtain wavelet coefficients of each layer; determining a threshold value of the wavelet coefficient of each layer according to the wavelet coefficient; performing quantization processing on the wavelet coefficient of each layer; and reconstructing the wavelet coefficient subjected to quantization processing to obtain a denoised signal. A spectral analysis method is adopted, the optimal decomposition layer number is determined in a self-adaptive mode, and the problem that the operation process is tedious when a trial-and-error method is adopted is solved.

The invention belongs to the field of load recognizing, and discloses a dynamic load recognizing method based on the wavelet multiresolution analysis to solve the problems existing in the study on the field of load recognizing at present. The method comprises the first step of solving recognizing parameters; the second step of carrying out wavelet reconstruction on a load through a wavelet basis function based on a time domain convolution model to obtain a wavelet response function; the third step of carrying out wavelet transform on the response and the wavelet response function to obtain a system response in a wavelet domain and a wavelet response function in the wavelet domain; the last step of calculating the weigh coefficient, reversely solving the load and finishing recognition. According to method, non-stable loads such as impact and mutation can be recognized, the recognition precision is high, the method is not sensitive to interference among the multi-path loads comprising quick and slow change in a multi-input and multi-output system, and the multi-path loads can be distinguished and recognized; the ration / qualitative determining method for the recognition parameters is provided, and the dynamic load recognizing method based on the wavelet multiresolution analysis can be used for determining the parameters.

The invention belongs to the technical field of underwater acoustic communication, and discloses a mobile frequency hopping underwater acoustic communication Doppler factor estimation method, which comprises the following steps of synchronizing communication signals at a receiving end, and intercepting a chip; carrying out doppler frequency offset factor measurement on the intercepted chip to obtain a Doppler frequency offset factor estimation result, wherein the pre-selected wavelet basis function is used as a wavelet basis function for denoising; decomposing the Doppler frequency offset factor estimation result to obtain an approximation coefficient and a detail coefficient; calculating a threshold value; performing soft threshold denoising on the detail coefficient; reconstructing the Doppler frequency offset factor; grouping Doppler frequency offset factor optimization results, and taking the median of each group to represent the group; predicting the median of the next group of Doppler frequency offset factors; and selecting the next group of chips, and repeating the steps to predict the median of the next group of Doppler frequency offset factors. According to the method, theoptimal de-noising wavelet basis function is selected, estimation errors caused by inaccurate synchronization and inaccurate chip selection are eliminated, and a more accurate Doppler frequency offset factor estimation result is obtained.

The invention provides a power microwave communication system wavelet noise reduction method. The method comprises the steps that b, according to different communication signals, a wavelet primary function is selected and wavelet decomposition level is determined; c, the signals are filtered to acquire a wavelet coefficient; d, according to the preset wavelet decomposition level, threshold quantization processing is carried out on the wavelet coefficient to acquire a quantized wavelet coefficient; and e, a reconstructed signal is acquired through wavelet reconstruction filtering. According to the invention, the original signals are decomposed into a series of approximate components and detail components through wavelet decomposition; detail component processing and wavelet reconstruction are carried out to extract useful communication signals; weak signal characteristics are extracted in a strong noise background; noise in the signals can be effectively reduced and even eliminated; wireless communication demodulation signal to noise ratio and signal relevance can be significantly improved; phase distortion and bit error rate are reduced; the power microwave communication performance is improved; and the problems of large electromagnetic background noise and many signal harmonic components of power communication are solved.

The invention relates to a multi-focus image fusion method based on two-dimensional empirical mode decomposition (EMD) and a genetic algorithm. At first, two-dimensional empirical mode decomposition (EMD) is performed on a source image, and therefore, the problem of weak correlation of local features of image fusion based on wavelet transform can be solved, and the problem of difficulty in wavelet basis function selection in a traditional wavelet method can be solved; high / low frequency selection is performed on obtained intrinsic mode function (IMF) components according to T-test, and then, fusion is performed on low-frequency components through adopting a regional information entropy maximum criterion, and regional correlation calculation is performed on high-frequency components, and components of which the correlations are in different threshold ranges are fused, and the selection of thresholds is searched through adopting the genetic algorithm, and therefore, the defects of experience determination of regional matching thresholds can be avoided; and finally, two-dimensional empirical mode decomposition (EMD) inverse transformation is performed on fused components so as to obtain fusion results. Thus, based on the combination of the two-dimensional empirical mode decomposition (EMD) and the genetic algorithm, and the multi-focus image fusion method can greatly improve the quality of fused images and has important significance and great use value in subsequent processing and image display of an application system.

The invention discloses a wavelet denoising method with variable threshold value, comprising five steps. The method comprises: Step 1, inputting the original image and adding the corresponding Gaussian noise; step 2, selecting a wavelet basis function and determining that number of layers O of wavelet decomposition: decomposing the noise image S to obtain a low-frequency coefficient A1 of the first lay, horizontal and vertical high-frequency coefficients H1 and V1, and diagonal high-frequency coefficient D1; decomposing A1 to obtain The second layer low frequency coefficient A2, the horizontaland vertical high frequency coefficients H2 and V2, and the diagonal high frequency coefficient D2; decomposing A2 to obtain The third layer low frequency coefficients A3, horizontal and vertical high frequency coefficients H3 and V3 and diagonal high frequency coefficients D3 ; decomposing sequentially until O layer; step 3, selecting the combined wavelet threshold and wavelet threshold functionwith straight line (shown in the description) as asymptotic line to process wavelet coefficients; step 4, carrying out wavelet reconstruction on the wavelet coefficients after threshold quantization;Step 5, outputting the denoised image. The invention can improve the precision of wavelet transform processing noise signal, effectively improve the de-noising effect of the image, and obtain high-quality de-noising image.

The invention discloses a rotating machine fault diagnosis method based on wavelet packet decomposition. The method comprises the following steps: 1) collecting vibration signals of a rotating machinein a normal state and a fault state; 2) selecting a wavelet basis function for fault feature extraction; 3) according to the selected wavelet basis function, obtaining sub-signals of different frequency bands of the vibration signal through wavelet packet decomposition; 4) calculating a fuzzy entropy value of the sub-signal to obtain a fault feature vector; 5) performing feature importance sorting according to the correlation, and selecting a set number of fault feature vectors with top sorting results according to the sorting results; 6) using a classifier to construct a fault diagnosis model, dividing the selected fault feature vector and the category label into a training set and a test set, and using the training set as the input of the model to train the model; and 7) inputting the test set into the fault diagnosis model to obtain a fault diagnosis result. According to the method, high-quality fault features can be effectively extracted, and the accuracy of fault diagnosis is improved.

The invention discloses a method for rapidly and nondestructively detecting green tea water content based on wavelet transformation. The method comprises the steps of: obtaining a diffuse reflection spectrum of tea leaf samples within a short-wave near infrared spectrum range being 888-1007nm, conducting calculation and transformation to obtain an absorbance spectrum, adopting a db2 wavelet basisfunction to conduct discrete wavelet transformation, extracting a three-scale low-frequency wavelet coefficient to obtain nineteen wavelet characteristic coefficients and accordingly calculating the water content of the samples. The method for rapidly and nondestructively detecting the green tea water content based on the wavelet transformation can rapidly and effectively monitor the dynamic change of the water content during green tea processing and can realize the rapid, nondestructive and low-cost detection of the water content during green tea processing.

The invention discloses an in-service bridge support damage diagnosis method based on a vehicle braking effect, and belongs to the technical field of bridge support damage diagnosis. The objective ofthe invention is to solve the problems that an existing in-service bridge support damage diagnosis method cannot accurately perform structural damage positioning and is not high in measurement precision. According to the invention, the accurate bridge finite element model is established by using a model correction method, and a vehicle brake numerical simulation method is provided, so that a braketest scheme suitable for the bridge to be detected can be formulated more accurately. A proper wavelet basis function decomposition level is selected by utilizing a cost function, wavelet packet decomposition is performed on a free attenuation signal acquired at the pier top of the pier, and a damage index is constructed to diagnose the damage position and degree of the support. The method is simple to operate, economical and practical, has certain robustness and noise immunity, and does not greatly influence the bridge structure.

The invention discloses a wavelet packet noise reduction method for a meteor trail communication system, used for performing noise reduction on meteor trail communication signals. The method comprisesthe following steps of: selecting a proper wavelet basis function, performing wavelet decomposition on the signals by using an orthogonal wavelet decomposition algorithm, performing threshold quantization on the wavelet decomposition coefficients by using soft threshold functions, performing signal reconstruction by using a reconstructed function with a specific structure, and realizing the feature extraction of weak signals in the background of strong noise. The method disclosed by the invention is capable of obviously improving demodulation signal-to-noise ratio of the meteor trail communication system, enhancing the detection and utilization abilities for the weak signals, shortening the communication waiting time and improving the information transmission capacity, and is especially suitable for application scenarios in which the meteor trail communication signals are drown by strong background noise.

The invention discloses a method for rapidly predicting content of soil organic matters based on eleven spectrum wavelet coefficients. The method is characterized by comprising the following steps: (1) collecting a soil diffuse reflectance value (R) with the spectrum range of 400-2450nm at the wavelength interval of 1nm; (2) converting the soil diffuse reflectance value into a soil absorbance value through a formula of A=-log(R); (3) performing discrete wavelet transform on the soil absorbance value by selecting a sym7(Symlets7) wavelet base function, extracting scale-5 low-frequency wavelet coefficients, and obtaining 75 low-frequency wavelet coefficients; (4) compressing the 75 low-frequency wavelet coefficients by utilizing a successive projections algorithm, and extracting the 9th, 11th, 22nd, 27th, 35th, 39th, 48th, 61st, 64th, 65th and 67th characteristic wavelet coefficients, totaling eleven wavelet coefficients; and (5) substituting the eleven characteristic wavelet coefficients into a multiple linear regression calculation formula, and calculating to obtain the content of soil organic matters. The method can rapidly predict the content of soil organic matters and is suitable for development and utilization of portable test instruments.

The invention discloses a method for diagnosing a failure of a wind-powered rotary support based on wavelet analysis. The method is characterized by comprising the following steps of a) extracting an acceleration signal and a torque signal of the early failure of the wind-powered rotary support through an acceleration sensor and a torque sensor; b) transmitting the torque signal through a transmitter, and converting the transmitted torque signal and the acceleration signal through a current and voltage converting plate; c) selecting a proper wavelet basis function in an NI data acquisition module, and performing multiscale decomposition on a failure signal by a wavelet analysis method; d) extracting fine characteristics of the failure signal from each scale decomposition reconstruction waveform and a frequency spectrum of the scale decomposition reconstruction waveform; and e) determining the failure type or the time when the failure occurs. The acceleration signal and the torque signal serve as characteristic parameters for the first time, and the failure signal of the wind-powered rotary support is acquired, so that the traditional problem of limitation under the condition of low speed of a vibration signal is solved.

The invention discloses a pulse signal classification method based on wavelet packet conversion and hidden markov models. The method includes the following steps that a db4 wavelet is adopted as a wavelet basis function of wavelet packet conversion, and the wavelet packet conversion is carried out on two kinds of collected pulse signals to obtain wavelet packet decomposition coefficients of various frequency bands; an optimal frequency band is selected according to a local area discriminant base algorithm; an optimal energy feature vector is selected by means of a Fisher criterion; one part of the two kinds of pulse signals is selected to serve as training signals, the other part of the two kinds of the pulse signals serves as testing signals, and the optimal feature vectors of the two kinds of signals are figured out according to the method; the optimal energy feature vector of the training signals serves as a continuous hidden markov observation vector to train two hidden markov models; the optimal energy feature vector of the testing signals is respectively input into the trained two models, the probability values P(O | lambada i) of the optimal energy feature vectors are worked out according to a forward-backward algorithm, the probability values are compared, and classification of the pulse signals is completed.

The invention discloses a power grid equipment data flow cleaning method based on an association rule, which comprises the following steps of: calculating association strength of historical data of each data sequence in a data flow by utilizing an Apriorri algorithm, and outputting an association relationship among different data sequences; using an abnormal data screening algorithm based on a sliding time window to detect the data sequences with weak correlation strength one by one; carrying out abnormal data identification processing on the data sequence with higher correlation degree at thesame time; and applying the neural networks with multiple wavelet basis functions to data cleaning to complete combined prediction. The data flow cleaning method is accurate in power grid equipment risk assessment and stable and reliable in data.

The invention discloses a wavelet constant modulus blind equalization method in a MIMO (Multiple Input Multiple Output) system from the aspect of reducing?inter channel?and inter signal?correlation, so as to overcome defects that the traditional constant modulus blind equalization method in the MIMO system is weak in inter channel interference inhibition ability, slow in the convergence rate and large in steady-state errors. According to the method, singular value decomposition is used for reducing the correlation of channel output signals; an orthogonal wavelet base function is used for carrying out orthogonal wavelet transformation on MIMO channel output signals and autocorrelation of input signals of a blind equalizer is reduced; on the basis of fully considering the inter channel correlation, a cost function of MIMO system constant modulus blind equalization algorithm is redefined, and cross-correlation of output signals of the blind equalizer is reduced; and a variable step size function is used for controlling a blind equalization?weight vector?updating process, the convergence rate quickened, and the steady-state errors are reduced.