64 results about "Low frequency modulation" patented technology

The invention relates to a method, a device and equipment for estimating the state of a power battery of a sweeper. The method comprises the following steps: performing impedance characteristic acquisition on a power battery of the sweeper under various working conditions, and performing random low-frequency modulation multi-sine excitation signal measurement to obtain battery working condition characteristic data; performing fitting operation through the equivalent circuit model to obtain a battery health factor model; performing charging process feature extraction analysis to obtain a battery charging curve reconstruction model; carrying out extended Kalman filtering operation to obtain a battery state-of-charge value; performing segment processing and Gaussian process regression calculation on voltage and current data in the charging process to obtain a battery health state value; and updating the polynomial mapping coefficient of the battery health factor model and the weight parameter of the battery charging curve reconstruction model to obtain a target health factor model and a target charging curve reconstruction model. According to the invention, the adaptability and accuracy of the state estimation method in the whole life cycle of the battery are ensured.

The invention discloses a wind generating set fault early warning method and system, and the method comprises the steps: S1, collecting vibration waveform original data and temperature original data, and carrying out the preprocessing of the data; s2, designing a heterogeneous double-current network to extract high-frequency impact features and low-frequency modulation features respectively; s3, calculating vibration characteristics according to the high-frequency impact characteristics and the low-frequency modulation characteristics; s4, extracting a temperature change rate sequence and temperature time sequence characteristics; s5, generating an attention weight matrix of vibration to temperature and an attention weight matrix of temperature to vibration, calculating a fault sensitive mask, and generating fault sensitive features; and S6, compressing the fault sensitive features, calculating a three-level early warning probability vector and a unit fault early warning category, and giving an alarm. According to the method, the problem of early warning lag caused by dependence on fixed threshold value alarm of a traditional wind generating set fault early warning method can be solved.

The invention belongs to the technical field of oil and gas well perforation, and relates to a high-speed communication method and system for single-core cable perforation. The method comprises the following steps: performing low-frequency modulation on underground perforation data; performing high-frequency modulation on underground logging data; synthesizing carrier signals; the driving power of the carrier synthesis signal is amplified, so that the carrier synthesis signal is transmitted to the ground end on the single-core cable; the ground terminal carries out carrier signal conditioning on the carrier synthesis signal and then carries out amplification processing; carrying out carrier signal separation, and respectively outputting a low-frequency modulation signal and a high-frequency modulation signal; and outputting perforation data and logging data. According to the invention, perforation signals and underground logging data can be transmitted on the single-core cable at the same time, and the exploitation efficiency of an oil-gas well is improved; according to the single-core cable perforation communication system, continuity and low delay characteristics of perforation operation signals are kept, discrete data can be selectively transmitted, the efficiency of single-core cable perforation communication is improved, and mutual influence of perforation or logging communication faults is effectively avoided.

The invention discloses a training system for color resolution capability. The training system comprises a training parameter setting module, a training image generation module, a man-machine interaction module, a training process control module and a data recording and analyzing module. The training parameter setting module sets training related parameters; the training image generation module generates a color training image; the man-machine interaction module presents a training image, obtains the identification condition of the training image by a training individual, monitors the training environment and the training state of the training individual, and stimulates auxiliary training through a low-frequency modulated intermediate-frequency current; the training process control module controls a training process; the data recording and analyzing module can record and analyze training data and evaluate the overall training process. Through targeted training image design, rigorous training process control and synchronous combination of training auxiliary hardware equipment, training of the color resolution capability is effectively realized, and the color resolution capability of a biological individual is improved.

The invention relates to the technical field of exercise direction distinguishing ability training, in particular to an exercise direction distinguishing ability training system which comprises a training parameter setting module, a training image generating module, a man-machine interaction module, a training process control module and a data recording and analyzing module. The training parameter setting module sets training related parameters; the training image generation module generates a training image; the man-machine interaction module presents a training image, obtains the identification condition of a training individual on the training image, monitors a training environment and a training state, and assists training through a low-frequency modulated intermediate-frequency current; the training process control module controls a training process; the data recording and analyzing module can record and analyze training data and evaluate the overall training process. Through targeted visual stimulation design, rigorous training process control and synchronous combination of training auxiliary equipment, training of the motion direction distinguishing ability is effectively achieved, and the motion direction distinguishing ability of a biological individual is improved.

The invention discloses a seismic data acquisition device based on a ground 3C-DAS optical cable and an acquisition method thereof, is applied to the field of seismic exploration, and aims to solve the problems of high layout cost, too long construction time and low operation efficiency of an existing ground three-component detector. According to the method, a 3C-DAS armored optical cable is buried under a shallow earth surface in a parallel or orthogonal mode, a ground multichannel low-frequency DAS modulation-demodulation instrument collects phase data of Rayleigh scattering light reflected back from the 3C-DAS armored optical cable and carries out modulation-demodulation processing, and high-density distributed three-component seismic data distributed along the 3C-DAS armored optical cable is obtained; according to the system, high-density distributed three-component ground seismic data along the armored optical cable can be rapidly and efficiently acquired at low cost; the invention also provides a processing method for the collected high-density distributed three-component ground seismic data, an analysis method for a shallow surface velocity structure with high spatial resolution, and an inversion imaging method for longitudinal waves, transverse waves and surface waves.

To provide an LED driver and an LED lighting system, capable of reducing the discomfort of users.SOLUTION: An LED driver includes a DC-DC converter 11 and a strobe circuit 12. The strobe circuit generates a low-frequency modulation signal. The DC-DC converter is connected to the strobe circuit and provides an adjustable operating current to an LED light source according to a DC electrical signal and the low-frequency modulation signal. The operating current includes a DC current signal corresponding to the electrical DC signal and a low-frequency AC current signal corresponding to the low-frequency modulation signal. A frequency of the low-frequency AC current signal is 25-100 Hz, and a current ripple factor, equaling a difference of a maximum value and a minimum value of the operating current divided by a sum of the maximum value and the minimum value, is less than 8%.SELECTED DRAWING: Figure 3

The invention belongs to the technical field of infrared thermal imaging defect detection, and particularly discloses an in-situ infrared thermal imaging detection device based on low-frequency modulation magnetic field excitation, which comprises a magnet yoke, an infrared thermal imager, a computer, a modulation power supply and an excitation coil group, the excitation coil group at least comprises a first excitation coil, a second excitation coil, a third excitation coil and a fourth excitation coil; the modulation power supply respectively introduces low-frequency alternating currents with different phases and direct currents with different directions into the first excitation coil, the second excitation coil, the third excitation coil and the fourth excitation coil; through composite excitation of an orthogonal direct-current magnetic field and a low-frequency rotating alternating-current magnetic field, the device can form a rotating low-frequency alternating-current magnetic field in a to-be-detected pipeline, the rotating magnetic field uniformly sweeps a detection area in a plurality of periods, the influence of eddy current distribution direction sensitivity is eliminated, and large-area uniform heating is realized; the problem of low efficiency of a traditional spot scanning or line scanning mode is avoided, and the coverage area of single detection is remarkably increased.

The invention relates to the technical field of vibration monitoring, and discloses a steam turbine cold end fault diagnosis method and system based on a fusion time sequence convolutional network, and the method comprises the steps: eliminating the frequency spectrum drift caused by the rotation speed fluctuation through band-limited filtering and order domain resampling, and automatically extracting the specific low-frequency modulation component of a cold end through the envelope square and a periodogram, and finally, the asymmetric dynamic response characteristics of the rotor system are quantified by third-order spectrum phase bias, and the phase change rule is adaptively learned by the pre-trained convolutional network. According to the method, the problem that traditional amplitude spectrum analysis is high in misjudgment rate under the strong noise, variable speed and unsteady environment is effectively solved, the fault features show higher stability and distinguishability in the phase domain, and therefore high-robustness and high-precision automatic diagnosis of the turbine cold end rotor faults is achieved.

The invention provides a biaxial coupling error suppression method of a nuclear magnetic resonance atomic gyroscope, and belongs to the technical field of nuclear magnetic resonance atomic gyroscopes. A nuclear magnetic resonance atomic gyroscope inertial measurement system is started, and a magnetic field is applied to the system, so that the system works in a nuclear magnetic resonance state; applying a low-frequency modulation magnetic field with a certain amplitude on a y axis orthogonal to the x axis, and outputting a measured magnetic field amplitude of the x axis under the condition by using a magnetometer embedded in the nuclear magnetic resonance atomic gyroscope; a PID module of a phase-locked amplifier in a system demodulation signal system is used as an input quantity of PID control, a target value is set as 0, an output quantity is set as a carrier demodulation frequency, and the suppression of coupling errors is realized by adjusting the carrier demodulation frequency to be consistent with an atomic frequency; when the nuclear magnetic resonance atomic gyroscope works normally, the alkali metal spin frequency is locked, it is guaranteed that the carrier demodulation frequency is consistent with the alkali metal spin frequency, additional auxiliary and calibration equipment does not need to be added, and the method has important significance in the field of inertial measurement.

The invention relates to the technical field of integrated photoelectron and electro-optical modulation, in particular to a thin-film lithium niobate electro-optical modulator structure and a preparation method of the thin-film lithium niobate electro-optical modulator structure. The buried layer is located on the upper surface of the substrate; the partially-etched thin film lithium niobate layer comprises a partially-etched thin film lithium niobate flat plate layer and a thin film lithium niobate ridge waveguide, the partially-etched thin film lithium niobate flat plate layer is located on the upper surface of the buried layer and exposes a part of the buried layer, and the thin film lithium niobate ridge waveguide is located on the partially-etched thin film lithium niobate flat plate layer; the partially-etched buffer layer is located on the partially-etched thin film lithium niobate layer, and the side wall of the partially-etched thin film lithium niobate flat plate layer is exposed; and the metal electrode is in electric contact with the side wall of the partially etched thin film lithium niobate flat plate layer. Therefore, the problems of low-frequency modulation waveform distortion, direct-current bias point drift and the like caused by electro-optical relaxation and electric field shielding of a related thin-film lithium niobate electro-optical modulator are solved.

The invention provides an optical pump atomic magnetometer device for deep space weak magnetic and zero magnetic detection, which comprises a magnetometer probe and an electronics box, and adopts an internal modulation or external modulation method to realize laser constant power control. The internal modulation scheme is as follows: the circuit realizes laser constant power control by synchronously controlling the driving temperature and current of the laser. The external modulation scheme is that optical power control devices such as an acousto-optic modulator or an electrically controlled attenuator are added to the output end of the laser to carry out closed-loop control on the laser power. According to the optical pump magnetometer with the stable optical power, a low-frequency modulation and narrow linewidth Mz magnetic resonance method is adopted, long-term accurate measurement can be conducted on weak magnetic field scalar quantity which is lower than geomagnetic field intensity and equivalent to Mz resonance linewidth, and near-zero magnetic field intensity detection can be achieved by adding a small compensation magnetic field on a magnetometer probe and a modulation resolving scheme of the small compensation magnetic field. The device is suitable for detection of deep space weak magnetic field environments such as earth high orbit, moon, Mars, asteroid, various satellites, interplanetary space and the like, and can also be applied to medical and basic scientific research.

The invention discloses a digital jitter zero-offset frequency locking system and method, and the system comprises the steps that a laser processing unit carries out the optical frequency mixing of reference laser transmitted by a reference laser and local oscillator laser with jitter frequency, and inputs a mixed optical signal to a balance detector; the balance detector outputs a beat frequency signal to the digital phase-locked loop according to the mixed optical signal; the digital phase-locked loop processes an error signal in the beat frequency signal to obtain a low-frequency modulation signal and a high-frequency modulation signal, the working current of the local oscillator laser is controlled through the low-frequency modulation signal, and the working voltage of the phase modulator is controlled through the high-frequency modulation signal. Phase errors cannot be generated due to frequency deviation between reference light and local oscillation light, so that transfer from the frequency deviation to phase noise is eliminated, then error signals are allowed to be filtered and processed in a plurality of modes by constructing a digital phase-locked loop, the performance under low-power frequency locking is improved, and long-term stability degradation is avoided.

The invention provides an all-optical line switching system based on an optical switch network, and the system comprises an optical switch which is provided with a plurality of optical switch communication modules. An optical line switching matrix is arranged in the optical switch, and the optical line switching matrix comprises a plurality of optical ports; each optical switch communication module is connected with an optical switch communication module of another optical switch or an external switching node communication module, and optical signals transmitted by the optical switch communication module and the optical switch communication module or the switching node communication module of another optical switch are communication light containing data optical signals and management optical signals. The communication light is formed by low-frequency modulation and superposition of a management light signal to a data light signal and / or formed by the management light signal and the data light signal in a wavelength division multiplexing mode; and after receiving the communication light, the optical switch communication module separates the data optical signal from the management optical signal, and transmits the data optical signal to the corresponding optical port. According to the invention, data optical signals and management optical signals can be transmitted simultaneously.

The invention provides an intermediate-frequency carrier and low-frequency modulated wave composite output control method, which comprises the following steps of: acquiring spatial intensity change distribution of a composite waveform in optic nerve tissue, and acquiring an actual measurement value of a carrier intensity distribution field by combining a microtube skeleton polymerization rate and an extension orientation angle of growth cone propulsion, forming basic data of an initial distribution field; extracting candidate output of an optimal waveform modulation combination from the adjusted envelope gradual change speed control combination, comparing an actual path of directional extension with an expected damage repair track through track overlapping degree measurement, and determining a quantization result of the deviation degree of the extension path; and performing deviation degree classification processing on the extension path deviation degree quantification result, obtaining a high deviation subset in a classification result, generating a compensation signal in combination with an aggregation rate change development trend, and obtaining a compensated composite waveform output combination.

The invention discloses an industrial steam extraction combined heat and power generation frequency modulation system and an industrial steam extraction combined heat and power generation frequency modulation method adapting to a variable load condition, and relates to the technical field of combined heat and power generation. The problems that the heat storage capacity is limited and the frequency modulation capacity is low due to the fact that only reheating hot section steam is utilized at the present stage are abandoned, and in the sliding temperature operation mode, due to the fact that the temperature of main steam and the temperature of reheating steam can be changed by unit variable loads, the change trend of the steam extraction amount is completely opposite to the steam extraction temperature in the frequency modulation process, and the heat storage capacity is limited. Therefore, the outlet temperature of the heated molten salt is controlled by changing the sensible heat exchange amount of the main steam superheating heat exchanger and the reheating hot section steam superheating heat exchanger, and through the control strategy that the temperature is dynamically adjusted to be stable through the flow branches and the molten salt flow distribution of the main steam superheating heat exchanger and the reheating hot section steam superheating heat exchanger, the heat exchange efficiency of the molten salt is improved. The stability of the temperature of the heat storage medium in the heat storage process is greatly improved, and finally efficient operation of the system is achieved.

The application provides a kind of dual-channel low-frequency modulation intermediate frequency bipolar pulse signal generation system, it is related to integrated circuit technical field, including: power supply and signal isolation circuit, signal generation master module and signal generation slave module respectively connected with power supply and signal isolation circuit;The circuit structure in signal generation master module and signal generation slave module is identical;Signal generation master module includes the first step-down power supply circuit connected in turn, first MCU controller circuit, first intermediate frequency bipolar pulse signal generation circuit and first signal output interface, it also includes the first output current detection circuit and the first low-frequency current signal generation circuit connected with first MCU controller circuit, and first low-frequency current signal generation circuit is also connected with first intermediate frequency bipolar pulse signal generation circuit.The technical scheme of the application overcomes the problem that the medical device in the prior art cannot stimulate multiple acupoints simultaneously.

The invention discloses an underwater image enhancement and restoration method based on a double-domain adaptive fusion module. The method comprises the following steps: feature extraction; replacing the module; performing frequency decomposition; performing high-frequency treatment; performing low-frequency modulation; reconstructing a frequency domain; performing normalization processing; performing statistical enhancement; performing nonlinear enhancement; carrying out double-source gating fusion; performing multi-scale enhancement; and network backflow and image restoration. The method has remarkable innovation points in network structure position selection, frequency modeling branch processing strategies, normalized branch modeling modes and fusion mechanism design, and can effectively improve the fineness, stability and robustness of underwater image restoration.

The application discloses a method and system for primary frequency modulation optimization based on reinforcement learning, and relates to the technical field of power dispatching and power generation. The method comprises the following steps: constructing a hierarchical reinforcement learning framework, taking system frequency deviation, frequency modulation demand prediction and unit operation state as input based on a high-level policy network, and outputting a dynamic frequency modulation weight vector; taking the dynamic frequency modulation weight vector and real-time frequency deviation as input based on a bottom-level policy network, outputting output adjustment instructions of each frequency modulation unit, and performing real-time verification; if the verification fails, triggering a strategy re-planning mechanism to iteratively calculate new instructions. The application solves the technical problem that the existing frequency modulation method cannot effectively respond to the frequency fluctuation caused by large-scale renewable energy access, resulting in slow response speed and low frequency modulation accuracy. The application achieves the technical effect of optimizing through hierarchical reinforcement learning, realizing efficient power grid frequency regulation under renewable energy access, and improving the efficiency and accuracy of primary frequency modulation.

The invention discloses an energy storage system scheduling method and device and a storage medium. The method comprises the following steps: acquiring various bidding behaviors of each target object in a frequency modulation market; the various bidding behaviors are obtained according to historical bidding records of the corresponding target objects in the frequency modulation market; according to the current market data of the frequency modulation market, calculating the income of each type of bidding behavior in the frequency modulation market, so as to predict the bidding information of each target object in the frequency modulation market; inputting the bidding information of each target object in the frequency modulation market into a frequency modulation market clearing model to predict bid winning information of the target energy storage system in the frequency modulation market; the frequency modulation market clearing model takes the lowest frequency modulation service cost as a target; inputting the bid winning information into a pre-constructed decision model, and calculating to obtain an optimal scheduling strategy of the target energy storage system; the decision model aims at maximizing the sum of benefits in the electric energy market and the frequency modulation market. According to the invention, the scheduling capability of a single energy storage system can be optimized.

Embodiments of the present disclosure provide a method and system for detecting abnormal instructions in a hardware device, which is applied to the technical field of network security. The method comprises receiving a mixed signal sent by a DCS device, wherein the mixed signal is obtained by mixing a side phase waveform and a low-frequency modulation signal by the DCS device, the side phase waveform is obtained by executing a DCS instruction by the DCS device, and the low-frequency modulation signal is obtained by a control device according to a first signature value corresponding to the DCS instruction; extracting a second signature value according to the mixed signal; verifying the second signature value according to the first signature value, and sending a verification result to the control device, wherein the verification result comprises a legal instruction or an illegal instruction. In this way, the identifiable execution of abnormal instructions can be realized without modifying the control device.

The invention discloses a speaker recognition model frequency modulation trigger injection method, particularly relates to the technical field of voice signal processing and artificial intelligence safety, and is used for solving the problems that an existing speaker recognition model back door injection method is insufficient in concealment, poor in black box environment adaptability, prone to distortion of trigger characteristics in the physical transmission process and poor in safety. And mainstream defense strategies such as model fine tuning, pruning and spectrum detection are difficult to resist. The method comprises the following steps: constructing a piecewise linear periodic low-frequency modulation curve for original voice, generating a smooth phase track according to a frequency and phase relationship, constructing a sinusoidal modulator to generate a hidden sample, mixing normal samples in proportion and uniformly marking, training a model by a black box, testing reproduction modulation and inputting the model; according to the method, the defects of insufficient concealment, black box adaptability, physical robustness and defensive resistance of an existing method are overcome, high-concealment, high-robustness and wide-application backdoor injection is realized, and reliable support is provided for security evaluation of a speaker recognition system.

The application discloses an underwater image enhancement and restoration method based on a dual-domain adaptive fusion module, and comprises the following steps: feature extraction, module replacement, frequency decomposition, high-frequency processing, low-frequency modulation, frequency domain reconstruction, normalization processing, statistical enhancement, nonlinear enhancement, dual-source gate fusion, multi-scale enhancement, network backflow and image restoration. The application has significant innovation points in network structure position selection, frequency modeling branch processing strategy, normalization branch modeling mode and fusion mechanism design, and can effectively improve the fineness, stability and robustness of underwater image restoration.

Provided are a method and an apparatus for lens focusing, a computer device, and a storage medium. The method includes: acquiring a test image obtained by a lens shooting a reference image at a current focusing position, and determining a low-frequency modulation transfer function value of the test image; in response to determining that the low-frequency modulation transfer function value meets a preset value range condition, determining a high-frequency modulation transfer function value of the test image, determining a movement step according to the high-frequency modulation transfer function value, and controlling the lens to move according to the movement step; and using the next focusing position which the lens moves to according to the movement step as a new current focusing position to focus the lens for one time.

The application relates to the technical field of signal processing and brain wave synchronization, and provides a brain wave induction signal synthesis and modulation method and system.The method comprises the following steps: receiving user configuration, wherein the configuration at least specifies a target brain wave frequency and a modulation topology type; in response to user selection, generating a harmonic waveform synthesized based on a golden ratio constant as a carrier or a low-frequency modulation signal; applying the waveform signal to audio processing according to the selected topology type, wherein the topology type comprises binaural beat, monaural beat, isochronous tone sequence, Shepard ascending sequence and external audio modulation; enabling a smooth interpolation mechanism for real-time parameter changes of the user during the processing; and finally outputting a processed audio signal.The application solves the problems of single function, monotonous listening feeling and harsh switching in the prior art, and significantly enhances the flexibility, comfort and application range of the brain wave synchronization audio.

The invention relates to an M3C low-frequency port modeling method, device, equipment and medium based on multi-frequency coupling, and the method comprises the steps: carrying out the non-single frequency multiplication sequence decomposition of voltage, current and modulation signals in an M3C bridge arm loop under a static coordinate system, extracting the steady-state component of each frequency component, and constructing an M3C steady-state frequency domain model; based on the model, analyzing frequency coupling characteristics of the power frequency component, the low frequency component and the circulating current component in the bridge arm current, and establishing a frequency-crossing conversion relation between the bridge arm current and the power frequency current, and between the bridge arm current and the low frequency current and the circulating current; based on the cross-frequency conversion relation, performing frequency domain modeling on the low-frequency side control signal, and establishing a low-frequency modulation small signal model; the modulation effect of a control loop is introduced, a bridge arm modulation small signal comprehensive model is established, equivalent coupling factors of power frequency power grid impedance and a low-frequency transformer are considered, and an admittance model of an M3C low-frequency port is deduced. The method has the effect of improving the judgment accuracy of the dynamic relation between the bridge arm current and the port voltage under the low-frequency and frequency doubling coupling condition.

The application discloses a multi-frequency fusion ranging method and system based on echo quality adaptive control, relates to the field of multi-frequency ranging, and comprises the following steps: high and low frequency modulated light signals are transmitted to a target and echoes are received, echo signal characteristics of each frequency are obtained through demodulation, multi-frame echo quality evaluation quantities are calculated, and a hysteresis-continuous multi-frame judgment mechanism is adopted to identify that the current ranging state is normal or weak light. In the normal state, a high frequency dominant ranging strategy is executed; in the weak light state, the configuration parameters of the receiving front end are adjusted, and the judgment mechanism is used to adaptively switch to a low frequency dominant ranging strategy. According to the single frequency distance value of each frequency and the multi-frequency fusion weight corresponding to the selected strategy, the final ranging result is calculated. The application combines the hysteresis and continuous multi-frame judgment with the low frequency dominant mode, constructs a state machine cooperative link under weak light, realizes the stable linkage of the front end parameters, phase processing and multi-frequency fusion, realizes the smooth switching of the ranging strategy, and improves the ranging stability and accuracy under weak light.

The invention relates to an optical accelerometer control method based on nonlinear optical coupling, and the method comprises the steps: building a nonlinear optical coupling module, carrying out the laser initial calibration, and obtaining system parameters; and establishing a blue detuning light field to drive the mechanical oscillator to enter a pre-oscillation-starting state. During operation, the output optical power spectrum of the resonant cavity is acquired in real time, and the laser wavelength and the coupling spacing are dynamically adjusted by using a PID feedback algorithm, so that dynamic optimization adaptation of optical coupling is realized. The system monitors laser input power in real time and intelligently switches working modes according to the laser input power, parameter optical mechanical oscillation is started in a low-power interval to improve sensitivity, and the risk of mode interlocking is avoided by reducing power or adjusting coupling in a high-power interval. And finally, establishing a linear relation between the amplitude of a low-frequency modulation sideband signal generated by a mechanical oscillator in an output optical power spectrum and the external acceleration by resolving the low-frequency modulation sideband signal, so as to accurately calculate the magnitude and the direction of the acceleration. Dynamic adaptation, wide-power-range noise suppression and high-precision segmented solution are effectively realized, and the measurement accuracy is remarkably improved.

The application discloses a kind of wind generating set fault early warning method and system, comprising: S1: collecting vibration waveform original data and temperature original data, and respectively pre-processing;S2: design isomorphic double-flow network respectively extracts high-frequency impact feature and low-frequency modulation feature;S3: according to high-frequency impact feature and low-frequency modulation feature, calculate vibration feature;S4: extract temperature rate of change sequence and temperature time series feature;S5: generate vibration on temperature attention weight matrix, temperature on vibration attention weight matrix, calculate fault sensitive mask, and generate fault sensitive feature;S6: to fault sensitive feature is compressed, again calculate three-stage early warning probability vector, unit fault early warning class, and issue alarm.The application can solve the problem of early warning lag caused by the dependence of traditional wind generating set fault early warning method on fixed threshold alarm.