3385 results about "Low noise" patented technology

The invention discloses a high-precision current detection circuit calibration system, and belongs to the technical field of current detection circuit calibration. Comprising a shunt resistor, a Hall sensor array, a chopping amplifier, a low-noise operational amplifier, a temperature sensor, an ADC acquisition unit, an RC low-pass filter, a signal conditioning unit, a main control unit, an environmental parameter adaptive compensation module and a gain adaptive adjustment module. After an original signal is obtained from a multi-channel sensor, a dynamic error model is constructed in combination with temperature data, a long-term imbalance trend is extracted by using a moving average or median filtering algorithm, and real-time compensation is performed through a pre-calibrated temperature-drift function, so that the system maintenance period is remarkably prolonged.

The invention discloses an automatic anchor point searching and processing method for grid-connected test data of a photovoltaic inverter, and belongs to the technical field of automatic test of a power system. According to the method, a three-phase voltage and current signal output by a power grid simulator and an inverter power instruction signal are aligned through a high-precision time synchronization device; wavelet transform multi-scale noise reduction and moving average filtering combined preprocessing is adopted to improve the signal-to-noise ratio; identifying a voltage zero crossing point candidate set, a drop starting point candidate set and a recovery termination point candidate set based on a self-adaptive dynamic threshold value; transient energy characteristic verification is introduced for a voltage drop starting point; effective anchor points are confirmed through time window association of power instruction step changes. According to the method, the problems of low efficiency of manual key event point identification, misjudgment caused by noise interference, grid event and inverter response time sequence correlation missing and the like are solved, and the automation degree of test data analysis, anchor point positioning precision and control response time sequence analysis reliability are remarkably improved.

The invention relates to the technical field of wireless communication, and provides a radio frequency low-noise amplifier which comprises an input matching module, a power amplification module, an output matching module, an RC attenuation module and an RC feedback module, the first end of the first inductor serves as the input end of the output matching module, the first end of the first inductor is connected with the output end of the power amplification module, the control end of the switch and the first end of the first adjustable resistor, and the second end of the first inductor is connected with the first end of the second inductor and the first end of the third capacitor. The output end of the switch is connected with the first end of the first capacitor and the first end of the second capacitor. The second end of the second capacitor is connected with the second end of the first adjustable resistor and the second end of the second inductor and used for being connected with a power supply. The second end of the first capacitor is connected with the second end of the third capacitor and serves as the output end of the output matching circuit. The radio frequency low-noise amplifier is adjustable in gain, low in noise and good in linearity performance.

The invention relates to a low-noise water purifier. The water purifier includes a shell, an integrated waterway plate and booster pumps. The integrated waterway plate is fixed to the shell and is connected to water outlets of the booster pumps through pressure stabilizing tanks. The pressure stabilizing tanks are disposed between the booster pumps and the integrated waterway plate. The integrated waterway plate is connected to a plurality of evenly arranged connection tubes and is connected to filter core base assemblies through the connection tubes in a one-one correspondence manner. Beneficial effects of the water purifier are that 1) water knocking abnormal sound of the water purifier is reduced, thus improving user experience, 2) the structure is simple, water leakage is prevented and the production efficiency is high, and 3) replacement of a consumable of the product is simpler and more convenient.

The invention discloses a defect identification system of an ultrasonic flaw detector, and relates to the technical field of nondestructive testing, the system comprises a signal acquisition and preprocessing module, a defect feature identification module, a physical modeling analysis module, a life prediction and evaluation module and an intelligent decision visualization module; the signal acquisition and preprocessing module adopts a multi-frequency-point phased array transducer array, obtains an original signal through low-noise amplification, band-pass filtering and analog-to-digital conversion, and outputs a time domain signal matrix through adaptive noise reduction and gain compensation processing; the multi-frequency-point phased array transducer array and the advanced signal processing technology are integrated, the defect recognition precision and efficiency are remarkably improved, the system obtains high-quality original signals through low-noise amplification, band-pass filtering and analog-to-digital conversion technologies at first, then the high-quality original signals are subjected to self-adaptive noise reduction and gain compensation processing, and the defect recognition accuracy is improved. The background noise interference is effectively eliminated, and the purity of the signal is ensured.

The invention relates to the technical field of earphone noise reduction processing, and discloses a motorcycle riding earphone noise reduction method combining different noise reduction algorithms, comprising the following steps: acquiring an original audio signal, and distributing the original audio signal to each processing module; a preprocessing signal is generated, voice activity information is determined, and environmental noise energy information is calculated; according to the environmental noise energy or the auxiliary information, adaptively adjusting a high-low noise energy threshold value; the original signals are input into the traditional and AI noise reduction module in parallel to generate two paths of noise reduction signals; based on the environment noise, the voice information and the threshold value, performing weighted fusion to generate an output signal; and carrying out equalization and compression processing on the output signal, and driving the loudspeaker to output. According to the invention, comprehensive judgment on environmental noise energy information, voice activity information and riding speed is introduced, an output signal of a traditional noise reduction module is set to be zero in a high-speed voice-free environment, and AI noise reduction is combined, so that tone quality distortion of a traditional noise reduction algorithm in a noise environment is effectively avoided.

The invention discloses a full waveform decomposition method based on a multi-branch convolutional neural network, which belongs to the technical field of airborne depth sounding laser radars, is used for airborne depth sounding laser radar echo signal decomposition, and comprises the following steps: obtaining and preprocessing an original echo sequence, constructing a multi-branch one-dimensional convolutional neural network model, and performing neural network training; and based on the trained multi-branch one-dimensional convolutional neural network, outputting a three-channel probability heat map, performing multi-peak sub-pixel decoding on the three-channel probability heat map, and outputting a peak accurate position, a peak accurate position normalized value and a peak confidence after de-weighting. According to the method, the multi-branch one-dimensional convolutional neural network model is constructed and multi-peak sub-pixel decoding is carried out, so that stable decomposition of full-waveform echoes and high-precision positioning of multi-echo peak values are realized under the conditions of multi-peak superposition, high noise and weak signals, and false detection and false peaks caused by noise are remarkably reduced.

The invention discloses a low-illumination image enhancement method suitable for a complex night operation scene. The method comprises the following steps: preprocessing a collected low-illumination RGB image to obtain a preprocessed image; multi-scale features are extracted from the shallow convolution projection and the multilayer axial converter unit, and cross-layer attention is used for weighted aggregation to form enhanced features; forward noise addition is executed according to noise scheduling to obtain potential representation representing low-illumination noise distribution, and deep semantic representation is extracted in combination with a U-Net encoder embedded in CBAM; in the decoding stage, details are reconstructed through jump connection and a multi-layer converter, meanwhile, reverse denoising of a diffusion model is introduced to gradually remove noise, and an enhanced image with high brightness and low noise is output; in the training process, bidirectional mapping from low illumination to normal illumination is constructed, a weighted target of self-encoding loss and enhanced loss is introduced to carry out joint optimization on a converter and diffusion parameters, and a low-illumination scene is monitored on a general GPU / edge device.

The invention belongs to the technical field of power equipment state monitoring and fault diagnosis, and discloses a high-voltage circuit breaker state quantity feature extraction method and system, and the method comprises the steps: extracting a local maximum value point and a local minimum value point from a denoised original monitoring signal; feature vectors are constructed and input into the trained neural network model, six-dimensional vectors are output in a Softmax output layer, and the six-dimensional vectors correspond to prediction probabilities of six orders in the optimal B-spline interpolation orders; based on the constructed envelope, adopting empirical mode decomposition to obtain a plurality of intrinsic mode function IMF components, performing Hilbert transform on each IMF component, constructing a state quantity feature set according to the instantaneous frequency and amplitude change of the analysis signal, and judging whether the high-voltage circuit breaker is in an abnormal state or not. According to the method, the vibration signals collected by the circuit breaker are subjected to multi-stage intelligent processing, high-precision and low-noise feature extraction is achieved, and fault diagnosis and state recognition of the circuit breaker can be effectively supported.

The invention discloses an Internet of Vehicles CAN bus intrusion detection method based on noise perception active learning, and belongs to the technical field of Internet of Vehicles safety and machine learning. The invention aims to solve the technical problems of false label noise interference, high manual labeling cost, high attack missing report rate caused by class imbalance and the like. The core of the method is to execute a noise sensing mixed query strategy in an iterative loop: firstly, generating a pseudo tag through clustering and correcting by using an integrated noise detector; secondly, calculating uncertainty scores and noise probabilities of the samples, fusing the uncertainty scores and the noise probabilities to obtain a comprehensive score, and preferentially selecting the samples with high uncertainty and low noise probabilities; and then adaptively selecting a sampling strategy according to the model performance and applying category balance constraint. The query batch is used to iteratively update the model while dynamically adjusting the classification threshold to reduce the missing report rate. According to the method, the influence of pseudo label noise can be effectively suppressed, and the attack detection precision and generalization capability are remarkably improved with extremely low labeling cost.

The invention discloses a band-gap reference source high-order curvature compensation method based on an operational amplifier, and belongs to the technical field of high-precision analog integrated circuit design, and the method comprises the steps: generating an absolute temperature proportion PTAT current and an absolute temperature complementary CTAT current; generating a high-order curvature compensation voltage based on the PTAT current and the CTAT current; amplifying the generated high-order curvature compensation voltage through an operational amplifier, and superposing the amplified high-order curvature compensation voltage to a band-gap reference output stage to obtain compensated band-gap reference output voltage; and carrying out band-gap reference output calibration on the compensated band-gap reference output voltage so as to ensure that the temperature drift coefficient of the output band-gap reference voltage in the preset temperature range is smaller than a preset threshold value. According to the invention, high-order elimination of base-emitter voltage nonlinearity can be realized in a wide temperature range, and a relatively low temperature coefficient and low noise are achieved.

Methods and devices to improve nonlinearity performance of low noise amplifiers (LNAs) are disclosed. The described methods and devices reduce the capacitive loading of the LNA amplifying devices on the bypass path of the LNAs when operating in the bypass mode. This is performed by decoupling the active devices from ground to put the amplifying devices in a floating state, thus minimizing the impact of the gate-source capacitances of the amplifying devices on the overall linear performance of the LNA operating in the bypass mode.

The invention discloses an ultrahigh-precision low-noise band-gap reference circuit, and relates to the technical field of integrated circuit design, the ultrahigh-precision low-noise band-gap reference circuit comprises a starting turn-off circuit, a reference core circuit, a high-order temperature compensation circuit and a digital trimming circuit, the starting turn-off circuit is connected with the reference core circuit, and the high-order temperature compensation circuit is connected with the digital trimming circuit through a Brokaw type band-gap reference topological structure. Accurate compensation of the voltage is achieved through deep negative feedback of the operational amplifier and a high-order temperature compensation circuit, and the temperature drift of the reference voltage is as low as 0.5-2 ppm / DEG C (-40 DEG C to 130 DEG C) and is increased by 5-10 times compared with a traditional band-gap reference (5-10 ppm / DEG C). The power consumption of the system can be saved by turning off the circuit when the circuit is not used, meanwhile, the digital trimming module is added to replace laser trimming in traditional engineering, the upper limit of the trimming range is improved, high-order temperature compensation and the digital trimming circuit are introduced, the precision of output voltage is improved, different devices are replaced, and the noise of the circuit is reduced. The problems that a traditional high-voltage band-gap reference temperature coefficient is high and noise is large are solved.

The invention relates to the field of crystal detection, and discloses a multifunctional comprehensive scintillation crystal and semiconductor material energy spectrum test system, which comprises a crystal adaptation module for judging the type and size of a placed crystal through an optical identification and parameter matching mechanism, and a signal acquisition module for amplifying, filtering and digitally processing an acquired signal, the signal correction module is used for correcting waveform distortion and baseline drift through a dynamic gain adjustment algorithm of a low-noise pre-amplifier and a multi-channel pulse analyzer; the data analysis module is used for generating an energy spectrum curve according to input signals and calculating energy resolution, counting rate, light attenuation time or carrier mobility, and the environment calibration module is used for dynamically evaluating shading degree, environment light, temperature and electromagnetic interference through an optical sensor and a temperature and humidity detector in the camera obscura and outputting updated calibration parameters. The method has the advantages of improving the detection efficiency, precision and consistency.

The invention belongs to the technical field of robot and intelligent equipment driving, and discloses an integrated servo push rod and a control method.A push rod module is integrally designed, and a high-density motor, a ball push rod and a band-type brake are arranged in the push rod module. The invention has the characteristics of high power density, small volume, high precision, high response speed, low noise, long service life, high reliability and the like. The inside of the push rod is contacted with the deep groove ball bearing at the tail end of the screw rod through the ball retainer, and the outside is contacted with the oil-free bushing, so that the lateral strength of the push rod is improved, friction is reduced, and operation efficiency is improved. A linear encoder is arranged at the end of the screw rod, accurate control of the tail end can be achieved, a magnetic encoder is arranged at the tail end of the motor, input and output end double-closed-loop can be achieved, precision can be improved, and errors of band-type brake in power-off response time can be compensated; the lead screw is connected with the motor through the diaphragm coupling in the motor, so that the axial space is saved, and the service life of the motor is prolonged.

Disclosed is a technology related to a low noise amplifier applied between an RF receiver and an RF switch. A phase difference matching circuit having the same phase difference as a positive gain amplifier is added to an output terminal of an attenuator of a low noise amplifier having a variable gain switching structure. In addition, a variable impedance circuit connected to an output terminal of the positive gain amplifier and an output terminal of a phase difference matching unit to finely adjust the phase difference of each output terminal may be further included.

The invention discloses a left auricle plugging device provided with a real-time atrial pressure monitoring chip, and particularly relates to the technical field of left auricle plugging, the left auricle plugging device comprises a plugging device body and a connector, the top end of the connector is provided with a signal supply module, and the signal supply module is located in the plugging device body; according to the left atrial appendage plugging device, after the inductor in the nut generates a voltage signal, the signal can be transmitted to the low-noise operational amplifier and amplified, meanwhile, the electromagnetic interference between the noise of a circuit and the outside world is restrained, and the plugging device can be used for plugging the left atrial appendage of the left atrial appendage of the left atrial appendage. The physical deformation is converted into corresponding electric signals, the electric signals are transmitted to an in-vitro receiving terminal through an antenna, and a doctor can indirectly judge the recovery condition of the atrial contraction function of the patient according to the pressure signal trend.

The utility model discloses an indoor tire road noise testing device, which comprises a sound insulation box body, a slope table, a tire driving system and a noise acquisition system, the sound insulation box body is of a closed structure. The slope table is arranged in the sound insulation box body, the slope table is provided with a slope adjusting device, the slope adjusting device is used for adjusting the slope angle of the slope table, and the slope of the slope table is used for placing a pavement test piece; the tire driving system comprises a driving assembly and a tire fixer, the tire fixer is rotatably mounted at the output end of the driving assembly, the tire fixer is arranged in the sound insulation box body and is used for mounting a test tire, and the driving assembly is used for driving the tire fixer to move to drive the test tire to impact the pavement test piece; the noise acquisition system is used for acquiring sound data in the sound insulation box body. The slope table is used for simulating the actual road surface condition, the noise condition of the designed road surface can be tested in the design stage, the actual effect is effectively guaranteed, and the problem that low-noise road surface design is difficult to verify in the design stage is solved.

The invention relates to the technical field of quantum sensing, and discloses a quantum magnetometer device based on diamond NV color center photoelectric detection magnetic resonance. Wherein the optical path module performs optical excitation and imaging positioning on a diamond sample through a laser and an objective lens. The circuit module integrates a microwave source, a lock-in amplifier, a low-noise current amplifier and a programmable power supply, applies a microwave field and bias voltage to a microwave antenna and an interdigital electrode on the surface of a sample respectively, and collects sample dependency photocurrent signals. And the operation module realizes automatic control and data processing of the system. According to the device, NV color center spin state information is converted into a photocurrent signal through a photoelectric detection magnetic resonance mechanism, an amplitude or frequency modulated signal is demodulated by using a lock-in amplification technology, information of a detection magnetic field is obtained through a high signal-to-noise ratio photoelectric detection magnetic resonance spectrum, and the sensitivity of the magnetic field is evaluated. Magnetic resonance signals are detected in an all-electricity mode, a complex optical system of traditional fluorescence detection is simplified, and the device has the advantages of being compact in structure, easy to integrate, high in measurement sensitivity and the like.

The invention relates to the technical field of digital-analog hybrid integrated circuit design, and particularly provides a FlexRay bus receiver with low noise and wide common-mode input range, which is characterized in that a common-mode voltage division module receives a differential signal output by a bus and carries out voltage division, and then a pre-amplification module carries out amplitude limiting amplification on the signal after voltage division; the hysteresis comparator module generates threshold hysteresis and generates a logic level signal based on the signal after amplitude limiting amplification, and the digital logic receiving module receives and processes the logic level signal, generates a receiving data signal and transmits the receiving data signal to an external controller. According to the invention, the differential signal pair received on the FlexRay bus is subjected to voltage division and amplitude limiting amplification, so that a wider common-mode input range is provided, the problem of large ground potential difference is solved, and the capability of suppressing common-mode noise and the communication stability are improved. Meanwhile, the hysteresis comparator module suppresses noise false triggering by generating threshold hysteresis, and the anti-noise capability of the receiver is remarkably improved.

The invention relates to a multi-scale adaptive detection method and system for an infrared image of an unmanned aerial vehicle, computer equipment, a storage medium and a computer program product. The method comprises the following steps: preparing and inputting infrared image data; performing reasoning and dynamic adaptive slicing processing on the infrared image data; building a detection framework model; performing model training and optimization based on the sliced infrared image data; and generating an image detection result. The problems of large target scale difference, low contrast, high noise, complex image, blurred target and the like in unmanned aerial vehicle infrared image target detection in the prior art are solved, and the detection performance is remarkably improved. Besides, a YOLOv8 + SAHI + ULSAM integrated model is adopted to carry out target detection on the infrared image of the unmanned aerial vehicle, a target under a complex background can be identified more accurately through the improved YOLOv8 model, the ULSAM module is light in weight, and the calculation burden cannot be remarkably increased.

The invention relates to the technical field of audio signal processing, and discloses a DRC control method for improving indoor riding reducibility of a motorcycle riding earphone, and the method comprises the steps: judging a macroscopic acoustic scene; analyzing a microcosmic acoustic state; and according to the scene, the state and the feedback correction vector, generating an audio processing parameter and applying the audio processing parameter to the audio signal. The method further comprises the steps of analyzing the output signal through an artifact monitor to detect and process artifacts, generating the correction vector and feeding the correction vector back to the parameter synthesizer, and forming closed-loop control. The system comprises a scene judgment module, a joint acoustic state encoder, a parameter synthesizer, a parameterized audio processing engine and an artifact monitor. Through combination of macroscopic scene judgment and microcosmic closed-loop self-adaption, the contradiction between noise reduction performance under high noise and high sound reducibility under low noise is solved, and the audio quality under different scenes is improved.

The invention provides a low-noise amplification circuit, a radio frequency front-end module and an electronic terminal.The low-noise amplification circuit comprises an input matching adjustment circuit, a common-source amplification circuit, a first inductor, an inter-stage feedback circuit, a common-gate amplification circuit, an input and output feedback circuit, an output matching adjustment circuit and an output attenuation circuit; the common-source amplification circuit comprises at least two first signal amplification units; each first signal amplification unit comprises a first field effect transistor; the common-gate amplification circuit comprises at least two second signal amplification units; and each second signal amplification unit comprises a second field effect transistor. According to the low-noise amplification circuit, transconductance nonlinearity of devices under different bias signals can be mutually counteracted, so that the linearity fluctuation of the low-noise amplification circuit at different temperatures is optimized, and the linearity of the low-noise amplification circuit is kept basically not to change along with the temperature change.

The invention relates to the technical field of voice recognition and intelligent household electrical appliance control, in particular to an induction cooker control method based on voice recognition, which comprises the following steps: synchronously acquiring a voice signal and a lip motion signal, and generating multi-modal sensing data through timestamp alignment; performing cross-modal anti-noise processing on the data, and outputting a fusion feature tensor fusing acoustic and visual features; semantic analysis and danger instruction recognition are executed, and a risk filtering control instruction containing temperature / fire parameters is generated; calling a thermodynamic dynamic model to calculate a power control parameter based on the instruction parameter; converting a power parameter into a space vector modulation waveform to drive a load circuit, and synchronously collecting the pot bottom temperature; and triggering an anti-noise process and thermodynamic model parameter updating according to the deviation between the actual temperature and the target temperature. Noise interference is reduced through multi-modal signal cross validation, dangerous instructions are intercepted in combination with knowledge graph constraints, the noise suppression capability and the temperature control precision are continuously optimized, and the safety and reliability of voice control in a complex environment are improved.

The invention provides a Q / V frequency band large-scale satellite communication load channel optimization method, which comprises a forward input link and a backward output link, and is characterized in that the forward input link comprises a V frequency band multi-beam antenna feed source array, a plurality of input preselectors, a V frequency band low noise amplifier input switch ring and a V / Ka integrated receiver assembly which are connected in sequence; the output end of the V-band multi-beam antenna feed source array is connected with the input end of the V / Ka integrated receiver assembly through a waveguide. The reverse output link comprises a Ka / Q frequency converter, a Q-band channel filter, a Q-band traveling wave tube amplifier, a change-over switch, a Q-band output duplexer or filter and a Q-band feed beam which are connected in sequence, and the output end of the Ka / Q frequency converter is connected to the input end of the Q-band channel filter through a waveguide; and the output end of the Q-band channel filter is directly connected to the input end of the Q-band traveling wave tube amplifier through the waveguide. Therefore, the method is low in link loss, high in space utilization rate and high in feasibility.

The utility model belongs to the field of paralleling reactors, and relates to a low-noise extra-high voltage paralleling reactor iron core structure which comprises iron core columns, an upper cross beam, an upper pressing beam, a lower foot pad and an iron yoke of a rectangular frame type structure, fourth pull screw rods are installed at the front ends and the rear ends of the left side and the right side of the periphery of the iron yoke, and fifth pull screw rods are installed at the front ends and the rear ends of the upper side and the lower side of the periphery of the iron yoke. VII pull screws are mounted at two corners of the upper end of the inner circumference of the iron yoke and close to the inclined seam of the iron yoke, first shock pads are mounted between the fourth pull screw, the fifth pull screw and the seventh pull screw and the iron yoke, insulating base plates and second shock pads are mounted between the upper cross beam, the upper pressing beam and the lower foot pad and the iron yoke, and an insulating rubber plate of an L-shaped structure is mounted on the outer side of the iron yoke and close to the inclined seam of the iron yoke. According to the utility model, the clamping force of the iron yoke is increased by adding the iron yoke clamping structure, and the transmission of noise is blocked while the vibration is reduced by adding the pull screw rod and arranging the insulating rubber plate at the inclined seam of the iron yoke; and by arranging a shock pad, noise generated by resonance of the pull screw rod is further reduced.

Front-end architectures for simultaneous transmit and receive operation are provided herein. In certain embodiments, a front-end module includes a bandpass filter, an antenna terminal for connecting to an antenna, a power amplifier, a low noise amplifier, and a plurality of switches for controlling access of the power amplifier and the low noise amplifier to the antenna terminal. The switches are operable in a transmit mode in which the switches connect an input of the power amplifier to the bandpass filter and an output of the power amplifier to the antenna terminal, and a receive mode in which the switches connect the input of the low noise amplifier to the antenna terminal through the bandpass filter.

The invention relates to the technical field of axial flow fans, in particular to an impeller with S-shaped blades and an axial flow fan applying the impeller. On the axial projection of the impeller, the peripheral contour of each blade is defined by a blade root edge, a blade front edge, a blade outer edge and a blade rear edge, and the blade root edge is connected with the hub; the blade front edge and the blade rear edge are both wavy curves, and the blade root edge and the blade outer edge are both arc-shaped lines with the axis O of the hub as the circle center. Each blade is provided with a pressure face and a suction face, the pressure faces and the suction faces are oppositely arranged, and on the radial projection of the impeller, the pressure faces are arranged in a wave shape in a protruding mode from one end of the blade root edge to one end of the blade rear edge towards the suction faces and then concavely arranged, so that the blades are of an S-shaped structure. The problem that a traditional blade structure cannot meet the use requirements of high efficiency and low noise can be solved.

The invention relates to the technical field of anomaly detection, in particular to an AI-based network traffic anomaly detection method, which comprises the following steps of: acquiring a difference value between a communication triple and a timestamp, dividing a time period to generate an aggregation interval, extracting a message to construct a density sequence, calculating a fluctuation offset degree, normalizing the density sequence to obtain a gradient margin, and identifying a low-margin interval feature. An anomaly ratio is generated in conjunction with the intersection interval. According to the method, data are reconstructed according to a triple by obtaining a source IP, a target IP, a port and a timestamp, a timestamp difference value is calculated, a minimum value is selected as a minimum communication interval to divide a time period to generate an interval sequence, the number of messages is extracted to construct a density sequence, and an adjacent difference value is calculated to obtain a fluctuation mean square offset degree quantization fluctuation characteristic. Normalizing the density sequence, calculating the gradient and amplitude to form a margin to reduce noise, judging the statistical duration of a low margin interval to obtain a stable feature recognition abnormal state, and generating a communication stability abnormal ratio in combination with a cross ratio to provide a quantitative score to improve the detection precision.

The embodiment of the invention provides a large language model optimization method based on variance reduction and momentum acceleration. The method comprises the following steps: a gradient estimation stage of a large language model: initializing a seed list and a projection list; executing multiple independent query iterations, calling a disturbance subprogram, generating a random seed for the large language model, and storing a table; in the disturbance subprogram, the determined gradient projection value is stored in a projection list, and iteration is carried out for next query; after executing multiple independent query iterations, storing a plurality of random seeds and a plurality of gradient projection values corresponding to the random seeds; in the weight updating stage of the large language model, a gradient norm subprogram is called for each layer of the large language model, and a random seed is obtained to reset a random number generator; and determining variance-reduced gradient estimation according to the gradient projection value extracted from the projection list and the reproduced disturbance vector. According to the embodiment of the invention, gradient information queried for multiple times is aggregated to generate low-noise gradient estimation, and fine tuning of a large language model is completed.