36 results about "Exponential weighting" patented technology

The invention discloses a multi-source navigation data fusion method for an unmanned loader, which comprises the following steps: data acquisition: acquiring environmental perception data of satellite navigation, an inertial measurement unit IMU, a wheel type odometer and a laser radar and camera in real time; performing space-time synchronization preprocessing, realizing multi-source data time synchronization, unifying environment sensing data to a body coordinate system, eliminating abnormal data, and complementing missing data; carrying out dynamic weight calculation, establishing an error model of each sensor, and adjusting a fusion weight by using an error reciprocal exponential weighting method; layering fusion is carried out, a satellite and an IMU are fused through bottom-layer extended Kalman filtering (EKF), a laser radar and a high-precision map are fused through middle-layer ICP, a middle-layer result and a wheel type odometer are integrated through high-layer federated filtering, and high-precision fusion is achieved; and outputting and optimizing a result, outputting navigation data, performing closed-loop optimization, monitoring the health degree of the sensor, and executing redundancy switching when a fault occurs. The system comprises a corresponding processing unit, and a storage medium stores a program for executing the method.

The invention provides a multi-modal feature fusion identification method and device for partial discharge of a high-voltage cable, and relates to the technical field of multi-modal feature fusion of partial discharge of the high-voltage cable. Multi-modal time sequence data of three-phase current, temperature and voltage to ground of the high-voltage cable are collected and subjected to preprocessing and time alignment to form a multi-dimensional input matrix; a physical relation matrix based on cable physical characteristics is constructed, physical constraints are injected into data through feature modulation, fused feature representation is generated, features are input to an LSTM auto-encoder trained by normal data, a sequence is reconstructed, a reconstruction error is calculated, finally, an exponential weighting algorithm is adopted to accumulate an error time sequence to obtain a comprehensive abnormal score, and the comprehensive abnormal score is obtained. And through comparison with a dynamic threshold value, accurate identification and partial discharge alarm of the insulation state abnormity are realized.

The invention discloses an elevator prediction scheduling method and system based on user habit self-learning, and the method comprises the steps: carrying out the statistical analysis based on historical elevator taking event data, and judging whether a floor combination meeting a preset condition exists or not; combining all the independent floors with the floors meeting the preset conditions to serve as a self-learning unit; for each self-learning unit, updating the elevator calling probability of the respective learning unit by using an exponential weighted average algorithm; when the elevator does not have the real-time task, all elevator calling probabilities corresponding to the current time window are inquired, and the self-learning units meeting the triggering condition are extracted as a target candidate set; calculating an optimal pre-stop layer by utilizing a mathematical model based on the target candidate set, and calculating a comprehensive probability corresponding to the optimal pre-stop layer; and calculating an income evaluation function based on the comprehensive probability, and generating an instruction to drive the elevator to run to an optimal pre-stop layer to enter a prediction waiting state when an obtained value meets a condition.

The present disclosure provides a hardware resource allocation method, device and equipment of an electronic device, which preliminarily identifies hardware power consumption mutation by sampling the hardware power consumption data of the electronic device in real time and determining the fluctuation index corresponding to the current sampling data. Further, the exponential weighted value of the current sampling data is determined, and when the current exponential weighted value and the exponential weighted value when the last fluctuation index meets the set condition, the mutation is determined as an effective mutation. Furthermore, when the fluctuation index and the exponential weighted value meet the condition at the same time, the future hardware power consumption data is further predicted, and the hardware resource demand information of the electronic device in the next target period is determined in combination with the fluctuation index. The fluctuation of the hardware power consumption data can cover the user's fine-grained operation, and by predicting the future hardware resource demand information, the resources can be prepared in advance, avoiding the situation of insufficient or lagging resources during scene switching, and ensuring smooth user experience.

This filtering method comprises steps in which: a Kalman gain is calculated; a system calculates an estimated value on the basis of the Kalman gain; a predicted value and a measured value of an error covariance are compared and analyzed to calculate the error covariance; an exponentially weighted average and an exponentially weighted covariance for the estimated value are calculated; and the exponentially weighted average and the exponentially weighted covariance of the calculated estimated value are reflected in the subsequent Kalman gain calculation. Therefore, performance superior to that of a conventional Kalman filter can be achieved in non-Gaussian noise conditions.

The application discloses an intelligent evaluation system and method for a knowledge base question answering application, and relates to the technical fields of artificial intelligence, natural language processing and multi-agent collaborative system. The system comprises a data preprocessing module, a data synthesis module, a data screening module, a data scoring module and an adaptive weight adjustment module. The data preprocessing module performs data preprocessing on enterprise original documents and generates a data knowledge base. The data synthesis module repeatedly asks and answers each paragraph of the data knowledge base by using a large language model to generate a question and answer pair. The data screening module screens out high-quality samples through screening agent collaboration and a MACA framework mechanism. The data scoring module scores the question and answer pair through scoring agent collaboration and a MACA framework mechanism. The adaptive weight adjustment module dynamically updates the weight of each screening and scoring dimension by using an exponentially weighted moving average algorithm.

The application discloses a method for removing and realizing strong vibration interference in a force signal of a polishing robot, two filter algorithms are connected in series to form a double Kalman filter, a first filter realizes filtering of Gaussian white noise and high-frequency vibration spectrum peak group noise; a single parameter is introduced, a gradually time-varying noise variance is designed in an exponential weighting mode to describe the colored noise characteristics, and an improved Kalman filter is designed to form a second filter to realize filtering and elimination of the colored noise. The application realizes effective filtering and elimination of complex noise composed of Gaussian white noise, vibration noise and colored noise superimposed in force measurement information in robot force control polishing.

The invention discloses a Bernoulli filter-based single target detection and DOA tracking method and system, and relates to the field of target tracking. The method solves the problems that a traditional tracking algorithm after detection cannot jointly detect the target when the underwater target repeatedly enters under the conditions of low signal-to-noise ratio and small snapshot, and the tracking effect is unstable. The method comprises the steps of constructing a Bernoulli random finite set model, defining a single target state as a dynamic description target state or a dynamic description target state, predicting the target existence probability by using a Bernoulli Markov process, and generating a prediction state distribution particle set composed of new particles and survival particles through particle filtering. A hydrophone array signal is decomposed into a noise subspace observation model and a signal subspace observation model, a generalized likelihood function after exponential weighting is used for updating a target posteriori existence probability and a particle weight, and an equal-weight particle set is generated through resampling. And judging whether the target exists or not according to the single-target posterior existence probability, thereby realizing single-target DOA joint detection and tracking.

The invention relates to the technical field of high-voltage relay life prediction, and discloses a high-voltage relay residual life prediction method based on an implicit non-linear scale transformation wiener process, and the method comprises the steps: carrying out the smoothing of the arcing time degradation data of a high-voltage relay through employing a moving average method based on the exponential weighting reinforcement recent influence; dividing different degradation stages of the high-voltage relay by using an important point clustering algorithm, extracting later-stage degradation data of arcing time, extracting a low-frequency intrinsic mode function component reflecting degradation characteristics of the high-voltage relay in the later-stage degradation data of the high-voltage relay through variational mode decomposition, and verifying feasibility of a wiener process; meanwhile, a degradation model is established for the later degradation stage of the high-voltage relay by using an implicit nonlinear scale transformation wiener process, model parameters are estimated, the influence of system noise is considered at the same time, and a drift coefficient is updated online through a Bayesian recursion method; and finally, predicting the residual life of the high-voltage relay based on the wiener process model in combination with the failure threshold value of the high-voltage relay, and evaluating the reliability.

The invention provides a civil aviation operation-oriented airspace probability modeling and risk quantification method, which uses ADS-B data to construct layered space-time probability density, adopts an anisotropic kernel which is widened along the course and tightened in the transverse direction, and realizes short-time smoothing by using a time kernel. Inverse detection probability correction is introduced into the sample weight to reduce the coverage deviation. The reference density is subjected to multi-day exponential weighting generation according to hour data so as to describe day and night laws and workday characteristics. Under an information geometry framework, a relative entropy item, a Fisher information item, a capacity penalty item and a conflict kernel item are combined into a unified risk functional, and calculation of density evolution and a congestion dissipation direction is realized by a Wasserstein gradient flow. The system is composed of a data preprocessing module, a detection probability correction module, a layered anisotropic space-time kernel density estimation module, a reference density generation module, a risk functional and gradient flow module and an increment updating and index output module. The technology has unified modeling and clear interpretable and engineering landing characteristics, and can be used for situation analysis and collaborative decision support of air routes, sectors and terminal areas.

The application discloses a kind of based on multi-modal data fusion's intelligent early warning system and method of accompanying, belong to wisdom medical treatment and artificial intelligence technical field, this system includes multi-modal data acquisition and pre-processing module, dynamic weight fusion module, intelligent risk assessment module and multidisciplinary collaborative intervention module, through hardware synchronization and software interpolation, the time-space alignment of multi-source sensor data is realized, the dynamic weight fusion mechanism combining multi-head self-attention and evidence theory is used to process modal conflict, risk dynamic assessment is carried out using sliding window exponential weighting and LSTM time series modeling, and based on medical knowledge graph and improved hungarian algorithm, the intelligent scheduling and closed-loop feedback optimization of nursing resource are realized.

The application discloses a kind of distributed power grid-connected control method and device, by accurately time alignment with power supply side power data and grid side voltage frequency data and combining static and dynamic phase compensation eliminates collection error;Power fluctuation cumulative grading and exponential weighted dynamic reference correction are used to evaluate power deviation under different severity, and additional credibility correction factor improves the reliability of deviation evaluation;Introduce bidirectional power flow link identification and transmission link loss nonlinear accumulation technology combined with load rate correction factor to realize accurate compensation of power regulation amount;By steady-state holding domain shrinkage analysis and disturbance propagation domain expansion analysis, the disturbance influence range is defined to construct adaptive response parameter set with overlapping transition characteristics;Fusion nonlinear grading gain strategy of linear regulation zone and limiting amplitude regulation zone is based on gain gradient mutation position Adaptive setting of grading regulation threshold value outputs grid-connected control instruction, improve the reliability and power regulation accuracy of distributed power grid-connected operation.

The application provides a non-volatile storage device cold and hot data shunt management method and device. In view of the mixed storage demand of real-time data and historical data in the industrial control scene, a dynamic shunt method based on logical space heat evaluation is proposed. By dividing the 512GB SSD user address space into 64000 logical spaces, using the exponential weighting algorithm to collect operation frequency data in real time, combining the DRAM history bidirectional list to generate the access order array, and calculating the comprehensive heat value. Based on the quantile coefficient, the threshold value is dynamically generated, the logical space with a heat value greater than the threshold value is allocated to the low-wear flash block with an erase number ≤5000, and the logical space with a heat value less than the threshold value is allocated to the high-wear block. The scheme realizes the substantial prolongation of the storage life, the reduction of the real-time instruction response delay, the adaptation to the mixed load characteristics of the high-frequency real-time and low-frequency archiving of industrial control, and ensures the balance between the data access efficiency and the storage reliability.

Novel tools and techniques are provided for implementing exponentially weighted scoring of calculated resource availability for automated network element selection, augmentation, and service delivery. After receiving a request for network services and measurement data associated with a plurality of network elements capable of providing the requested network services, a computing system may calculate a weighted score for each of a plurality of resources that is available on each network element. Each weighted score for a corresponding resource may be based on an exponential decay function of a difference between a limit value corresponding to an allocatable amount of the corresponding resource and a measured usage value based on the measurement data. For each network element, the computing system may calculate an availability score based on the weighted scores, and may output either the availability score and / or a resource allocation decision for that network element.

This application relates to the field of power equipment condition monitoring and intelligent operation and maintenance technology, specifically to a dynamic weighted safety assessment method for pumped storage units with coupled multi-condition operation. The method includes: real-time acquisition of multi-source raw data; preprocessing the multi-source raw data to obtain multi-source data; identifying the current micro-operating condition of the pumped storage unit based on the multi-source data; determining a benchmark weight vector based on the micro-operating condition, and correcting the benchmark weight vector based on normalized degradation degree using an exponential weighting function to obtain a dynamic weight vector; and performing a fuzzy comprehensive evaluation of the operating status of the pumped storage unit based on the dynamic weight vector to obtain a health index. The dynamic weighted safety assessment method for pumped storage units provided in this application can dynamically adjust the weights of assessment indicators according to micro-operating conditions, improving the accuracy of safety assessment for transient and steady-state processes, and achieving adaptive quantitative assessment of the health status of pumped storage units.

The invention discloses a lamp post vibration and inclination monitoring method and device and electronic equipment. The method comprises the following steps: acquiring inclination angle original data of a lamp post and associated information of the lamp post, wherein the associated information at least comprises geographical location information and local weather forecast information of the lamp post; establishing a baseline model based on an exponential weighted moving average algorithm, and calculating a current normal inclination angle baseline; calculating an exponential weighted moving variance and calculating a standard deviation according to the exponential weighted moving variance; according to the standard deviation, the normal fluctuation range of lamp pole inclination is determined by combining the current normal inclination angle base line; based on the association information and the normal fluctuation range, adaptively adjusting a preset initial alarm threshold; and comparing the inclination angle original data with the adaptively adjusted alarm threshold value, and if the inclination angle original data exceeds a range corresponding to the alarm threshold value, triggering an alarm and reporting alarm data. According to the invention, the situation of frequent alarm reporting or even false alarm can be avoided, and the operation and maintenance cost is reduced at the same time.

The invention discloses a variable step size self-adaptive elimination method for electromyographic signal electrocardio interference. The method comprises the steps that original myoelectricity observation signals and reference electrocardiosignals are obtained and preprocessed; constructing an adaptive FIR filter and a reference input vector, and calculating a full-frequency error signal; separating estimation error total power, electrocardio frequency band power and reference signal power in an exponential weighting mode; calculating a residual electrocardio energy ratio and a ratio of the electrocardio energy to the reference electrocardio power based on power estimation, constructing a basic step length according to the residual electrocardio energy ratio and the ratio, and performing electrocardio dominant detection to adjust the step length; the correlation between the reference signal and the error electrocardio component is further calculated, and step freezing control is implemented according to a correlation threshold value; and finally, carrying out normalized least mean square updating on the weight vector by adopting the final variable step size. According to the method, the step length can be adaptively adjusted according to the change of the residual electrocardio energy and the electromyographic energy, the waveform and spectral characteristics of the electromyographic signals are kept while the electrocardio interference is effectively suppressed, and excessive suppression or introduction of adaptive noise is avoided.

The invention discloses a light storage and charging integrated charging pile and a shared charging model construction method, relates to the technical field of charging pile operation optimization, and aims to solve the problems of inaccurate data processing, one-sided priority evaluation, rigid resource constraint and the like of the existing shared charging technology. According to the method, on the basis of time-space associated data of an optical storage and charging system, a user and a power grid, a high-quality dynamic data set is output through processing of a double-exponential weighting algorithm; on the basis, user charging priorities are quantified through a three-dimensional coupling algorithm, and a priority sequence is formed; combining the data and the sequence to construct an energy storage capacity and power grid interaction power constraint model dynamically adjusted along with the priority; a multi-objective optimization function of a priority adaptive weight is constructed, a charging pile distribution and power scheduling scheme is solved by adopting an adaptive particle swarm algorithm, full-process cooperation of data processing, priority evaluation, resource constraint, optimization objective and dynamic response is realized, and high efficiency, fairness and stability of shared charging are improved.

The application provides a multi-card cooperative Internet of Things communication link stability control method and system, which comprises device initialization processing and link activation; Internet of Things device identity authentication; real-time collection of signal quality, transmission quality and cell load data of the current serving cell and adjacent cell, and exponential weighted smoothing processing; if the current smoothed link quality is less than the signal quality threshold value and lasts for a preset time length, it is determined that the link is deteriorated; a cooperative prediction mechanism based on the Internet of Things device, the edge end and the cloud end is used to predict the network blind area and signal deterioration trend, trigger link switching and perform multi-link dynamic bandwidth allocation, hash shunting, disorder rearrangement, packet self-healing or redundant transmission. The application can identify the link deterioration risk in advance; effectively avoid the service interruption caused by switching after the device enters the blind area or the signal is deeply deteriorated, improve the stability of the communication link; solve the problems of business jam, data packet loss and reconnection time consumption in cross-operator switching.

The invention provides an equilibrium strategy generation method, device and equipment for a large-scale unmanned cluster game, relates to the technical field of large-scale autonomous system decision control, and aims to solve the defects of poor stability and low calculation efficiency when an equilibrium strategy is generated for a large-scale unmanned cluster in the prior art. And efficient collaboration and autonomous decision-making of a large-scale cluster in a complex environment are realized. Comprising the following steps: dividing devices included in a device cluster into a plurality of populations based on characteristic parameters of each device included in the device cluster; constructing an average field game model of the plurality of populations, wherein the average field game model is used for reflecting an interaction effect among the plurality of populations; iteration is carried out based on a target mode and the average field game model, a balance control strategy of the equipment cluster is obtained, and the target mode comprises at least one of exponential weighted aggregation, near-regularization, bootstrap regularization and heuristic acceleration; and controlling the equipment cluster to execute the target task based on the balance control strategy.

The invention provides a real-time self-adaptive beam forming method and device based on the maximum SNR, and the method comprises the following steps: carrying out the windowing preprocessing of time domain audio signals collected by M microphones, and carrying out the FFT conversion, and obtaining an M * F-dimensional frequency domain signal matrix X; updating an M * M-dimensional covariance matrix Rnow (f) of each frequency point by adopting an exponential weighted moving average mechanism based on the frequency domain signal matrix X; the inverse matrix Rnow-1 (f) of the covariance matrix Rnow (f) is subjected to recursive updating by adopting a Sherman-Morrison-Woodbury matrix inversion lemma, and the inverse matrix Rnow-1 (f) of the covariance matrix Rnow (f) is subjected to recursive updating by adopting the Sherman-Morrison-Woodbury matrix inversion lemma; calculating the SNR of each microphone channel in parallel, screening an optimal reference microphone and extracting a corresponding beam forming weight, and for each frequency point f, performing beam forming processing on the frequency domain signal X (f) by applying the weight to obtain an output frequency spectrum Y (f); and constructing a gain compensation mechanism based on the optimal reference microphone, performing frequency-point-by-frequency gain compensation on the output spectrum Y (f), performing IFFT (Inverse Fast Fourier Transform) to restore the output spectrum Y (f) into a time domain signal, and outputting enhanced voice. Therefore, the technical support of current voice interaction is adapted.

The application discloses a fan yaw control method and system based on turbulence self-adaptation and historical optimal direction, and the method comprises the following steps: collecting operation data in real time; calculating turbulence intensity and generating a self-adaptive yaw triggering threshold; updating the historical optimal power coefficient, yaw direction and cumulative yaw angle by using exponential weighted filtering; when the wind vane fails, directly taking the historical optimal direction as the first trial direction; and executing yaw output after jointly judging the wind speed, change rate, unit state and power coefficient stability. The application adjusts the triggering threshold by the turbulence intensity, avoids invalid trial under high turbulence, improves the first trial correctness by more than 60% through the historical optimal state memory, supports independent and cooperative dual-mode output, and forms a hierarchical control structure with the existing closed-loop algorithm. The method significantly improves the yaw control robustness, convergence efficiency and unit power generation under the condition that the wind vane fails.

The invention relates to the technical field of artificial intelligence model optimization and edge deployment, and discloses a dynamic quantization-based AI model lightweight deployment method, which is characterized by comprising the following steps of: 1, dynamically partitioning input data of a complex scene according to semantic / target attributes, the target detection scene is divided into a target area and a non-target area of different scales, and the semantic segmentation scene is divided into a high-confidence-coefficient core area and a low-confidence-coefficient edge area; and step 2, calculating a real-time quantization threshold value for each data partition by adopting a sliding window and exponential weighted average, adapting a quantization bit width according to a model hierarchy type, adopting INT8 quantization for a non-key layer, adopting FP16 quantization for a key layer, and retaining 10%-15% of FP32 core parameters. According to the AI model lightweight deployment method based on dynamic quantization, in YOLOv8 (target detection), the mAP of a low-illumination scene is improved from 67.2% to 75.8%; in U-Net (semantic segmentation), the IoU in a rainy day scene is improved from 71.5% to 78.3%, and the key layer feature similarity loss is controlled within 5%.

This invention discloses a method and system for improving the success rate of wide-lane ambiguity fixation in BeiDou dual-frequency positioning, belonging to the field of high-precision positioning technology. It solves the problems of existing methods being unable to adapt to dynamic changes in observation quality, suffering from old information contamination, and difficulties in setting the smoothing window length. The method includes: constructing a MW combined observation model for BeiDou dual-frequency positioning; establishing a time-decay-based weight allocation function to recursively smooth the wide-lane ambiguity floating-point solution sequence; performing integer fixing and reliability verification based on the smoothed wide-lane ambiguity floating-point solution estimate; and dynamically adjusting the exponential weighting factor according to the measurement residual to achieve adaptive smoothing. This invention introduces a time-decaying weight function to give higher weight to recent observations and establishes a recursive update model. Simultaneously, it introduces a residual-driven dynamic decay factor adjustment mechanism, effectively improving the reliability and timeliness of wide-lane ambiguity fixation in BeiDou dual-frequency positioning.

The invention provides an unmanned aerial vehicle radiation comparison direction finding method based on time decay moving average, and the method comprises the steps: enabling a direction finding processor to receive unmanned aerial vehicle target information containing an unmanned aerial vehicle frequency point and a detection time interval Td, and carrying out the initialization of direction finding data for an unmanned aerial vehicle target; for the ith directional antenna, if an effective unmanned aerial vehicle wireless signal, namely an effective receiving signal, is received within a specified time interval Ts, the average receiving signal intensity of the directional antenna is updated by using an exponential weighted moving average algorithm according to the signal intensity of the current receiving signal; for the jth directional antenna, if effective unmanned aerial vehicle wireless signals, namely effective receiving signals, are not received within a specified time interval Ts, the average intensity of the receiving signals of the directional antenna is updated by using an exponential weighted moving average algorithm according to the electromagnetic bottom noise in the environment; meanwhile, the latest update time of the directional antenna is set as the current time; and the single received signal strength estimation error and the jitter along with time or environment are effectively inhibited.

The invention provides a powder flow precise control and real-time adjustment method for quantitative powder packaging, which comprises the following steps of: identifying dynamic model parameters of an actuating mechanism by utilizing a recursive augmented least square method through periodically injecting lightweight excitation signals and accurately collecting input and output time sequence data; a preset nonlinear rule mapping relation is combined to automatically convert into an optimal PID control parameter, and exponential weighted average is adopted to realize smooth transition of the parameter, so that system disturbance is avoided; according to the method, the characteristic change of an executing mechanism can be matched in real time, and the powder flow adjusting precision and the system robustness are improved.

This invention discloses an adaptive transmission method for trunked intercom based on wireless local area networks (WLANs), comprising the following steps: collecting link-state and service-state indicators to form a state vector; calculating the carrying window and delay budget using sliding window statistics and exponential weighting, generating a resource profile including congestion and risk levels; layering voice frames and labeling key frames with the latest delivery timestamp; selecting representative terminals for group clustering based on access point identifier, channel, and link quality; and generating transmission parameters by calling an improved BBRv3 under resource profile constraints in the unicast pipeline; prioritizing the transmission of key frame copies, aggregating completion summaries, determining majority completion, and retransmitting missing key frames with targeted compensation and updating completion; switching the distribution mode and performing gray-scale rollback when delay exceeds limits, busy-occupancy surges, or roaming occurs; recording indicator and parameter versions, establishing a template library through scenario fingerprint matching, and storing versioned data. This invention improves the reliability of key voice delivery and reduces the risk of delay jitter and packet loss under congestion and roaming conditions.

The invention discloses a heterogeneous SAR image high-precision connection point matching method. The method comprises the following steps: extracting object space connection points as region guidance according to object space SAR image data; determining an initial position and a guide area of an image space connection point based on an object-image conversion relation, searching a local optimal connection point, and extracting a supplementary feature point according to an exponential weighted mean ratio of an image texture structure; constructing SAR image feature descriptors according to a gradient statistical histogram and accurately matching the SAR image feature descriptors with image spaces; further eliminating wrong matching points through a similarity threshold value and a random sample consistency algorithm to obtain high-precision connection points; according to the invention, the problems in the prior art that the different-orbit SAR image connection point matching method has obvious defects in coping with the problems of image isomerism and object image conversion position deviation and is difficult to meet the application requirements of high-precision SAR images are solved.

This invention relates to a battery fault detection method based on quantile and exponential weighting adaptive normalization, belonging to the field of battery fault detection technology. The method includes the following steps: S1: Collecting electric vehicle operating data, including voltage data and time parameters of each cell in the battery pack, and preprocessing it; S2: Performing Kalman filtering on the cell voltage data; S3: Normalizing the filtered cell voltage data, including using quantiles to determine the normal value range, applying an exponential decay function to normal values, and applying an exponential growth function to outliers; S4: Calculating the DTW distance between each pair of normalized cell voltage time series and calculating the standard score of the distance; S5: Setting a fixed threshold and determining whether the standard score of the cell voltage exceeds the set threshold, thereby determining whether each cell battery has failed.