43 results about "Sigma point" patented technology

The invention provides a maneuvering target intelligent tracking method based on a self-attention mechanism and physical constraints, and the method comprises the steps: carrying out the coordinate transformation of original radar measurement data, obtaining a trace point under a Cartesian system, and calculating a Sigma point set of a target state vector of each time step of an initial window; for the n sampling points, Sigma points are calculated; performing normalization processing, and inputting the normalized Sigma point sequence into the prediction model to generate a prediction result; configuring physical constraints in the loss function, wherein the physical constraints comprise speed constraint loss and acceleration constraint loss; the prediction result is combined with an insensitive Kalman filter, and the real-time correction capability of the insensitive Kalman filter is utilized. And outputting the final estimated values of the position, the speed and the acceleration for a radar tracking system to display. Compared with an IMM interactive multi-model method, the accuracy of the three-dimensional maneuvering target tracking performance is improved, and compared with an existing maneuvering target intelligent tracking method, the prediction accuracy and efficiency are improved.

The embodiment of the invention provides an aircraft motion state determination method based on Lie group unscented Kalman filtering, and the method comprises the steps: building a state model of a target aircraft, and a measurement model of the target aircraft; obtaining a final Lie group state estimation value of the target aircraft through iterative calculation according to the state model and the measurement model; and determining the motion state of the target aircraft according to the final Lie group state estimation value. According to the technical scheme, a Lie group unscented Kalman filtering navigation framework based on external measurement information is provided, specific forms of sigma point error propagation on Lie algebra and state updating on the Lie group are given, and a self-adaptive updating method of a state noise covariance matrix and a measurement noise covariance matrix is designed. Therefore, the self-adaptive Lie group unscented Kalman filtering method is provided, and the multi-sensor multi-beacon target state estimation precision is greatly improved by means of the self-adaptive Lie group unscented Kalman filtering method.

The invention provides an attitude monitoring method and system in navigation of an aircraft, and the method comprises the steps: determining a process noise covariance matrix through a high-frequency energy component of an angular velocity signal measured by a current gyroscope in each filtering period; predicting a prior state vector and a prior covariance matrix at the current moment according to the state vector at the previous moment, the angular velocity measured by the current gyroscope and the process noise covariance matrix; generating an asymmetric Sigma point set and determining a measurement noise covariance matrix; the asymmetric Sigma point set is substituted into a measurement model, and a predicted measurement value and a predicted measurement covariance are obtained through unscented transformation; calculating Kalman gain based on the predicted measurement covariance and the measurement noise covariance matrix, and performing weighted gating test on innovation formed by a real measurement value and a predicted measurement value; and updating a state vector and a covariance matrix on the basis of the adjusted Kalman gain and information so as to obtain attitude information of the aircraft at the current moment.

The invention discloses a coordinate system transformation fusion filtering tracking method and system for a dual-base-station radar, and relates to the technical field of radar target tracking. The method comprises the following steps: firstly, establishing a state and observation equation, and determining an initial state of a target based on prior information under Cartesian coordinates; then, obtaining a target motion state through one-step prediction, generating a sigma point by utilizing U transformation, and calculating a mean value and a covariance for constructing state space prediction; thirdly, fusing the measurement data and the prediction data by using a Kalman filter to obtain optimal state estimation and a covariance matrix; and finally, converting the updated state estimation back to the Cartesian coordinate system based on U transformation, and circulating the process until the tracking is finished. According to the method, the nonlinear filtering problem in the updating process can be avoided, and the tracking robustness and precision are improved.

The invention discloses a spherical simplex robust unscented Kalman filter and a PWM (Pulse Width Modulation) rectifier control method and device, which are characterized in that (2n + 1) Sigma points in standard unscented Kalman filtering are replaced by (n + 2) simplex vertexes which are symmetrically distributed on a unit hyper-sphere, so that the calculation complexity is remarkably reduced. A covariance matrix and a state quantity are dynamically adjusted in combination with a corrected observation noise covariance matrix and an adaptive factor adjustment mechanism, so that the estimation precision of the filter on the lumped disturbance quantity is remarkably improved, and the robustness and the tracking performance of the system on non-Gaussian noise and abnormal observation values are enhanced. And estimating the lumped disturbance quantity of the system in real time through spherical simplex robust unscented Kalman filtering, introducing the lumped disturbance quantity into a cost function for optimization solution, and selecting an optimal voltage vector to control the on-off state of the rectifier. According to the invention, the total harmonic distortion rate of the output current and the voltage fluctuation of the direct-current bus are obviously reduced under the steady-state condition, and higher response speed and stronger anti-interference capability can be shown under the dynamic working condition.

The invention provides a condenser state estimation method based on adaptive noise square root Kalman filtering. Firstly, mathematical models of a condenser evaporation area, a hot well water area and a cooling pipe area are established according to the operation mechanism of a thermal power generating unit condenser; an unscented Kalman filtering algorithm is selected as a basic algorithm for state estimation of a condenser system, and then an optimal adaptive noise square root unscented Kalman filtering algorithm is constructed. The method comprises the steps of filtering initialization, Sigma point set generation, time updating, Sigma point set reconstruction, measurement updating, filtering updating, adaptive factor construction, cross covariance matrix / square root correction, fading forgetting factor calculation and noise covariance matrix correction. And finally, estimating the pressure state and the liquid level state of the condenser by using an optimal adaptive noise square root unscented Kalman filtering algorithm.

The invention discloses an IGBT life prediction method based on adaptive unscented particle filtering of time sequence form perception optimization, and belongs to the technical field of health prediction and reliability engineering. The method aims at the problems that a traditional filtering algorithm is prone to falling into local optimum in a high-dimensional parameter space and prediction trajectory forms are discontinuous, and comprises the steps that firstly, an IGBT nonlinear degradation model is established through a state-space equation; performing Sigma point sampling and nonlinear propagation at a particle level by adopting an unscented transformation and particle filtering cooperation mechanism, and fusing process noise to realize dynamic state prediction; constructing a Gaussian likelihood function by using the deviation between a predicted value and an actual observed value to generate a system residual sequence; an FCD fusion index is innovatively introduced, the index fuses a Pearson correlation coefficient and a Frechet distance to synchronously quantify trend consistency and form similarity, and self-adaptive genetic optimization is driven accordingly; and repeating the behaviors before a preset termination condition is met, and continuously updating the parameters of the state-space equation.

The invention discloses a system inertia and primary frequency modulation coefficient estimation method and device, equipment and a medium. A Sigma point set is generated according to posterior state estimation including inertia, primary frequency modulation coefficient and frequency deviation at the last moment and covariance of the posterior state estimation; a prediction state and statistics thereof are obtained through nonlinear state transfer function transformation; obtaining a predicted measurement value and a statistic thereof through measurement function transformation, and calculating a cross covariance of a state and measurement; the prediction state is updated by using the actual frequency deviation and the Kalman gain calculated based on the measurement covariance and the cross covariance, and the current optimal state estimation is obtained; and finally, inertia and a primary frequency modulation coefficient are extracted from state estimation. According to the method, real-time, joint and online accurate identification of the inertia and the primary frequency modulation coefficient is realized, the problems of linearization error and parameter coupling neglect of a traditional method are effectively solved, the robustness and accuracy of estimation are remarkably improved, and reliable state sensing support is provided for stable analysis of the frequency of a power system.

The invention discloses a wide-working-condition steam turbine system operation data correction method based on digital twinning, and relates to the technical field of steam turbine equipment digitalization, and the method comprises the steps: collecting operation data of a wide-working-condition steam turbine system in real time; constructing a digital twinborn model of the steam turbine system; performing prior covariance estimation on the operation data according to the statistical distribution condition and the design parameters; generating a Sigma point by adopting an unscented Kalman filtering algorithm, iteratively updating a state vector and a covariance in combination with a state transition equation and an observation equation, and outputting an updated known variable; performing smoothing processing on the updated known variable in a plurality of time point regions by adopting an RLOESS algorithm to obtain corrected operation data; according to the method, the real dynamic characteristics of the wide-working-condition steam turbine are approached by building the digital twinborn model, the calculation complexity of dynamic data fusion of a steam turbine system can be remarkably reduced, and the problems that the convergence speed is low and calculation is complex due to strong nonlinearity of a traditional method are solved.

The application discloses a power distribution network prediction auxiliary state estimation method and system based on an improved iteration process, and the method comprises the following steps: establishing a nonlinear discrete time state space model; generating a group of Sigma points through an unscented transformation, and substituting the Sigma points into a state transition equation to calculate a state prediction value and a state prediction error covariance matrix at a current time; substituting the Sigma points into a measurement equation to obtain a measurement prediction value, and calculating a Kalman gain; calculating a process noise covariance matrix at a next time; and correcting the state prediction value and the state prediction error covariance matrix at the current time to obtain an optimal state estimation value and an error covariance matrix thereof. The application avoids the risk of losing semi-positive definiteness of the covariance matrix in the iteration process from the algorithm mechanism, significantly enhances the numerical stability and robustness of the UKF algorithm, and ensures reliable application in high-dimensional complex power distribution network state estimation.

The application belongs to the technical field of radio positioning and motion control, and specifically discloses a kind of exoskeleton control filtering methods based on UWB and IMU fusion, comprising: deploying UWB base station, UWB label and IMU module;First displacement of knee joint in horizontal direction is measured by UWB label, first forward speed is calculated, second displacement of knee joint in horizontal direction is measured by IMU module, and second forward speed is calculated;State equation and observation equation are constructed, Sigma point is generated and untraceable Kalman filtering iteration is executed, and prediction result is optimized until the confidence of prediction result reaches a preset value.The application applies PDOA algorithm to solve angle, the angle and angular velocity measured by IMU module are fused by untraceable Kalman filtering, the motion direction error caused by single sensor is avoided, and the attitude estimation precision of exoskeleton control is improved.The application is suitable for exoskeleton control.

The invention discloses a mine robot tracking method for trust adaptive event-driven unscented Kalman fusion filtering. Firstly, an RSS time-varying response radius-based trust adaptive event-driven anchor point scheduling and anchor point data transmission triggering mechanism is designed to adapt to distribution change scheduling of trust anchor points around a mine robot to be close to a preset number of trust anchor points and dynamically trigger data transmission as required. Then, randomly and uniformly distributed noise covariance is introduced to describe the uncertainty of mine robot motion modeling, and through iterative uniform sampling Sigma point weighted average, the consistency of square root unscented Kalman filter information estimation is ensured; and finally, a credible anchor point separation mechanism of K-means dimensionality reduction clustering is constructed, a credible measurement updating mine robot motion state estimation adaptive weight fusion strategy is constructed, and the accuracy, stability and robustness of mine robot tracking are improved.

The invention relates to the technical field of robot positioning, discloses a four-foot robot three-dimensional positioning method and system, a storage medium and a robot, and aims to solve the technical problems of height drift and foot end error accumulation of a four-foot robot in a visual degradation environment without a GPS (Global Positioning System) by a traditional positioning method. The method comprises the following steps: firstly, constructing an initial state vector containing motion states of a fuselage and four foot ends, and establishing a nonlinear continuous time state equation considering foot end dynamics; relying on an unscented Kalman filtering framework, a prior state is obtained through Sigma point sampling and time updating; a foot end observation value is corrected through foot bottom rolling compensation, a height absolute reference is provided by combining a local invariant elevation map, and multi-leg slippage detection and observation weight self-adaptive adjustment are synchronously completed; and finally, the optimal state is updated and output through observation. According to the invention, the visual sensor is used for topographic mapping instead of direct positioning, the height drift is effectively inhibited, the robustness and precision of positioning in a complex environment are improved, and the method is suitable for unstructured scenes such as rescue exploration and the like.

The invention belongs to the technical field of radio positioning and motion control, and particularly discloses an exoskeleton control filtering method based on UWB and IMU fusion, and the method comprises the steps: deploying a UWB base station, a UWB tag, and an IMU module; the first displacement of the knee joint in the horizontal direction is measured through the UWB tag, the first forward speed is calculated, the second displacement of the knee joint in the horizontal direction is measured through the IMU module, and the second forward speed is calculated; and constructing a state equation and an observation equation, generating Sigma points, executing unscented Kalman filtering iteration, and optimizing a prediction result until the confidence coefficient of the prediction result reaches a preset value. According to the invention, the angle is calculated by using the PDOA algorithm, and the angle and the angular velocity measured by the IMU module are fused through the unscented Kalman filtering, so that the motion direction error caused by a single sensor is avoided, and the attitude estimation precision of exoskeleton control is improved. The method is suitable for exoskeleton control.

The invention belongs to the technical field of rigid body posture estimation, and discloses a rigid body posture estimation method, system and device based on generalized correlation entropy geometric filtering, and the method comprises the steps: obtaining target motion information, and building a discrete nonlinear kinematics model; the discrete nonlinear kinematics model is initialized; sigma points of a posterior state covariance matrix and a process noise covariance matrix are generated respectively, and the sigma points are propagated and updated through manifold fast unscented transformation; obtaining a prediction state mean value, a priori state covariance matrix and a square root of the priori state covariance matrix; updating the central parameters based on the information of the historical moments; calculating a pseudo measurement matrix and updating the gain; correcting the state estimation by using the gain; judging whether convergence occurs or not, and if not, continuing iteration; otherwise, outputting a posterior state estimation value as a final rigid body posture estimation result at the moment. According to the method, the problem that the robustness and precision of rigid body posture estimation in a non-zero mean value and non-Gaussian noise environment are reduced is effectively solved.

The invention relates to the technical field of battery state prediction, in particular to a battery SOC and SOH prediction method and system based on multi-technology fusion. Comprising the following steps: S1, collecting real-time data of a battery, including numerical values of voltage, current and temperature; s2, integrating and preprocessing the data, wherein the preprocessing operation comprises denoising and standardization; s3, voltage discrimination: extracting key features from the voltage data, wherein the key features comprise a voltage value, a voltage derivative and a statistical feature; s4, physical constraints are embedded through PINNs, modeling is carried out on battery voltage behaviors, and predicted voltage is calculated; performing feature regression through XGBoost, predicting an intermediate variable, and constructing a single voltage prediction model through ensemble learning to calculate the voltage of a single battery; s5, unscented Kalman filtering is applied, the nonlinear relation is processed through Sigma points, and prediction values of SOC and SOH states are optimized and obtained; compared with the prior art, the prediction precision is improved, the robustness is enhanced, the interpretability is improved, real-time prediction is realized, and the real-time performance of the system is ensured.

The present invention discloses a fast sensor data fusion method based on sigma point belief propagation, comprising the following steps: step (1), predicting the state and covariance of target k at time t; step (2), performing sigma point sampling on the predicted state; step (3), transferring sigma points on the sampled samples in step (2); step (4), calculating iterative data association within all sensors s in parallel; step (5), performing confidence calculation after completing the iterative data association within the sensor; step (6), estimating the state and covariance of each target k at time t+1 based on the distribution of the state of target k at time t+1. The method solves the problems of high computational complexity and poor scalability of traditional multi-sensor information fusion algorithms, and reduces computational complexity based on the implementation of the sigma point belief propagation algorithm.

The invention discloses a DTW-UKF collaborative attitude estimation method for multi-source noise, and the method comprises the steps: 1, carrying out the preprocessing of IMU data, and constructing a static time sequence and a dynamic time sequence based on the preprocessed data; 2, calculating an alignment path of the dynamic time sequence and the static time sequence through a multi-dimensional DTW algorithm, and calculating a time local difference and an inter-axis difference degree; 3, adaptively adjusting the noise covariance of the UKF; 4, defining a state variable of the UKF, generating a Sigma point set, and predicting the state variable and a covariance thereof; 5, updating posteriori estimation and covariance of the UKF by combining observation data; and repeating the steps 2-5 to obtain attitude estimation at all moments. According to the method, the analysis capability of the DTW algorithm on the multi-dimensional time sequence is utilized, the influence of dynamic disturbance is overcome, the attitude estimation precision is improved, and a high-precision and high-robustness method is provided for motion attitude estimation.

The invention discloses a power distribution network prediction auxiliary state estimation method and system based on iteration process improvement. The method comprises the following steps: establishing a nonlinear discrete time state space model; generating a group of Sigma points through unscented transformation, substituting the Sigma points into a state transition equation, and calculating a state prediction value and a state prediction error covariance matrix at the current moment; substituting the Sigma point into a measurement equation to obtain a measurement predicted value, and calculating a Kalman gain; calculating a process noise covariance matrix at the next moment; and correcting the state prediction value and the state prediction error covariance matrix at the current moment to obtain an optimal state estimation value and an error covariance matrix thereof. According to the method, the risk that the covariance matrix loses the semi-positive qualitative property in the iteration process is avoided from the algorithm mechanism, the numerical stability and robustness of the UKF algorithm are remarkably enhanced, and reliable application in high-dimensional complex power distribution network state estimation is ensured.

Techniques for determining a collision probability between a candidate trajectory associated with a vehicle and an object (e.g., a vehicle or pedestrian) in the vehicle's environment. In some cases, the techniques include selecting a set of sampled states, where each sampled state represents a predicted state (e.g., a predicted position and / or orientation) of the object at a future time. The set of sampled states may then be used to determine the collision probability. In some cases, the set of sampled states associated with a future time t may be determined based on: (i) a probability distribution (e.g., a Gaussian distribution) associated with the predicted object state at the future time t, and / or (ii) a covariance of the probability distribution. In some cases, the set of sampled states selected based on a distribution includes a set of sigma points each associated with a probability.

Computer-implemented method for training a machine learning system. The machine learning system is configured to accept a sensor signal as input for anomaly detection and / or for sampling a trajectory of a traffic participant and / or for sampling of sensor signals and / or for determining a value characterizing a likelihood of a sensor signal with respect to a training dataset. The training includes: determining, by an encoder of the machine learning system and based on a training sensor signal, a first intermediate representation characterizing a mean of a latent distribution of a latent space and a second intermediate representation characterizing a variance and / or covariance of the latent distribution; determining, based on the first intermediate representation and the second intermediate representation, a plurality of sigma points with respect to the latent distribution; determining an output signal.

The invention relates to the technical field of hydrogen-electricity coupling, and provides a hydrogen-electricity coupling micro-grid energy management model adaptive updating method, which comprises the steps of generating a Sigma point based on unscented transformation, predicting a state and an observation value through a state and observation equation, and calculating statistical characteristics of the state and the observation value; according to the method, state-observation cross covariance and Kalman gain are further calculated, weighted fusion correction is carried out on state prior estimation by using actual observation residual, noise is effectively suppressed, nonlinear dynamics are accurately processed, and high-credibility state estimation is provided for energy management decision. A feature space is constructed based on real-time state estimation feedback information, fusion features are extracted through incremental principal component analysis and input into an energy management model to calculate a loss function, and a learning rate is adaptively adjusted by calculating first-order and second-order moments of gradient information in parameter vector loop iteration. The model parameters are updated according to the learning rate and the gradient of the loss function, and the problem that the model applicability in a traditional model updating mode is reduced is solved.

Approaches presented herein provide for the support of distorted cameras in 3D scene reconstruction. Objects in a scene can be represented by 3D Gaussian particles. To determine which 3D Gaussian particles contribute to individual pixels of an image to be rendered, an unscented transform-based approach can be used to project representative sigma points for the 3D Gaussian particles onto a 2D camera plane. The 3D Gaussian particles determined to potentially contribute to a given pixel can then have rays traced to determine a segment of intersection of the ray across a 3D Gaussian particle, and a value corresponding to the point of maximum response across that segment can be returned as a contribution value. The various contribution values for each pixel can then be blended to provide an output color value.

The invention discloses a self-adaptive UKF fusion positioning method for an outdoor cleaning robot. The method comprises the following steps: acquiring inertial measurement unit IMU data, point cloud data and RTK positioning data through a sensor; according to the system state of the robot, unscented Kalman filtering (UKF) is adopted for unscented transformation, IMU data serve as control input, a predicted sigma point set is obtained, and sigma points serving as the system state are updated; evaluating the quality of the point cloud data; according to the point cloud data quality evaluation and the RTK positioning data, a current positioning mode is adaptively selected through a state machine, and the positioning mode comprises a first mode with the point cloud data quality as the main mode, a second mode with the RTK positioning data as the main mode and a third mode fusing the point cloud data quality and the RTK positioning data; and according to a corresponding mode, observation updating and time-sharing dimension multiplexing are carried out based on unscented Kalman filter (UKF).

The application discloses a coordinate system transformation fusion filtering tracking method and system for a dual-base station radar, and relates to the technical field of radar target tracking. The method comprises the following steps: firstly, establishing state and observation equations, and determining an initial state of a target based on prior information in a Cartesian coordinate system. Then, a target motion state is obtained through one-step prediction, sigma points are generated by using U transformation, and a mean value and a covariance of state space prediction are calculated and constructed. Next, measurement data and prediction data are fused by using a Kalman filter to obtain optimal state estimation and a covariance matrix. Finally, the updated state estimation is converted back to the Cartesian coordinate system based on U transformation, and the process is repeated until the tracking is completed. The application can avoid the nonlinear filtering problem in the updating process, and improves the robustness and precision of tracking.

The invention discloses a battery SOC prediction method and system based on an improved particle filter algorithm, and belongs to the field of battery SOC prediction. The method comprises the following steps: acquiring voltage and current of a battery; constructing a battery equivalent model according to the voltage and the current of the battery; identifying each model parameter in the battery equivalent model; generating N particles in the state space, wherein the particles form a point set; determining an initial value of each particle, wherein the initial value comprises an initial SOC value represented by the particle; sOC prediction is carried out through an improved particle filtering algorithm, the battery SOC can be accurately calculated through the algorithm through the steps of generating an initial particle swarm, spreading Sigma points, calculating a prediction mean value and covariance, updating particle weights, carrying out self-adaptive resampling and the like, high precision and stability can be kept in a complex environment, and meanwhile, the SOC prediction accuracy is improved. The system quickly responds to and alarms battery loop abnormity, and safe operation of the system is ensured.

The invention relates to the technical field of attitude control, in particular to a propulsion control method, system and device applied to an underwater robot. According to the method, a sigma point of an unscented Kalman filtering process is obtained through an environment system state vector of a robot, and an adjustment limit vector of a sampling moment is obtained according to numerical fluctuation of different dimensions; a confusion degree evaluation index is obtained according to the unstable condition of continuous change of the adjustment limit vector; and performing self-adaptive regulation and control on the switching gain through the deviation condition of the current and initial limiting ranges and chaos degree evaluation, and outputting a control signal again in combination with the adjustment of the sampling frequency by the chaos degree evaluation. According to the method, in the process of adjusting the switching gain and the sampling frequency through the continuous environment interference chaos condition, the gain is improved, so that the system is rapidly converged, meanwhile, the limitation adjustment is considered, the jitter condition of control output is reduced, and the adjustment control of the attitude of the propulsion action of the AUV at each time is more accurate and efficient.

The application discloses a kind of based on SVD-ACUKF online model updating hybrid test method, belongs to the technical field of hybrid test.To solve the problem of insufficient parameter identification accuracy, inaccurate test substructure loading boundary, insufficient model updating accuracy and test error in online model updating hybrid test method, the initial value of the parameter to be identified is first determined, the displacement response of the overall structure is obtained based on ground motion, and then the displacement response of the physical substructure is obtained and tested to be loaded, to obtain the counterforce and displacement of the physical substructure, based on the state of the physical substructure at step k-1, 2n+1 Sigma points are generated, and they are sent to the numerical model of the equivalent physical substructure together with the displacement, 2n+1 times of nonlinear static analysis are completed to obtain the restoring force, and then the constitutive model parameters are obtained by using the counterforce and displacement of the physical substructure and the state estimation value of the equivalent physical substructure at step k-1, and then the finite element refined numerical model of the overall structure is updated, and the above process is repeated until the test is completed.

The invention discloses a building structure load identification method and equipment. The method comprises the steps of determining an initial state quantity and an initial state quantity covariance of a building structure; based on the UT transformation principle of an unscented Kalman filter, generating a group of Sigma points according to the state quantity and covariance at the previous moment, and writing the Sigma points into a matrix form; calculating information of each column of Sigma points in the matrix on the basis of the Alembert principle, the structural quality and the matrix information to obtain a plurality of accelerated speed predicted values with the number equal to that of the columns of the matrix; calculating a weighted average value of all acceleration predicted values based on a weight value of UT (Unscented Kalman Filter); constructing a target function about the unknown load of the structure based on the acceleration measurement value measured by the sensor and the weighted average value; based on a Levenberg-Marquardt optimization method, calculating a corresponding load estimation value when the target function obtains a minimum value; updating the state quantity and the state quantity covariance of the unscented Kalman filter based on the load estimation value; and circularly executing the steps until all time steps are calculated.

The invention discloses an SOC (State of Charge) estimation method of deep fusion neural network unscented Kalman filtering, relates to the technical field of all-vanadium redox flow batteries, and aims to realize high-precision estimation of the state of charge of the battery. The method comprises the following steps: establishing a second-order equivalent circuit model of a battery, and performing parameter identification on the second-order equivalent circuit model; obtaining a group of sigma points through unscented transformation, calculating propagation state estimation and propagation state prediction of each sigma point, obtaining a first-order statistical moment of prior state estimation, a first-order statistical moment of prior state prediction and a cross covariance through unscented transformation, and obtaining an observation difference and a state updating difference; and inputting the observation difference and the state updating difference into a neural network system to obtain a Kalman gain, and performing state updating on the priori state estimation through the Kalman gain to obtain prediction of the state of charge of the battery.