157 results about "Square error" patented technology

The invention belongs to the field of cable eccentricity detection, and particularly relates to a cable eccentricity detection method and system based on X rays. The method comprises the following steps: calculating a gradient magnitude diagram and a gradient direction diagram through an X-ray image of a cable, screening a point with the local maximum gradient magnitude and low neighborhood divergence as a contour starting point, performing contour tracking to generate a contour point set, and selecting a next contour point based on a tangential prediction direction; performing ellipse fitting on the contour point set to obtain a geometric center, long and short axis parameters and a root-mean-square error of a fitting ellipse, dividing the fitting ellipse into an inner candidate ellipse and an outer candidate ellipse according to a long axis, and matching the inner candidate ellipse meeting the condition for each outer candidate ellipse; and screening out an outer candidate ellipse and an inner candidate ellipse which meet conditions from the candidate pairs, and calculating the eccentricity of the cable to be measured based on geometric center coordinates of the outer candidate ellipse and the inner candidate ellipse. According to the invention, the accuracy and reliability of cable eccentricity measurement results can be improved.

The invention discloses a fog boundary layer parameterization scheme correction method based on a multi-scale physical coupling network, relates to the crossing field of numerical weather forecast and artificial intelligence, and aims to improve the performance of different boundary layer parameterization schemes in a WRF mode, construct a fusion framework of a physical mode and deep learning, adopt a double-branch heterogeneous network structure, and improve the performance of a fog boundary layer parameterization scheme. Establishing a nonlinear mapping relation between different parameterization scheme deviations and large eddy simulation turbulence characteristics by combining a space-time attention mechanism; by correcting systematic deviation existing in a parameterization scheme, the analysis capability of a turbulence structure is improved, and physical description of a boundary layer process is optimized; compared with a traditional boundary layer scheme, the method has the advantages that the fog zone simulation precision is remarkably improved, the fog zone visibility, the liquid water content and the root-mean-square error of a liquid water path are reduced by 66.7%, 50% and 58.3% respectively, the problem that turbulence intermittent characterization is insufficient under a stable boundary layer in a traditional method is effectively solved, and a new normal form is provided for refined forecasting of the fog generation and elimination process.

The invention discloses an improved vibration compensation method suitable for an atom interference gravimeter. The method comprises the following steps: acquiring an atom interference signal and a vibration voltage signal output by a seismometer; initializing population parameters of the improved whale optimization algorithm, wherein the population parameters comprise a search space of a population size, a maximum number of iterations, a gain coefficient K and a delay coefficient tau; the method comprises the following steps: generating an initial whale population by adopting a method of combining Logistic chaotic mapping and Latin hypercube sampling LHS; calculating a phase shift caused by vibration noise and an atomic interference phase delta phi, and calculating the fitness of the whale individual by taking a root-mean-square error of compensated interference fringes as a target function; iteratively updating the population position through an improved whale optimization algorithm, outputting an optimal gain coefficient and an optimal delay coefficient, and obtaining an atomic interference signal after vibration compensation; according to the atomic interference gravimeter vibration compensation method, the global search capability, the local development precision and the stability of a vibration compensation algorithm are improved.

The invention discloses a determination method, an application method and a related system of a grain yield prediction model, and relates to the field of grain yield prediction, and the determination method comprises the steps: obtaining time-space dynamic data; preprocessing and normalizing climate data, management data and soil data in the spatial-temporal dynamic data to obtain a spatial-temporal data set; respectively inputting the spatio-temporal data set into a plurality of machine learning models to obtain outputs of the plurality of machine learning models; respectively calculating a root-mean-square error and a decision coefficient of each machine learning model based on grain annual output data corresponding to the spatio-temporal data set and the output of each machine learning model; abandoning the machine learning model with the determination coefficient smaller than a preset value to obtain a screened machine learning model; and based on the screened machine learning model and the corresponding root-mean-square error, constructing a weighted average grain yield prediction model. According to the invention, a scientific basis can be provided for grain production under different climate scenes and management strategies.

The invention discloses a soil water content inversion method based on improved combination roughness, and relates to the field of remote sensing, and the method comprises the steps: synchronously obtaining a Sentinel-1 radar image and a Sentinel-2 optical image, and extracting different polarization backscattering coefficients, local incident angles and normalized water body indexes through preprocessing; the method comprises the following steps: generating a bare soil simulation backscattering coefficient data set, removing vegetation scattering contribution by using a water cloud model, obtaining a real bare soil backscattering coefficient, constructing a training set and a verification set containing actually measured soil water content, constructing a lookup table based on double models, and calculating the water content of the bare soil by minimizing a root-mean-square error between simulation and the real backscattering coefficient. Global search is carried out in a preset parameter space to determine an optimal earth surface root mean square height and a correlation length, a novel polynomial combination roughness is constructed, a physical correlation between the roughness and a backscattering coefficient is established, a dual-polarization empirical equation set is constructed, and simultaneous solution is carried out after parameters are optimized according to a criterion; according to the method, a more reasonable inversion result of the soil water content in a large range can be obtained.

The invention discloses a battery capacity prediction and state evaluation method and system based on multi-model collaborative learning, and belongs to the technical field of battery management. The method comprises the following steps: constructing a database containing multiple lithium ion battery long-term cycle data, and classifying according to a capacity attenuation trend; cleaning and preprocessing short-term cycle data of the to-be-tested battery; matching the to-be-tested data with the long-term attenuation trend in the database by using a clustering algorithm, and determining an optimal matching trend; distributing weights for the data in the matching trend by adopting a correlation algorithm, and generating initial capacity attenuation prediction; performing sequence correction on the preliminary prediction in combination with meta-learning and a related model, and generating a smooth future attenuation trend conforming to a physical law; and outputting a capacity prediction and health state evaluation result, and evaluating the prediction precision through a root-mean-square error and an average absolute percentage error. The method significantly improves the precision and generalization ability of long-term capacity prediction, and is suitable for various scenes such as electric vehicles, energy storage systems, consumer electronics and the like.

The invention relates to a hydraulic pump motor degradation evaluation method and system based on an LSTM neural network and similarity calculation. The hydraulic pump motor degradation evaluation method comprises the steps that key indexes representing hydraulic pump motor degradation and input parameters influencing hydraulic pump motor working conditions are extracted from test data in different working conditions; selecting and dividing the obtained original test data into a training data set and a test data set; performing normalization and data screening on samples in the training data set to obtain data enhancement samples for model training; samples in the test data set are only subjected to normalization processing and are used for model testing; setting model parameters based on data features, obtaining a volumetric efficiency calculation result, and calculating a root-mean-square error; determining operation conditions of the hydraulic pump motor in the test data, and analyzing change trends of degradation index parameters of the hydraulic pump motor in different time periods; the similarity of the degradation index parameters of the hydraulic pump motor in different time periods is calculated, and the degradation degree is obtained; and respectively multiplying a prediction result of the neural network and a similarity calculation result by a weight coefficient of 0.5 to realize degradation evaluation of the hydraulic pump motor.

The invention relates to the technical field of intelligent agriculture and computer vision, in particular to a small-sample high-density chicken counting framework based on a deep learning Mamba structure, which comprises a feature extraction network, a support-query enhancement module and a decoder. According to the method, a multi-scale feature extraction network based on a residual block (ResNet Block) is introduced, so that local detail information is effectively reserved; and then, a support-query enhancement module is constructed by introducing a Mamba structure, and context interaction between support features and query features is effectively enhanced by utilizing the long sequence modeling capability of linear complexity of the support-query enhancement module, so that the problems of individual overlapping and boundary fuzziness in a high-density chicken flock scene are solved. Experimental results on a PoultryCount real breeding data set show that the mean absolute error (MAE) and the root-mean-square error (RMSE) of the method are reduced compared with those of an existing method, and the chicken counting precision and generalization ability under the condition of a small number of labeled samples are remarkably improved.

The invention belongs to the technical field of photoelectric stabilized platform intelligent control, and particularly relates to a photoelectric stabilized platform composite anti-interference method based on improved active-disturbance-rejection control, which realizes multi-band disturbance cooperative suppression through an improved extended state observer (ESO) and non-linear state error feedback (NLSEF) core architecture. For the problems of phase distortion and parameter sensitivity under traditional ADRC high-frequency disturbance, an ALMRBF-ESO observer is constructed by adopting a dynamic regularization constrained radial basis function neural network and an adaptive damping adjustment mechanism, gradient dispersion limitation of a traditional RBF in a high-frequency domain is broken through, and disturbance root-mean-square errors are reduced; by reconstructing a sliding mode gain equation of nonlinear tracking error feedback and implanting a parameter self-correction mechanism, dynamic response of a system is accelerated, overshoot is reduced at the same time, and the parameter drift rate of a controller is stabilized at 0.12. According to the method, on the basis of completely reserving ADRC robustness, the problem of collaborative optimization of ESO broadband disturbance observation precision and NLSEF dynamic tracking performance is solved.

The invention relates to an indoor cross-scene multi-band sub-terahertz channel prediction method, a storage medium and software, and belongs to the technical field of wireless communication. In order to solve the problems of poor adaptability to nonlinear features, high calculation complexity and insufficient generalization ability in traditional channel modeling, a channel prediction model is constructed based on a back propagation neural network, and a particle swarm optimization algorithm is utilized to optimize a network initial weight and a threshold value. According to the method, system parameters and environment characteristics are fused, an indoor sub-terahertz channel simulation data set is constructed, and the indoor sub-terahertz channel simulation data set is subjected to normalization processing and then used for model training and verification. And finally, evaluating the prediction performance according to indexes such as a root-mean-square error, a mean absolute error and a decision coefficient. The method can improve the accuracy and generalization ability of channel characteristic prediction, is suitable for various indoor scenes, and provides technical support for 6G communication system deployment.

The invention provides an abrasive flow polishing device and method for gear tooth surface homogenization. According to the method, tooth-shaped drainage pieces with the same shape as the gear are installed at the two ends of the upper portion and the lower portion of the gear, drainage section design is introduced, an abrasive flow medium path runner model is constructed, main parameters influencing an experiment are sampled and grouped through a Latin hypercube sampling method, abrasive flow polishing simulation is conducted in a grouped mode, and a Gaussian regression process is utilized to obtain an abrasive flow polishing simulation result. Selecting the minimum root-mean-square error of each group of PiVi data, determining the group of experimental parameters, obtaining a group of abrasive flow machining process parameters of optimal inlet pressure, back pressure and drainage section length, and finally guiding the experiment; according to the method, the influence of the inlet effect is relieved, the abrasive flow medium path runner model is constructed, a set of optimal abrasive flow machining process parameter combination is obtained, and the method is suitable for spur gears and bevel gears and high in universality.

The invention discloses a pressure vessel lower chamber model order reduction method combining sparse grids and POD, and the method comprises the steps: adding one to the grade of the sparse grids after iteration initialization, and obtaining test grid points according to important grid points; executing the lower chamber full-order computational fluid mechanics model corresponding to the test grid points to obtain a first snapshot; reconstructing a second snapshot by using the mapping relation and the POD modal matrix; calculating a root-mean-square error of the two snapshots, and if the root-mean-square error is smaller than a threshold value, completing construction of a reduced-order model Otherwise, the grid points with the errors meeting the requirements are stored in a non-important grid point set, the grid points which do not meet the requirements are marked as important grid points, and the snapshot matrix is updated. Performing singular value decomposition on the snapshot matrix, updating the mapping relation, judging whether the maximum sparse grid grade is reached or not, and continuing iteration if the maximum sparse grid grade is not reached. According to the method, the field distribution of the thermal hydraulic parameters in the lower chamber can be quickly obtained through matrix algebraic operation, and the method can be used for application scenes such as parameterization analysis, design optimization and digital twinning.

The invention discloses a Cuckoo search algorithm-based grid-connected inverter multi-objective optimization control method, which comprises the following steps of: constructing a three-level grid-connected inverter mathematical model, and calculating to obtain an inverter side current reference value; constructing a current prediction model of the inverter side; constructing a value function control item of the inverter side current by adopting a value function in a square error form; constructing a cost function by taking the neutral-point potential balance and the switching frequency as control items; and iteratively calculating a weight factor by adopting an improved self-adaptive cuckoo retrieval algorithm to realize multi-objective optimization control of the grid-connected inverter. According to the multi-objective optimization control method for the grid-connected inverter based on the cuckoo search algorithm, the problem that the control performance is limited due to the fact that weight factors are difficult to set in multi-objective optimization of the grid-connected inverter in the prior art is solved.

The invention discloses a J-A model parameter identification method, system, equipment and medium based on RBF and an improved brownish bear algorithm, and belongs to the technical field of power system optimization, and the method comprises the steps: building a Jiles-Atherton hysteresis reverse model of a current transformer, determining a to-be-identified parameter vector, and building a model with a root-mean-square error between actually measured magnetic field intensity and simulated magnetic field intensity as a target function, training a radial basis function neural network model, expanding data through linear interpolation processing, obtaining a predicted magnetic induction intensity value, inputting an objective function and radial basis function prediction data into an improved brownish bear optimization algorithm, and iteratively optimizing model parameters through hierarchical population position updating and fitness evaluation until convergence conditions are met. And outputting an optimal parameter identification result. According to the method, high-precision and high-efficiency identification of hysteresis model parameters is realized, the generalization capability and robustness of the system are improved, and reliable technical support is provided for hysteresis characteristic analysis of a complex physical system.

The invention provides an SAR image vegetation coverage inversion method and system based on multi-dimensional feature optimization and model optimization, a storage medium and electronic equipment, and the method comprises the following steps: carrying out the preprocessing of a Sentinel-2 image, calculating a normalized difference vegetation index (NDVI) through a wave band calculation formula, and carrying out the calculation of the NDVI; a reference vegetation coverage (FVC) of a research area is obtained by combining a pixel bipartite model (DPM), and then an optimal SAR feature combination is selected by applying an improved genetic algorithm ET-GA, using XGBoost as an adaptability evaluation function and using a root-mean-square error (RMSE) as an evaluation index. According to the method, SAR features are extracted through multiple feature decomposition methods, multi-dimensional feature optimization, intelligent algorithm optimization and a deep learning model are combined, the core problems of feature redundancy, insufficient precision, low efficiency and the like of SAR data in vegetation parameter inversion are solved, and an innovative technical scheme is provided for the field of remote sensing quantitative inversion.

Co-channel beacon transmissions are provided with at least one of spectral redundancy and temporal redundancy. A receiver produces a snapshot of a superposition of received co-channel beacon transmissions. Subcarrier demodulation, code nulling, or a Class-C linear minimum-mean-square error (MMSE) operation separates multiples ones of the co-channel beacon transmissions or eliminates inter-symbol interference and inter-subcarrier interference in the snapshot. Receiver operations can be performed at a network user, a network node, or a network operations center.

The invention discloses a hierarchical constraint ionosphere chromatography method based on hybrid interpolation and an adaptive relaxation factor. The method comprises the following steps: firstly, determining an inversion area, setting a virtual observation station in the inversion area, generating virtual STEC data of the virtual observation station through a hybrid interpolation method, and integrating actual STEC data and the virtual STEC data to obtain observation STEC data; calculating the intercept of each ray in the voxel according to the actual observation value in the inversion area and the coordinates of the virtual observation station, and constructing a coefficient matrix; obtaining an electron density initial value through an IRI-2016 model, and then iterating the electron density value by using the coefficient matrix and observation STEC data based on a multiplication algebraic reconstruction algorithm; adding constraints in the horizontal direction and applying constraints in the vertical direction; and taking the constrained electron density as an electron density initial value of a new round of iteration until the root-mean-square error is smaller than a preset threshold value.

The invention discloses a high-voltage cable on-line monitoring method and device, and relates to the technical field of on-line monitoring. The method comprises the following steps: after a high-voltage cable insulating layer is extruded, acquiring and processing electric field intensity data to obtain processed data; constructing an ideal electric field distribution reference model, comparing the processed data to calculate an electric field matching root-mean-square error, and if the error exceeds a first threshold value, performing early warning and marking an insulation abnormal area; detecting after the shielding layer is formed to obtain excitation end input power and receiving end output power, calculating electromagnetic signal power attenuation, and marking a shielding layer defect area if the electromagnetic signal power attenuation exceeds a second threshold value; calculating a spatial cross validation index according to the error and the attenuation amount, and marking a cross defect area if the spatial cross validation index exceeds a third threshold value; and constructing a defect evolution trend probabilistic model, inputting a real-time error, an attenuation amount and a cross validation index, outputting a defect deterioration posterior probability, and carrying out graded early warning. According to the method, defects of all links of production are covered through full-process real-time monitoring, and the limitation that traditional sampling detection cannot cover production batches is changed.

The invention provides a multi-modal optimal noise scheduling-based molecular generation method and device, and the method comprises the steps: representing any one joint noise scheduling function in a predefined joint noise scheduling function space as a preset target form through a time re-parameterization function, and building a unified parameterization frame of continuous modal noise and discrete modal noise; minimizing a pre-constructed generalized loss function through a pre-trained model, and training based on the generalized loss function to obtain a generation model for evaluating any joint noise scheduling function in the joint noise scheduling function space; based on an L2 norm square error evaluation result of each point of the discretized time grid by the generative model, searching and determining a minimum cumulative cost path from a starting point to an end point through a dynamic programming algorithm; and determining optimal joint noise scheduling according to the minimum cumulative cost path, and generating a drug molecular structure based on the optimal joint noise scheduling. Through the method provided by the invention, the geometric effectiveness and the calculation efficiency of molecule generation are improved.

Disclosed in the present invention are a fault diagnosis method and system for blade icing and blade mass imbalance of wind turbines. The method comprises the steps of: acquiring operation data and blade icing information of a plurality of wind turbines; labeling the operation data with an icing state label on the basis of the blade icing information, so as to obtain fault data; extracting fault features in the fault data, and ranking the fault features according to the degree of importance, so as to generate an optimal feature set; on the basis of a criterion of minimizing a squared error, selecting optimal features in the optimal feature set to generate an optimal decision tree; and performing classification on the basis of the optimal decision tree, so as to obtain a diagnosis test result including fault information. The present invention has the advantages of a high level of diagnostic accuracy, etc.

The invention discloses an underwater sound double-expansion channel parameter estimation method based on Newton orthogonal matching pursuit. The method comprises the following steps: firstly, realizing frame synchronization by using a cross-correlation peak of an LFM signal and a received signal; a coarse estimation value of a Doppler coefficient is obtained on a discrete time frequency grid through an ambiguity function method by utilizing the self-correlation characteristic of m-sequence pilot frequency, multipath time delay of a channel is detected by utilizing a matched filtering peak value of a signal after frequency offset compensation, and a coarse estimation result of the channel is obtained; executing Newton iteration optimization in a continuous parameter space by adopting a Newton orthogonal matching pursuit algorithm to realize super-resolution off-grid estimation of the Doppler coefficient of the first data block; and finally, proposing an iterative block tracking strategy, initializing the compensation of the rear block through the estimated value of the front block, and updating the time-varying Doppler coefficient by using the balanced signal feedback to realize the dynamic tracking of the time-varying Doppler coefficient. According to the invention, the Doppler estimation root-mean-square error is obviously reduced, and the calculation efficiency is high.

The invention belongs to the technical field of industrial automation and computer vision, and discloses a grabbing point pose calculation method based on a target detection model and application of the grabbing point pose calculation method. The grabbing point pose calculation method comprises the following steps: step 100, according to a target center area obtained by processing a grayscale image of a to-be-grabbed object through a target detection model, extracting a corresponding sub-area point cloud from a three-dimensional point cloud of the to-be-grabbed object; step 200, performing rough matching on the sub-region point cloud and the template point cloud through the FPFH feature vector and the RANSAC to obtain a rough matching result; step 300, performing fine matching on the subarea point cloud and the template point cloud through the rough matching result and the GICP to obtain a fine registration result; step 400, judging whether a precise registration result meets a preset requirement or not according to the interior point root-mean-square error and the overlap ratio, if so, entering the next step, and otherwise, terminating the process; and 500, according to the fine registration result and coordinate transformation, the grabbing point pose of the to-be-grabbed object is obtained through calculation. The process is simple in steps and easy to implement and control.

The invention discloses a geophysical prospecting signal denoising method based on VMD (variational mode decomposition) combined with an improved wavelet threshold function, and belongs to the technical field of mineral exploration geophysical prospecting signal process.The method includes the steps that firstly, a decomposition mode number K of a signal is determined through VMD in a self-adaptive mode, a plurality of IMFs (intrinsic mode components) are obtained, and then according to the frequency characteristic and noise distribution of each IMF component, a wavelet threshold function is obtained; and carrying out targeted noise suppression by adopting an improved wavelet threshold function containing an adjustment parameter alpha, and finally, carrying out linear superposition reconstruction on all the processed IMF components to obtain a de-noised geophysical prospecting signal. Experimental verification shows that compared with a traditional method, the method has the advantages that the signal-to-noise ratio and the correlation coefficient of the noisy geophysical prospecting signals can be effectively increased, the root-mean-square error can be effectively reduced, the inherent defects of the traditional method are overcome, the geologic features of the geophysical prospecting signals can be effectively reserved, the method is suitable for various mineral exploration scenes, and reliable data support is provided for anomaly recognition in mineral exploration.

The invention discloses an intelligent segmental arc correlation method based on a minimum optical admissible domain, belongs to the field of intelligent calculation optimization of space target monitoring, solves the problem of large calculation amount of segmental arc correlation, and comprises the following steps: constructing the minimum optical admissible domain by a first observation segmental arc; uniformly sampling in the minimum optical tolerance domain, performing orbit propagation on each sampling point under a J2 perturbation model to obtain theoretical optical observation data of a second observation arc section, and calculating an angle root-mean-square error of the second observation arc section to obtain an angle loss function; using a DBSCAN clustering algorithm to identify wave troughs; selecting an initial value of a simplex method in each wave trough to obtain an angle optimal solution; a probability loss function is obtained through Bayesian theorem calculation; substituting the angle optimal solution into a probability loss function, and obtaining a probability optimal solution by using an LM algorithm; a threshold value is set according to the probability loss function obeying chi-square distribution, and when the probability optimal solution is smaller than the threshold value, the two arc segments are associated successfully; according to the invention, the calculation cost is saved, and the space target observation accuracy is improved.

The invention discloses an electricity price prediction method and system based on a converter and a bidirectional gating cycle network, and belongs to the technical field of power systems and artificial intelligence prediction.The method comprises the steps that a historical electricity price data set is acquired, the historical data set is constructed, and the historical data set is preprocessed; the preprocessed data is input into a Transform encoder layer in a hierarchical multi-head attention mechanism; a bidirectional gating recurrent neural network optimized through a global attention mechanism; a cross attention mechanism is combined with the output of a Transform encoder and the output of a BiGRU layer; and performing performance evaluation on a prediction result by adopting an absolute mean error, a mean square error and a root-mean-square error. According to the invention, while the time sequence dynamic modeling capability is maintained, the physical constraint information of the power system is effectively integrated, and the generalization capability of the electricity price prediction model for the multi-source uncertainty in the high-proportion renewable energy penetration scene is significantly improved.

The invention provides a multi-unmanned aerial vehicle optimal game limited accelerated reinforcement learning control method and device, and relates to the technical field of accelerated reinforcement learning. The method comprises the following steps: constructing an evaluation neural network, and approaching a performance index function of a hyperbolic tangent function and a game neighbor item, the optimal limited control input of the unmanned aerial vehicle and the limited control input of the unmanned aerial vehicle under the worst condition; based on the input parameters, constructing a Hamilton-Jacobian error equation; through an error equation, constructing a summation square error which contains current information and past information and is provided with adjusting parameters; designing a weight updating law for evaluating the neural network; calculating the weight of the next iteration according to a weight updating law; and solving an error between two adjacent iteration weights, comparing the error with a preset threshold value, if the error is smaller than the preset threshold value, stopping iteration, and outputting a solution approaching the optimal game consistency control problem of the distributed unmanned aerial vehicle system. According to the invention, the cooperation efficiency between the unmanned aerial vehicles can be improved.

The invention relates to the technical field of electromagnetic scattering characteristic analysis and modeling, in particular to a scattering center establishing and correcting method based on a geometric model and reference data. The method comprises the following steps: inputting a target geometric or grid file, and extracting geometric features of a plane, an edge and a curved surface to construct an initial scattering center model; inputting electromagnetic scattering reference data of the same target, solving a scattering center correction coefficient by adopting a pseudo-inverse or Tikhonov regularization method, and optimizing an initial model; and outputting an optimized scattering center model with angle and frequency expansibility. According to the method, the electromagnetic scattering characteristics of metal and coating type complex structure targets in single-station, double-station and full-polarization scenes can be uniformly represented, and the problems of insufficient model precision, incomplete types and high data dependence degree in the prior art are solved; the method achieves the excellent effects that the RCS root-mean-square error is smaller than 2dB and the high-resolution imaging structure similarity exceeds 90% in typical target modeling, and is suitable for the fields of target characteristic simulation and characteristic extraction and recognition.

The invention relates to an ultrasonic water meter zero flow detection method based on machine learning, and the method comprises the steps: based on the measurement principle of a time difference method ultrasonic flowmeter, collecting time difference data of each sound channel of a multichannel ultrasonic water meter before and after a valve is closed at different flow points, and classifying the time difference data; preprocessing the original data, calculating the statistics of each group of time difference data, and constructing the statistics into a feature data set; performing parameter initialization on the improved subtraction average optimizer; dividing the feature data set into a training set and a test set, taking a root-mean-square error of a predicted value and an actual value as a fitness function, and taking a parameter corresponding to the lowest fitness value as an optimal solution; training a support vector machine by using the optimized parameters to obtain an ultrasonic water meter zero flow detection model; based on the ultrasonic water meter zero flow detection model, ultrasonic water meter zero flow detection is conducted. According to the invention, accurate metering of the ultrasonic water meter is realized.

The invention discloses an off-line reinforcement learning action exploration agent method based on expected reward regularization, and belongs to the field of reinforcement learning. The problem that reliable track splicing and strategy generalization are difficult to realize in a complex task by the existing method is solved. The method comprises the following steps: constructing state loss and action loss based on sequence modeling, and carrying out iterative training of states and actions; designing an RTG loss function based on a weighted square error; a double-Q learning framework is adopted to maintain two independent Q functions, conservative constraints of conservative Q learning are applied in the Q function updating process, and action exploration optimization is achieved in combination with Boltzmann distribution; combining state loss, action loss, RTG regularization loss and Q value loss to form a joint optimization objective function; performing noise disturbance sampling on the plurality of RTG candidate values to generate diversified action prediction; and value evaluation is performed on the candidate actions based on a double-conservative Q function, and the action with the highest Q value is selected for execution. The method is mainly used in the intelligent agent exploration field.

The invention discloses a power distribution network operation state analysis method and system, and relates to the field of power distribution network operation and maintaining.The key points of the technical scheme are that a graph structure is constructed according to operation data, and a feature matrix of the graph structure is extracted; constructing a graph neural network, training the graph neural network in a two-stage training mode by taking the feature matrix as input and taking the topological connection state and the node electrical quantity predicted value as output, and when the maximum training frequency is reached, obtaining the topological connection state and the node electrical quantity predicted value. Outputting an operation state analysis model capable of predicting the communication state of a power distribution network branch and the node electrical quantity capability; wherein the first-stage training mode adopts self-supervised pre-training to train the graph neural network, and the second-stage training mode is training of a weighted loss function of mean square error loss and square error loss of a power flow equation residual error; and collecting current operation data of the target power distribution network, inputting the current operation data into the operation state analysis model, and analyzing an analysis result of the target power distribution network.