23 results about "Mean square difference" patented technology

The invention discloses a multivariable time sequence prediction method and system based on an implicit neural network. The method comprises the following steps: 1) collecting data and preprocessing the data; 2) performing window division on the standardized or normalized multivariable time sequence and determining the length of a to-be-predicted window; 3) performing variable correlation coding on the input window to obtain a variable-level feature vector; 4) the implicit neural network based on time attention predicts target parameters by using the variable features in the step 3), and implicit neural representation of the target sequence is modeled through the parameters; 5) taking the output of the implicit nerve representation and the original input window as the input of the multi-head attention predictor, and obtaining a prediction result through cross-sequence cross attention calculation performed in the implicit space and multi-layer perceptron conversion output dimension; and 6) training and optimizing model parameters, calculating a mean square error of a prediction result and a real result, taking the mean square error as a loss function, carrying out back propagation to optimize trainable parameters of the variable correlation coding module, the implicit neural network module multi-head attention predictor and the multi-layer perceptron, and then repeating the steps 3) to 6) to obtain the multi-head attention predictor. Until the preset number of iterations is reached or the error of the model on the verification set meets the requirement of early stop; and 7) performing prediction by using a model of training convergence, and performing reverse normalization on a prediction result to obtain a final prediction result. The method has good generalization, and meanwhile, the interpretability of the attention mechanism is remarkably improved by generating the hidden space characteristics of the trend component and the season component.

The invention discloses a weak radar signal target tracking method, system and device and a medium, and belongs to the field of radar signal processing, and the method comprises the steps: constructing a state space model, predicting a target state and an error covariance at a moment k based on model parameters and a target state filtering value and an error covariance filtering value at a moment k-1, and obtaining a target state and an error covariance; target motion information is obtained according to the echo signals at the moment k, whether dead pixels or leakage points exist or not is judged, if yes, the predicted position value in the target state at the moment k is extracted to replace the original position value, the target position error value and the mean square error are calculated again and replace the original error value and the mean square error, and the target position is obtained. Updating the filtering value into the predicted target state and error covariance at the moment k, and if not, calculating the Kalman gain and combining with the position value at the moment k, and performing filtering correction on the predicted target state and error covariance at the moment k; the system, the equipment and the medium are used for implementing the method. According to the method, the robustness and the tracking precision are improved, and the calculation complexity and calculation resource occupation are reduced.

The invention discloses a sparse view angle 3D-DSA reconstruction method based on a three-dimensional Poisson generative model. The method comprises the following steps: obtaining pairing data of a sparse view angle 2D-DSA projection drawing and a 3D-DSA reconstruction image; extracting features of a projection image by using a projection domain encoder, then converting features of two-dimensional projection into a three-dimensional image domain according to a geometrical relationship of the cone beam CT, and then obtaining a prior image by using an image decoder; the method comprises the following steps: constructing a three-dimensional Poisson generation model, adding noise to a three-dimensional image patch in a training stage to obtain a disturbance image, outputting a network reconstruction image by taking a prior image as a condition and the three-dimensional image patch before noise addition as a target image, and calculating loss of the output image and the target image to update network parameters; the mean square error loss and the mean square error loss of the maximum intensity projection images of the three orthogonal planes are used during loss calculation; in the sampling stage, random noise is used as input, a prior image is used as a condition, noise of a noise image is continuously denoised within a limited step length, and finally a reconstructed 3D-DSA image is obtained.

The invention belongs to the field of water-wind-light multi-energy complementary optimal scheduling, and particularly discloses a water-wind-light optimal scheduling method and system based on a two-stage dual-population evolutionary algorithm, and the method comprises the steps: taking the water level of a reservoir as a decision variable, taking the maximum power generation amount and the minimum residual load mean square deviation as objective functions, and constructing a multi-objective scheduling model; the two populations are initialized, and the water-wind-light multi-target scheduling model is solved; the two populations are initially in a fixed division stage, and operators A and B are respectively adopted for iterative updating; when a switching condition is met, switching to a self-adaptive cooperation stage for iterative updating, determining the selection probability of the operators A and B according to population performance at the moment, and selecting the operators based on the selection probability for iterative updating; the switching condition is that after the population is iteratively updated each time, if the population diversity is smaller than a threshold value or the evaluation frequency reaches the threshold value, stage switching is carried out. According to the method, the contradiction between convergence and diversity in water-wind-light multi-objective optimization can be solved, and accurate optimization is realized.

The invention relates to the technical field of monitoring and early warning, in particular to a non-coal mine slope monitoring and early warning equipment threshold setting method based on an AI algorithm, and the method comprises the following steps: obtaining the change trend of a slip rate through introducing time sequence derivative analysis, and screening an abnormal turning feature section; the method comprises the following steps of: establishing a mapping relation between curvature and grade by combining a current equipment grade boundary parameter, constructing a curvature threshold interval adaptive to equipment state change, fusing and training a neural network model by utilizing Euclidean distance of a plurality of disturbance factor groups and mean square error characteristics in a time window, performing accurate classification on disturbance density grades, and calculating the disturbance density. The calculation of the level correction weight is completed through the proportional mapping between the disturbance level and the preset level, and the weight correction is performed on the initial level threshold parameter, so that each equipment threshold has the self-adaptive adjustment capability for the current state, and the dynamic response capability of the early warning threshold setting and the anomaly recognition accuracy are further improved.

The present invention relates to a method and system for constructing a spindle simulation component considering damage gradient. First, based on the simplified model of the spindle structural component, the damage gradient path at the critical point is determined, as well as the target damage values of each center point on the damage gradient path. Then, a spindle simulation component is constructed based on the structure at the critical point, and the reciprocal of the mean square error between the simulated damage values and the target damage values of each center point on the damage gradient path is used as the fitness function. The genetic algorithm is adopted to optimize the parameters of the spindle simulation component to make it consistent with the damage gradient of the spindle structural component.

The invention provides a music time coding method for artificial neural network modeling. The method aims at the coding requirement of music beat time, molecules are decomposed according to a fixed denominator for multi-bit independent classification coding, and the method comprises the following steps: setting the fixed denominator as a composite number and decomposing the composite number into a prime factor combination; the molecules are mapped into vectors, the value range of each bit is determined by a cardinal number of a prime factor combination, and the vectors are converted into one-hot coding sub-vectors with redundant components removed. The technical scheme of the invention solves the problem that the traditional mean square error (MSE) ignores the internal correlation of scores and has high sparsity in the topology recognition task of an optical music recognition system (OMR), and compared with a traditional coding method adopting floating point numerical output and a mean square error loss function under the conditions of the same data set, model architecture and training hyper-parameters, the method provided by the invention has the advantages that the coding efficiency is greatly improved; according to the method, the comprehensive error index is reduced by about 21%, and the model precision and the training efficiency are improved.

A difference analysis method for microwave detection result evaluation comprises the following steps: carrying out microwave detection on a defect-free sample to obtain a reference S parameter; performing microwave detection on the to-be-detected sample to obtain S parameters of the same type; carrying out wavelet transform de-noising processing on the obtained two groups of S parameters to eliminate the influence of local fluctuation points on calculation; performing mean square error MSE calculation on the two groups of de-noising processing S parameters; evaluating the defect condition of the to-be-detected sample according to the value of the mean square error MSE; the larger the mean square error MSE value is, the larger the S parameter difference of the to-be-detected sample and the reference defect sample is, and the higher the defect degree of the to-be-detected sample is. And continuing to detect other samples to be detected, and carrying out detection and difference analysis to complete defect evaluation of the samples to be detected. Defect evaluation is carried out by utilizing the difference value analysis method provided by the invention, the frequency band does not need to be manually selected, and the whole frequency band is defaulted, so that the detection complexity is greatly reduced.

This disclosure relates to the field of bridge engineering, specifically to a design method, system, and medium for HPC-RC combined eccentrically compressed columns. The method includes: defining a design space; selecting optimal design data combinations based on engineering specification constraints, bearing capacity constraints, and cost-effectiveness values ​​to construct a design data sample set; constructing a two-branch heterogeneous neural network model, the model including an input layer, a shared feature extraction layer, a diameter prediction classification branch, and a reinforcement area prediction regression branch; constructing a loss function composed of focal loss and mean square error loss; training the two-branch heterogeneous neural network model using the design data sample set; and using the trained two-branch heterogeneous neural network model to predict the diameter and total reinforcement area of ​​the eccentrically compressed column. This disclosure improves the stability and accuracy of the design, reduces the computational burden, increases computational speed, and reduces computational complexity.

The invention discloses a chaotic system synchronous regulation and control method based on a reserve pool computing network, and the method specifically comprises the steps: dividing groups based on the symmetry of nodes of the reserve pool computing network, training the nodes of different groups with different system data, and introducing Lorenz and Rossler chaotic systems as RC training data sources; in the training stage, input data are firstly mapped to a reservoir through an input weight, and reservoir neurons update weight state information in time after receiving the input data; after training is completed, output data serve as input data to be fed back to a reservoir, the reservoir operates autonomously, and RC is used for simulating dynamic characteristics of oscillator nodes in a complex network; and introducing a mean square error function RMS to judge the stability of the synchronous group. According to the method, the coupling strength and other parameters among the group nodes are regulated and controlled by analyzing the characteristics of the network topology structure, so that the overall dynamic state of the group nodes is influenced, and stable zero-time-delay synchronization among the calculation nodes of the reserve pools in the group is realized.

According to the method for measuring the surface type parameters of the transparent thin plate, outlier identification and correction are carried out on collected surface coordinates and thickness data of the transparent thin plate before and after polishing through an abnormal value detection algorithm and adjacent non-abnormal point linear interpolation replacement, and then noise reduction processing is carried out on the corrected data by adopting a dynamic noise reduction combination. The method comprises the following steps of: acquiring surface coordinates and thickness data of a transparent thin plate, performing interpolation prediction on the unacquired surface coordinates and thickness data of the transparent thin plate, finally performing data fitting, generating a thickness thermodynamic diagram, calculating the bending degree, the warping degree, the total thickness change and the local thickness change of the transparent thin plate, simultaneously generating a removal quantity thermodynamic diagram, calculating the kurtosis, the skewness, the mean square error and the local deviation value, and judging the polishing uniformity; the method is high in key defect recognition accuracy, high in interpolation adaptability and high in parameter acquisition precision, multi-dimensional quantitative analysis of the polishing uniformity of the transparent thin plate is achieved by introducing indexes such as kurtosis, skewness, mean square deviation and local deviation, and data support is provided for technological parameter adjustment.

The present invention relates to a method for correcting working condition parameters of a cascade wind tunnel experiment based on a deep neural network, and belongs to the field of aerospace experimental measurement and artificial intelligence. Floating working condition parameters, constructing multiple sets of incoming flow parameters and performing numerical calculations of the cascade flow field; using the normalized working condition parameters after floating as input, and using the corresponding physical quantities of the measuring point positions in the front 50% of the blade chord length of the numerical flow field as output, to construct a deep neural network; using the mean square error of the deep neural network prediction results and the experimental measurement results as the loss function, using the automatic differentiation algorithm to backpropagate the loss function gradient to obtain the gradient of the mean square error with respect to the working condition parameters; using the gradient optimization algorithm to correct the working condition parameters. The present invention can effectively correct the incoming flow boundary conditions of the cascade wind tunnel experiment, can effectively overcome the separation zone prediction deviation generated by the turbulence model, does not require the special selection of the turbulence model, and is particularly suitable for the reconstruction and inversion of the experimental flow field using the flow field numerical simulation method.

The invention discloses a storage and calculation integrated chip-based neural network quantification method and device. The method comprises the following steps of performing dimension expansion processing on an original base library vector and an original target vector; based on the mean value and the mean square error of each dimension of the dimension expansion base library vector, normalizing the dimension expansion base library vector and the dimension expansion target vector; sorting data of each dimension of the normalized base library vector to obtain two identification points in a sorting result of each dimension, determining a base library data core distribution interval corresponding to the dimension based on the identification points, and taking the base library data core distribution interval as a reference to obtain a base library data core distribution interval corresponding to the dimension; mapping data of a dimension corresponding to a bottom library data core distribution interval in the normalized bottom library vector to a first hardware adaptive quantization interval; and determining a corresponding target data core distribution interval according to the first number, the second number, the maximum value and the minimum value in the normalized target vector, and mapping the normalized target vector to a second hardware adaptive quantization interval by taking the target data core distribution interval as a reference.

The invention discloses a precipitation prediction method based on a GRU and an LSTM neural network. The precipitation prediction method comprises the following steps: selecting meteorological station data continuously monitored for a long time; removing abnormal values of the meteorological station data, and complementing missing values; selecting first 85% of historical data for model parameter training, and selecting last 15% of recent data for model generalization ability test; normalizing the three precipitation index dimensions of the training set and the test set respectively; combining an algorithm model by using GRU and LSTM components; inputting the preprocessed training set data into the network, using a mean square error as a loss function, adopting an Adam algorithm to optimize the loss function, and iteratively updating network parameters until the model converges; data to be predicted are input into the trained model after being preprocessed in the first step, and a precipitation prediction result is output. Sequence data information is processed through multiple influence factor data and the superposed GRU and LSTM combination, the gradient disappearance problem is avoided, the prediction effect is good, precision is high, and applicability is good.

The present invention discloses a CSI compression and feedback method and system for large-scale MIMO based on neural networks. The present invention uses a new deep neural network structure to complete the feedback of CSI, and this network is called Dilated Attention Inception Net (DAINet). The CSI data to be compressed and feedback is sequentially input into the pre-denoising module and encoder of the trained DAINet after 2D DFT transformation. After compressing and outputting a one-dimensional codeword for feedback, it is received and decoded by the trained decoder at the feedback receiving end to obtain the CSI matrix in the angular delay domain, and finally the restored value of the CSI matrix in the spatial frequency domain is obtained through 2D IDFT transformation; in the case of ideal channel estimation and non-ideal channel estimation, considering the accuracy rate and training cost comprehensively, the proposed DAINet is superior to the existing methods. Specifically, in the case of ideal channel estimation, the NMSE (Normalized Mean Square Error) of DAINet is about 1 / 2 of AnciNet and 1 / 12 of CsiNet; in the non-ideal case, the NMSE of DAINet is about 9 / 10 of AnciNet and 1 / 2 of CsiNet. The number of network parameters of DAINet is about 80% of AnciNet and 130% of CsiNet.

The application discloses a source-load power prediction method and device considering different granularities and a storage medium. The method comprises the following steps: collecting source-load historical data in a to-be-predicted area; pre-processing the source-load historical data to obtain multiple groups of data groups with different granularities; inputting the multiple groups of pre-processed data groups with different granularities into a pre-constructed and trained MultiGNet prediction model respectively to obtain a prediction result of source-load bilateral power; wherein the MultiGNet prediction model is obtained by extracting semantic dependency features of different granularity data groups and establishing a connection between different granularity data groups through a cross-granularity learning module, analyzing feature relationships of different granularity data groups through a granularity attention module to obtain optimal weights of different granularity data groups, and training source-load historical data with a minimum mean square error loss function as a target. The application fully utilizes rich information of different granularity data for prediction, and improves the accuracy of source-load power prediction.

The application belongs to the field of water, wind and light multi-ability complementary optimal scheduling, and particularly discloses a water, wind and light optimal scheduling method and system based on a two-stage double-population evolution algorithm, which comprises the following steps: constructing a multi-objective scheduling model by taking reservoir water level as a decision variable and taking maximum power generation and minimum residual load mean square deviation as objective functions; initializing two populations and solving the water, wind and light multi-objective scheduling model; the two populations are in a fixed division stage at the beginning and are respectively updated by operators A and B; when the switching condition is met, the two populations are switched to an adaptive cooperation stage for iterative updating, at this time, the selection probability of operators A and B is determined according to the population performance, and the operators are selected based on the selection probability for iterative updating; the switching condition is that the population diversity is less than a threshold value or the evaluation number reaches a threshold value after each iterative updating of the population, and the stage is switched. The application can solve the contradiction between convergence and diversity in water, wind and light multi-objective optimization and realize precise optimization.

The application discloses a high-resolution artistic face landmark detection method based on deep learning, first, an artistic face data set and corresponding landmark annotations used for model training and evaluation are prepared; a global low-resolution predicted landmark detection map and a regional high-resolution predicted landmark detection map are obtained respectively; the global low-resolution and regional high-resolution predicted landmark detection maps are optimized through a mean square deviation loss; finally, the high-resolution predicted landmark detection map of each region is restored to a global coordinate system, that is, face landmark detection is realized. The application shows excellent effect in face landmark detection of paintings and prints and other artworks. The application proposes a combination of a global encoding-decoding network and a regional encoding-decoding network to realize rough and refined marking of artistic face landmarks, and the marking effect is more excellent than that of existing methods.

The application discloses a kind of centroid extraction method and extraction system of autocollimator image, wherein based on the Gaussian filter of mean square deviation σ is filtered to image;According to the edge of the image refined by multistage calculation, the edge detection of image is completed, and edge pixel point is determined;Ellipse centroid is fitted to edge pixel point;Based on each ellipse centroid is normalized, and the centroid of image is calculated, at this time, the edge of the image refined by filtering processing under multistage calculation, and corresponding edge pixel point is determined, ellipse centroid is fitted to edge pixel point, and based on each ellipse centroid is normalized, by this method, the influence of unstable gray on centroid extraction can be overcome, and the centroid extraction precision of autocollimator image is improved.

The present invention discloses a method for ADC digital background calibration based on an intelligent optimization algorithm, comprising: after receiving M groups of output data sent by an analog-to-digital converter, constructing M groups of interpolation output codes using a Lagrange interpolation algorithm; constructing a target optimization function based on the M groups of output data and the M groups of interpolation output codes, and performing an optimization search using a particle swarm optimization algorithm to obtain M groups of weights so that the mean square error between each group of output data and its corresponding interpolation output code is minimized; and the average value of the M groups of weights is output as the optimal weight for this optimization search and as the central value for the next optimization search. By combining the interpolation and particle swarm optimization algorithms as a framework to optimize the search for the target optimization function, with the goal of minimizing the mean square error between the M groups of output data and their corresponding interpolation output codes, the optimal weight is obtained, thereby improving the accuracy of the successive approximation ADC and simply and efficiently implementing the digital background calibration of the ADC.

The present application provides a track fusion method, a fusion device, a processor, and a fusion system. The method includes: obtaining multiple first track data sets and multiple second track data sets; calculating the sum of the mean square errors of all track data subset pairs in each track data set group to obtain multiple total mean square error sums, wherein the track data set group includes multiple track data subset pairs, a track data subset pair includes a first track data set and a second track data set, and the first track data set and the second track data set of any two track data subset pairs are different; using a track fusion algorithm to fuse the two tracks corresponding to the optimal track data subset pair to obtain multiple optimal tracks, wherein the optimal track data subset pair is the track data subset pair with the smallest total mean square error and the corresponding track data set group. This method solves the problem of low accuracy of the track similarity calculation method used in the prior art for track association.

This invention provides a method for optimizing and scheduling federated learning clients based on a three-dimensional grid, comprising the following steps: S1, defining three-dimensional coordinates based on the client's training speed, amount of training data, and mean square error of the training data, and dividing the space into three-dimensional grid cells; S2, assigning the client to the corresponding grid cell based on its three-dimensional attributes; S3, during each training round, a central server determines the grid cell based on a set three-dimensional subscript probability and selects the corresponding client from it until the number of clients reaches the training round requirement; S4, the central server distributes the global model parameters to the selected clients for local training; S5, the central server aggregates the trained model parameters and model accuracy data to update the global model; S6, repeating S1-S5 until the model reaches the training accuracy. Compared to existing algorithms, this invention considers various heterogeneities in federated learning, improving the training speed and model accuracy of federated learning.