235 results about "Extreme learning machine" patented technology

The invention discloses a physical field solving method based on a Bayesian physical information extreme learning machine, and the method comprises the steps: constructing a single-layer full-connection neural network, carrying out the random initialization, and fixing the weight of an input layer; based on a partial differential equation of a physical system and boundary conditions thereof, defining a training loss item containing physical information; a physical system solving problem is converted into a linear least square problem, and a linear equation set is constructed; supposing that an output layer weight parameter obeys Gaussian prior distribution with the mean value being zero, and controlling a covariance matrix by an adjustable hyper-parameter; constructing a Gaussian likelihood function based on the observation data, and calculating posterior distribution of the output weight in combination with the prior distribution; carrying out iterative optimization on the hyper-parameter by adopting an evidence maximization method to obtain a mean value and a covariance of posterior distribution; based on posterior distribution, adopting a Monte Carlo integral method to generate prediction output of the physical system; and performing uncertainty quantization based on the variance of prediction output, and outputting a prediction value and a confidence interval thereof.

The invention discloses a computer network security data processing method and system based on artificial intelligence, and the method comprises the steps: building a multi-channel deep learning fusion model, extracting spatial local features in a traffic sequence through employing a 1D-CNN one-dimensional convolutional neural network, capturing a long-range time sequence dependence relation between log events, and carrying out the recognition of the long-range time sequence dependence relation between log events; modeling the user operation behavior sequence based on an LSTM (Long Short-Term Memory) network, and fusing the feature weight by using an attention mechanism to obtain a fused feature vector; inputting the fusion feature vector into a classifier established based on an OS-ELM online sequence extreme learning machine to perform real-time threat assessment, and outputting a probability index of network attacks occurring in a short time in the future; and generating a cooperative defense decision according to the network attack probability index, and sending the cooperative defense decision to security equipment for execution. Excessive defense or insufficient protection is avoided, and the cooperation efficiency of safety equipment is remarkably improved.

The invention discloses a self-adaptive ultrasonic measurement method and system based on multi-channel cooperation, and relates to the technical field of ultrasonic measurement. The method comprises the following steps: collecting original flight time signals of each channel; meanwhile, environment data are collected in real time; original flight time signals are processed, effective signal arrival pre-flight time is extracted, and a fusion data set is constructed; constructing a two-dimensional sound channel spectrogram based on the multi-sound channel time sequence of the pre-flight time, and performing feature extraction based on a lightweight convolutional neural network to generate an abnormal confidence vector; analyzing the sound channel state based on the abnormal confidence vector; constructing a physical information neural network, and analyzing the corrected sound velocity value of each sound channel and the two-dimensional sound velocity field distribution on the section of the whole pipeline; training an online sequence extreme learning machine model in combination with historical measurement data; and based on the final fusion weight of each sound channel and the corresponding sound channel flow velocity, carrying out weighted fusion to generate a flow velocity optimal estimation value. And the measurement precision and robustness are improved.

The invention discloses a tool fracture and fatigue intelligent detection method based on vibration signal analysis, and the method comprises the following steps: S1, installing a vibration sensor, and collecting the vibration signal of a tool in real time; s2, the collected tool vibration signals are preprocessed, and noise in the signals is removed; s3, performing time-frequency analysis on the preprocessed vibration signals, and extracting time-frequency features in the signals; s4, performing deep feature learning on the extracted time-frequency features to form deep features; s5, the depth features are classified and analyzed, and the health state of the cutter is output; s6, according to the health state optimization feature extraction and prediction result of the cutter, generating learning output; s7, evaluating the health state of the cutter in real time according to the learning output, and pushing alarm information; and S8, according to the alarm information, predicting the service life of the cutter and optimizing a cutter replacement and maintenance strategy. According to the method, short-time Fourier transform and Hough transform are combined, and the extreme learning machine is applied, so that intelligent detection on the fracture and fatigue of the cutter is realized.

The invention provides an online sequential preposed interference layer extreme learning machine, a classification method, equipment and a medium, and relates to the technical field of machine learning. The online sequential preposed interference layer extreme learning machine comprises an input layer, an interference layer, a hidden layer and an output layer, the input layer is used for receiving original sample data, the interference layer reduces the sample distribution complexity through nonlinear kernel mapping, the hidden layer is located between the interference layer and the output layer, and the output layer is located between the interference layer and the output layer. And the output layer is used for further extracting features of the samples mapped by the interference layer and calculating a final classification result based on the output of the hidden layer. According to the method, the performance of the extreme learning machine in classification and regression tasks can be remarkably improved, particularly, higher generalization ability and stability are shown when nonlinear distribution samples are processed, meanwhile, dependence on parameter selection is reduced, and the method is suitable for real-time data processing scenes such as intelligent medical treatment, fault detection and financial prediction.

The invention belongs to the field of non-intrusive power load monitoring, and discloses a non-intrusive power load identification method, which comprises the following steps of: constructing a load identification framework utilizing frequency domain characteristics, and aims to solve the problem that transient characteristics cannot be effectively applied in non-linear load identification in non-intrusive load monitoring. A current feature extraction technology based on fast Fourier transform and Hilbert-Huang transform and a classification method based on an extreme learning machine as a main body are used for training and testing a public data set to verify that the method is used for extracting transient load features and identifying nonlinear loads.

The invention provides a soil ammonium nitrogen content hyperspectral prediction method based on an improved extreme learning machine, and relates to the technical field of hyperspectral prediction. The method comprises the following steps: firstly, collecting and treating a soil sample for soil spectral measurement and NH4 < + >-N content determination; measuring soil spectral reflectivity data; preprocessing the soil spectral reflectivity data to form a spectral reflectivity data set; then carrying out characteristic wave band selection by adopting a sequential forward selection algorithm; an improved butterfly optimization algorithm IBOA is adopted to optimize model parameters of an extreme learning machine ELM, and then a hyperspectral prediction model used for predicting the NH4 < + >-N content of the soil is constructed; and finally, the ELM model after parameter optimization is selected to construct a hyperspectral prediction model to predict the NH4 < + >-N content. The method not only provides theoretical and technical support for soil ammonium nitrogen content monitoring, but also provides important reference and guidance for soil nitrogen cycle research and soil management.

The invention discloses an SOH prediction method and system based on an internal resistance-temperature correlation model. The method comprises the following steps: acquiring a pre-established battery internal resistance-temperature correlation model; operating temperature data and a current internal resistance value of a to-be-measured battery are collected in real time, and the operating temperature data are input into the internal resistance-temperature correlation model to obtain a reference internal resistance predicted value at the temperature; calculating the relative deviation between the current internal resistance value and the reference internal resistance predicted value, and constructing a multi-dimensional health state feature vector in combination with the cycle index of the battery; and inputting the multi-dimensional health state feature vector into a pre-trained extreme learning machine (ELM) prediction model, and outputting a health state quantitative index and a residual life prediction result of the battery. According to the embodiment of the invention, the accuracy and reliability of battery health state estimation under different temperature working conditions can be improved, and more accurate battery life management is realized.

The invention discloses a power distribution network construction material demand analysis and prediction method, which comprises the following analysis and prediction steps: S1, collecting historical material data of a power distribution network construction project, the historical material data comprising project attribute data and material usage data; s2, preprocessing the material consumption data, including data cleaning, standardization and normalization processing; s3, carrying out dimension reduction processing on the material data based on a principal component analysis method; s4, performing clustering analysis on the material project attributes based on a clustering analysis algorithm; s5, parameters of the extreme learning machine are optimized based on a particle swarm algorithm, the optimized extreme learning machine is adopted to construct a power distribution network construction material demand prediction model, and the number and amount of power distribution network construction material demands are predicted; the system can effectively improve the material management level of a power enterprise, and improves the efficiency and quality of a material demand plan.

The invention provides a method, a system and a device for predicting the residual service life of a proton exchange membrane fuel cell. The method comprises the following steps: step 1, collecting monitoring data of a charge-discharge process of the fuel cell; 2, carrying out feature selection on the monitoring data of the charging and discharging process; step 3, using robust local mean decomposition based on dynamic time warping improvement to decompose the monitoring data of the charging and discharging process of the fuel cell; 4, reconstructing a sub-sequence by adopting an entropy control principal component polymerization method; 5, establishing a dynamic adaptive weighted extreme learning machine of the life prediction model and the residual service life prediction model; step 6, improving a projection iterative optimization algorithm; and 7, optimizing the dynamic adaptive weighted extreme learning machine of the residual service life prediction model to obtain a final prediction result. The problem that the residual service life of the fuel cell cannot be accurately predicted is solved, and the prediction accuracy of the residual service life of the cell is improved.

The invention discloses a foundation pit horizontal displacement probability prediction method and system based on a sparse Bayesian extreme learning machine, and belongs to the technical field of civil engineering monitoring and artificial intelligence crossing. According to the method, a probability model containing input and output noise is constructed, feature selection and uncertainty quantification are automatically carried out by using a sparse Bayesian framework, and probability prediction of horizontal displacement at the position where a sensor is not arranged is realized. The method can output the prediction mean value and the confidence interval, effectively solves the problems of data sparsity and noise, and improves the prediction reliability and the engineering decision support capability. The method has the advantages of being high in automation degree, high in anti-interference capacity, suitable for actual engineering monitoring and the like.

The invention provides an anti-saturation fault-tolerant control method and device for a double-flexible-arm space robot, relates to the field of robot control, and solves the multiple challenges that in the prior art, a flexible mechanical arm faces continuous and difficult vibration suppression, an actuator is prone to failure, torque is limited, task time is constrained and the like. The dynamic state of the system is unstable; and the control capability is limited. The method comprises the steps that a hypothesis modal method and a momentum conservation theorem are combined to derive a double-flexible-arm space robot kinetic equation in a Lagrange form; a flexible system singular perturbation technology is utilized to obtain a slow-varying subsystem for representing rigid motion characteristics of a mechanical body and a quick-varying subsystem for representing residual vibration characteristics of a rod piece; for the slow-varying subsystem, designing a fixed-time anti-saturation fault-tolerant controller based on an extreme learning machine and a torque output function; and designing an anti-saturation controller based on state variable negative feedback for the quick change subsystem. The method is used in the robot control process.

The invention discloses a synthetic aperture radar online trajectory planning method based on an adaptive network. The method comprises the following steps: S1, establishing a bistatic SAR system model and a task coordinate system; s2, dividing a task space; s3, establishing a multi-objective optimization model; s4, performing independent evolution in each task block by using a joint multi-objective evolutionary algorithm of a block propagation strategy, and performing joint optimization through an inter-block elite individual sharing mechanism; s5, generating a trajectory sample data set; s6, introducing a cooperative differential evolution optimization mechanism based on a regularization extreme learning machine algorithm, and constructing a large-scale adaptive network; s7, carrying out real-time track prediction and online updating; and S8, outputting an optimal trajectory result meeting the constraint. According to the method, the problems of long time consumption, high calculation complexity and insufficient network generalization performance of a multi-objective evolutionary algorithm in real-time trajectory planning are solved, and the real-time performance and precision of trajectory planning are improved.

The invention discloses a battery life prediction method and system based on an adaptive generative adversarial network. The method comprises the steps of collecting battery use data through a sensor of a battery management system; performing manual labeling on the collected data, wherein labeling categories comprise'normal ', 'abnormal' and'performance reduction '; constructing a data set in combination with the data and the annotations; generating additional training data by adopting a generative adversarial network based on adaptive chaos optimization, and adding the training data into the data set to complete data expansion; reading the data set, and training a battery life prediction model by adopting an extreme learning machine algorithm based on a quantum topological phase to obtain a classification result; and inputting a real-time sample into the battery life prediction model for battery life prediction. According to the method, the accuracy and the reliability of a battery life prediction result can be improved, nonlinear characteristics of battery performance degradation can be more effectively captured, and higher efficiency and accuracy are provided when complex data are processed.

The invention provides an efficient data acquisition method and system for an HPLC (High Performance Liquid Chromatography) dual-mode communication module based on edge calculation, and relates to the technical field of HPLC dual-mode communication data acquisition. The method comprises the steps of collecting multi-dimensional resource data of an HPLC dual-mode communication module in real time, performing feature extraction by adopting a local linear embedding algorithm, constructing a resource state evaluation model based on an extreme learning machine, evaluating a resource load state of the communication module in real time, and establishing a multi-target optimization model. Solving by adopting a non-dominated sorting genetic algorithm III to obtain an acquisition strategy solution set, selecting an optimal acquisition strategy, converting the optimal acquisition strategy into a control instruction, and adjusting data acquisition parameters in combination with predictive control and a feedback correction mechanism. According to the method, the problems of resource waste, single index and the like of an existing scheme are effectively solved by performing adaptive denoising, feature extraction and resource state accurate evaluation on the multi-dimensional data and combining multi-objective optimization of data space-time association mining and acquisition efficiency, energy consumption and reliability.

The application discloses a kind of analog circuit fault diagnosis method based on wavelet scattering and ensemble learning, to solve the problem that analog circuit fault response aliasing leads to difficult fault diagnosis.First, the circuit response is divided into multiple frequency band subsets by wavelet scattering transform, and the fault features of the subsets are enhanced using Fisher discriminant analysis.Second, each frequency band subset is sent to different extreme learning machines under Bagging integration, and the classification accuracy of each fault mode in the frequency band subset is used as the class weight of the extreme learning machine.Then, the output values of each extreme learning machine are weighted to obtain the fused output result, and the fault class is determined accordingly.Finally, two example circuits are simulated, and the simulation results show that the diagnostic accuracy is 100%, indicating that the method is feasible and effective, and can realize fault classification and positioning.

The application provides a sedimentation tank flocculation sedimentation control method and device, electronic equipment and storage medium, including: obtaining raw water quality target parameters; based on the raw water quality target parameters, feature extraction and noise reduction processing are performed through an adaptive mode decomposition algorithm to obtain a feature vector; the feature vector is processed by using an extreme learning machine model optimized by an improved genetic algorithm to obtain a coagulant dosage prediction value; based on the coagulant dosage prediction value, coagulant is added to the raw water, and the operating parameters of the sedimentation tank are monitored; based on the operating parameters, the dry sludge amount in the sedimentation tank is determined through a dry sludge amount model; if it is determined that the sludge discharge condition is met based on the dry sludge amount and a pre-set sludge level threshold value, sludge is discharged; the effluent turbidity of the sedimentation tank is monitored in real time during the sludge discharge process, and the coagulant dosage is dynamically adjusted based on the effluent turbidity through the adaptive mode decomposition algorithm. In this way, the coagulant dosage accuracy and process synergy are improved, and the sedimentation tank is discharged on demand.

This invention relates to the field of navigation and timing technology, specifically to a GNSS timing data prediction method and related equipment. The method involves acquiring Roland timing data when GNSS timing data is invalid and preprocessing it. The preprocessed data is then input into a trained single-hidden-layer extreme learning machine model. This model calculates the hidden-layer output matrix by randomly initializing the input layer weight matrix and bias vector, and solves for the output weights using regularized least squares. The model outputs a normalized GNSS timing data prediction value, which is then denormalized to obtain the final GNSS timing prediction result. This method combines the high stability of the Roland system with the efficient computational power of the extreme learning machine to achieve accurate GNSS timing data prediction. It is suitable for enhancing the resilience of positioning, navigation, and timing systems in critical infrastructure scenarios, ensuring the continuous operation of the system even when GNSS signals are interfered with or fail.

The application discloses a kind of HBA-ELM-based wind turbine fault diagnosis method and device, belong to wind power generation technical field, the method includes: obtaining target data collected in target wind turbine by sensor, and the target data is preprocessed;The target data after preprocessing is input into the preset extreme learning machine model, and the fault type probability distribution is output;The fault type probability distribution is input into the preset fault classification output model, and the fault type diagnosis value is output;Wherein, the model parameter of the extreme learning machine model is based on the improved honey badger algorithm optimization generation of dynamic self-adaptive strategy into fusion.This application can effectively identify the fault type of wind turbine, improve the diagnosis precision and generalization ability, provide reference basis for the operation and maintenance of wind farm.

The invention discloses a multi-energy and carbon emission correlation analysis method and system for a high-energy-consumption enterprise, and relates to the technical field of energy management and carbon emission accounting, and the method comprises the steps: carrying out the preprocessing of historical multi-source data of a high-energy-consumption target enterprise, and constructing a training set and a test set which are distributed in a balanced manner; then introducing a shuffled frog-leaping optimization algorithm to perform global optimization on the network structure and parameters of the deep extreme learning machine, determining the optimal hidden layer structure and weight configuration of the model through an iteration mechanism of subgroup division, local jump and merge sorting, and constructing a multi-energy consumption prediction model with efficient deep feature extraction capability; a coupling feature matrix generated based on real-time multi-source data is input into a multi-energy consumption prediction model to obtain a multi-energy consumption prediction result, finally, association analysis is performed by combining real-time data, the prediction result and an energy carbon emission coefficient, the association strength of multi-energy and carbon emission is quantified, and the energy consumption prediction accuracy is improved. And more accurate and reliable data support is provided for energy structure optimization and emission reduction decision making of high-energy-consumption enterprises.

The application discloses a lithium battery residual service life prediction method based on a fusion data-driven model, and the method comprises the following steps: extracting the constant current charging duration, the constant voltage charging duration, the vertical slope at the corner of the constant current charging curve and the vertical slope at the corner of the constant current discharging curve as health characteristics; taking the four health characteristics as inputs and corresponding battery capacities as outputs to train a chaos sparrow-extreme learning machine model and a least square support vector regression model; obtaining the battery capacity prediction values by using the two trained models respectively, and performing weighted fusion to obtain the final lithium battery capacity prediction value; and finally obtaining the residual service life of the lithium battery in combination with the lithium battery capacity curve. The CSSA-ELM-LSSVR fusion algorithm can fully utilize the CSSA-ELM to extract the overall trend of the lithium battery degradation process, and utilize the LSSVR to obtain the local nonlinear characteristics, so that the accurate prediction of the residual service life of the lithium battery is realized, and the good robustness is also achieved.

The invention relates to a partial discharge fault diagnosis method and device, equipment, a medium and a product. The method comprises the following steps: performing feature extraction on sample partial discharge data of sample gas insulated switchgear under different fault types to obtain sample partial discharge features; constructing an initialized population; population individuals in the initialized population correspond to parameter combinations of to-be-trained parameters; updating individual positions of population individuals in the initialized population according to the sample partial discharge features to optimize the initialized population and obtain a target population; according to a target parameter combination in the target population, initializing model parameters of the extreme learning machine model; and performing model training on the initialized extreme learning machine model according to the sample partial discharge characteristics to obtain a fault diagnosis model which is used for diagnosing the fault type of the partial discharge signal. By adopting the method, the fault type of partial discharge can be accurately diagnosed.

The invention relates to a coal gun sound collaborative identification method and system based on voiceprint and vibration monitoring, and belongs to the technical field of coal mine safety. According to the invention, through the high-sensitivity voiceprint sensor and the three-axis vibration sensor, voiceprint and vibration information in a coal mine tunnel is synchronously acquired. And after the data is subjected to noise reduction preprocessing and multi-dimensional feature extraction, a classification model is established by adopting an extreme learning machine, and accurate recognition of the coal gun sound is realized. The system fuses voiceprint and vibration characteristics to accurately identify the coal gun sound, so that the problems of high subjectivity, low reliability and the like of manual identification are effectively solved, and technical support is provided for safe production of a coal mine.

The application provides a cable joint insulation deterioration non-contact diagnosis method based on infrared thermal imaging. Firstly, the insulation surface layer of the cable joint and the cable at both ends is subjected to infrared thermal imaging, so that the surface layer temperature of multiple symmetrical regions on both sides of the center of the cable joint and the surface layer temperature of the cable at both ends of the joint are non-contact collected. Secondly, a deep learning network based on a double-hidden-layer self-encoding extreme learning machine is constructed to mine deep-level implicit features in the surface layer temperature data, and the extracted deep implicit features are used as inputs of a random forest diagnosis model. Then, a quantum rotation gate with a nonlinear dynamic self-adaptive rotation angle is proposed to improve the update strategy of a quantum firework algorithm and is used for parameter optimization of the diagnosis model. Finally, a dataset is constructed by combining the joint surface layer infrared temperature and the insulation medium loss tangent value, and the diagnosis model is trained and tested on site.

This invention discloses a cable defect identification method and system based on an online sequence extreme learning machine, relating to the field of cable inspection technology. The method includes the following steps: when other defects exist in the cable under test, the processor acquires the location information of the wireless radio frequency signal transmitting unit and the wireless radio frequency signal receiving unit, activates the nearest image detection unit to collect an image of the cable under test based on the location information, and sends it to the processor. The processor converts the image of the cable under test into a corresponding partial discharge spectrum, constructs and trains an online sequence extreme learning machine recognition model, and processes the partial discharge spectrum using the online sequence extreme learning machine recognition model to obtain the defect identification result. This invention can simultaneously identify possible moisture defects, appearance defects, or other physical defects in the cable, improving the identification efficiency.

The application discloses application of a tensor ELM based on LDA and SSA in HSI classification, and comprises the following steps: step 1: inputting test tensor samples, training tensor samples, label samples and parameters; step 2: data processing part; step 4: obtaining tensor sample output of a hidden layer; step 5: initialization process, for each mode i; step 8: obtaining initial value of W (i) Step 16: obtaining final optimized classification result of a prediction test sample: test tensor sample A i After singular spectrum analysis denoising and a hidden layer of an extreme learning machine, output tensor B i of the hidden layer is obtained, and then, Γ i ∈R C×1 is obtained, and the serial number of the maximum element in Γ i is the classification result of the test tensor sample A i . The application can effectively utilize spatial sequence information of a hyperspectral image, can fuse spatial and spectral information into a tensor, can optimize the spatial and spectral information together, and can propose a tensor ELM model, so that the classification precision is higher and the calculation speed is faster in actual classification and other applications.

The invention discloses a high-proportion new energy short-circuit current prediction method based on a GWOGAC-IELMsin mixed model. The invention relates to a high-proportion new energy power grid short-circuit current intelligent prediction method. The method comprises the following steps: constructing a hybrid prediction model GWOGAC-IELMsin based on an improved grey wolf optimizer and an incremental extreme learning machine; fault transient signal features are rapidly extracted in a 0.2 ms time window through wavelet transform, a high-dimensional input vector containing a fault phase angle, a voltage and current instantaneous value and a change rate thereof, d / q axis current reflecting new energy power electronic equipment features and other features is constructed, and key features are screened through grey correlation analysis; optimizing an input weight and a hidden layer offset parameter of the incremental extreme learning machine by adopting an improved grey wolf optimization algorithm; and finally, dynamically adjusting the network structure through an incremental learning mechanism until the prediction precision requirement is met. The method effectively solves the three technical problems that in high-proportion new energy power grid short-circuit current prediction, a traditional model is insufficient in transient feature capture, an optimization algorithm is prone to local optimization, and prediction speed and precision are difficult to balance.

A kind of power grid stability evaluation method and system of large-scale electric vehicle access, the method first collects the power grid data set related to electric vehicle in power grid and carries out classification annotation, obtains original power grid data training set;Then based on the generation of antithesis network algorithm training data expansion model of riemannian manifold feature diffusion, obtain the expanded power grid data training set;Based on the limit learning machine algorithm of nonlinear system coupling learning, the expanded power grid data training set is input into the power grid stability evaluation model for training;Finally, the new power grid data set sample related to electric vehicle in power grid is input into the trained power grid stability evaluation model for evaluation, and the evaluation grade of power grid stability is obtained.The present application carries out stability evaluation modeling for the highly nonlinear and dynamic change characteristics of electric vehicle data in power grid, comprehensively reflects the multidimensional influence of electric vehicle access to power grid, and realizes the stability evaluation of large-scale electric vehicle access to power grid.

The present application relates to a blasting block size prediction method based on a machine learning and extreme learning machine hybrid model, comprising the following steps: A) obtaining original data of blasting, determining variable parameters thereof, and calculating skewness values of the variable parameters; B) classifying the variable parameters according to the skewness values, transforming the variable parameters of each class based on a set transformation method to generate new variable parameters, and processing the new variable parameters based on at least one machine learning method to generate a new feature data set; C) determining the number of hidden layer neurons of an extreme learning machine model according to the new feature data unit, obtaining the best extreme learning machine model network structure accordingly, and saving the corresponding weight parameter values in the best extreme learning machine model network structure; and D) predicting the average block size of blasting based on the new feature data set, the best extreme learning machine model network structure and the weight parameter values. The blasting block size prediction method based on the machine learning and extreme learning machine hybrid model can reduce the amount of calculation and has high precision.

The invention belongs to the technical field of load prediction, and particularly relates to a residential quarter short-term load prediction method, which comprises the following specific steps of: acquiring load data and meteorological parameters of a target community, and processing the load data and the meteorological parameters into sliding window characteristics; and inputting the sliding window features into a pre-trained hybrid prediction model, wherein the hybrid prediction model outputs a short-term load prediction result. According to the invention, through combination of fast nonlinear mapping of the extreme learning machine layer and potential distribution learning of the variational auto-encoder layer, efficient feature extraction of complex load fluctuation is realized; a residual module of the residual network further strengthens the expression ability of deep time sequence coupling through a residual connection structure, and significantly improves the adaptability of the model to load changes in different power consumption scenes in peak days and valley days.