57 results about "Recursive prediction" patented technology

The invention relates to the technical field of pipeline welding automation, and discloses a pipeline multi-layer and multi-pass welding track self-adaptive planning method. The method comprises the steps that weld joint track poses, welding electrical parameters and temperature information are collected in real time, and a sliding time window data sequence is constructed; establishing an interlayer constraint model and a trajectory prediction model based on the sequence, and generating a plurality of trajectory deviation predictions through short-term recursive prediction and long-term sequence prediction; a residual error reciprocal weighted fusion strategy is combined with a welding seam forming quality evaluation function, fusion track deviation is obtained, and uncertainty is evaluated; the fusion deviation is superposed to an original planned trajectory, and a self-adaptive correction trajectory is generated through a multi-objective optimization model; and the welding robot is controlled to execute track correction and real-time feedback updating. According to the method, through dual-time scale prediction and trajectory-process parameter collaborative optimization, the problem of insufficient welding seam forming precision caused by lack of dynamic correction in traditional static planning is effectively solved, and the welding quality and efficiency are remarkably improved.

The invention discloses a high-density NAND storage medium self-adaptive repairing method which comprises the following steps: a main control chip separates a target feature vector representing the real aging trend of a storage unit from original read data containing random physical noise; the main control chip deduces the target feature vector by using a full-integer recursive prediction model to obtain a health state prediction result of the storage unit in a future preset time period; wherein the health state prediction result comprises an optimal read reference voltage offset and an estimated bit error rate growth curve; and the main control chip adjusts the charge distribution pattern and programming voltage parameters of the data on the storage unit in the data writing stage according to the health state prediction result. By means of the mode, the problems of firmware assembly line blocking and performance jitter caused by floating point operation and huge model parameter loading can be solved under the limited hardware environment that the main control chip only supports integer operation and on-chip cache is extremely small.

The invention provides an LSTM underwater robot modeling method based on TPE hyper-parameter optimization, and relates to the technical field of underwater robot modeling, and the method comprises the steps: carrying out the time sequence feature extraction through a memory unit comprising a forgetting gate, an input gate and an output gate, and predicting the state change amount delta Yt'at a t + 1 moment through an output layer; performing iterative optimization on the number of layers and the number of units of the LSTM network and training the model to obtain an optimized LSTM model; evaluating the optimized LSTM model through a multi-step cumulative prediction error, inputting an initial real state into the model to carry out T-step recursive prediction, updating a current state by utilizing a prediction state increment in each step, and finally calculating an average position error and an attitude error in a T-step window on a verification set; and selecting the hyper-parameter combination with the minimum comprehensive error of the verification set as a final model parameter, and completing the dynamic modeling of the underwater robot. According to the method, the problem that the model is inaccurate due to excessive parameterization in a nonlinear dynamic model can be solved.

The invention discloses a regional carbon emission prediction method and system based on deep learning, and belongs to carbon emission prediction. The method comprises the following steps: acquiring historical carbon emission time sequence data and multi-dimensional constraint boundary data of a target area, and mapping the historical carbon emission time sequence data into an input feature vector; outputting an initial future carbon emission evolution trajectory vector; identifying and extracting a constraint violation dimension; constructing an input feature compensation tensor; generating an updated input feature vector; inputting the updated input feature vector into the depth space-time prediction model again for recursive prediction; and outputting a target carbon emission trajectory vector meeting the multi-dimensional constraint boundary data and a corresponding feature adjustment parameter set. According to the method, by automatically analyzing the multi-dimensional management constraint and fusing the multi-dimensional management constraint into the iterative optimization closed loop of the prediction model, the problem that the prediction model in the prior art only optimizes mathematical statistical errors and is seriously disjointed with multi-dimensional targets such as cost control and path feasibility in real management is solved.

The invention relates to the technical field of environment monitoring, in particular to an urban carbon emission intelligent prediction method based on multi-modal data and machine learning integration, which comprises the following steps: feature acquisition and preprocessing: acquiring multi-source heterogeneous data and processing the multi-source heterogeneous data into a standardized data set; constructing an LSTM (Long Short Term Memory) and XGBoost hybrid prediction model; predicting future data on the basis of existing data, and predicting input features of future time steps of the weight integration prediction model on the basis of historical data in a standardized data set; and predicting carbon emission in the future. Industrial structure indexes, environment variables and social and economic substitution indexes are integrated through time characteristic engineering, and emission driving factors in different urban environments can be comprehensively captured; a recursive prediction mechanism is established, development trajectories of different cities are considered, reliable emission prediction can be carried out, meanwhile, time consistency is kept, and prediction uncertainty is quantified; interference influences under different policies are quantified, and a quantitative basis is provided for carbon management decisions.

The invention relates to the technical field of computer software, and discloses a large-scale multi-agent collaborative reasoning method and a social network simulation system. The method comprises the steps that users in a social network are simulated through intelligent agents, figure portraits are distributed to all the intelligent agents, and an intelligent agent social network topology is constructed; calculating entropy values of the vertical field numerical values of all the intelligent agents in the previous time step; selecting an intelligent agent with a preset proportion as a core intelligent agent of the next time step; the core agent is driven by a large language model; the conventional agent is driven in parallel by adopting a graph attention situation deduction model based on a graph attention network, and a standing numerical value of a next time step is recursively predicted by aggregating static features and dynamic features of neighbor agents; mapping the standing text of the core agent into a standing value through a scoring device, and replacing the predicted standing value of the current core agent; according to the method, an agent interaction mechanism conforming to social laws can be more effectively simulated.

The invention belongs to the technical field of distributed photovoltaic power prediction and power distribution network operation, and discloses a distributed photovoltaic power ultra-short-term prediction method based on multi-source data fusion. The method comprises the following steps: collecting and aligning historical power and electric parameters of a photovoltaic station, meteorological monitoring, short-term and imminent forecasting, voltage of a power distribution network, power flow and states of a voltage regulating device under a unified time dimension; based on a photovoltaic output mechanism model and a linearized power flow sensitivity model or an equivalent electrical distance model, constructing a power constraint interval and a baseline power track which meet voltage and line thermal stability constraints; and through residual normalization, weighted adjacent matrix and space-time recursion prediction model modeling and prediction of a multi-station residual field, parameter increment is updated in combination with a data quality index and an error threshold, and distributed photovoltaic cluster ultra-short-term power prediction giving consideration to power distribution network security constraints and data quality is realized.

The invention relates to the technical field of electro-hydraulic servo systems, in particular to a control method and system for a single-cylinder electro-hydraulic position servo system based on reinforcement learning, and the method comprises the steps: building a single-cylinder electro-hydraulic position servo system model; constructing a DDPG algorithm architecture, and defining a reward function R; constructing a target future trajectory prediction model based on dynamic recursive prediction and multi-modal information fusion, and generating a target future trajectory; training an intelligent agent of the DDPG algorithm; the observation data of the intelligent agent comprises a current position, a control error of the system, an accumulated error of the system and a future trajectory of the target; and based on the trained intelligent agent, the output action is controlled to obtain continuous action signals within a set threshold range, and control over the single-cylinder electro-hydraulic position servo system is completed. Through the control method and the control system, the highly nonlinear dynamic characteristic in the single-cylinder electro-hydraulic position servo system can be fully dealt with, and the control precision, the stability and the robustness of the intelligent agent under the complex working condition are improved.

The invention relates to the technical field of artificial intelligence and computer vision, in particular to a time sequence remote sensing image prediction method and related equipment, and the method comprises the steps: firstly obtaining historical multi-frame remote sensing images of a target area, and constructing a time-space sequence data set; the model is a ViT-Informer model, and a ViT spatio-temporal feature extractor captures spatio-temporal correlation features among images through a self-attention mechanism to generate a spatio-temporal joint coding sequence; the Informer time sequence predictor adopts a probability sparse attention mechanism to process long sequence dependence, and multi-step recursive prediction is achieved. The model can effectively model a remote sensing image spatio-temporal evolution law, and a prediction image at a specified moment in the future is generated through iterative reasoning.

Embodiments of the present disclosure relate to a combination of a conditional and marginal model, where the conditional model provides its conditional frame prediction as input to the marginal model. Various embodiments leverage an incremental diffusion process to insert the predicted frame by mixing it with noise and starting the diffusion process part way or at some intermediate level. Some embodiments also minimize the propagation of errors introduced in the process of video generation by recursive prediction of video frames.

The application discloses a PID control optimization method for a water turbine regulating system based on a PINN, which comprises the following steps: constructing and training a physical information neural network; deploying the trained physical information neural network to an online inference system, and simultaneously, equipping a PID controller as a main control loop of the water turbine regulating system; the PID controller outputs a real control signal, and performs state evolution based on a multi-state mechanism model and outputs a real-time state feedback signal; the real-time state feedback signal is input into the physical information neural network for state recursive prediction and output of a predicted state feedback signal; the real-time state feedback signal and the predicted state feedback signal are continuously compared, and the accuracy and reliability of the physical information neural network under the current operating condition are evaluated; a bypass verification loop is constructed based on the evaluation result, and the main control loop and the bypass verification loop jointly form a closed-loop intelligent control system. The application improves the control accuracy and operation reliability of the water turbine regulating system under nonlinear and time-varying operating conditions.

The invention discloses a power battery remaining service life prediction method based on big data analysis, and belongs to the technical field of battery health management. The method comprises the steps that battery operation time sequence data, vehicle working condition data and environment data are synchronously collected and preprocessed; second-level transient characteristics and stroke-level working condition characteristics are extracted, and calendar aging accumulation characteristics for quantifying high-temperature standing and high-charge state keeping duration are constructed; constructing a physical information enhanced deep residual space-time network prediction model, and guiding model learning to accord with a physical rule by introducing a physical consistency constraint term based on an experience degradation model into a loss function; and a transfer learning strategy combining cloud pre-training and edge side lightweight fine tuning is adopted to realize personalized adaptation of the model and dynamic recursive prediction of the remaining service life. According to the invention, the prediction precision and the model generalization ability are improved, and the precise modeling of the cyclic aging and calendar aging coupling effect is realized.

The application discloses a material performance evolution trend prediction method and device and a storage medium, relates to the technical field of material science, and mainly aims at improving the determination efficiency and accuracy of the material performance evolution trend. The method comprises the following steps: in response to a material performance evolution trend prediction signal of a target material in an aging process, obtaining aging environment information of the target material and historical material performance of a previous aging time point corresponding to a current aging time point; taking the current aging time point and a plurality of future aging time points as to-be-predicted time points, inputting the historical material performance and the aging environment information into a preset performance prediction model to gradually perform material performance recursive prediction of each to-be-predicted time point, and sequentially obtaining material performance of each to-be-predicted time point; and determining the material performance evolution trend of the target material in the aging process based on the historical material performance and the material performance of each to-be-predicted time point.

The invention provides a spatio-temporal network multi-physics field prediction method and system based on physical information and double attention, and the method comprises the steps: obtaining thermal process multi-physics field spatio-temporal sequence data generated based on high-fidelity three-dimensional CFD simulation, and carrying out the fusion alignment of the data and actual measurement time sequence data of limited measurement points of an industrial site, forming a standardized multi-channel training sample; constructing a basic physical information depth space-time prediction model, wherein the basic physical information depth space-time prediction model comprises a double attention fusion module, a physical perception hybrid encoder and a physical constraint multi-field decoder; training the basic physical information depth space-time prediction model by adopting the standardized multi-channel training sample to obtain a target physical information depth space-time prediction model; and inputting to-be-predicted data into the target physical information depth space-time prediction model by adopting a sliding window, and recursively predicting future multi-physical field evolution. The problems that traditional numerical simulation is poor in real-time performance, a pure data driving model is insufficient in physical consistency, and whole-field measurement of an industrial site is difficult are effectively solved.

The invention belongs to the technical field of channel prediction, discloses a Transform channel prediction method based on complex value modeling and multi-domain fusion, and constructs a channel prediction neural network model integrating complex value modeling, multi-dimensional feature extraction, cross-domain fusion and parallel time modeling. Wherein the channel prediction neural network model comprises a complex value perception feature module, a multi-dimensional attention module, an iterative attention fusion module and an improved Transform structure. According to the Transformer channel prediction method based on complex value modeling and multi-domain fusion, the error accumulation problem of traditional recursive prediction can be effectively solved, the cross-domain feature expression, time dynamic modeling and long-time prediction precision of the channel state information is comprehensively improved, and the method is suitable for efficient prediction of the channel state information.

The invention discloses a multi-model adaptive enterprise carbon emission general prediction method based on reinforcement learning, and relates to the technical field of enterprise carbon emission prediction. Comprising the following steps that multiple base models with different modeling characteristics are pre-trained, a reinforcement learning meta-model based on near-end strategy optimization is constructed, and the reinforcement learning meta-model outputs the weight and error correction value of each base model through a strategy network; according to the method, a multi-model adaptive integrated architecture based on reinforcement learning is utilized, a gradient boosting decision tree, a random forest and a long-short-term memory network are taken as pre-training base models, the weight of each base model is dynamically adjusted through reinforcement learning meta-models, an error correction term and a recursive prediction mode are combined, and maximization of long-term accumulated rewards is taken as an optimization target; the technical problems that in an existing enterprise carbon emission prediction method, a single model is difficult to capture complex features of a multi-energy system, the generalization ability of the model is limited, and short-term error accumulation influences prediction precision and is insufficient in robustness are solved.

The invention provides a flying robot trajectory control method and system based on data-driven generalized iterative prediction. The method comprises the following steps: S1, constructing a flying robot data-driven model; s2, by introducing a nominal dynamic linearization system, decoupling pseudo partial derivative estimation of the system from external disturbance, and estimating a pseudo partial derivative parameter at the current moment; s3, designing a model prediction framework, and predicting pseudo-partial derivatives of multiple iterations in the future by using historical pseudo-partial derivative parameters in combination with multi-layer recursion through a multi-stage hierarchical prediction method; s4, designing a generalized iterative prediction controller, establishing a performance index function containing tracking errors and control input increments, and solving an optimal control input sequence in real time through an optimization algorithm to realize position control of the flying robot; and S5, designing an iterative extended state observer, estimating external disturbance in the system, introducing a disturbance compensation item into a control law, and enhancing the robustness and anti-disturbance capability of the system in an interference environment.

The present application relates to a kind of residual agricultural film water pollution prediction methods based on time delay embedding timing recursive network, comprising the following steps: S1: regional multi-source data acquisition and dataset construction, provide input data for prediction model;S2: build timing recursive prediction model, based on the timing recursive of multilayer gate control cycle unit GRU and time delay embedding mechanism builds prediction model, establishes the dynamic transmission relationship between residual agricultural film, soil characteristics and water pollution three kinds of medium;S3: pollution transmission chain construction and dynamic prediction, in prediction model, introduce intermediary variable and environmental driving factor, build from residual agricultural film to soil characteristics, again to water pollution, realize the dynamic prediction and response analysis of pollution process.The present application can show the time lag relationship between pollution release and water quality response, simulate the timing recursive process of residual agricultural film pollution between multiple media, provide reliable technical support for water pollution risk early warning and management decision.

The invention discloses a hypersonic flow field density and speed high-precision prediction method, system and device based on physical information constraint U-Net, and a medium. The method comprises the following steps: obtaining value distribution of instantaneous transverse and longitudinal speeds and densities in target wake flow fields with different Mach numbers; forming a data set by the densities, the transverse speeds and the longitudinal speeds of two continuous moments and the density, the transverse speed and the longitudinal speed of the third moment in the target wake flow field at different moments, and dividing the data set into a training set and a test set; constructing a physical information U-Net flow field prediction model, and defining a continuity equation in an N-S equation as a loss function of the model; training the physical information U-Net flow field prediction model to obtain a physical information U-Net flow field prediction model of which the weight is trained; predicting input data by using the physical information U-Net flow field prediction model of which the weight is trained, and obtaining a prediction result through recursive prediction; the system, the equipment and the medium are used for implementing the method. By means of the method, the output data can meet the physical mechanism, and the accuracy is high.

The invention discloses a hemodynamics calculation method based on an infinitesimal neural operator, which comprises the following steps: realizing time shift prediction of an intravascular flow field through a pre-trained prediction network based on an initial flow field, vascular geometry and boundary conditions; the prediction network realizes efficient calculation of hemodynamic analogue simulation of different computational domains by designing infinitesimal neural operators with infinitesimal correlation and invariable displacement; the prediction network comprises a lifting layer, an inner block layer and a projection layer; the prediction network is trained in a supervised learning mode, a mean square error of multi-step recursive prediction is minimized under a multi-step circulation framework, and a universal time shift operator is learned by decomposing a hemodynamics problem into subproblems related to infinitesimal elements, so that two-dimensional or three-dimensional rapid simulation is realized. Compared with the prior art, the method can adapt to various computational domains without retraining, the computational efficiency and precision are remarkably improved, and the method is suitable for the field of medical blood flow simulation.

The invention discloses a long-term dynamics prediction method based on a track. The method comprises the following steps: collecting state-action sequence data; dividing the collected data into a plurality of sub-tracks, and constructing a training data set; designing a trajectory prediction model which directly predicts the state of the future time sequence by taking the initial state, the control parameters and the displayed future time sequence as input; and training the trajectory prediction model to accurately predict the long-term trajectory. The invention discloses a trajectory-based long-term dynamic prediction method, which is used for directly predicting a future multi-step state to replace single-step recursive prediction, remarkably reducing long-term prediction errors and improving data efficiency and calculation speed. The method can be widely applied to the fields of robot control, automatic driving and the like.

The invention discloses a multi-component reaction solute transport simulation method based on a physical information neural network. Establishing a multi-component convection-dispersion-reaction (ADR) equation model; constructing a double sub-network, and loading a Runge-Kutta coefficient so as to complete implicit time advance; calculating a solute concentration gradient change rate and a reaction intensity change rate which are respectively used as two types of self-adaptive control indexes, and dynamically adjusting a time step length; and training iteration is carried out by adopting an Adam and L-BFGS combined optimization algorithm, so that recursive prediction of the spatial and temporal distribution of the pollutant concentration is realized. According to the method, the time resolution can be automatically adjusted, the calculation efficiency and the numerical stability are both considered, compared with a traditional PINN and time marching PINN (TM-PINN) method, the long-term prediction precision and the convergence performance are remarkably improved, and the method is suitable for underground water pollution migration simulation and multi-component reaction-diffusion process modeling.

The application discloses a high-density NAND storage medium adaptive repair method, which comprises the following steps: a master control chip separates a target feature vector representing a real aging trend of a storage unit from original read data containing random physical noise; the master control chip deduces the target feature vector by using a full integer recursive prediction model to obtain a health state prediction result of the storage unit in a future preset time period; wherein the health state prediction result comprises a best read reference voltage offset and an estimated bit error rate growth curve; and the master control chip adjusts the charge distribution form and the programming voltage parameter of data on the storage unit in the data writing stage according to the health state prediction result. In the above manner, the firmware pipeline blockage and performance jitter problems caused by floating point operation and large model parameter loading can be eliminated in a restricted hardware environment where the master control chip only supports integer operation and the on-chip cache is extremely small.

The invention belongs to a sky wave radar target detection technology, and provides a sky wave radar long-time accumulation Doppler compensation method, which comprises the following steps: initializing a covariance matrix based on a measurement sequence, and adaptively estimating the number of frequency modulation modes contained in a signal; estimating an initial amplitude and an initial frequency of each mode through a weighted covariance fitting method; constructing a state vector for each mode, wherein the state vector comprises the amplitude and the current phase of the current moment and the amplitude and the phase of the historical moment; constructing a state transition equation based on a polynomial prediction model, and performing recursive prediction and updating on the state vector by using unscented Kalman filter (UKF); extracting instantaneous amplitude and instantaneous frequency of each mode according to the updated state vector; doppler compensation is carried out on radar echoes based on the instantaneous amplitude and the instantaneous frequency, long-time coherent accumulation is carried out on compensated signals, the compensated distance-Doppler spectrum is output, and the method has the advantages of low calculation complexity, good real-time performance and high resolution.

The invention relates to the technical field of computer data processing, in particular to a controller fault detection and fault-tolerant result output method based on operation data, which comprises the following steps: acquiring an operation quantity measurement value in a preset sampling period to generate an operation observation record; inputting the operation observation record into a multivariable state space recursive prediction model based on extended Kalman filtering to obtain a corresponding predicted operation amount; generating three residual errors according to the difference value of the measured value and the predicted value, and extracting residual error morphological characteristics in a sliding window of continuous N sampling periods to form a characteristic vector; checking the consistency of the feature vector and pre-stored normal and two fault mode feature sets to obtain a consistency score; and comparing the output label according to the threshold value and the score, and generating fault-tolerant result output data. According to the method, recursive prediction, residual morphological feature extraction and mode consistency check are carried out on multiple operation quantities of the controller, and fault-tolerant result data are output to improve the accuracy of fault recognition.

The invention discloses a hybrid model predictive control method and system for controlling a motor, and relates to the technical field of machine learning, and the method comprises the following steps: S1, employing a machine learning technology to construct a hybrid dynamic model; s2, performing multi-time-domain state prediction by using the hybrid dynamic model; s3, performing real-time cost quantification by using the prediction state sequence; s4, performing control parameter solving by using the optimal state response path; and S5, performing model error compensation by using the execution parameter set. Multi-time-domain state prediction constructed based on a machine learning technology is set, continuous variable evolution and discrete mode conversion of a motor system are described in a unified state space at the same time, the structure is based on a unified hybrid dynamic function, recursive prediction of continuous-discrete variables is achieved, and the reliability of the motor system is improved. And a Kalman prediction correction technology is combined to carry out error updating on a continuous state, and a logic constraint is applied to a discrete state, so that the prediction precision and the dynamic response capability of a machine learning system under a complex working condition are improved.

The application relates to a clock difference prediction method and device, computer equipment and a readable storage medium, comprising: obtaining a to-be-processed clock difference sequence; performing data preprocessing on the to-be-processed clock difference sequence to obtain a standardized residual error sequence; processing the standardized residual error sequence by using a probability-weighted fuzzy time sequence model to obtain a probability-weighted fuzzy time rule set; performing recursive prediction in combination with the fuzzy time rule set to obtain a prediction value set matched with the fuzzy time rule; performing defuzzification processing on the prediction value set to obtain a prediction residual error; performing inverse standardization processing on the prediction residual error and combining the prediction residual error with a trend item fitted and extracted in the data preprocessing process to obtain a target clock difference prediction result. The application eliminates data noise and magnitude difference through data preprocessing, fully captures sequence time sequence characteristics by using a probability-weighted fuzzy time sequence, can effectively improve clock difference prediction accuracy, and provides a reliable clock difference prediction solution for high-precision real-time positioning of a satellite navigation system.

The invention relates to a clock error forecasting method and device, computer equipment and a readable storage medium. The method comprises the following steps: acquiring a to-be-processed clock error sequence; performing data preprocessing on the to-be-processed clock error sequence to obtain a standardized residual error sequence; processing the standardized residual error sequence by using a probability weighted fuzzy time sequence model to obtain a probability weighted fuzzy time rule set; recursive prediction is carried out in combination with the fuzzy time rule set, and a predicted value set matched with the fuzzy time rule is obtained; performing defuzzification processing on the prediction value set to obtain a prediction residual error; and carrying out inverse standardization processing on the prediction residual error, and combining the prediction residual error with a trend term extracted by fitting in a data preprocessing process to obtain a target clock error prediction result. According to the method, data noise and magnitude difference are eliminated through data preprocessing, and sequence time sequence characteristics are fully captured by adopting the probability weighted fuzzy time sequence, so that the clock correction forecasting precision can be effectively improved, and a reliable clock correction forecasting solution is provided for high-precision real-time positioning of a satellite navigation system.

The application discloses a kind of based on mesoscopic level features asphalt mixture dielectric constant stratified recursive prediction method, including obtaining the dielectric constant measured data of each component in asphalt mixture multistage dispersion system, asphalt mixture is characterized as the three-phase composite medium consisting of asphalt, aggregate and air, and the volume fraction of each phase medium is calculated;According to the composition relationship of mesoscopic level, asphalt mixture is divided into mortar layer, mortar layer and mixture layer from bottom to top, based on the dielectric constant measured data of each component and the volume fraction of three-phase composite medium, the dielectric constant estimation equation of mortar layer, mortar layer and mixture layer is established respectively, and the index parameter and high dielectric phase critical volume fraction in the estimation equation are solved, so as to obtain the dielectric constant estimation model of each level;Further, the recursive way from bottom to top is used to obtain the composite dielectric constant prediction value of asphalt mixture.The application can improve the stability and precision of asphalt mixture dielectric constant prediction, and provide reliable parameter support for engineering detection results such as ground penetrating radar.

The invention relates to a performance parameter prediction method and system based on a multilevel quantile recurrent neural network, and belongs to the field of data processing. Comprising the steps that an equipment parameter table is acquired for data preprocessing to obtain target data, and the target data is a detailed table of time and prediction parameters; inputting target data into the trained multilevel quantile recurrent neural network model to obtain model output, placing a decoder with shared parameters in each loop layer in an encoder to create multi-step time sequence prediction, and calculating loss by adopting a quantile loss function, a global multi-layer perceptron and a local multi-layer perceptron are used for sensing at the decoder; and obtaining prediction parameters based on model output. According to the method, on the basis of the recurrent neural network, multi-step time sequence prediction and quantile regression are combined, a bifurcated sequence training strategy is used, accurate quantiles are predicted, the model is more stable, error accumulation of the recurrent prediction strategy is avoided, and the defect that the accumulated error of long-time prediction is too large is overcome.