47 results about "Ensemble prediction" patented technology

The invention discloses a medium and long term probabilistic ensemble forecasting method for enteromorpha green tide, and belongs to the field of enteromorpha green tide forecasting. The method comprises the following steps: a, acquiring longitude, latitude and initial occurrence time of an initial occurrence position of enteromorpha green tide based on historical monitoring data, and setting a potential jettisoning area and an initial window period; constructing a space-time probability density function of a jettisoning area, and generating an initial field set scheme of enteromorpha green tide medium and long term drift prediction; b, acquiring multiple groups of medium-and-long-term marine environment forecast driving field data, and constructing and forming a multi-driving field set scheme; and c, substituting the initial field set scheme and the multi-drive field set scheme into a Lagrange drift prediction model, performing cyclic simulation, obtaining a drift trajectory of each enteromorpha green tide particle under all set schemes, and generating an enteromorpha green tide medium and long term set prediction result. According to the method, medium and long-term forecasting of enteromorpha green tide can be carried out, the accuracy and reliability of forecasting are remarkably improved, and a scientific basis is provided for disaster prevention and reduction decisions.

The invention discloses a cloud cluster structure dynamic analysis and forecasting method based on local set rapid assimilation, and the method comprises the steps: carrying out the hour-by-hour forecasting of the evolution process of different types of cloud cluster structures in a target region through employing a local wrf-chema mode, and obtaining a mode short-term and imminent forecasting result; acquiring real-time observation data of the evolution process of different types of cloud cluster structures in the target area; calculating short and temporary forecast errors of evolution processes of different types of cloud clusters; generating disturbance ensemble forecasting members based on short and temporary forecasting errors of evolution processes of different types of cloud clusters; the perturbation ensemble prediction members are combined with real-time observation data through a 4D-LETKF ensemble assimilation method, the perturbation ensemble members are updated, and a more accurate cloud cluster structure evolution prediction result is obtained. According to the method, the uncertainty characterization capability of ensemble forecasting can be enhanced while the accuracy of short-term and imminent forecasting is remarkably improved, and the assimilation capability of multi-source data is quickly improved.

The invention provides a heterogeneous grid four-dimensional set variation assimilation forecasting method and device which can be applied to the technical field of atmospheric science and numerical mode application. The method comprises the following steps: pre-processing weather prediction data and a plurality of weather observation data of a target area to obtain prediction pre-processing data and observation pre-processing data; performing grid division on the map of the target area to obtain a grid map; mapping the prediction preprocessing data to a gridding map to obtain an initial field gridding map; based on the initial field gridding map, performing ensemble forecasting on the weather of the target area by using a cold start method to obtain ensemble forecasting data; based on the ensemble forecast data and the multiple weather observation data, performing four-dimensional variation assimilation analysis on the gridding map to obtain an assimilation analysis field; and based on the assimilation analysis field, predicting the weather of the target area to obtain a non-uniform grid four-dimensional set variation assimilation result.

The invention relates to the technical field of water conservancy projects and emergency management, and discloses a cascade reservoir group flood emergency scheduling method and system fused with dynamic risk assessment, and the method comprises the steps: a dynamic risk assessment stage, employing an ensemble forecast product of numerical weather forecast to drive a hydrological model to generate a flood set, calculating the instantaneous dam overtopping risk probability of each reservoir, and calculating the flood overtopping risk probability of each reservoir; a risk chain type propagation network model is constructed to evaluate the overall accident probability of the system; and the intelligent decision support stage is triggered when the overall failure probability of the system exceeds a risk threshold value, similar historical cases are retrieved by adopting a graph neural network based on the water conservancy field knowledge graph, and an optimal disposal scheme is generated and recommended through a multi-attribute utility evaluation model. According to the method, dynamic quantification of risks, accurate identification of systematic risks and intelligence of emergency decision making are achieved, the problems that traditional risk assessment is static and isolated, and emergency decision making depends on experience are solved, and the flood control emergency response capacity of the cascade reservoir group is improved.

A model weight mixing-based online knowledge distillation method and system, comprising: 1) the server collects labeled data, which is divided into a training set and a validation set; 2) select N student models with the same architecture, and construct a weight mixing model through linear weighting of student parameters; 3) select data augmentation to expand the training set into N+1 parts, which are respectively input into N student models and the weight mixing model to obtain N+1 predicted outputs and their average ensemble prediction; 4) calculate the classification loss of each student and the weight mixing model; 5) calculate the knowledge distillation loss between the output of each student and the ensemble prediction; 6) calculate the parameter optimization target of each student and update the parameters of the N students; 7) repeat steps 3-6, after each round, reconstruct the mixed weight model according to the updated student, and every fixed round Δ, fuse the parameters of the mixed weight model into the student; 8) select the model with the best performance on the validation set and deploy it to the terminal device.

The invention discloses a reservoir group risk adaptive scheduling method and system for coupling ensemble forecast uncertainty, and the method comprises the steps: fusing monitoring and meteorological ensemble forecast data in real time, and outputting an initialization field through assimilation processing; a complete scene set is generated through parallel hydrological-hydrodynamic simulation, and a dynamic risk probability field is calculated and clustered to obtain a weighted typical scene set; then, a rolling optimization model containing dynamic probability constraints is constructed, and an optimal instant scheduling instruction is solved and output; and updating data in a rolling manner according to a fixed period and repeating the process. The system adopts a modular design and comprises a data synchronization assimilation module, a parallel simulation and risk quantification module, a rolling optimization decision module, an instruction execution and feedback control module and a system support layer. According to the invention, the scheduling strategy adapts to the risk dynamic change in real time, the safety and benefit are balanced, the scheduling intelligence and standardization level are improved, and the engineering potential safety hazard is reduced.

The invention discloses a photovoltaic snow covering depth and power loss forecasting method and system based on multi-source data fusion, and belongs to the technical field of photovoltaic power forecasting. The problems of high-precision accumulated snow influence assessment, short-term risk early warning and power loss quantitative prediction of a photovoltaic system in cold and alpine regions are solved. The method comprises the following steps: selecting three numerical weather forecasting modes, namely a global forecasting system operated by the National Environmental Forecasting Center, a wind energy and solar energy forecasting system of the China Meteorological Administration and a unified mode weather forecast of the British Meteorological Administration, acquiring weather forecasting data, performing unified space-time interpolation, abnormal value correction and unit conversion, and then performing set averaging to obtain a set average value; ensemble forecast data is obtained; based on a temperature-irradiance condition and a friction force mechanism, an accumulated snow covering model is constructed, a photovoltaic panel snow covering range and photovoltaic power loss are obtained, early warning information of snow covering time, duration and power loss is generated through matching of station longitude and latitude and national grid forecasting, and photovoltaic snow covering depth and power loss forecasting based on multi-source data fusion is completed.

This invention discloses a multi-scale power prediction and uncertainty quantification method for photovoltaic-storage-charging microgrids, belonging to the field of power dispatching technology. The method includes: acquiring historical operational data and meteorological data of the photovoltaic-storage-charging microgrid; extracting different time-frequency features through a multi-scale decomposition algorithm; establishing a nonlinear mapping relationship through a deep neural network prediction model to determine the optimal network structure and hyperparameter combination; quantifying prediction uncertainty using a Bayesian deep learning framework to obtain prediction confidence intervals; training multiple base learners with different structures and obtaining ensemble prediction results through an ensemble learning algorithm; and obtaining a power prediction solution for the photovoltaic-storage-charging microgrid by constructing prediction intervals and combining them with point prediction accuracy indicators. This invention achieves high-precision power prediction and accurate uncertainty quantification for photovoltaic-storage-charging microgrids, providing reliable decision support for the safe and stable operation of distribution networks.

The application discloses a CMA-BJ-EN convection-resolvable ensemble prediction method and system, which comprises the following steps: collecting observation data of a target area, performing global ensemble data preprocessing on the observation data to obtain an initial analysis field, and performing preprocessing and disturbance on the observation data to obtain disturbed observation data; performing ensemble data assimilation on the initial analysis field and the observation data based on three-dimensional variation, controlling a prediction initial value, a disturbed prediction initial value and obtaining side boundary conditions, performing integral prediction by adopting a convection-resolvable numerical mode, and introducing random physical process disturbance in the integral process to obtain a prediction result of an ensemble member; and performing seasonal post-processing on the prediction result to generate an ensemble prediction product.

The application discloses a runoff set prediction method based on spatial heterogeneity and variable attention, comprising the following steps: step 1, data collection and processing; step 2, spatial partitioning of a basin; step 3, adaptive feature screening of a prediction period; step 4, two-stage runoff sequence decomposition; step 5, variable attention modeling; and step 6, two-stage reconstruction based on physical interaction coupling. The application can fully utilize the internal spatial heterogeneity information of a basin, extract multi-scale time sequence features, realize collaborative modeling of meteorological driving and runoff response, fully utilize the internal spatial heterogeneity information of the basin, realize high-quality feature construction for multi-scale prediction, and significantly improve the learning ability of the model for complex hydrological processes and the prediction accuracy of extreme runoff events.

This invention discloses a classification method for ultra-high carbon steel microcrystals based on spatial attention and ensemble prediction, relating to the field of metallic materials technology. The method includes: acquiring K improved EfficientNet-B7 models; these K improved EfficientNet-B7 models are obtained by training with K-fold cross-validation based on the initial improved EfficientNet-B7 model structure; the initial improved EfficientNet-B7 model adds a spatial attention module between the output of the sixth feature extraction stage and the input of the seventh feature extraction stage; classifying the microcrystal image of ultra-high carbon steel to be classified using each improved EfficientNet-B7 model to obtain the probability distribution of each category; arithmetically averaging the multiple probability distributions predicted by the multiple improved EfficientNet-B7 models to obtain a fused probability vector; and selecting the category corresponding to the maximum probability in the fused probability vector as the category of the microcrystal image to be classified. This method improves the classification accuracy of ultra-high carbon steel microcrystal structures.

The invention discloses a high-resolution rainfall ensemble forecasting method, system and equipment and a medium. The method comprises the following steps: acquiring a historical atmosphere reanalysis data set and a multi-source fusion rainfall data set; the constructed deterministic neural network learning model is trained to output target duration accumulated average rainfall; training a probability precipitation model constructed by a variational auto-encoder model and a denoising diffusion probability model; acquiring meteorological data at a to-be-predicted moment, and inputting the meteorological data into the deterministic rainfall model to obtain an accumulated average rainfall prediction result within a target duration; sampling from Gaussian distribution to obtain noise, and performing denoising processing in a hidden space by using the trained denoising diffusion model; the denoising result is restored by the trained variational auto-encoder model and added with the output of the deterministic model to obtain the accumulated rainfall of the target time length; and constructing ensemble forecast based on the multiple groups of target duration accumulated rainfall forecast. According to the method, high-resolution rainfall distribution can be generated, the extreme rainfall prediction accuracy is improved, and the uncertainty of rainfall is evaluated.

The application discloses a cold vortex path set prediction method and device, relates to the technical field of meteorological data processing, and comprises the following steps: determining the real-time position of a cold vortex based on the latest zero field data of a global model; selecting N member paths with the minimum path error in each member cold vortex path of set prediction to perform arithmetic averaging, so as to obtain a corrected path, wherein the path error is the distance between the current time cold vortex center position of set prediction and the real-time position; obtaining the prediction data of the center position and path of the cold vortex of all prediction members of set prediction, and performing path clustering through a DSBCAN algorithm. Cold vortex path prediction products are determined based on path clustering and the corrected path, and objective and accurate prediction results can be obtained based on the products combined with the results after the 500-hPa situation field of an initial date is clustered. Further, the defects in the prior art are solved.

The present application relates to an interpretable and machine learning-based climate model preferably combined with a collection estimation method, belonging to the technical field of climate models, the method comprising: obtaining area-weighted average ozone column total of reanalysis data and CMIP6 model data, and constructing a data set; based on the data set, respectively adopting a traditional method and a decision tree SHAP method to screen the CMIP6 model; based on the CMIP6 model data screened by the traditional method and the decision tree SHAP method, a machine learning model is constructed; based on the constructed machine learning model, the optimal screening method and the corresponding optimal machine learning model are determined; based on the optimal screening method and the corresponding optimal machine learning model, the Antarctic ozone under different greenhouse gas emission scenarios is estimated. Through the present application, more accurate Antarctic ozone recovery estimation results can be obtained.

The invention discloses a cascade hydropower station reservoir runoff prediction and optimal scheduling method and system based on land-gas coupling, and the method comprises the steps: fusing satellite remote sensing, glacier quality change, numerical weather forecast and other multi-source data through data assimilation, and generating a meteorological and hydrological ensemble forecast; inputting the ensemble forecast into the graph neural network model, and outputting short-term runoff probability distribution and a middle-term expectation sequence; a hierarchical optimization scheduling strategy is adopted; a two-stage stochastic programming model is established based on probability distribution to formulate a water level datum line; in actual operation, forecast distribution is updated based on latest observation data, and a distributed robust optimization model is adopted to generate a robust real-time scheduling instruction in a rolling manner; integrally calibrating parameters of the prediction model and the decision model on line through a variational inference method; according to the method, full-process closed loop from multi-source data fusion and probabilistic prediction to robust decision and adaptive learning is realized, and the precision and physical rationality of runoff forecasting and the comprehensive benefit and robustness of scheduling decision are improved.

The invention discloses a cloud analysis and random perturbation combined method for regional convective scale ensemble forecast in the technical field of meteorological data processing, and the method comprises the steps: obtaining cloud analysis data, such as blackbody brightness temperature, total cloud amount and radar three-dimensional networking reflectivity, of a monitoring region and other data needed by analysis assimilation based on a regional WRF mode; aiming at the uncertainty source of a cloud analysis system, the method comprises the following steps of: randomly disturbing multi-source observation data input in the process of forming an initial field of each ensemble forecasting member cloud in ensemble forecasting; aiming at the uncertainty of a micro-physical parameterization scheme of the cloud analysis system, three-dimensional disturbance is carried out on key dynamics and thermodynamics sensitive parameters of cloud analysis, combined random disturbance of two processes of an initial value and the parameterization scheme is formed, and the initial value and cloud micro-physical parameterization uncertainty of the cloud analysis system and the interaction of the initial value and the cloud micro-physical parameterization uncertainty are further reflected. Convection triggering and rainfall evolution processes can be changed, and the cloud and rainfall forecasting accuracy of a mesoscale service ensemble forecasting system is improved.

The invention provides an emergency communication spectrum prediction method based on a gating adaptive time sequence diagram integrated network, which relates to the technical field of communication, and comprises the following steps: acquiring multi-band historical spectrum data, and dividing a training sample and a prediction label according to a time step; constructing a self-adaptive dynamic graph structure fusing priori knowledge and a learnable matrix, and combining gating time sequence convolution, graph convolution, channel alignment convolution and residual connection to form a gating time sequence graph network; carrying out iterative training by utilizing the training sample and the dynamic graph structure, and obtaining an integrated prediction model through period storage and average weight snapshot; and inputting a to-be-tested sample into the model and outputting a frequency spectrum prediction value. The method has the advantages that spatio-temporal feature extraction is fused, and prediction precision and scene adaptability are improved; the generalization and robustness of the model in the environment change are enhanced by the self-adaptive dynamic graph structure; the parallel gating time sequence convolution is adopted to reduce the calculation complexity and the reasoning time delay, and the millisecond-level response requirement of emergency communication is met.

The invention relates to the technical field of equipment predictive maintenance, in particular to an industrial equipment residual life prediction method based on an SRenbPI algorithm, which comprises the following steps: preprocessing industrial equipment source domain data, including constructing a heteroscedasticity noise simulation data set and introducing real industrial sensor data, and setting standardized input through normalization, sliding window construction and label; in the training stage, a bootstrap sampling strategy is adopted to construct an integrated regression model, noise influence is quantized through a scaling residual formula, and a residual set is formed; in the prediction stage, noise scale estimation and residual quantile are combined to generate an unequal-width prediction interval, meanwhile, a dynamic updating mechanism is introduced, a residual set is updated in real time to adapt to data distribution changes after a plurality of test samples are processed every time, and the model does not need to be trained again. According to the method, the adaptive defect of a traditional static residual set in industrial equipment full-life-cycle monitoring is effectively overcome, and the prediction interval precision and real-time performance are remarkably improved.

The invention discloses a method for improving sand and dust numerical forecasting through assimilation-aerosol-cloud-radiation, and the method comprises the steps: building a sand and dust storm forecasting mode, calculating sand and dust optical parameters based on a spherical particle extinction theory, and achieving the bidirectional feedback of sand and dust to weather; a sand and dust aerosol-cloud interaction ice nucleus nucleation mechanism is given, a sand and dust CCN type aerosol homogeneous freezing process is realized, a radiation variable temperature rate is calculated and fed back to a weather mode power process, and bidirectional feedback of weather-driven sand and dust and influence of sand and dust on weather is formed; constructing an ensemble forecasting module, performing time and space related disturbance on a meteorological initial field, a boundary condition and an aerosol initial concentration, and performing ensemble forecasting after a mode is input; and establishing an ensemble assimilation analysis correction module, constructing a background error covariance matrix by utilizing mode ensemble forecasting, disturbing observation at an analysis moment, and assimilating and correcting the initial concentration of the mode aerosol based on a localized ensemble Kalman filtering method to obtain a forecasting result.

The present application belongs to the technical field of deep learning and meteorological prediction, and comprises the following specific steps: taking the reservoir basin in the central part of China as the research object, selecting the control prediction results of 5 prediction modes, dividing the prediction data of the central reservoir according to the latitude and longitude information of the ground observation station, taking the data information near the station as the influencing factor, and performing regularization processing on the data in the small space range; the regularized prediction data of each mode is respectively put into each grouping module to obtain the characteristics of each mode prediction, and finally the accurate precipitation prediction near each station is obtained through the output module after effectively fusing the features such as terrain and station latitude and longitude. The present application can use multiple sub-networks to extract different mode characteristics respectively, and jointly act, give different weights, and can effectively fuse the prediction data of each mode to generate more accurate precipitation value near the station, so as to achieve better correction effect.

A method and system for predicting product defect rate, the method comprising: providing a dataset, wherein each data point in the dataset corresponds to a design combination and includes at least one first type of data field and at least one second type of data field; training a first model using the first type of data field of a portion of the dataset and training a second model using the second type of data field; integrating another portion of the dataset into the trained first / second model to obtain multiple first prediction results and multiple second prediction results; training an ensemble prediction model using the multiple first prediction results and multiple second prediction results; and integrating new data into the first model and the second model to obtain new first prediction results and new second prediction results, and integrating the two prediction results into the ensemble prediction model to obtain a defect rate prediction result for the new data.

The application provides a multi-source field fusion method and system for meteorological gridded data, comprising: collecting gridded meteorological element data of multiple data sources in the same time and space range; performing two-dimensional Fourier transform on the gridded meteorological element data of each data source respectively to obtain complex plane wave data corresponding to each gridded meteorological element data; dividing the complex plane wave data corresponding to each gridded meteorological element data into low-wave-number large-scale data and high-wave-number small-scale data according to a preset scheme; averaging the low-wave-number large-scale data, superimposing any one of the high-wave-number small-scale data, and then performing two-dimensional inverse Fourier transform to obtain fused multi-source gridded new meteorological element data; the application improves the small-scale prediction effect of weather prediction by flattening high-frequency small-scale processes through averaging, simultaneously embodies the advantages of traditional ensemble prediction, reduces errors through the averaging of multiple prediction fields, and improves the prediction accuracy.

This invention relates to the interdisciplinary field of fatigue fracture mechanics and machine learning, and discloses a method and computer program product for predicting fatigue crack propagation rate. The invention first constructs a heterogeneous candidate model pool, and then selects three preferred models from the pool through Bayesian optimization and competitive screening strategies. These three preferred models are then adjusted to achieve unified scheduling, forming an HPD-Stacking ensemble prediction model as the first layer, which, together with the meta-learner in the second layer, performs ensemble learning. Finally, a trained overall model is obtained for prediction. This invention effectively integrates the physical consistency of fracture mechanics laws with the data mining capabilities of ensemble learning. Even when faced with data of varying quality, it can still achieve high-precision prediction of fatigue crack propagation behavior, avoiding overfitting and non-physical fluctuations inherent in traditional single models.

The invention discloses a method and device for correcting a medium and long term power prediction result through ensemble forecast data, and relates to the technical field of wind power generation. The method comprises the following steps: superposing random noise on historical wind speed data, inputting the data into a long-short-term memory neural network, and generating an initial power prediction sequence in a future time period; performing normalization processing on the initial power prediction sequence and the corresponding ensemble prediction data, and respectively marking the initial power prediction sequence and the corresponding ensemble prediction data as false data and true data; inputting the preprocessed false data and true data into a discrimination model, and iteratively updating model parameters by adopting a small-batch gradient descent method and an Adam optimization algorithm; and constructing a loss function by discriminating the false data probability output by the model, and updating the parameters of the LSTM prediction model. According to the technical scheme, medium and long term power prediction of LSTM and other machine learning models is corrected through ensemble forecast data, the comprehensive loss function is matched with Adam optimization to suppress overfitting, and the power error can be remarkably reduced.

The invention provides a combined temperature forecast correction method and system, and relates to the technical field of weather forecast. According to the method, three kinds of errors are systematically corrected, and the climate mode temperature forecast deviation is corrected. The method comprises the following steps: firstly, considering an error source of temperature forecast, and respectively disassembling observation temperature data and original forecast temperature data in a training period into three independent components, namely a mean term, a trend term and a residual term; and using conditional Gaussian correction to optimize the residual term error, and keeping the rank correlation structure of the original ensemble forecast. And according to requirements, performing mean value correction, trend correction and residual error correction on the original forecast temperature data, combining a mean value correction result, a trend correction result and a residual error correction result of the original forecast temperature data, and outputting a combined temperature forecast correction result. The method is used for temperature forecast correction and has the advantages of being good in correction effect, flexible to use, efficient in calculation and the like.

The application provides a multi-model multi-boundary estuary saltwater intrusion ensemble probability prediction method and system, wherein the method is coupled based on a plurality of calibrated and verified numerical models, a plurality of runoff boundaries and a plurality of wind field boundary conditions, an ensemble prediction system of multiple source uncertainty disturbances is constructed, and saltwater intrusion evolution results under a plurality of prediction scenarios are generated. Further based on historical observation data and model performance evaluation indexes, a sub-item confidence probability weighting method of the numerical model, the runoff boundary and the wind field boundary is established, and probability distribution estimation of the ensemble prediction result is realized. The method can quantitatively represent the probability and uncertainty of the occurrence of a saltwater intrusion event in a water source, and improve the accuracy and reliability of the estuary saltwater intrusion prediction.

A product defect rate prediction method and system. The product defect rate prediction method comprises: providing a data set, wherein each piece of data in the data set separately corresponds to a design combination, and comprises at least one first-type data field and at least one second-type data field; using the first-type data field of a part of the data set to train a first model, and using the second-type data field to train a second model; inputting the remaining part of the data set into the trained first / second model to obtain a plurality of first prediction results and a plurality of second prediction results; using the plurality of first prediction results and the plurality of second prediction results to train an ensemble prediction model; and inputting new data into the first model and the second model to obtain a new first prediction result and a new second prediction result, and inputting the two prediction results into the ensemble prediction model to obtain a defect rate prediction result of the new data.

This invention relates to the field of intelligent monitoring technology for hydropower projects, and discloses a method, device, medium, and equipment for intelligent analysis and early warning of grouting in hydropower projects. The method includes acquiring real-time monitoring data during the grouting process and constructing an ensemble prediction model using random forest, support vector machine, and XGBoost as base learners (Stacking); iteratively optimizing the hyperparameters of the ensemble prediction model using a particle swarm optimization algorithm; predicting the unit grouting volume during the grouting process using the model; generating explanatory information using the SHAP algorithm for feature importance interpretation; generating a grouting quality early warning signal based on the prediction results and explanatory information; and outputting the grouting quality early warning signal to a monitoring terminal. This method effectively improves the prediction accuracy, dynamic adaptability, and decision interpretability of the grouting process in hydropower projects, thereby enhancing the intelligence, precision, and risk control level of grouting construction in hydropower projects.

The embodiment of the invention provides a regional wind power icing back-and-reserve capacity clustering set prediction method and device and a storage medium. The method comprises the following steps: dividing wind power plants in a region into a plurality of wind power plant clusters based on meteorological characteristic and topographic characteristic parameters; for each cluster, respectively establishing an individual icing reserve capacity prediction model of each wind power plant in the wind power plant cluster, and establishing an icing reserve capacity reference prediction model of the wind power plant cluster; obtaining an individual reserve capacity prediction value of each wind power plant in each wind power plant cluster through an individual icing reserve capacity prediction model; determining a set initial value of the wind power plant cluster according to the similarity between the meteorological characteristics and topographic characteristic parameters of each wind power plant in the wind power plant cluster; obtaining a reference prediction value of each wind power plant cluster and correcting the reference prediction value to obtain an icing reserve capacity prediction value of each wind power plant cluster; and summing the predicted values of the icing reserve capacity of all the wind power plant clusters to obtain a total predicted value of the regional wind power icing reserve capacity.

To provide a technique capable of performing ensemble prediction with high accuracy even when a distribution of information related to a prediction target locally changes.SOLUTION: The prediction device is configured to, based on an evaluation result obtained by evaluating performance of each of a plurality of models with reference to evaluation information including model input information for evaluation and a true value corresponding to the model input information, and the evaluation information, A weight update unit that updates some or all of the plurality of first weight vectors and some or all of the plurality of second weight vectors, and an integration unit that integrates a prediction result predicted by each model with reference to model input information included in prediction target information related to a prediction target, using a weight vector selected from the plurality of first weight vectors and the plurality of second weight vectors on the basis of the prediction target information, and SELECTED DRAWING: Figure 1