CN-121998195-A - Power supply and demand factor probability prediction method and system in extreme meteorological scene

CN121998195ACN 121998195 ACN121998195 ACN 121998195ACN-121998195-A

Abstract

The invention provides a power supply and demand factor probability prediction method and a power supply and demand factor probability prediction system under an extreme meteorological scene, comprising the steps of obtaining power supply and demand factor historical data and historical real meteorological data, carrying out standardized processing on the historical real meteorological data according to an extreme meteorological early warning standard, and constructing an extreme meteorological scene library, namely extracting multidimensional features aiming at the extreme meteorological data and the power supply and demand factor historical data; the method comprises the steps of constructing a multi-mode fusion prediction model, training, verifying and testing the multi-mode fusion prediction model by utilizing multi-dimensional characteristics to obtain a target prediction model, obtaining actual weather forecast data, and inputting the actual weather forecast data into the target prediction model to obtain a prediction result comprising a power supply and demand factor prediction mean value, confidence information and probability density information. According to the invention, the power supply and demand factor prediction with high precision, reliability and scene suitability under extreme meteorological scenes is realized by improving links such as data processing, feature engineering, model architecture, training evaluation and the like.

Inventors

DAI CHANGCHUN
LI YONGBO
Qian Hanhan
QI HUI
ZHANG WEISHI
JI CHAO
HAO YUXING
XIE DAOQING
FU JINGYU
ZHOU TAO
JIANG HAILONG
ZHAO XUETING
ZHANG WEI
Cheng Honggu
LIN ZHEMIN

Assignees

安徽电力交易中心有限公司

Dates

Publication Date: 20260508
Application Date: 20260129

Claims (10)

1. A method for predicting power supply and demand factor probability in an extreme meteorological scenario, the method comprising: acquiring historical data and historical real meteorological data of power supply and demand factors, carrying out standardized processing on the historical real meteorological data according to extreme meteorological early warning standards, and constructing an extreme meteorological scene library: Extracting a multi-dimensional feature set comprising extreme weather core features, power supply and demand history evolution features and extreme weather-power supply and demand cross-correlation features aiming at the extreme weather data and the power supply and demand factor history data in the extreme weather scene library; constructing a multi-mode fusion prediction model comprising a feature coding module, an attention fusion module and a probability prediction output module; training, verifying and/or testing the multi-mode fusion prediction model by utilizing the multi-dimensional feature set to obtain a target prediction model for power supply and demand factor probability prediction; And acquiring actual weather forecast data, and inputting the actual weather forecast data into the target prediction model to obtain a prediction result containing the power supply and demand factor prediction mean value, the confidence information and/or the probability density information.
2. The method of claim 1, wherein the acquiring historical real weather data, the classifying the historical real weather data according to extreme weather prediction criteria, generating an extreme weather dataset consistent with an actual weather forecast data format, comprises: based on the time granularity consistent with the actual weather forecast data, acquiring multi-dimensional historical real weather data in a target period aiming at a target area; determining grading thresholds corresponding to various extreme weather scenes according to the extreme weather prediction standards; screening an extreme weather sample set in the historical real weather data based on the grading threshold; referring to the data format of the actual weather forecast data, carrying out format reconstruction on the extreme weather data set to obtain an extreme weather data set with field types and time granularity consistent with the actual weather forecast data; and labeling the scene labels of the extreme weather data sets to form a multi-dimensional extreme weather scene library, wherein the scene labels comprise weather types and early warning grades.
3. The method of claim 1, wherein the extracting a multi-dimensional feature set comprising extreme weather core features, power supply and demand history evolution features, and extreme weather-power supply and demand cross-correlation features, comprises: Extracting extreme weather core features representing the event intensity, the duration characteristic, the change trend and/or the spatial distribution of the extreme weather event based on the extreme weather data, wherein the event intensity comprises at least one of extreme values, average values and peak occurrence time of various extreme weather parameters, the duration characteristic comprises the duration time of the extreme weather event, the change trend comprises an intensity level change sequence of the extreme weather event and/or a change amplitude of weather parameters in adjacent time periods, and the spatial distribution comprises a standard deviation of wind speed and/or a precipitation spatial gradient between sites; And Extracting power supply and demand historical evolution characteristics representing time sequence evolution characteristics, fluctuation characteristics and/or supply and demand balance characteristics based on the power supply and demand factor historical data, wherein the time sequence evolution characteristics comprise at least one of power load, historical synchronous value, historical average value and historical extremum of wind power/photovoltaic power, the fluctuation characteristics comprise fluctuation rate, trend characteristics and abrupt change characteristics of the power supply and demand factors, and the supply and demand balance characteristics comprise the difference value of power grid power supply capacity and power load, the proportion of wind power/photovoltaic power to total power supply quantity and/or the change of the duty ratio of industrial load to resident load; And analyzing the nonlinear correlation between the extreme weather event and the power supply and demand factor according to the extreme weather data and the power supply and demand factor historical data to obtain the extreme weather-power supply and demand cross correlation characteristic.
4. A method according to claim 3, wherein said analyzing the nonlinear association between the extreme weather event and the power supply and demand factor based on the extreme weather data and the power supply and demand factor history data to obtain the extreme weather-power supply and demand cross-correlation feature comprises: calculating correlation coefficients of the extreme weather parameters and the power supply and demand factors under different hysteresis durations to obtain hysteresis correlation characteristics, and/or, Calculating sensitivity coefficients of the extreme weather parameters and the power supply and demand factors at different intensity levels by piecewise linear regression to obtain sensitivity characteristics, and/or, Constructing scene-specific cross features for different terminal meteorological scenes, and/or, And extracting the time difference between the ending time of the extreme weather event and the time when the power supply and demand factor is recovered to the normal level and the load/power change rate in the recovery process as recovery characteristics.
5. The method of claim 1, wherein after the extracting the multi-dimensional feature set comprising extreme weather core features, power supply and demand history evolution features, and extreme weather-power supply and demand cross-correlation features, the method further comprises: And adopting a two-step method to perform feature screening on the multi-dimensional feature set to obtain a preferable multi-dimensional feature set, wherein the method specifically comprises the following steps of: Calculating importance scores of all the characteristics in the multidimensional characteristic set on the power supply and demand factor prediction result based on a tree model, sorting the importance scores, and removing the characteristics lower than a first preset threshold value to obtain a preliminary screening characteristic set; And calculating mutual information entropy among all the features in the preliminary screening feature set, and aiming at the feature pairs with the mutual information entropy higher than a second preset threshold value, removing the features with lower importance scores in the feature pairs to obtain a preferred multi-dimensional feature set.
6. The method of claim 1, wherein constructing a multi-modal fusion prediction model comprising a feature encoding module, an attention fusion module, and a probabilistic prediction output module comprises: The method comprises the steps of constructing a characteristic coding module integrating a convolutional neural network architecture, a long-short-period memory network architecture and a full-connection layer, wherein the convolutional neural network architecture is used for coding spatial distribution characteristics of extreme weather, the long-short-period memory network architecture is used for coding time sequence characteristics of power supply and demand, and the full-connection layer is used for coding cross-correlation characteristics of the extreme weather and the power supply and demand; Constructing an attention fusion module adopting a multi-head attention mechanism, wherein the multi-head attention mechanism is used for carrying out self-adaptive weighted fusion on the characteristics after multi-mode coding; And the Bayesian neural network structure is used for outputting a prediction result containing the power supply and demand factor prediction mean value, the confidence information and/or the probability density information by introducing the probability distribution of the parameters.
7. The method according to claim 1, wherein training, verifying and/or testing the multi-modal fusion prediction model using the multi-dimensional feature set to obtain a target prediction model for power supply and demand factor probability prediction comprises: Dividing the multi-dimensional feature set into a training sample set, a verification sample set and a test sample set according to a preset proportion, and Constructing a mixed loss function combining mean square error loss and negative log likelihood loss, wherein the mean square error loss is used for optimizing the accuracy of a prediction mean value, and the negative log likelihood loss is used for optimizing the fitting degree of probability distribution; Model training is carried out based on the training sample set and the mixed loss function, and an Adam optimizer is adopted to initialize super parameters so as to realize parameter learning of the multi-mode fusion prediction model; cross-verifying based on the verification sample set to realize hyper-parameter adjustment of the multi-mode fusion prediction model, so as to obtain the target prediction model; and carrying out reliability evaluation on the target prediction model based on the test sample set, judging whether the coverage rate of the prediction interval and the average bandwidth error meet the preset standard condition, and if the coverage rate and the average bandwidth error do not meet the preset standard condition, retraining the target prediction model.
8. The method of claim 7, wherein the method further comprises: And adding a Dropout layer into a feature coding module of the multi-mode fusion prediction model, and introducing L2 regularization constraint into the mixed loss function to avoid the phenomenon of over-fitting in the model training process.
9. The method according to any one of claims 1-8, wherein the confidence information includes confidence intervals corresponding to different confidences, and the probability density information includes a probability density curve.
10. A power supply and demand factor probability prediction system in an extreme meteorological scenario, the system comprising: The data acquisition module is used for acquiring historical real meteorological data, carrying out standardized processing on the historical real meteorological data according to extreme meteorological early warning standards, and constructing an extreme meteorological scene library: The feature extraction module is used for extracting a multi-dimensional feature set comprising extreme weather core features, power supply and demand historical evolution features and extreme weather-power supply and demand cross correlation features aiming at the extreme weather data in the extreme weather scene library; The model construction module is used for constructing a multi-mode fusion prediction model comprising a feature coding module, an attention fusion module and a probability prediction output module; The model training module is used for training, verifying and testing the multi-mode fusion prediction model by utilizing the multi-dimensional feature set to obtain a target prediction model for power supply and demand factor probability prediction; The probability prediction module is used for acquiring actual weather forecast data, inputting the actual weather forecast data into the target prediction model, and obtaining a prediction result containing the power supply and demand factor prediction mean value, the confidence information and/or the probability density information.

Description

Power supply and demand factor probability prediction method and system in extreme meteorological scene Technical Field The invention relates to the technical field of power system prediction, in particular to a power supply and demand factor probability prediction method and system in an extreme meteorological scene. Background Under the drive of a double-carbon target, the global energy structure is accelerated to be transformed into low carbonization, the installation scale of renewable energy sources such as wind power, solar energy and the like is continuously enlarged, the renewable energy sources gradually become one of the core forces of power supply, and the international energy agency predicts that renewable energy sources exceeding coal become the largest global power source in the earliest 2025 year. Meanwhile, the global climate change aggravates the frequent and intensity-rising extreme weather events such as typhoons, heavy rain, extreme high temperature, strong cold tides and the like, and the safety and stability operation of the novel power system form a serious challenge. Meteorological conditions become key factors affecting supply and demand balance of a power system, namely, on one hand, extreme weather directly impacts the output stability of renewable energy sources, such as extremely low wind speed and extremely low solar radiation can cause wind-light power generation 'dark valley', and extremely high temperature can reduce the output power of a photovoltaic module, on the other hand, extreme weather can cause rigid surge of power load to form 'winter and summer dual peak' characteristics, for example, every 1 ℃ rise of the highest air temperature can cause 4.5% of the highest load of power, extremely high temperature and drought superposition can cause insufficient water and electricity output, supply and demand contradiction is further increased, and 2022-year Sichuan electricity limiting event is a typical case. Under the background, the accurate prediction of power supply and demand factors (including power load, wind-solar power generation power and the like) in extreme meteorological scenes becomes a core premise for guaranteeing power grid dispatching optimization, improving power supply reliability and promoting new energy consumption, and is a key technical support for constructing a novel climate adaptability power system. Along with the deep application of the artificial intelligence technology in the energy field, the electric power supply and demand prediction is gradually updated from classical methods such as traditional trend extrapolation, ARIMA model and the like to a prediction scheme based on deep learning, the prediction precision is remarkably improved, and a part of AI-driven schemes can improve the prediction precision to more than 90%. The current industry research hotspots focus on multi-source data fusion (such as meteorological data, electric power data and user behavior data) and multi-model collaborative prediction, and aim to further improve the adaptation capability of a prediction model to a complex scene. However, in the prior art, an obvious short plate still exists under the special scene of extreme weather, the precision and the reliability of a conventional prediction model are greatly reduced due to the strong randomness, the strong destructiveness and the influence mechanism complexity of the extreme weather event, and the high requirement of power system emergency dispatching cannot be met, so that a targeted technical innovation is needed to break through the bottleneck. Disclosure of Invention Therefore, the invention provides a power supply and demand factor probability prediction method and system in an extreme meteorological scene, and aims to solve the technical problem that the power supply and demand factor prediction in the extreme meteorological scene is difficult to meet the core requirements of power system scheduling decision on prediction precision, reliability and scene suitability in the prior art. In order to achieve the above purpose, the invention adopts the following technical scheme: According to a first aspect of the present invention, the present invention provides a power supply and demand factor probability prediction method in an extreme meteorological scenario, the method comprising: acquiring historical data and historical real meteorological data of power supply and demand factors, carrying out standardized processing on the historical real meteorological data according to extreme meteorological early warning standards, and constructing an extreme meteorological scene library: Extracting a multi-dimensional feature set comprising extreme weather core features, power supply and demand history evolution features and extreme weather-power supply and demand cross-correlation features aiming at the extreme weather data and the power supply and demand factor history data in the extreme weather scene library; constructi