CN-122000865-A - Photovoltaic power self-adaptive confidence interval prediction method, device, medium and equipment

CN122000865ACN 122000865 ACN122000865 ACN 122000865ACN-122000865-A

Abstract

The application discloses a photovoltaic power self-adaptive confidence interval prediction method, a device, a medium and equipment, and relates to the technical field of power prediction, wherein the method comprises the steps of obtaining historical photovoltaic power generation data and historical meteorological data; the method comprises the steps of carrying out correlation analysis on historical photovoltaic power generation data and historical meteorological data by a Person correlation coefficient analysis method to obtain a key meteorological data sequence, carrying out model training on a deep learning network model based on a CNN-Informer hybrid architecture by using the key meteorological data sequence and the historical photovoltaic power generation data as training samples to obtain a photovoltaic power self-adaptive confidence interval prediction model, carrying out prediction on photovoltaic power generation data and meteorological data obtained in real time by using the photovoltaic power self-adaptive confidence interval prediction model to obtain a prediction interval, and controlling the charging and discharging states of an energy storage system by taking the upper limit and the lower limit of the prediction interval as reference boundaries of charging and discharging power of the energy storage system. The application can improve the accuracy and reliability of interval prediction.

Inventors

BAO ZHEN
LI ANQI
CHEN LEI

Assignees

国能智深控制技术有限公司

Dates

Publication Date: 20260508
Application Date: 20251226

Claims (10)

1. A photovoltaic power adaptive confidence interval prediction method, comprising: acquiring historical photovoltaic power generation data and historical meteorological data of different time periods; Carrying out correlation analysis on the historical photovoltaic power generation data and the historical meteorological data by adopting a Pelson correlation coefficient analysis method to obtain a preset number of key meteorological data sequences which are strongly correlated with photovoltaic power generation power; The key meteorological data sequence and the historical photovoltaic power generation data are used as training samples to carry out model training on a deep learning network model based on a CNN-Informer hybrid architecture, so that a photovoltaic power self-adaptive confidence interval prediction model meeting preset conditions is obtained; The photovoltaic power self-adaptive confidence interval prediction model is adopted to predict photovoltaic power generation data and meteorological data acquired in real time, and a prediction interval of photovoltaic power in a future period is obtained; And controlling the charge and discharge states of the energy storage system based on the upper limit and the lower limit of the prediction interval as reference boundaries of the charge and discharge power of the energy storage system.
2. The method according to claim 1, wherein the performing correlation analysis on the historical photovoltaic power generation data and the historical meteorological data by using a pearson correlation coefficient analysis method to obtain a predetermined number of key meteorological data sequences which are strongly correlated with photovoltaic power generation power specifically comprises: performing data preprocessing on the historical photovoltaic power generation data and the historical meteorological data; taking the preprocessed historical meteorological data as an independent variable, taking the historical photovoltaic power generation data as a dependent variable, and calculating pearson correlation coefficients of the historical meteorological data and the historical photovoltaic power generation data with different dimensions; absolute value calculation processing is carried out on each pearson correlation coefficient, so that absolute pearson correlation coefficients corresponding to each pearson correlation coefficient are obtained; the absolute pearson correlation coefficient values are determined as a key meteorological data sequence according to the historical meteorological data corresponding to the first predetermined number of pearson correlation coefficients which are arranged in order from top to bottom.
3. The method of claim 1, wherein the model training the deep learning network model based on the CNN-Informer hybrid architecture using the key meteorological data sequence and the historical photovoltaic power generation data as training samples to obtain a photovoltaic power adaptive confidence interval prediction model meeting a preset condition, specifically comprises: Firstly, adopting a CNN feature extraction layer of the deep learning network model to perform feature extraction on the key meteorological data sequence and the historical photovoltaic power generation data to obtain a historical high-dimensional feature tensor; step two, adopting an encoder of Informer layers of the deep learning network model to encode the historical high-dimensional characteristic tensor to obtain a historical context vector sequence; Thirdly, decoding by adopting a Informer-layer decoder of the deep learning network model according to the historical context vector sequence and a future time step length to obtain a historical photovoltaic power future sequence representation; Performing parallel calculation processing on the future sequence representation of the historical photovoltaic power by adopting an output layer of Informer layers of the deep learning network model to obtain historical photovoltaic power predicted values corresponding to different quantiles; fifthly, updating model parameters of the deep learning network model by adopting a preset quantile loss function and a preset model loss function according to the historical photovoltaic power predicted value and the actual photovoltaic power corresponding to the future time step length; And step six, repeatedly executing the step one to the step five to update the updated deep learning network model until the model converges, and obtaining the photovoltaic power self-adaptive confidence interval prediction model meeting the preset condition.
4. The method of claim 3, wherein the encoding the historical high-dimensional feature tensor by the Informer-layer encoder using the deep-learning network model to obtain the historical context vector sequence specifically comprises: performing sine-cosine position coding on the historical high-dimensional characteristic tensor to obtain a first coding vector; performing depth calculation processing on the first coding vector by adopting a probability sparse self-attention method to obtain a second coding vector; distilling the second coded vector by adopting a rolling and maximum pooling mode to obtain a third coded vector; And carrying out residual calculation and normalization processing on the third coding vector to obtain the historical context vector sequence.
5. The method of claim 3, wherein the decoder employing Informer layers of the deep learning network model decodes from the historical context vector sequence and future time steps to obtain a historical photovoltaic power future sequence representation, comprising: Step one, determining historical prediction data and an initial token before a current time point to be predicted by adopting a masking multi-head self-attention method; Step two, taking future representation of the photovoltaic power of the current time point to be predicted as a query target, and querying key value pairs of the historical context vector sequence to obtain the predicted historical power representation of the current time point to be predicted; Step three, nonlinear transformation processing is carried out on the predicted historical power representation by adopting a feedforward neural network, so that the predicted historical power characteristic representation of the current time point to be predicted is obtained; Performing residual calculation and layer normalization processing on the prediction history power characteristic representation to obtain a history prediction value corresponding to a current time point to be predicted, wherein the history prediction value comprises a first history prediction value corresponding to a target median, a second history prediction value corresponding to a pessimistic score and a third history prediction value corresponding to an optimistic score; Step five, filling the historical predicted value in a masked future position of the historical context vector sequence corresponding to the current time point to be predicted so as to update the historical context vector sequence; And step six, repeatedly executing the step one to the step five to predict the future representation of the photovoltaic power at the next time point to be predicted based on the updated historical context vector sequence so as to obtain the future sequence representation of the historical photovoltaic power.
6. The method of claim 1, wherein prior to controlling the state of charge and discharge of the energy storage system based on the upper and lower predicted intervals of the predicted interval as reference boundaries for the charge and discharge power of the energy storage system, the method further comprises performing a performance evaluation of the predicted interval for a future period of time; the performance evaluation of the prediction interval of the future period specifically includes: Collecting the actual photovoltaic power value of a future period; performing evaluation index calculation based on the photovoltaic power actual value and the prediction interval to obtain a multi-dimensional evaluation index, wherein the multi-dimensional evaluation index comprises an interval coverage index, an interval average width index and an interval coverage width index; And evaluating the prediction interval according to the multi-dimensional evaluation index.
7. The method of claim 1, wherein the predicting the photovoltaic power generation data and the meteorological data obtained in real time by using the photovoltaic power adaptive confidence interval prediction model to obtain a prediction interval of the photovoltaic power of the future period specifically comprises: adopting a CNN feature extraction layer of the photovoltaic power self-adaptive confidence interval prediction model to perform feature extraction on the photovoltaic power generation data and the meteorological data to obtain a target dimension feature tensor; Adopting a Informer-layer encoder of the photovoltaic power self-adaptive confidence interval prediction model to encode the target high-dimensional characteristic tensor to obtain a target context vector sequence; Decoding by adopting a Informer-layer decoder of the photovoltaic power self-adaptive confidence interval prediction model according to the target context vector sequence and a future time step length to obtain a target photovoltaic power future sequence representation; And carrying out parallel calculation processing on the future sequence representation of the target photovoltaic power by adopting an output layer of the photovoltaic power self-adaptive confidence interval prediction model to obtain photovoltaic power predicted values corresponding to different target fractional numbers so as to obtain a predicted interval of the photovoltaic power in a future period.
8. A photovoltaic power adaptive confidence interval prediction device, comprising: the acquisition module is used for acquiring historical photovoltaic power generation data and historical meteorological data in different time periods; The correlation analysis module is used for carrying out correlation analysis on the historical photovoltaic power generation data and the historical meteorological data by adopting a Pelson correlation coefficient analysis method to obtain a preset number of key meteorological data sequences which are strongly correlated with photovoltaic power generation power; The model training module is used for carrying out model training on a deep learning network model based on a CNN-Informer hybrid architecture by adopting the key meteorological data sequence and the historical photovoltaic power generation data as training samples to obtain a photovoltaic power self-adaptive confidence interval prediction model meeting preset conditions; the prediction module is used for predicting photovoltaic power generation data and meteorological data acquired in real time by adopting the photovoltaic power self-adaptive confidence interval prediction model to obtain a prediction interval of photovoltaic power in a future period; and the control module is used for controlling the charge and discharge states of the energy storage system based on the upper limit and the lower limit of the prediction interval as reference boundaries of the charge and discharge power of the energy storage system.
9. A storage medium storing a computer program which, when executed by a processor, implements the steps of the photovoltaic power adaptive confidence interval prediction method of any of the preceding claims 1-7.
10. An electronic device comprising at least a memory, a processor, the memory having stored thereon a computer program, the processor, when executing the computer program on the memory, implementing the steps of the photovoltaic power adaptive confidence interval prediction method of any of the preceding claims 1-7.

Description

Photovoltaic power self-adaptive confidence interval prediction method, device, medium and equipment Technical Field The invention relates to the technical field of power prediction, in particular to a photovoltaic power self-adaptive confidence interval prediction method, device, medium and equipment. Background Photovoltaic power generation is an important component of clean energy, and is widely applied in the global field due to the characteristics of renewable, low pollution and low carbon. However, photovoltaic power generation output is affected by solar irradiance, temperature, cloud cover and other natural factors, and has remarkable fluctuation, intermittence and uncertainty. As the installed scale of new energy continues to expand and replace traditional fossil energy is accelerated, random fluctuation of power of the new energy forms a serious challenge for real-time balance, safe and stable operation and electric energy quality of an electric power system. The problems of voltage out-of-limit, line overload, power dumping and the like are easily caused by the high-proportion distributed photovoltaic access on the power distribution network level, the reserve capacity requirement is increased due to the inaccurate predictability of the photovoltaic output on the system scheduling level, the system operation cost is raised, and the phenomenon of 'light abandoning' still occurs at time, so that the resource waste is caused. However, the existing method still has the following bottleneck, which limits the efficiency of directly serving the prediction result to the downstream advanced applications such as energy storage dispatching, power grid optimization and the like, and the loss function design of the partial interval prediction model is difficult to obtain effective balance between the interval coverage rate and the interval width, or has the problems of incapability, and the like, so that the training stability and the final interval quality are affected. Aiming at multi-step prediction tasks, the capturing capability of the existing model on long-term dependency, complex periodic patterns and abrupt change features in the photovoltaic power time sequence still has room for improvement, and the reliability and practicability of the prediction interval on a long time scale are restricted. Particularly, the output results of most interval prediction methods at present cannot fully consider how to perform efficient and seamless connection with decision models such as downstream energy storage system optimal control, power grid risk scheduling and the like. These downstream applications typically require that the predicted results provide clear, reliable, and dynamically changing power fluctuation boundaries as key constraints for their optimization model. The lack of the existing methods in terms of section quality and adaptive capacity makes it difficult to provide a decision boundary that is directly reliable, thereby affecting the final performance of the overall "predictive-control" chain. Disclosure of Invention In view of the above, the invention provides a photovoltaic power self-adaptive confidence interval prediction method, a device, a medium and equipment, and aims to solve the problems that the point prediction result is strong in uncertainty and difficult to be directly used for accurate control of an energy storage system due to the fact that the random fluctuation of power generation is ignored in the existing photovoltaic power prediction method. In order to solve the above problems, the present application provides a photovoltaic power adaptive confidence interval prediction method, which includes: acquiring historical photovoltaic power generation data and historical meteorological data of different time periods; Carrying out correlation analysis on the historical photovoltaic power generation data and the historical meteorological data by adopting a Pelson correlation coefficient analysis method to obtain a preset number of key meteorological data sequences which are strongly correlated with photovoltaic power generation power; The key meteorological data sequence and the historical photovoltaic power generation data are used as training samples to carry out model training on a deep learning network model based on a CNN-Informer hybrid architecture, so that a photovoltaic power self-adaptive confidence interval prediction model meeting preset conditions is obtained; The photovoltaic power self-adaptive confidence interval prediction model is adopted to predict photovoltaic power generation data and meteorological data acquired in real time, and a prediction interval of photovoltaic power in a future period is obtained; And controlling the charge and discharge states of the energy storage system based on the upper limit and the lower limit of the prediction interval as reference boundaries of the charge and discharge power of the energy storage system. Optionally, the performing