Photovoltaic characteristic curve prediction method based on probabilistic graph model and scene classification

Abstract

The invention discloses a photovoltaic characteristic curve prediction method based on a probability graph model and scene classification, and the method comprises the following steps: abstracting photovoltaic changes of different characteristic scenes in one day, dividing the interval time, and setting sampling points; collecting data, preprocessing the data, and determining a distribution form of random disturbance by adopting a statistical method; conjugate prior distribution Dirichlet distribution of multinomial distribution is introduced, and the occurrence frequency of each feature scene in the data is used as a parameter of multinomial distribution; introducing a conditional probability to obtain a final scene probability, taking the final scene probability as a parameter of multi-term distribution, and generating a sequence of occurrence of each feature scene; 5, simulation of future photovoltaic changes is finally generated according to the generated feature scene sequence, the method is based on the reality situation and fits the reality, the method is more robust in the long-time simulation process, and the result is more reliable; and a longer strategy can be made for the charging pile.

Description

technical field [0001] The invention relates to the technical field of photovoltaic characteristic prediction, in particular to a photovoltaic characteristic curve prediction method based on a probability graph model and scene classification. Background technique [0002] Photovoltaic is unstable, which is not good news for light-storage-charge integrated charging piles, because when photovoltaics are uncertain, it is difficult to formulate a strategy for charging piles to maximize their charging and discharging actions. It is also difficult to maximize the economic benefits of using light energy. Therefore, it is very important to generate and simulate the future photovoltaic changes more accurately. [0003] Many previous studies have mentioned the concept of "scenario", because photovoltaics will show different characteristics under different scenarios. For example, there may be "multiple peaks" in cloudy weather conditions, while generally only one peak occurs in clear...

AI Technical Summary

Problems solved by technology

[0005] At present, there are the following defects: each scene is acquired through an algorithm, even if each feature scene is determined, it is difficult for these scenes to correspond to the actual situation

Therefore, it is difficult to determine the characteristic scene of the day from the beginning, and it is impossible to complete the formulation of the charging and discharging strategy for the day; when identifying the characteristic scene, it is only based on the data level, and the data distance between the photovoltaic change trend curves will be Scenes divided by photovoltaic change curves

This will amplify the influence of photovoltaic value, while ignoring the influence of photovoltaic change trend; only stay at the level of real-time optimization, even if some research has realized the photovoltaic prediction for the next 15 minutes through the random forest algorithm, it still has not formed a robust Simulation of photovoltaic changes in the future for a long time; prediction accuracy is poor

We mentioned in the second article that because the influence of the trend is ignored in the clustering process, when using LSTM and other algorithms to make predictions, the law of trend changes will be regarded as random disturbances, thereby increasing the prediction error, resulting in Poor prediction accuracy