Photovoltaic power prediction method

Abstract

The invention relates to the technical field of computer technology, in particular to a photovoltaic power prediction method, which comprises the following steps: acquiring historical data and prediction day meteorological data, and training photovoltaic prediction models to obtain a first photovoltaic prediction model; obtaining a reference evaluation index of the first photovoltaic prediction model through precision evaluation; inputting the prediction day meteorological data into a first photovoltaic prediction model to obtain first prediction data; acquiring historical time sequence data of a prediction day in real time, and when the number of samples in the historical time sequence data reaches a first preset value, performing precision evaluation on the first photovoltaic prediction model based on the historical time sequence data and the first prediction data to obtain a first evaluation index; comparing the first evaluation index with the reference evaluation index in real time to judge the prediction precision of the first photovoltaic prediction model, and determining a final prediction model based on the prediction precision; and inputting the meteorological data of the prediction day into the final prediction model to obtain a prediction result. The method has the advantage of improving the prediction precision.

Description

Technical Field

[0001] The present application relates to the technical field of computer technology, and in particular to a photovoltaic power prediction method. Background Art

[0002] With the transformation of the global energy structure and the popularization of the concept of sustainable development, photovoltaic power generation, as a clean and renewable energy form, has received increasing attention and application. However, the inherent intermittent, random and volatile characteristics of photovoltaic power generation make it a considerable challenge to the stable operation of the power system after large-scale photovoltaic power generation is connected to the power grid. This instability may not only affect the quality of power supply, but also bring potential risks to the safe and stable operation of the power grid.

[0003] In order to meet these challenges, it is necessary to accurately predict the power of photovoltaic power generation so that the operation mode of the power grid can be reasonably arranged and corresponding countermeasures can be taken. Through the prediction of photovoltaic power generation, the power dispatching department can better balance the relationship between power generation and power consumption, thereby ensuring the stability and reliability of the power system.

[0004] At present, the commonly used photovoltaic power generation prediction methods mainly include physical prediction methods and statistical prediction methods. The physical prediction method mainly relies on information such as meteorological data, satellite images, and physical parameters of photovoltaic equipment to predict the power of photovoltaic power generation by establishing a prediction model. The statistical prediction method mainly uses statistics and analysis of historical data to find the implicit connection between photovoltaic power generation and other factors, so as to make predictions.

[0005] In recent years, with the rapid development of artificial intelligence technology, statistical forecasting methods have gradually become mainstream. Advanced machine learning algorithms such as time series models, support vector machines, and neural networks have also been widely used in the field of photovoltaic power forecasting. However, these methods still have some problems in practical applications. For example, when the forecasting model is only based on historical data for analysis without the supplement of new data, the model may become solidified. In addition, due to the accuracy of meteorological data and the randomness of photovoltaic power generation itself, the solidified forecasting model may have a large forecast error, thus affecting the accuracy of the forecast. Summary of the invention

[0006] In order to solve the above problems, the present application provides a photovoltaic power prediction method.

[0007] In a first aspect, the present application provides a photovoltaic power prediction method, which adopts the following technical solution: A photovoltaic power prediction method comprises the following steps: Obtain historical data and forecasted daily meteorological data, and obtain a first photovoltaic prediction model after training the photovoltaic prediction model, and obtain reference evaluation indicators of the first photovoltaic prediction model through accuracy evaluation; Inputting the forecast day meteorological data into the first photovoltaic forecast model to obtain first forecast data; Acquire historical time series data of the forecast day in real time, and when the number of samples in the historical time series data reaches a first preset value, perform accuracy evaluation on the first photovoltaic forecast model based on the historical time series data and the first forecast data to obtain a first evaluation index; Comparing the first evaluation index with the reference evaluation index in real time to determine the prediction accuracy of the first photovoltaic prediction model, and determining the final prediction model based on the prediction accuracy; The forecast day meteorological data is input into the final forecast model to obtain the forecast results.

[0008] With such a setting, after obtaining the first photovoltaic prediction model through historical data, the indicators of the first photovoltaic prediction model are evaluated by obtaining the historical time series data of the prediction day, and the actual prediction accuracy of the first photovoltaic prediction model on the prediction day is checked, and then the actual prediction accuracy is judged, and the final prediction model is determined based on the judgment results.

[0009] In one embodiment, the step of obtaining a first photovoltaic prediction model after training the photovoltaic prediction model and obtaining a reference evaluation index of the first photovoltaic prediction model through accuracy evaluation specifically includes: Selecting a first model training set and a first model test set from historical data based on the meteorological data of the forecast day and a cluster analysis model of similar days; Inputting the first model training set into a preset photovoltaic prediction model for training to obtain a first photovoltaic prediction model; The first photovoltaic prediction model is evaluated based on the first model test set to obtain an initial evaluation index, and a reference evaluation index is obtained based on the initial evaluation index.

[0010] With such a setting, the data for model training is obtained by means of similar day clustering analysis, which can improve the prediction accuracy of the first photovoltaic prediction model.

[0011] In one embodiment, the step of comparing the first evaluation index with the reference evaluation index in real time to determine the prediction accuracy of the first photovoltaic prediction model and determining the final prediction model based on the prediction accuracy specifically includes: determining in real time whether the accuracy of the first evaluation index is greater than the reference evaluation index, and if so, using the first photovoltaic prediction model as the final prediction model; If not, training the photovoltaic prediction model based on the historical time series data to obtain a second photovoltaic prediction model; Inputting the forecast day meteorological data into the second photovoltaic forecast model to obtain second forecast data; Based on the second prediction data and the historical time series data of the sample time series after the first preset value, the accuracy of the second photovoltaic prediction model is evaluated to obtain a second evaluation index, the second evaluation index and the reference evaluation index are compared in real time to judge the prediction accuracy of the second photovoltaic prediction model, and the final prediction model is determined based on the prediction accuracies of the first photovoltaic prediction model and the second photovoltaic prediction model.

[0012] In this way, when the accuracy of the first evaluation index is greater than the reference evaluation index, it means that the prediction accuracy of the first photovoltaic prediction model meets expectations, so it can be used as the final prediction model. When the accuracy of the first evaluation index is not greater than the reference evaluation index, the second photovoltaic prediction model is obtained through historical time series data training, and then the accuracy of the second photovoltaic prediction model is evaluated through the index evaluation, and finally the final prediction model is determined by the prediction accuracy results of the first photovoltaic prediction model and the second photovoltaic prediction model.

[0013] In one embodiment, the step of training the photovoltaic prediction model based on the historical time series data to obtain the second photovoltaic prediction model specifically includes: Based on the historical time series data, the second model training set is selected from the historical data through the similar day clustering analysis model; The second model training set is combined with the historical time series data and input into a preset photovoltaic prediction model for training to obtain a second photovoltaic prediction model.

[0014] In one of the embodiments: the condition for obtaining the second evaluation index is: judging whether the number of samples in the historical time series data reaches a second preset value, wherein the second preset value is greater than the first preset value.

[0015] With this arrangement, when the number of samples reaches a certain number, the second evaluation index is calculated. This can avoid the situation where the evaluation accuracy is affected by data fluctuations when the number of samples is too small.

[0016] In one embodiment, the step of determining the final prediction model based on the prediction accuracy of the first photovoltaic prediction model and the second photovoltaic prediction model specifically includes: Determine in real time whether the accuracy of the second evaluation indicator is greater than the reference evaluation indicator before the number of samples in the historical time series data reaches a third preset value; If so, when the accuracy of the first evaluation index is not greater than the reference evaluation index, the second photovoltaic prediction model is used as the final prediction model; when the accuracy of the first evaluation index is greater than the reference evaluation index, the first photovoltaic prediction model and the second photovoltaic prediction model are merged to obtain the final prediction model; If not, the final prediction model is determined based on the comparison result of the first evaluation index and the reference evaluation index.

[0017] In this way, the third preset value is used as the latest prediction time of the historical time series data. The historical time series data obtained after this time is no longer used as a reference for prediction. This is mainly to provide enough time for data acquisition and processing, as well as model training. When the accuracy of the second evaluation index is greater than the reference evaluation index at the third preset value, and the accuracy of the first evaluation index is also greater than the reference evaluation index, it means that the prediction accuracy of the first and second photovoltaic prediction models is relatively high. Therefore, the two are fused as the final prediction model. When the accuracy of the second evaluation index is greater than the reference evaluation index at the third preset value, and the accuracy of the first evaluation index is not greater than the reference evaluation index, the accuracy of the second photovoltaic prediction model is relatively high, while the accuracy of the first photovoltaic prediction model is relatively low. Therefore, the second photovoltaic prediction model is used as the final prediction model. If the accuracy of the second evaluation index is not greater than the reference evaluation index before the third preset value, it means that the prediction result of the second photovoltaic prediction model is not accurate, and further adjustments need to be made based on the first photovoltaic prediction model.

[0018] In one embodiment, the step of fusing the first photovoltaic prediction model with the second photovoltaic prediction model to obtain a final prediction model specifically includes: Calculating weights of the first photovoltaic prediction model and the second photovoltaic prediction model based on the first evaluation index and the second evaluation index; The final prediction model is obtained by weighting the first photovoltaic prediction model and the second photovoltaic prediction model through weights.

[0019] In this way, the first photovoltaic prediction model and the second photovoltaic prediction model are fused by weighting. Since both the first photovoltaic prediction model and the second photovoltaic prediction model have good accuracy, and the first photovoltaic prediction model has a more stable prediction accuracy, although the second photovoltaic prediction model has better prediction accuracy, its anti-interference ability is weaker due to its higher specificity. Therefore, after the two are fused, a more stable and accurate prediction effect can be obtained.

[0020] In one embodiment, the step of determining the final prediction model based on the comparison result of the first evaluation index and the reference evaluation index specifically includes: Determine whether the accuracy of the first evaluation index is greater than the reference evaluation index. If not, retrain the photovoltaic prediction model based on the historical time series data to obtain a third photovoltaic prediction model, and use the third photovoltaic prediction model as the final prediction model. If so, the first photovoltaic prediction model and the second photovoltaic prediction model are fused based on the time series to obtain a final prediction model.

[0021] In this way, when the accuracy of the first evaluation index is not greater than the reference evaluation index, it means that the accuracy of the first and second photovoltaic prediction models is not enough. Therefore, the third photovoltaic prediction model is obtained after retraining and is used as the final prediction model. Otherwise, it means that the first photovoltaic prediction model still has a high prediction accuracy, and the second photovoltaic prediction model has a good prediction accuracy for the data before the first preset value. Therefore, the first photovoltaic prediction model and the second photovoltaic prediction model are fused based on the time series, so that the final prediction model is mainly based on the first photovoltaic prediction model and supplemented by the second photovoltaic prediction model.

[0022] In one embodiment, the step of fusing the first photovoltaic prediction model and the second photovoltaic prediction model based on time series to obtain a final prediction model specifically includes: Calculate the weights of the first photovoltaic prediction model and the second photovoltaic prediction model based on the first evaluation index, the second evaluation index, and the first preset value; The final prediction model is obtained by weighting the first photovoltaic prediction model and the second photovoltaic prediction model through weights.

[0023] In this way, by introducing the first preset value and combining the first evaluation index with the second evaluation index to set the weight, the effect of taking the first photovoltaic prediction model as the main and the second photovoltaic prediction model as the auxiliary can be better achieved.

[0024] In one embodiment, the weighted formula of the step of calculating the weights of the first photovoltaic prediction model and the second photovoltaic prediction model based on the first evaluation index, the second evaluation index, and the first preset value is: The weight of the first photovoltaic prediction model = the first evaluation index / (the first evaluation index+the first preset value / the total number of samples*the second evaluation index); The weight of the second photovoltaic prediction model = the second evaluation index / (the first evaluation index+the first preset value / the total number of samples*the second evaluation index). BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Schematic diagram of the steps of the photovoltaic power prediction method of this embodiment. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0027] A photovoltaic power prediction method, such as Figure 1 As shown, the following steps are included: S101, acquiring historical data and forecasted day meteorological data, and obtaining a first photovoltaic prediction model after training the photovoltaic prediction model, and obtaining a reference evaluation index of the first photovoltaic prediction model through accuracy evaluation.

[0028] In this step, the historical data at least includes historical photovoltaic power generation data and historical meteorological data. The meteorological data includes radiation illumination, ambient relative humidity, real-time wind speed, ambient temperature, etc. The historical photovoltaic power generation data and historical meteorological data are sorted based on time series. The sampling time of each data in the historical data can be freely defined, usually with 15 minutes as a sampling interval. The daily sampling time is set according to the photovoltaic setting location, usually 6~7 o'clock as the first sampling time, and 18~19 as the last sampling time.

[0029] The training method of the first photovoltaic prediction model is specifically as follows: S201. Filter a first model training set and a first model test set from historical data based on the forecast day meteorological data and a similar day clustering analysis model.

[0030] Because the photovoltaic power generation capacity will have obvious differences under different weather conditions, the weather type of the photovoltaic forecast day is usually distinguished into three types: sunny, cloudy and rainy. Similar days under the same weather type are selected based on the cosine similarity value between the forecast day and the historical day. The cosine similarity analysis formula is: ; Where: —cosine similarity value; n—time node; N—sampling time series length; —Predict the feature vector of time node n on the same day; —The feature vector of time node n on the ith prediction day.

[0031] when →0, it means that the similarity of the two data is high and can be attributed to days with similar weather types. In order to ensure the effectiveness of historical data training, historical data of multiple similar days with high cosine similarity are selected from historical meteorological data. The selected data are split into two parts: the first model training set and the first model test set. Usually, the model training set accounts for 70% to 80% of the total data.

[0032] S202: Then input the first model training set into a preset photovoltaic prediction model for training to obtain a first photovoltaic prediction model.

[0033] The photovoltaic prediction model preset in this embodiment uses the Long Short Term Memory (LSTM) network. Due to its deep learning function, it can deeply mine the implicit correlation information between various influencing factors and photovoltaic power generation historical data, and can achieve the purpose of improving the prediction accuracy of photovoltaic power generation. LSTM is improved on the basis of recursive neural network. t Forget Gate F t and output gate O t It has greatly solved the problems of gradient vanishing and gradient exploding in the training process of the original network model and improved the learning accuracy. In addition, due to its loop feedback memory logic unit, it reduces the length requirement of the input time series and can use historical time series of any length for training and modeling.

[0034] The LSTM network works as follows: ; ; ; ; ; ; Where: , , —Output values ​​of forget gate, input gate and output gate; , , , —weight linear transformation matrix; σ(), tanh() —activation function, hyperbolic tangent function; b—bias vector; —Input value at time t; —The hidden layer state value at time t-1.

[0035] After the photovoltaic power generation and its related meteorological characteristics enter the LSTM unit, each gate structure performs operations and is judged by σ() to determine whether it is activated. The unit state C and the hidden layer state h will update the value. After nonlinear transformation, Processed Phase superposition , through calculation and Finally, the output value of the LSTM network is obtained , which is the photovoltaic power generation power prediction result.

[0036] S203 . Evaluate the first photovoltaic prediction model based on the first model test set to obtain an initial evaluation index, and obtain a reference evaluation index based on the initial evaluation index.

[0037] The initial evaluation index can be obtained from the relative root mean square error (RRMSE), the goodness of fit determination coefficient (R2), and the mean absolute percentage error (MAPE). It can be one of them or a comprehensive evaluation through multiple indicators. In this embodiment, the goodness of fit determination coefficient (R2) is used for evaluation, and its expression is: ; Where: — the Kth observation, — the kth predicted value, — the mean of all observations, n — the number of observations.

[0038] The value range of R2 is [0,1]. When the result is closer to 0, the model fitting effect is worse, and when the result is 1, the model fitting effect is the best. Generally, the larger the R2, the more perfect the model fitting is.

[0039] After obtaining the initial evaluation index, the initial evaluation index can be directly used as a reference evaluation index, or it can be used as a reference evaluation index after lowering the value of the initial evaluation index. It depends on the actual needs. However, it should be noted that the closer the value of the reference evaluation index is to the initial evaluation index, the higher the accuracy of the final prediction result will be.

[0040] S102: Input the forecast day meteorological data into a first photovoltaic forecast model to obtain first forecast data.

[0041] S103, acquiring historical time series data of the prediction day in real time, and when the number of samples in the historical time series data reaches a first preset value, performing accuracy evaluation on the first photovoltaic prediction model based on the historical time series data and the first prediction data to obtain a first evaluation index.

[0042] In this step, the historical time series data includes not only the meteorological data of the forecast day, but also the photovoltaic power data corresponding to the meteorological data of the forecast day. The purpose of setting the first preset value is to ensure a large number of samples. After the number of samples reaches the requirement, the goodness of fit determination coefficient (R2) is used for evaluation to obtain the first evaluation index.

[0043] S104 . Compare the first evaluation index and the reference evaluation index in real time to determine the prediction accuracy of the first photovoltaic prediction model, and determine the final prediction model based on the prediction accuracy.

[0044] This step specifically includes: S301. Determine in real time whether the accuracy of the first evaluation index is greater than the reference evaluation index. If so, use the first photovoltaic prediction model as the final prediction model; if not, train the photovoltaic prediction model based on historical time series data to obtain a second photovoltaic prediction model.

[0045] The training principle of the second photovoltaic prediction model is the same as that of the first photovoltaic prediction model, and the specific steps are as follows: Based on the historical time series data, the second model training set is screened from the historical data through the similar day clustering analysis model.

[0046] The second model training set is combined with the historical time series data and input into a preset photovoltaic prediction model for training to obtain a second photovoltaic prediction model.

[0047] In this step, since there is no need to set a test set during the training process of the second photovoltaic prediction model, all useful data can be used as training data.

[0048] S302: Input the predicted day meteorological data into a second photovoltaic prediction model to obtain second prediction data.

[0049] S303, based on the second prediction data, and the historical time series data of the sample time series after the first preset value, the second photovoltaic prediction model is evaluated for accuracy to obtain a second evaluation index, the second evaluation index and the reference evaluation index are compared in real time to determine the prediction accuracy of the second photovoltaic prediction model, and the final prediction model is determined based on the prediction accuracy of the first photovoltaic prediction model and the second photovoltaic prediction model.

[0050] In this step, the condition for obtaining the second evaluation index is: judging whether the number of samples in the historical time series data reaches a second preset value, wherein the second preset value is greater than the first preset value, and preferably, the second preset value is greater than or equal to twice the first preset value. It should also be noted that the historical time series data used in the evaluation of the second evaluation index should be sample data collected after the first preset value.

[0051] Among them, the step of determining the final prediction model based on the prediction accuracy of the first photovoltaic prediction model and the second photovoltaic prediction model specifically includes: judging in real time whether the accuracy of the second evaluation index is greater than the reference evaluation index before the number of samples in the historical time series data reaches the third preset value. If so, when the accuracy of the first evaluation index is not greater than the reference evaluation index, the second photovoltaic prediction model is used as the final prediction model; when the accuracy of the first evaluation index is greater than the reference evaluation index, the first photovoltaic prediction model and the second photovoltaic prediction model are merged to obtain the final prediction model; if not, the final prediction model is determined based on the comparison result of the first evaluation index and the reference evaluation index.

[0052] Among them, the third preset value is greater than the second preset value, and the third preset value is set according to the accuracy requirement of the required comparison. The larger the third preset value is, the higher the accuracy of the final prediction model is, but relatively, the amount of data that can be predicted by the final prediction model is smaller.

[0053] Specifically, the steps of fusing the first photovoltaic prediction model and the second photovoltaic prediction model to obtain the final prediction model specifically include: The weights of the first photovoltaic prediction model and the second photovoltaic prediction model are calculated based on the first evaluation index and the second evaluation index.

[0054] The final prediction model is obtained by weighting the first photovoltaic prediction model and the second photovoltaic prediction model through weights.

[0055] In the above steps, the weighted formula for weight calculation is: The weight of the first photovoltaic prediction model = the first evaluation index / (the first evaluation index + the second evaluation index) The weight of the second photovoltaic prediction model = the second evaluation index / (the first evaluation index + the second evaluation index).

[0056] The step of determining the final prediction model based on the comparison result of the first evaluation index and the reference evaluation index specifically includes: judging whether the accuracy of the first evaluation index is greater than the reference evaluation index. If not, the photovoltaic prediction model is trained based on the historical time series data to obtain a third photovoltaic prediction model again, and the third photovoltaic prediction model is used as the final prediction model. Among them, the training principle of the third photovoltaic prediction model is the same as the training principle of the second photovoltaic prediction model. If so, the first photovoltaic prediction model and the second photovoltaic prediction model are fused based on the time series to obtain the final prediction model.

[0057] The above steps specifically include: The weights of the first photovoltaic prediction model and the second photovoltaic prediction model are calculated based on the first evaluation index, the second evaluation index, and the first preset value.

[0058] The final prediction model is obtained by weighting the first photovoltaic prediction model and the second photovoltaic prediction model through weights.

[0059] Among them, the weighted formula of the step of calculating the weights of the first photovoltaic prediction model and the second photovoltaic prediction model based on the first evaluation index, the second evaluation index, and the first preset value in the above steps is: The weight of the first photovoltaic prediction model = the first evaluation index / (the first evaluation index+the first preset value / the total number of samples*the second evaluation index); The weight of the second photovoltaic prediction model = the second evaluation index / (the first evaluation index+the first preset value / the total number of samples*the second evaluation index).

[0060] S105, inputting the forecast day meteorological data into the final forecast model to obtain the forecast result.

[0061] The embodiments of this specific implementation method are all preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Therefore, all equivalent changes made based on the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A photovoltaic power prediction method, characterized in that: The following steps are involved: Obtain historical data and forecasted daily meteorological data, and obtain a first photovoltaic prediction model after training the photovoltaic prediction model, and obtain reference evaluation indicators of the first photovoltaic prediction model through accuracy evaluation; Inputting the forecast day meteorological data into the first photovoltaic forecast model to obtain first forecast data; Acquire historical time series data of the forecast day in real time, and when the number of samples in the historical time series data reaches a first preset value, perform accuracy evaluation on the first photovoltaic forecast model based on the historical time series data and the first forecast data to obtain a first evaluation index; Comparing the first evaluation index with the reference evaluation index in real time to determine the prediction accuracy of the first photovoltaic prediction model, and determining a final prediction model based on the prediction accuracy; The meteorological data of the forecast day is input into the final forecast model to obtain the forecast results.

2. A photovoltaic power prediction method according to claim 1, characterized in that: The step of obtaining a first photovoltaic prediction model after training the photovoltaic prediction model and obtaining a reference evaluation index of the first photovoltaic prediction model through accuracy evaluation specifically includes: Selecting a first model training set and a first model test set from historical data based on the meteorological data of the forecast day and a cluster analysis model of similar days; Inputting the first model training set into a preset photovoltaic prediction model for training to obtain a first photovoltaic prediction model; The first photovoltaic prediction model is evaluated based on the first model test set to obtain an initial evaluation index, and a reference evaluation index is obtained based on the initial evaluation index.

3. A photovoltaic power prediction method according to claim 1 or 2, characterized in that: The step of comparing the first evaluation index with the reference evaluation index in real time to determine the prediction accuracy of the first photovoltaic prediction model, and determining the final prediction model based on the prediction accuracy specifically includes: determining in real time whether the accuracy of the first evaluation index is greater than the reference evaluation index, and if so, using the first photovoltaic prediction model as the final prediction model; If not, training the photovoltaic prediction model based on the historical time series data to obtain a second photovoltaic prediction model; Inputting the forecast day meteorological data into the second photovoltaic forecast model to obtain second forecast data; Based on the second prediction data and the historical time series data of the sample time series after the first preset value, the accuracy of the second photovoltaic prediction model is evaluated to obtain a second evaluation index, the second evaluation index and the reference evaluation index are compared in real time to judge the prediction accuracy of the second photovoltaic prediction model, and the final prediction model is determined based on the prediction accuracies of the first photovoltaic prediction model and the second photovoltaic prediction model.

4. A photovoltaic power prediction method according to claim 3, characterized in that: The step of training the photovoltaic prediction model based on the historical time series data to obtain the second photovoltaic prediction model specifically includes: Based on the historical time series data, the second model training set is selected from the historical data through the similar day clustering analysis model; The second model training set is combined with the historical time series data and input into a preset photovoltaic prediction model for training to obtain a second photovoltaic prediction model.

5. A photovoltaic power prediction method according to claim 3, characterized in that: The condition for obtaining the second evaluation index is: judging whether the number of samples in the historical time series data reaches a second preset value, wherein the second preset value is greater than the first preset value.

6. A photovoltaic power prediction method according to claim 3 or 5, characterized in that: The step of determining the final prediction model based on the prediction accuracy of the first photovoltaic prediction model and the second photovoltaic prediction model specifically includes: Determine in real time whether the accuracy of the second evaluation indicator is greater than the reference evaluation indicator before the number of samples in the historical time series data reaches a third preset value; If so, when the accuracy of the first evaluation index is not greater than the reference evaluation index, the second photovoltaic prediction model is used as the final prediction model; when the accuracy of the first evaluation index is greater than the reference evaluation index, the first photovoltaic prediction model and the second photovoltaic prediction model are merged to obtain the final prediction model; If not, the final prediction model is determined based on the comparison result of the first evaluation index and the reference evaluation index.

7. A photovoltaic power prediction method according to claim 6, characterized in that: The step of fusing the first photovoltaic prediction model with the second photovoltaic prediction model to obtain a final prediction model specifically includes: Calculating weights of the first photovoltaic prediction model and the second photovoltaic prediction model based on the first evaluation index and the second evaluation index; The final prediction model is obtained by weighting the first photovoltaic prediction model and the second photovoltaic prediction model through weights.

8. A photovoltaic power prediction method according to claim 6, characterized in that: The step of determining the final prediction model based on the comparison result of the first evaluation index and the reference evaluation index specifically includes: Determine whether the accuracy of the first evaluation index is greater than the reference evaluation index. If not, retrain the photovoltaic prediction model based on the historical time series data to obtain a third photovoltaic prediction model, and use the third photovoltaic prediction model as the final prediction model. If so, the first photovoltaic prediction model and the second photovoltaic prediction model are fused based on the time series to obtain a final prediction model.

9. A photovoltaic power prediction method according to claim 8, characterized in that: The step of fusing the first photovoltaic prediction model and the second photovoltaic prediction model based on time series to obtain a final prediction model specifically includes: Calculate the weights of the first photovoltaic prediction model and the second photovoltaic prediction model based on the first evaluation index, the second evaluation index, and the first preset value; The final prediction model is obtained by weighting the first photovoltaic prediction model and the second photovoltaic prediction model through weights.

10. A photovoltaic power prediction method according to claim 9, characterized in that: The weighted formula of the step of calculating the weights of the first photovoltaic prediction model and the second photovoltaic prediction model based on the first evaluation index, the second evaluation index, and the first preset value is: The weight of the first photovoltaic prediction model = the first evaluation index / (the first evaluation index+the first preset value / the total number of samples*the second evaluation index); The weight of the second photovoltaic prediction model = the second evaluation index / (the first evaluation index+the first preset value / the total number of samples*the second evaluation index).

Images