CN107563565A - A kind of short-term photovoltaic for considering Meteorology Factor Change decomposes Forecasting Methodology - Google Patents

Abstract

The invention discloses a short-term photovoltaic decomposition and prediction method considering changes in meteorological factors, including: S1 decomposing the photovoltaic output time series through a singular spectrum analysis method to obtain low-frequency series, high-frequency series and noise series; S2 using the Pearson correlation coefficient method Determine the main meteorological factors affecting photovoltaic output, and analyze their sensitivity to photovoltaic output; S3 establish a forecast model considering meteorological factors for the low-frequency sequence and high-frequency sequence combined with sensitivity; S4 obtain the low-frequency sequence prediction value and high-frequency sequence according to the prediction model Sequence prediction value, and obtain the photovoltaic output prediction value according to the low-frequency sequence prediction value and the high-frequency sequence prediction value. The present invention decomposes the photovoltaic output into different sub-sequences through the singular spectrum analysis method to analyze the characteristics of each sequence separately; obtains the influence degree of the unit variation of different meteorological factors on the photovoltaic output through correlation analysis and sensitivity analysis, so as to more accurately Forecast PV output.

Description

A Short-Term Photovoltaic Decomposition Prediction Method Considering Changes in Meteorological Factors

technical field

The invention belongs to the technical field of intermittent renewable energy forecasting such as wind power and photovoltaics, and more specifically relates to a short-term photovoltaic decomposition forecasting method (Singular Spectrum Analysis Method Considering Meteorological Factors, referred to as SSA-MF method) considering changes in meteorological factors.

Background technique

With the development of a high proportion of renewable energy, intermittent renewable energy such as wind power and photovoltaics has been increasingly popularized and applied. However, intermittent renewable energy sources such as wind power and photovoltaics have strong randomness and volatility, which makes the safety, stability and economic operation of the power system face important challenges. Therefore, how to accurately predict intermittent renewable energy such as wind power and photovoltaics has important practical guiding significance for the dispatching and operation of power systems.

At present, there are three main types of photovoltaic output prediction: physical methods, statistical methods, and combination methods of the above methods. The physical method is to establish a physical model based on factors such as the detailed geographical location of the photovoltaic module and the photoelectric conversion efficiency, and directly use the meteorological data as input to predict according to the power generation principle of the photovoltaic system. Its effectiveness depends on the degree of grasp of the internal structure of the research object and the laws it follows, as well as the accuracy of the model parameters. It involves many links, the process is complicated, and the parameters are difficult to solve. Statistical methods are based on the use of certain statistical methods to analyze historical photovoltaic output data, find inherent laws in the data and use them for prediction. It mainly includes time series method, regression analysis method, gray prediction method and meta-heuristic series method. The essence of the meta-heuristic method is to simulate the work and rest rules of organisms, and use some algorithm to train the sample data to obtain the relationship between the predicted conditions and the predicted quantities. Meta-heuristic methods mainly include neural network, support vector machine, genetic algorithm, fuzzy system and so on. Among them, the neural network method has a strong nonlinear fitting ability, and can map any complex nonlinear relationship, which is very similar to the characteristics of the photovoltaic power generation system, so it is very suitable for short-term prediction of the output of photovoltaic power plants. However, a single neural network cannot adapt to the variety of weather types, and the prediction effect is not good. In addition, the traditional BP neural network training adopts the gradient descent method, which is easy to fall into the local minimum and the convergence speed is slow. When the fuzzy system predicts the photovoltaic output, the establishment of fuzzy reasoning rules requires a large amount of historical data and sufficient expert experience. The combination method utilizes the information provided by different models and exerts their respective advantages, and selects the appropriate way to combine in order to improve the prediction effect. Compared with the former two methods, the modeling of the combination method is more complicated than that of the single method, and the realization process is more difficult.

To sum up, when the above methods are used to predict the photovoltaic output, they use the historical data to process and model the prediction, without considering the photovoltaic output characteristics reflected in the sub-sequences after the data decomposition, and not digging out some hidden information and information of the photovoltaic output. Inherent laws, so it is difficult to achieve a better prediction effect.

Contents of the invention

In view of the defects of the prior art, the purpose of the present invention is to provide a short-term photovoltaic decomposition prediction method considering changes in meteorological factors, aiming to solve the problem of low prediction accuracy in the prior art.

The invention provides a short-term photovoltaic decomposition prediction method considering changes in meteorological factors, comprising the following steps:

S1: Decompose the photovoltaic output time series by singular spectrum analysis method to obtain low frequency series, high frequency series and noise series, and remove the noise series;

S2: Use the Pearson correlation coefficient method to determine the main meteorological factors affecting photovoltaic output, and analyze the sensitivity of the main meteorological factors to photovoltaic output;

S3: For the low-frequency sequence and the high-frequency sequence and in combination with the sensitivity, respectively establish a high-frequency sequence prediction model and a low-frequency sequence prediction model considering meteorological factors;

S4: Obtain the predicted value of the high-frequency sequence according to the prediction model of the high-frequency sequence, obtain the predicted value of the low-frequency sequence according to the prediction model of the low-frequency sequence; and obtain the photovoltaic value according to the predicted value of the low-frequency sequence and the predicted value of the high-frequency sequence output forecast.

Further, step S1 is specifically:

S11: Transform the photovoltaic output time series into a matrix form, and decompose the matrix into the sum of d equivalent sub-matrices;

S12: Decompose the d sub-matrices obtained After grouping, the low-frequency matrix Z _low , high-frequency matrix Z _high and noise matrix Z _noise are obtained, and the low-frequency matrix Z _low , high-frequency matrix Z _high and noise matrix Z _noise are respectively diagonally averaged and restored to the original sequence form The low-frequency sequence P _l , the high-frequency sequence _{Ph and the noise sequence P n are} obtained after the sequences are reconstructed.

Furthermore, in step S2, the Pearson correlation coefficient method is used to determine the main meteorological factors affecting photovoltaic output as follows:

S21: Select temperature, radiation, wind speed and rainfall as meteorological factors;

S22: According to the formula Calculate the Pearson correlation coefficient between photovoltaic output and temperature, irradiance, wind speed or rainfall respectively;

S23: Determine the main meteorological factors affecting photovoltaic output according to the size of the Pearson correlation coefficient;

in, Pearson correlation coefficient absolute value of The closer to 1, the higher the degree of linear correlation between the two variables.

Furthermore, in step S3, the prediction model of the high-frequency sequence considering meteorological factors is established as follows:

(1) Select the reference date and benchmark value of the high-frequency series:

Take the day before the day to be predicted as the reference day for the high-frequency sequence, and use the high-frequency sequence of photovoltaic output on the reference day as the benchmark value of the high-frequency sequence on the day to be predicted;

(2) The Pearson correlation coefficient between different meteorological factors and photovoltaic output is used as the weight coefficient of the meteorological factors affecting the change of photovoltaic output;

(3) According to the sensitivity of meteorological factors to the change of photovoltaic output, the temperature difference and radiation difference between the day to be predicted and the reference day, and according to the formula P _high ＝P' _high + α ₁ ΔP ₁ + α ₂ ΔP ₂ on the high frequency of photovoltaic output The sequence P _high is corrected;

Among them, P _high is the high-frequency sequence of photovoltaic output on the day to be predicted, P' _high is the high-frequency sequence of photovoltaic output on the reference day, ΔP ₁ is the change in the high-frequency sequence of photovoltaic output due to temperature change, and ΔP ₂ is the change in the high-frequency sequence of photovoltaic output due to radiation α ₁ is the weight coefficient of temperature affecting the high-frequency sequence change of photovoltaic output, and α ₂ is the weight coefficient of radiation affecting the high-frequency sequence change of photovoltaic output.

Furthermore, when the daily temperature to be predicted and the reference daily temperature are in the same sensitivity interval, ΔP ₁ =S _t (t-t'); when the daily temperature to be predicted and the reference daily temperature are in two different sensitivity intervals,

Among them, t is the daily temperature value to be predicted, t' is the reference daily temperature value, S _t is the sensitivity of the interval of the daily temperature to be predicted, S' _t is the sensitivity of the interval of the reference daily temperature, Represents the temperature value at the common endpoint of the two intervals.

Furthermore, in step S4, the photovoltaic output prediction value P=P _low +P _high is obtained according to the low-frequency sequence prediction value P _low and the high-frequency sequence prediction value P _high .

The present invention decomposes the photovoltaic output into different sub-sequences through the singular spectrum analysis method, and can analyze the characteristics of each sequence separately; through the correlation analysis and sensitivity analysis, the influence degree of the unit change of different meteorological factors on the photovoltaic output can be obtained, so that Predict photovoltaic output more accurately, provide favorable data reference for dispatching decision makers, thereby reducing the impact of photovoltaic output access on the power system.

Description of drawings

Fig. 1 is a schematic diagram of the SSA-MF method provided by the embodiment of the present invention;

Fig. 2 is a technical flow chart of singular spectrum analysis provided by an embodiment of the present invention;

Figure 3 is the decomposition sequence of photovoltaic output in April 2014; (a) is the historical photovoltaic output; (b) is the low-frequency sequence of historical photovoltaic output; (c) is the high-frequency sequence of historical photovoltaic output; (d) is the noise sequence of historical photovoltaic output ;

Figure 4 is the actual data and forecast data curve (SSA-MF) of photovoltaic output on May 12; (a) is the low-frequency sequence prediction value of photovoltaic output on May 12; (b) is the high-frequency sequence prediction of photovoltaic output on May 12 value; (c) is the predicted value of photovoltaic output on May 12.

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

Conventional forecasting models and methods are often difficult to adapt to intermittent changes in photovoltaic output, and the decomposed subsequences have not been deeply excavated and analyzed, and the processing of meteorological factors related to photovoltaic output is also relatively complicated, making it difficult to realize or the prediction accuracy is not high . The purpose of the present invention is to overcome the following two limitations of conventional photovoltaic output forecasting methods: (1) when meteorological factors are not considered, the prediction accuracy is not high, and it is difficult to achieve the desired forecasting effect; (2) when meteorological factors are considered, the weather The processing method and process of factors are relatively complicated, and it is difficult to realize in practical application. For this reason, the present invention provides a short-term photovoltaic decomposition prediction method that considers changes in meteorological factors; the method in the present invention can simplify the processing process of the influence of complex meteorological factors on photovoltaic output through correlation analysis and sensitivity analysis, and can use historical photovoltaic output data Decomposition is carried out, and the decomposed subsequences are analyzed and predicted separately, so that the prediction results have higher accuracy and ensure the safe and stable operation of the power system after photovoltaic access.

Figure 1 shows the idea of SSA-MF method, a singular spectrum analysis method for short-term photovoltaic output prediction considering meteorological factors provided by the present invention, including the following steps:

(1) Decompose the photovoltaic output time series by singular spectrum analysis technology to obtain low frequency series, high frequency series and noise series.

The Singular Spectrum Analysis (SSA) method is a technique for time series analysis and forecasting. Singular Value Decomposition (SVD for short) can be performed on the signal to obtain the trend characteristics, periodic characteristics and white noise characteristics of the original signal, etc. , which is beneficial to the analysis of the original signal. Figure 2 is a flow chart of singular spectrum analysis technology, which mainly includes two complementary stages of decomposition and reconstruction. SSA decomposition is to convert the original time series P into a matrix form first, and then use SVD decomposition to obtain d sub-matrices equivalent to the original matrix. The SSA reconstruction is to group the sub-matrices after SVD decomposition to obtain low-frequency, high-frequency and noise matrices, and then diagonally average them to obtain low-frequency, high-frequency and noise sequences.

Step (1) is specifically:

(11): SSA decomposition: The basic idea of the SSA decomposition method is to transform the original time series into a matrix form, and then decompose the matrix into the sum of multiple equivalent sub-matrices. SSA decomposition is mainly divided into two steps: embedding operation (Embedding) and SVD decomposition (Singular Value Decomposition).

(111): Embedding operation: The embedding operation is to transform the original one-dimensional time-series photovoltaic output P=(P ₁ ,P ₂ ,L,P _N ) with a length of N (N>2) into a multi-dimensional time-series photovoltaic output matrix Z=[ A mapping operation of Z ₁ , Z ₂ , L, Z _K ], that is, P=(P ₁ , P ₂ , L, P _N )→Z=[Z ₁ , Z ₂ , L, Z _K ]...( 1); where Z _i (i=1,2,L,K) is a certain column of matrix Z, Z _i =(P _i ,P _i+1 ,L,P _i+L-1 ) ^T ∈ R ^L , there are L dimensions in total, L is the embedding dimension (2≤L≤N), K=N-L+1. Usually, the selection of L should not exceed 1/3 of the entire sequence length. The matrix Z is called the trajectory matrix (Trajectory Matrix), that is: So far, the transformation from one-dimensional photovoltaic output sequence to multi-dimensional photovoltaic output matrix is completed.

(112): SVD decomposition: SVD decomposition decomposes the trajectory matrix Z of formula (2) into d sub-matrices d is the rank of matrix Z, and makes the sum of d submatrices equal to matrix Z, namely: Calculate according to formula (4) In the formula, λ ₁ , λ ₂ , L, λ _L (λ ₁ ≥ λ ₂ ≥ _L ≥ λ L ≥ 0) are the eigenvalues of S=ZZ ^T , U ₁ , U ₂ , L, U _L are the standard eigenvectors Orthogonal system.

V _j is obtained from formula (5) Among them, U _j and V _j represent the left and right eigenvectors of the trajectory matrix Z, respectively, is the singular value of trajectory matrix Z, set is the singular spectrum of the matrix Z, together form a characteristic ring So far, the SVD decomposition is completed, and d sub-matrices corresponding to the singular spectrum of the matrix Z are obtained.

(12): SSA reconstruction: SSA reconstruction is the d sub-matrix obtained by decomposing SVD first Carry out grouping to obtain the low frequency/high frequency/noise matrix, which are respectively recorded as Z _low , Z _high and Z _noise ; then the low frequency/high frequency/noise matrix is diagonally averaged and restored to the reconstructed sequence in the form of the original sequence, that is, the low frequency Sequence P _l , high frequency sequence _{Ph and noise sequence P n .} SSA reconstruction mainly includes two steps: Grouping and Diagonal Averaging.

(121): Grouping: Grouping is performed according to the contribution rate η of the first r (r≥0) singular values to the sum of the singular values of matrix Z, and the large jumps in the singular values. For example, suppose λ' and λ" are two unequal eigenvalues of matrix S, and both λ' and λ" are much smaller than their previous eigenvalues λ' ₀ and λ" ₀ . When λ _j ≥ λ', The matrix Z _j corresponding to λ _j is regarded as a low-frequency sub-matrix; when λ'> λ _j >λ", the matrix Z _j corresponding to λ _j is regarded as a high-frequency sub-matrix; when λ _j ≤ λ', the corresponding matrix Z j of λ _j The matrix Z _j of is regarded as a noise sub-matrix. The specific grouping depends on the actual situation. That is, the d sub-matrices obtained by formula (3) can be divided into low-frequency/high-frequency/noise matrices as shown in formula (6). The calculation formula of contribution rate η is shown in formula (7):

(122): Diagonal averaging: further transform the low-frequency/high-frequency/noise matrix determined by grouping in step S121 into a low-frequency/high-frequency/noise sequence with a length of N. The high-frequency matrix Z _high is used as an example for illustration below.

Suppose Z _high is a matrix of a×b, Z _ij is any element of Z _high , denote a ^* =min(a,b), b ^* =max(a,b), N=a+b-1, and When a<b, otherwise, The reconstruction sequence RC=(rc ₁ ,rc ₂ ,…,rc _N ) corresponding to the above grouping matrix can be obtained by the following formula:

From formula (8), the low-frequency sequence P _h can be obtained, and the low-frequency sequence P _l and the noise sequence P _n can be obtained similarly.

(2) Use the Pearson correlation coefficient method to determine the main meteorological factors affecting photovoltaic output.

The specific implementation method is as follows: Correlation analysis is performed on the time series of photovoltaic output and different meteorological factors. The present invention adopts the Pearson correlation coefficient method shown in formula (9). 1) select meteorological factors, the present invention studies different meteorological factors such as temperature, radiation, wind speed, rainfall; 2) calculate the difference between photovoltaic output and meteorological factors such as temperature, radiation, wind speed, rainfall, etc. according to formula (9) respectively The calculation results are shown in Table 1; 3) According to the size of the correlation coefficient in 2), determine the main meteorological factors affecting photovoltaic output.

in, correlation coefficient absolute value of The closer to 1, the higher the degree of linear correlation between the two variables.

Table 1 Correlation coefficient between meteorological data and light output data

(3) According to the main weather factors determined in step (2), analyze the sensitivity of each main weather to photovoltaic output.

Analyze the sensitivity of the main meteorological factors to the change of photovoltaic output. The sensitivity of photovoltaic output to meteorological factors refers to the change of photovoltaic output under the change of unit meteorological factors.

Analyze the sensitivity of the main meteorological factors to the change of photovoltaic output. The sensitivity of photovoltaic output to meteorological factors refers to the change of photovoltaic output under the change of unit meteorological factors. The specific operation process is as follows: According to the main meteorological factors determined in 2 (the temperature and radiation intensity are taken as examples below), the sensitivity of photovoltaic output to temperature and radiation intensity is analyzed respectively, and the solution results are shown in Table 2 and Table 3 Show.

Table 2 Sensitivity of photovoltaic output to temperature changes

Temperature range/℃ Sensitivity/kW Temperature range/℃ Sensitivity/kW - - [26, 27) 13.996 [16, 17) 15.186 [27, 28) 13.877 [17, 18) 15.067 [28, 29) 13.758 [18, 19) 14.948 [29, 30) 13.640 [19, 20) 14.829 [30,31) 13.521 [20, 21) 14.710 [31, 32) 13.402 [21, 22) 14.591 [32,33) 13.283 [22, 23) 14.472 [33, 34) 13.164 [23, 24) 14.353 [34, 35) 13.045 [24, 25) 14.234 [35, 36) 12.926 [25, 26) 14.115 - -

Table 3 Sensitivity of photovoltaic output to changes in irradiance intensity

(4) For low-frequency series and high-frequency series, respectively establish forecasting models considering meteorological factors.

Considering that the steps of modeling and forecasting low-frequency/high-frequency series are the same, the following takes the forecasting process of high-frequency series as an example to illustrate.

The specific method is as follows:

a. Select the reference day and benchmark value of the high-frequency series, take the day before the day to be predicted as the reference day of the high-frequency series, and use the high-frequency series of photovoltaic output on the reference day as the benchmark value of the high-frequency series on the day to be predicted.

b. The Pearson correlation coefficient between different meteorological factors and photovoltaic output is used as the weight coefficient of the meteorological factors affecting the change of photovoltaic output.

c. According to the sensitivity of meteorological factors to the change of photovoltaic output, the temperature difference and radiation difference between the day to be predicted and the reference day, the high-frequency sequence P _high of photovoltaic output is corrected according to formula (10).

P _high ＝P' _high + α ₁ ΔP ₁ + α ₂ ΔP ₂ ... (10); where: P _high , P' _high , ΔP ₁ and ΔP ₂ are the high-frequency sequence of photovoltaic output on the day to be predicted, and the reference Daily photovoltaic output high-frequency sequence, photovoltaic output high-frequency sequence variation caused by temperature changes, and photovoltaic output high-frequency sequence variation caused by radiation changes. α ₁ and α ₂ are the weight coefficients of temperature and radiation affecting the high-frequency sequence change of photovoltaic output, respectively.

It can be seen from formula (10) that the corrected amount includes ΔP ₁ and ΔP ₂ , and the present invention adopts the following method to determine the values of the two, and ΔP ₁ is taken as an example for illustration below.

(a) When the daily temperature to be predicted is in the same sensitivity range as the reference daily temperature:

ΔP ₁ =S _t (t-t')...(11);

(b) When the daily temperature to be predicted and the reference daily temperature are in two different sensitivity intervals, if two adjacent intervals are taken as an example, then In the formula: t and t' represent the temperature values of the day to be predicted and the reference day, respectively; S _t and S' _t represent the sensitivity of the respective intervals of the temperature of the day to be predicted and the reference day; Represents the temperature value at the common endpoint of the two intervals.

(5) Superimpose the predicted value of the low-frequency sequence and the predicted value of the high-frequency sequence obtained in step (4) according to formula (13) to obtain the predicted value of photovoltaic output.

P=P _low +P _high ... (13); wherein, P _low and P _high are predicted values of low frequency sequence and high frequency sequence respectively.

The implementation examples of the present invention will be described in further detail below in conjunction with the accompanying drawings.

The invention proposes a method suitable for forecasting intermittent renewable energy sources such as wind power and photovoltaics; specifically, it provides a singular spectrum analysis method for short-term photovoltaic output forecast considering meteorological factors. This method can decompose the photovoltaic output into different sub-sequences through singular spectrum analysis technology (Singular Spectrum Analysis, referred to as SSA method), and can analyze the characteristics of each sequence separately; through correlation analysis and sensitivity analysis, the units of different meteorological factors can be obtained The degree of influence of the change on the photovoltaic output can be used to more accurately predict the photovoltaic output and provide favorable data reference for dispatching decision makers, thereby reducing the impact of photovoltaic output access on the power system.

Implementation step 1: Obtain the data of photovoltaic output, temperature, radiation, wind speed, rainfall from May 1, 2013 to May 30, 2014, and the data of temperature and radiation in May 2014. The present invention uses the data of the past year to predict the photovoltaic output in the next day. In the invention, the data from May 1, 2013 to May 30, 2014 is used as the prediction sample, and the data in May 2014 is used as the test sample.

Implementation step 2: Use SSA technology to decompose the time series of photovoltaic output, and obtain the low-frequency sequence, high-frequency sequence and noise sequence of photovoltaic output, as shown in Figure 1. Since the noise sequence is reconstructed from a sub-matrix with a small proportion of eigenvalues and has little effect on the original data, consider removing the noise sequence. Therefore, the present invention focuses on predicting low-frequency sequences and high-frequency sequences.

Implementation step 3: Use the Pearson correlation coefficient method to determine the main meteorological factors affecting the change of photovoltaic output. According to the calculation results in Table 1, temperature and radiation intensity are the main meteorological factors affecting photovoltaic output.

Implementation step 4: Calculate the sensitivity of the main meteorological factors, namely temperature and irradiance intensity, to changes in photovoltaic output, as shown in Table 2 and Table 3.

Implementation step 5: According to the result of sensitivity calculation, model and predict the low-frequency sequence and high-frequency sequence respectively. The prediction results of low-frequency sequence and high-frequency sequence are obtained, as shown in Figure 4(a) and 4(b).

Implement step 6: superimpose the low-frequency sequence and high-frequency sequence obtained in step 5 to obtain the prediction result of photovoltaic output in Figure 4(c).

Those skilled in the art can easily understand that the above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention, All should be included within the protection scope of the present invention.

Claims (6)

1. A short-term photovoltaic decomposition forecasting method considering meteorological factors variation, is characterized in that, comprises the following steps: S1: Decompose the photovoltaic output time series by singular spectrum analysis method, obtain low frequency series, high frequency series and noise series, and eliminate the noise series at the same time; S2: Use the Pearson correlation coefficient method to determine the main meteorological factors affecting photovoltaic output, and analyze the sensitivity of the main meteorological factors to photovoltaic output; S3: For the low-frequency sequence and the high-frequency sequence and in combination with the sensitivity, respectively establish a high-frequency sequence prediction model and a low-frequency sequence prediction model considering meteorological factors; S4: Obtain the predicted value of the high-frequency sequence according to the prediction model of the high-frequency sequence, obtain the predicted value of the low-frequency sequence according to the prediction model of the low-frequency sequence; and obtain the photovoltaic value according to the predicted value of the low-frequency sequence and the predicted value of the high-frequency sequence output forecast. 2. The short-term photovoltaic decomposition prediction method according to claim 1, wherein step S1 is specifically: S11: Transform the photovoltaic output time series into a matrix form, and decompose the matrix into the sum of d equivalent sub-matrices; S12: Decompose the d sub-matrices obtained After grouping, the low-frequency matrix Z _low , high-frequency matrix Z _high and noise matrix Z _noise are obtained, and the low-frequency matrix Z _low , high-frequency matrix Z _high and noise matrix Z _noise are respectively diagonally averaged and restored to the original sequence form The low-frequency sequence P _l , the high-frequency sequence _{Ph and the noise sequence P n are} obtained after the sequences are reconstructed. 3. The short-term photovoltaic decomposition prediction method as claimed in claim 1 or 2, characterized in that, in step S2, the Pearson correlation coefficient method is used to determine the main meteorological factors affecting photovoltaic output are specifically: S21: Select temperature, radiation, wind speed and rainfall as meteorological factors; S22: According to the formula Calculate the Pearson correlation coefficient between photovoltaic output and temperature, irradiance, wind speed or rainfall respectively; S23: Determine the main meteorological factors affecting photovoltaic output according to the size of the Pearson correlation coefficient; in, Pearson correlation coefficient absolute value of The closer to 1, the higher the degree of linear correlation between the two variables. 4. The short-term photovoltaic decomposition prediction method as claimed in any one of claims 1-3, wherein the prediction model of the high-frequency sequence that considers meteorological factors is set up in step S3 as follows: (1) Select the reference date and benchmark value of the high-frequency series: Take the day before the day to be predicted as the reference day for the high-frequency sequence, and use the high-frequency sequence of photovoltaic output on the reference day as the benchmark value of the high-frequency sequence on the day to be predicted; (2) The Pearson correlation coefficient between different meteorological factors and photovoltaic output is used as the weight coefficient of the meteorological factors affecting the change of photovoltaic output; (3) According to the sensitivity of meteorological factors to the change of photovoltaic output, the temperature difference and radiation difference between the day to be predicted and the reference day, and according to the formula P _high = P _h ' high + α _{1 ΔP 1} + α ₂ ΔP ₂ for the high photovoltaic output The frequency sequence P _high is corrected; Among them, P _high is the high-frequency sequence of photovoltaic output on the day to be predicted, P _h ' high is the high-frequency sequence of photovoltaic output on the reference day, _{ΔP 1} is the change in high-frequency sequence of photovoltaic output due to temperature change, and ΔP ₂ is the change in high-frequency sequence of photovoltaic output due to radiation α ₁ is the weight coefficient of temperature affecting the high-frequency sequence change of photovoltaic output, and α ₂ is the weight coefficient of radiation affecting the high-frequency sequence change of photovoltaic output. 5. The short-term photovoltaic decomposition prediction method as claimed in claim 4, characterized in that, when the daily temperature to be predicted and the reference day temperature are in the same sensitivity interval, ΔP ₁ =S _t (t-t'); When the temperature and the reference day temperature are in two different sensitivity intervals, Among them, t is the daily temperature value to be predicted, t' is the reference daily temperature value, S _t is the sensitivity of the interval of the daily temperature to be predicted, S _t ' is the sensitivity of the interval of the reference daily temperature, Represents the temperature value at the common endpoint of the two intervals. 6. The short-term photovoltaic decomposition prediction method according to any one of claims 1-5, characterized in that in step _S4 , the _photovoltaic output prediction value P= P _low +P _high .