Photovoltaic power generation power prediction method based on modal switching and parallel weight adjustment

Abstract

The present application belongs to the technical field of new energy prediction of power system, and specifically relates to a photovoltaic power generation power prediction method based on mode switching and parallel weight adjustment, comprising the following steps: constructing a data set by using historical photovoltaic data of a photovoltaic power station, determining fluctuations of the data set according to global horizontal irradiance and time sequence, optimizing and adjusting weights of different categories of fluctuation period data through correlation analysis and feature extraction, constructing a parallel LSTM neural network, training the data after weight optimization and adjustment to obtain three trained parallel modes, and finally inputting data at a prediction time, determining the data at the time, determining the mode to be entered by the data at the time, and predicting power by using the corresponding mode. Compared with the prior art, the prediction result is more accurate and has high prediction accuracy, and can better solve random fluctuations caused by various interferences.

Description

Technical Field

[0001] This invention belongs to the field of new energy prediction technology for power systems, specifically, it relates to a photovoltaic power generation prediction method based on mode switching and parallel weight adjustment. Background Technology

[0002] In terms of meteorological data processing, existing research includes ultra-short-term photovoltaic power generation forecasting techniques that consider cloud cover. These techniques address the impacts of cloud clusters at different heights, cloud movement, and power attenuation coefficients under cloud cover on photovoltaic power plants while ensuring forecast accuracy. Furthermore, in terms of data processing, existing technologies introduce concepts such as clustering and similar days to categorize historical photovoltaic data into different weather types, reducing the influence of different weather conditions on forecasts and thus improving forecast accuracy.

[0003] The shortcomings or deficiencies of the above-mentioned technologies are: they mainly focus on the impact of single meteorological factors on photovoltaic power and classify historical photovoltaic data into different types to improve prediction accuracy, but rarely consider the problem of missing meteorological data of some types due to insufficient detection methods for photovoltaic power plants, as well as the impact of random and specific fluctuations in photovoltaic power series in the time domain on photovoltaic prediction and the problem of easily ignoring important information.

[0004] Therefore, given the complex power output process of photovoltaic power plants, the frequent disturbances caused by various meteorological factors, and the insufficient and unmeasurable detection methods for photovoltaic power plants, which leads to low prediction accuracy, those skilled in the art urgently need to propose a new photovoltaic power generation prediction method based on mode switching and parallel weight adjustment. Summary of the Invention

[0005] The purpose of this invention is to address the problems of low prediction accuracy caused by the complex output process of photovoltaic power plants, frequent disturbances from various meteorological factors, and insufficient or unmeasurable detection methods. This problem is solved by adopting a photovoltaic power generation prediction method based on mode switching and parallel weight adjustment.

[0006] To achieve the above-mentioned objectives, the present invention proposes the following technical solution:

[0007] The photovoltaic power generation prediction method based on mode switching and parallel weight adjustment includes the following steps:

[0008] S1. Constructing the dataset: Historical photovoltaic data from photovoltaic power plants is extracted, and a dataset is constructed with the time sampling points of the historical photovoltaic data as rows and different meteorological factors as columns. The dataset includes the feature value matrices of the historical photovoltaic data [X1, X2, ..., X...]. q ] Matrix of average values ​​of each feature Standard deviation matrix of each feature [δ1,δ2,...,δ q]; where q is the dimension of the data features;

[0009] S2. Determine the fluctuation of the dataset: Based on the global horizontal irradiance, divide the dataset into high fluctuation period data and low fluctuation period data. Then, based on the time series as the judgment condition, divide the high fluctuation period data into specific high fluctuation period data and random high fluctuation period data.

[0010] S3. Weight Optimization and Adjustment: Correlation analysis is used to determine the impact of different meteorological factors on photovoltaic power, and the input sequences of data for different types of fluctuation periods are determined. Then, through feature extraction, the weights of the input sequences are optimized and adjusted to increase the weight of important information.

[0011] S4. Modal Training: Construct a parallel LSTM neural network, perform modal training on the data after weight optimization and adjustment to obtain three trained parallel modes. The data trained during random high fluctuation periods is denoted as Mode 1, the data trained during specific high fluctuation periods is denoted as Mode 2, and the data trained during low fluctuation periods is denoted as Mode 3.

[0012] S5. Power Prediction: Input the prediction time data, determine the mode that the data should enter at that time, and use the corresponding mode to perform power prediction. Then, inversely normalize the predicted power in the mode to obtain the final prediction result.

[0013] Preferably, in step S1, the historical photovoltaic data includes time data measured every 5 minutes within the last two years, power data, performance ratio data, wind speed data, wind direction data, temperature data, humidity data, rainfall data, global horizontal radiation data, diffuse horizontal radiation data, global tilted radiation data, and scattered tilted radiation data.

[0014] Preferably, step S1 further includes preprocessing of the dataset, the preprocessing including the following steps:

[0015] Step S11, according to the 3-sigma principle, perform abnormal data detection using the following formula (1):

[0016]

[0017] In equation (1), X i =[X1,X2,...,X q [i] represents the i-th data point in each eigenvalue matrix of the historical photovoltaic data; δ is the i-th data point in the matrix of average values ​​of each feature of the historical photovoltaic data; i =[δ1,δ2,...,δ q [i] represents the i-th data point in the standard deviation matrix of each feature of the historical photovoltaic data; p represents the number of data samples;

[0018] S12: The feature data that conforms to equation (1) are identified as outliers, and the outlier data is processed using the mean substitution method to obtain the processed data matrix [y1, y2, ..., y i ];

[0019] S13: For the data matrix [y1,y2,...,y] i Normalization is performed using the following formula (2):

[0020]

[0021] In equation (2), v n Represents the normalized data; v i Indicates the original data; v min Indicates the minimum value of the data; v max This indicates the maximum value of the data.

[0022] Preferably, step S2, determining the fluctuation of the dataset, includes the following steps:

[0023] S21: The global horizontal irradiance fluctuation is defined according to the following formula (3):

[0024]

[0025] In equation (3), F t F represents the global horizontal irradiance at time t. t-1 α represents the global horizontal irradiance at time t. t Let α represent the fluctuation value of the global horizontal irradiance at time t. t The threshold is 30;

[0026] S22: When α t When α < 30, the dataset is determined to be data from a period of low fluctuation. t When the value is ≥30, the dataset is determined to be data from a period of high volatility.

[0027] S23: Calculate the sun's position and describe its trajectory using the following formula (4);

[0028]

[0029] In equation (4), φ represents the solar zenith angle, and λ represents the solar altitude angle. The latitude of the observation point is represented by γ, the solar declination angle is represented by η, and the hour angle is represented by the current time.

[0030] S24: When the solar trajectory corresponding to the high fluctuation period data satisfies the following formulas (5) and (6), it is determined to be specific high fluctuation period data; otherwise, it is random high fluctuation period data.

[0031]

[0032]

[0033] In equations (5) and (6), θ represents the angle between the photovoltaic panel and the ground.

[0034] Preferably, in step S3, the following formula (7) is used to perform correlation analysis on the meteorological factor data of different categories of fluctuation period data:

[0035]

[0036] Select meteorological factor data with |P|≥0.4 as the input sequence;

[0037] In equation (7), A i Data representing meteorological factors, Representing A i The average value of B i Represents power data, Representing B i The average value.

[0038] Preferably, in step S3, the feature extraction includes the following steps:

[0039] S31: Calculate the weights between elements based on equation (8)

[0040]

[0041]

[0042] In equation (8), This represents the query for element i at time t. Represents the power element j at time t. D(·) represents the correlation between query element i and power element j at time t, and D(·) represents the dot product function. represents the weight between query element i and power element j at time t, and k represents all query elements at time t;

[0043] S32: Weighted summation of meteorological data at the same time using formula (9)

[0044]

[0045] In the formula, Q t It represents the new element after weighted summation of all elements at time t.

[0046] Preferably, in step S4, the modality training includes the following steps:

[0047] S41: Construct three parallel LSTM neural networks, and add decision conditions and optimization channels before the LSTM neural networks;

[0048] S42: Substitute historical photovoltaic data into the judgment conditions in step S2 and enter different optimization channels;

[0049] S43: Perform correlation analysis and feature extraction within the optimized channel to achieve weight optimization and adjustment;

[0050] S44: Input the data with optimized weights into the LSTM neural network, set the LSTM parameters, and control the prediction loss value to be less than 0.2;

[0051] S45: Perform training to obtain three parallel modes.

[0052] Preferably, in step S5, the power prediction includes the following steps:

[0053] S51: Input the predicted time data and calculate the fluctuation value α between the current time and the previous time using formula (3). t ;

[0054] S52: If α t ≥30, measure the angle θ between the photovoltaic panel and the ground, determine the latitude and longitude of the photovoltaic power station, and use formula (4) to determine the solar zenith angle φ and solar altitude angle γ at that moment; and determine whether the relationship between θ and the solar zenith angle φ and solar altitude angle λ satisfies formula (5) and formula (6). If it satisfies, enter mode two; if it does not satisfy, enter mode one.

[0055] S53: If α t If the value is less than 30, then mode three is entered;

[0056] S54: Based on the mode to be entered by the data at this moment, the meteorological factors of the data at this moment are optimized and adjusted by weight using formulas (7), (8) and (9);

[0057] S55: Input the data with optimized weights into the mode, adjust its weights with historical photovoltaic data at different times, and predict the power generation of the photovoltaic power station at that moment.

[0058] Preferably, in step S5, the predicted power in the mode is denormalized according to the following formula (10):

[0059] v = 0.5[v n (v max -v min )+v max +v min (10)

[0060] In equation (10), v represents the data after inverse normalization.

[0061] The present invention also includes other steps that enable normal use, all of which are conventional technical means in the field. In addition, steps not limited in the present invention all adopt conventional means in the prior art.

[0062] Compared with the prior art, the beneficial effects of the present invention are:

[0063] The present invention provides more accurate and precise predictions and can better address random fluctuations caused by various disturbances. Attached Figure Description

[0064] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0065] Figure 1 This is an overall process flow diagram of the present invention in the embodiments.

[0066] Figure 2 This is a comparison chart of the predicted value and the actual value of the present invention in the embodiment.

[0067] Figure 3 This is a comparison chart of the prediction results of different models in the embodiments.

[0068] Figure 4 This is a comparison chart of the absolute errors of the prediction results of different models in the embodiment. Detailed Implementation

[0069] The present invention will now be clearly described with reference to the accompanying drawings and specific embodiments. This description is merely illustrative and is not intended to limit the scope of the invention. Any modifications, equivalent substitutions, or improvements made by those skilled in the art based on the embodiments of the present invention without inventive effort to obtain all other embodiments should be included within the scope of protection of the present invention.

[0070] Example

[0071] like Figure 1 As shown, the photovoltaic power generation prediction method based on mode switching and parallel weight adjustment provided by the present invention includes the following steps:

[0072] S1. Constructing the dataset: Historical photovoltaic data from photovoltaic power plants is extracted, and a dataset is constructed with the time sampling points of the historical photovoltaic data as rows and different meteorological factors as columns. The dataset includes the feature value matrices of the historical photovoltaic data [X1, X2, ..., X...]. q ] Matrix of average values ​​of each feature Standard deviation matrix of each feature [δ1,δ2,...,δ q ]; where q is the dimension of the data features;

[0073] S2. Determine the fluctuation of the dataset: Based on the global horizontal irradiance, divide the dataset into high fluctuation period data and low fluctuation period data. Then, based on the time series as the judgment condition, divide the high fluctuation period data into specific high fluctuation period data and random high fluctuation period data.

[0074] S3. Weight Optimization and Adjustment: Correlation analysis is used to determine the impact of different meteorological factors on photovoltaic power, and the input sequences of data for different types of fluctuation periods are determined. Then, through feature extraction, the weights of the input sequences are optimized and adjusted to increase the weight of important information.

[0075] S4. Modal Training: Construct a parallel LSTM neural network, perform modal training on the data after weight optimization and adjustment to obtain three trained parallel modes. The data trained during random high fluctuation periods is denoted as Mode 1, the data trained during specific high fluctuation periods is denoted as Mode 2, and the data trained during low fluctuation periods is denoted as Mode 3.

[0076] S5. Power Prediction: Input the prediction time data, determine the mode that the data should enter at that time, and use the corresponding mode to perform power prediction. Then, inversely normalize the predicted power in the mode to obtain the final prediction result.

[0077] Preferably, in step S1, the historical photovoltaic data includes time data measured every 5 minutes within the last two years, power data, performance ratio data, wind speed data, wind direction data, temperature data, humidity data, rainfall data, global horizontal radiation data, diffuse horizontal radiation data, global tilted radiation data, and scattered tilted radiation data.

[0078] Step S1 further includes preprocessing of the dataset, which includes the following steps:

[0079] Step S11, according to the 3-sigma principle, perform abnormal data detection using the following formula (1):

[0080]

[0081] In equation (1), X i =[X1,X2,...,X q [i] represents the i-th data point in each eigenvalue matrix of the historical photovoltaic data; δ is the i-th data point in the matrix of average values ​​of each feature of the historical photovoltaic data; i =[δ1,δ2,...,δ q [i] represents the i-th data point in the standard deviation matrix of each feature of the historical photovoltaic data; p represents the number of data samples;

[0082] S12: The feature data that conforms to equation (1) are identified as outliers, and the outlier data is processed using the mean substitution method to obtain the processed data matrix [y1, y2, ..., y i ];

[0083] S13: For the data matrix [y1,y2,...,y] i Normalization is performed using the following formula (2):

[0084]

[0085] In equation (2), v n Represents the normalized data; v i Indicates the original data; v min Indicates the minimum value of the data; v max This indicates the maximum value of the data.

[0086] Furthermore, in step S2, determining the fluctuation of the dataset includes the following steps:

[0087] S21: The global horizontal irradiance fluctuation is defined according to the following formula (3):

[0088]

[0089] In equation (3), F t F represents the global horizontal irradiance at time t. t-1 α represents the global horizontal irradiance at time t. t Let α represent the fluctuation value of the global horizontal irradiance at time t. t The threshold is 30;

[0090] S22: When α t When α < 30, the dataset is determined to be data from a period of low fluctuation. t When the value is ≥30, the dataset is determined to be data from a period of high volatility.

[0091] S23: Calculate the sun's position and describe its trajectory using the following formula (4);

[0092]

[0093] In equation (4), φ represents the solar zenith angle, and λ represents the solar altitude angle. The latitude of the observation point is represented by γ, the solar declination angle is represented by η, and the hour angle is represented by the current time.

[0094] S24: When the solar trajectory corresponding to the high fluctuation period data satisfies the following formulas (5) and (6), it is determined to be specific high fluctuation period data; otherwise, it is random high fluctuation period data.

[0095]

[0096]

[0097] In equations (5) and (6), θ represents the angle between the photovoltaic panel and the ground.

[0098] Furthermore, in step S3, the following equation (7) is used to perform correlation analysis on the meteorological factor data of different categories of fluctuation period data:

[0099]

[0100] Select meteorological factor data with |P|≥0.4 as the input sequence;

[0101] In equation (7), A i Data representing meteorological factors, Representing A i The average value of B i Represents power data, Representing B i The average value.

[0102] The feature extraction in step S3 includes the following steps:

[0103] S31: Calculate the weights between elements based on equation (8)

[0104]

[0105]

[0106] In equation (8), This represents the query for element i at time t. Represents the power element j at time t. D(·) represents the correlation between query element i and power element j at time t, and D(·) represents the dot product function. represents the weight between query element i and power element j at time t, and k represents all query elements at time t;

[0107] S32: Weighted summation of meteorological data at the same time using formula (9)

[0108]

[0109] In the formula, Q t It represents the new element after weighted summation of all elements at time t.

[0110] In this embodiment, the modality training in step S4 includes the following steps:

[0111] S41: Construct three parallel LSTM neural networks, and add decision conditions and optimization channels before the LSTM neural networks;

[0112] S42: Substitute historical photovoltaic data into the judgment conditions in step S2 and enter different optimization channels;

[0113] S43: Perform correlation analysis and feature extraction within the optimized channel to achieve weight optimization and adjustment;

[0114] S44: Input the data with optimized weights into the LSTM neural network, set the LSTM parameters, and control the prediction loss value to be less than 0.2;

[0115] S45: Perform training to obtain three parallel modes.

[0116] Specifically, in step S5, the power prediction includes the following steps:

[0117] S51: Input the predicted time data and calculate the fluctuation value α between the current time and the previous time using formula (3). t ;

[0118] S52: If α t ≥30, measure the angle θ between the photovoltaic panel and the ground, determine the latitude and longitude of the photovoltaic power station, and use formula (4) to determine the solar zenith angle φ and solar altitude angle γ at that moment; and determine whether the relationship between θ and the solar zenith angle φ and solar altitude angle λ satisfies formula (5) and formula (6). If it satisfies, enter mode two; if it does not satisfy, enter mode one.

[0119] S53: If α t If the value is less than 30, then mode three is entered;

[0120] S54: Based on the mode to be entered by the data at this moment, the meteorological factors of the data at this moment are optimized and adjusted by weight using formulas (7), (8) and (9);

[0121] S55: Input the data with optimized weights into the mode, adjust its weights with historical photovoltaic data at different times, and predict the power generation of the photovoltaic power station at that moment.

[0122] Finally, in step S5, the predicted power in the mode is denormalized according to the following formula (10):

[0123] v = 0.5[v n (v max -v min )+v max +v min (10)

[0124] In equation (10), v represents the data after inverse normalization.

[0125] To verify the predictive effectiveness of the above method, this embodiment uses data from a distributed solar photovoltaic power station in Uluru, central Australia, for case study verification and analysis. The power data of this photovoltaic power station is updated every 5 minutes. Power data from the period of March 1, 2019 to March 1, 2021 is selected, and an example is shown to illustrate the actual power generation curve for one day as follows: Figure 2 As shown by the mid-star line.

[0126] The prediction results of this invention are compared with those of models that do not consider mode partitioning (DL model), ATT-LSTM model, and LSTM model. To avoid the influence of other factors, this embodiment uses the same input data for different models. The prediction results of each model are as follows: Figure 3 As shown, the absolute errors of prediction results from different models are compared as follows: Figure 4 As shown.

[0127] The following conclusions can be drawn from the above comparison results:

[0128] (1) The prediction results of the prediction method proposed in this invention are closer to the true value than the prediction results of other models, and the absolute error is also more stable and closer to the X-axis, indicating that the prediction results of this invention are more accurate and can better solve the random fluctuations caused by various interferences.

[0129] (2) The DL model prediction results are better than those of other models, which proves the superiority of selecting appropriate meteorological data for prediction by conducting correlation analysis on the data.

[0130] (3) Compared with the LSTM model, the ATT-LSTM model predicts the results more closely to the true values, indicating that the data after weight adjustment and reorganization can improve the prediction accuracy.

[0131] The embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A photovoltaic power generation prediction method based on mode switching and parallel weight adjustment, characterized in that, Includes the following steps: S1. Constructing the dataset: Historical photovoltaic data from photovoltaic power plants is extracted, and a dataset is constructed with the time sampling points of the historical photovoltaic data as rows and different meteorological factors as columns. The dataset includes the feature value matrices of the historical photovoltaic data. Matrix of average values ​​of each feature Standard deviation matrix of each feature ;in, q The dimension of the data features; S2. Determine the fluctuation of the dataset: Based on the global horizontal irradiance, divide the dataset into high fluctuation period data and low fluctuation period data. Then, based on the time series as the judgment condition, divide the high fluctuation period data into specific high fluctuation period data and random high fluctuation period data. The fluctuation determination of the dataset includes the following steps: S21: The global horizontal irradiance fluctuation is defined according to the following formula (3): （3） In equation (3), Represents the global horizontal irradiance at time t. Represents the global horizontal irradiance at time t. The value representing the fluctuation of global horizontal irradiance at time t is set as follows: The threshold is 30; S22: When When <30, the dataset is determined to be data from a low-fluctuation period. When the value is ≥30, the dataset is determined to be data from a period of high volatility. S23: Calculate the sun's position and describe its trajectory using the following formula (4); （4） In equation (4), Represents the zenith angle of the sun. Represents the solar altitude angle. Represents the latitude of the observation point. Represents the solar declination angle. The hour angle represents the current time; S24: When the solar trajectory corresponding to the high fluctuation period data satisfies the following formulas (5) and (6), it is determined to be specific high fluctuation period data; otherwise, it is random high fluctuation period data. （5） （6） In equations (5) and (6), Represents the angle between the photovoltaic panel and the ground; S3. Weight Optimization and Adjustment: Correlation analysis is used to determine the impact of different meteorological factors on photovoltaic power, and the input sequences of data for different types of fluctuation periods are determined. Then, through feature extraction, the weights of the input sequences are optimized and adjusted to increase the weight of important information. S4. Modal Training: Construct a parallel LSTM neural network, perform modal training on the data after weight optimization and adjustment to obtain three trained parallel modes. The data trained during random high fluctuation periods is denoted as Mode 1, the data trained during specific high fluctuation periods is denoted as Mode 2, and the data trained during low fluctuation periods is denoted as Mode 3. The modality training includes the following steps: S41: Construct three parallel LSTM neural networks, and add decision conditions and optimization channels before the LSTM neural networks; S42: Substitute historical photovoltaic data into the judgment conditions in step S2 and enter different optimization channels; S43: Perform correlation analysis and feature extraction within the optimized channel to achieve weight optimization and adjustment; S44: Input the data with optimized weights into the LSTM neural network, set the LSTM parameters, and control the prediction loss value to be less than 0.2; S45: Perform training to obtain three parallel modes; S5. Power Prediction: Input the prediction time data, determine the mode that the data should enter at that time, and use the corresponding mode to perform power prediction. Then, inversely normalize the predicted power in the mode to obtain the final prediction result.

2. The photovoltaic power generation prediction method based on mode switching and parallel weight adjustment according to claim 1, characterized in that, In step S1, the historical photovoltaic data includes time data measured every 5 minutes over the past two years, power data, performance ratio data, wind speed data, wind direction data, temperature data, humidity data, rainfall data, global horizontal radiation data, diffuse horizontal radiation data, global tilted radiation data, and scattered tilted radiation data.

3. The photovoltaic power generation prediction method based on mode switching and parallel weight adjustment according to claim 2, characterized in that, Step S1 also includes preprocessing of the dataset, which includes the following steps: Step S11, according to the 3-sigma principle, perform abnormal data detection using the following formula (1): （1） In equation (1), This refers to the i-th data point in the eigenvalue matrix of the historical photovoltaic data. This refers to the i-th data point in the matrix of average values ​​of each feature of the historical photovoltaic data. This refers to the i-th data point in the standard deviation matrix of each feature of the historical photovoltaic data. p The number of samples in the data; S12: The feature data that conforms to equation (1) are identified as outliers, and the outlier data is processed using the mean substitution method to obtain the processed data matrix. ; S13: Data matrix Normalization is performed using the following formula (2): （2） In equation (2), This represents the normalized data; Represents the original data; This represents the minimum value of the data. This indicates the maximum value of the data.

4. The photovoltaic power generation prediction method based on mode switching and parallel weight adjustment according to claim 3, characterized in that, In step S3, the following equation (7) is used to perform correlation analysis on meteorological factor data for different categories of fluctuation periods: （7） choose Meteorological factor data are used as the input sequence; In equation (7), A i Data representing meteorological factors, represent A i The average value, B i Represents power data, represent B i The average value.

5. The photovoltaic power generation prediction method based on mode switching and parallel weight adjustment according to claim 4, characterized in that, In step S3, the feature extraction includes the following steps: S31: Calculate the weights between elements based on equation (8) （8） In equation (8), This represents the query for element i at time t. Represents the power element j at time t. This represents the correlation between query element i and power element j at time t. Represents the dot product function. represents the weight between query element i and power element j at time t, and k represents all query elements at time t; S32: Weighted summation of meteorological data at the same time using formula (9) （9） In the formula, It represents the new element after weighted summation of all elements at time t.

6. The photovoltaic power generation prediction method based on mode switching and parallel weight adjustment according to claim 5, characterized in that, In step S5, the power prediction includes the following steps: S51: Input the predicted time data and calculate the fluctuation value between the current time and the previous time using formula (3). ; S52: If ≥30, measure the angle between the photovoltaic panel and the ground. Determine the latitude and longitude of the photovoltaic power station, and use formula (4) to determine the solar zenith angle at that moment. and solar altitude angle ; and judge With the zenith angle of the sun and solar altitude angle If the relationship satisfies formulas (5) and (6), then proceed to mode two; otherwise, proceed to mode one. S53: If If the value is less than 30, then mode three is entered; S54: Based on the mode to be entered by the data at this moment, the meteorological factors of the data at this moment are optimized and adjusted by weight using formulas (7), (8) and (9); S55: Input the data with optimized weights into the mode, adjust its weights with historical photovoltaic data at different times, and predict the power generation of the photovoltaic power station at that moment.

7. The photovoltaic power generation prediction method based on mode switching and parallel weight adjustment according to claim 6, characterized in that, In step S5, the predicted power in the mode is denormalized according to the following formula (10): （10） In equation (10), This represents the data after denormalization.

Images