Prediction method and device for photovoltaic power generation

Abstract

The invention relates to a photovoltaic power generation prediction method and equipment, and the method comprises the steps: obtaining the prediction input data of a photovoltaic power station in each time period of a to-be-predicted date, and the prediction input data comprise prediction meteorological data corresponding to the time period and historical output reference data; inputting the prediction input data into a slice prediction model corresponding to each time period to obtain a first prediction result of each time period; based on point prediction and / or interval prediction of the day-ahead photovoltaic power generation amount, multi-decision information with meteorological characteristics is obtained, and the obtained multi-decision information comprises the day-ahead predicted total power generation amount and / or the interval of the day-ahead predicted total power generation amount at the date to be predicted; and correcting the first prediction result of each time period according to the multi-decision information. By adopting the method provided by the invention, the problems of low precision and large error of photovoltaic power generation prediction period by period in the prior art can be improved.

Description

Technical Field

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

[0002] Photovoltaic power generation has the advantages of being clean, environmentally friendly, and sustainable, but it is greatly affected by weather conditions, and the intensity and time of light are not fixed. Accurate prediction of photovoltaic power generation can help grid operators better plan and manage power systems, ensuring reliable access to photovoltaic power generation and stable operation of the grid.

[0003] The determining factor for the accuracy of photovoltaic power generation prediction is weather forecast information. When predicting photovoltaic power generation in each period a day before, the prediction accuracy of short-term weather information in each period is low, which leads to low precision and large error in photovoltaic power generation prediction in each period. Summary of the invention

[0004] Based on this, a photovoltaic power generation prediction method, device, computer equipment and storage medium are provided to improve the problems of low accuracy and large error in photovoltaic power generation prediction in each time period in the prior art.

[0005] On the one hand, a photovoltaic power generation prediction method is provided, comprising:

[0006] Obtaining forecast input data of the photovoltaic power station for each time period on the date to be forecasted, wherein the forecast input data includes forecast meteorological data corresponding to the time period and historical output reference data;

[0007] Input the prediction input data into the slice prediction model corresponding to each time period to obtain the first prediction result of each time period, wherein the slice prediction model corresponding to the time period is trained based on the historical output data of the photovoltaic power station at the same time period every day in the historical time period and the corresponding historical meteorological data;

[0008] Based on the point prediction and / or interval prediction of the photovoltaic power generation on the previous day, multiple decision information with meteorological characteristics is obtained, wherein the multiple decision information includes the total power generation predicted on the previous day and / or the interval of the total power generation predicted on the previous day on the date to be predicted, wherein the point prediction includes inputting the meteorological mean data of the date to be predicted into the power generation point prediction model to obtain the total power generation predicted on the previous day, wherein the power generation point prediction model is trained based on the meteorological mean data and actual output data of each day in a historical time period; the interval prediction includes obtaining the interval of the total power generation predicted on the previous day based on the prediction error of the point prediction;

[0009] The first prediction result of each time period is corrected according to the multiple decision information to obtain the second prediction result of each time period.

[0010] In one embodiment, obtaining the interval of the day-ahead predicted total power generation based on the prediction error of the point prediction includes:

[0011] According to the prediction error conditional probability distribution, a lower error limit and an upper error limit under a preset confidence level are obtained, wherein the prediction error conditional probability distribution is obtained according to the actual output data and the corresponding predicted output data of each day in a historical time period;

[0012] According to the lower error limit and the upper error limit, on the basis of the total power generation predicted on the day before, an interval of the total power generation predicted on the day before is obtained.

[0013] In one of the embodiments, the multiple decision information also includes a forecast error-corrected total power generation;

[0014] The obtaining of multiple decision information with meteorological characteristics also includes:

[0015] Obtaining the forecast error corrected total power generation according to the integral of the forecast error conditional probability distribution between the error lower limit and the error upper limit;

[0016] The first prediction result of each time period is corrected by taking the total power generation predicted on the day before, the interval of the total power generation predicted on the day before, and the total power generation corrected by the prediction error as multi-decision information.

[0017] In one embodiment, the first prediction result of each time period is modified according to the multiple decision information to obtain the second prediction result of each time period, including:

[0018] The multiple decision information and the first prediction result of each time period are input into a pre-trained correction model to obtain the second prediction result of each time period.

[0019] In one embodiment, inputting the prediction input data into the slice prediction model corresponding to each time period to obtain the first prediction result of each time period includes:

[0020] According to the normalized forecast input data, forecast the normalized forecast output data corresponding to the time period;

[0021] According to the non-normalized predicted input data, predict and obtain non-normalized predicted output data corresponding to the time period;

[0022] When the non-normalized predicted output data and the normalized predicted output data corresponding to a time period are both smaller than the segmentation threshold of the time period, the non-normalized predicted output data is determined to be the first prediction result of the time period; otherwise, the normalized predicted output data is determined to be the first prediction result of the time period.

[0023] In one embodiment, inputting the prediction input data into the slice prediction model corresponding to each time period to obtain the first prediction result of each time period includes:

[0024] The predicted input data of each time period is used as non-normalized input data, and input into the non-normalized slice prediction model corresponding to each time period to obtain the non-normalized predicted output data of each time period;

[0025] Performing normalization processing on the predicted input data corresponding to each time period to obtain normalized input data for each time period;

[0026] Inputting the normalized input data into the normalized slice prediction model corresponding to each time period to obtain the normalized predicted output data of each time period;

[0027] When the non-normalized predicted output data and the normalized predicted output data corresponding to a time period are both less than the segmentation threshold of the time period, the non-normalized predicted output data is determined to be the first prediction result of the time period; otherwise, the normalized predicted output data is determined to be the first prediction result of the time period;

[0028] Among them, the non-normalized slice prediction model corresponding to the time period is trained based on the non-normalized historical output data and the non-normalized historical meteorological data of the time period; the normalized slice prediction model corresponding to the time period is trained based on the normalized historical output data and the normalized historical meteorological data of the time period.

[0029] In one embodiment, inputting the meteorological mean data of the date to be predicted into the power generation point prediction model to obtain the total power generation predicted on the day before includes:

[0030] Determine the target weather type for the location of the photovoltaic power station on the date to be predicted;

[0031] The meteorological mean data is input into the power generation point prediction model corresponding to the target weather type of the date to be predicted to obtain the total power generation predicted on the day before, wherein the power generation point prediction model corresponding to the target weather type is trained based on a first data set corresponding to the target weather type in a historical time period, and the first data set includes the meteorological mean data corresponding to the target weather type in a historical time period, and the actual output data corresponding to the meteorological mean data.

[0032] In one embodiment, the prediction error conditional probability distribution is obtained based on actual output data and corresponding predicted output data for each day in a historical time period, including:

[0033] For each target weather type, determine a corresponding Copula function; the variable of the Copula function is a second data set corresponding to the target weather type, and the second data set includes the actual output data and the predicted output data;

[0034] Calculating a correlation coefficient according to the actual output data corresponding to the target weather type and the corresponding predicted output data, and calculating a Copula parameter of the Copula function corresponding to the target weather type according to the correlation coefficient;

[0035] According to the actual output data corresponding to the target weather type, an actual output marginal density function and an actual output marginal distribution function corresponding to the target weather type are calculated, and according to the predicted output data corresponding to the target weather type, a predicted output marginal distribution function corresponding to the target weather type is calculated;

[0036] According to the Copula function corresponding to the target weather type, the Copula parameters, the actual output marginal density function, the actual output marginal distribution function and the predicted output marginal distribution function, the prediction error conditional probability distribution under the target predicted output value is calculated.

[0037] In one embodiment, the first data set or the second data set corresponding to the target weather type in the historical time period is determined according to the following steps:

[0038] Using a rain flow counting method to extract features from a first actual output value sequence determined by the actual output data to obtain corresponding counting feature data; and / or performing differential processing on a second actual output value sequence determined by the actual output data to obtain corresponding output trend feature data;

[0039] Based on the counting feature data and / or the output trend feature data, the actual output data belonging to the same target weather type and the meteorological mean data corresponding to the actual output data are divided into the same data set to obtain the first data set corresponding to each target weather type;

[0040] Based on the counting characteristic data and / or the output trend characteristic data, the actual output data belonging to the same target weather type and the predicted output data corresponding to the actual output data are divided into the same data set to obtain the second data set corresponding to each target weather type.

[0041] On the other hand, a photovoltaic power generation prediction device is provided, comprising a processor and a memory, wherein a computer program is stored in the memory, and the processor executes the computer program to implement the method described above.

[0042] The above-mentioned photovoltaic power generation prediction method and device obtain the prediction input data of the photovoltaic power station in each time period on the date to be predicted, input the prediction input data into the slice prediction model corresponding to each time period, and obtain the first prediction result of each time period; then based on the point prediction and / or interval prediction of the photovoltaic power generation on the day before, obtain multiple decision information with meteorological characteristics, wherein the point prediction includes inputting the meteorological mean data of the date to be predicted into the power generation point prediction model to obtain the predicted total power generation on the day before, and the power generation point prediction model is trained according to the daily meteorological mean data and actual output data in the historical time period; the interval prediction includes obtaining the interval of the predicted total power generation on the day before based on the prediction error of the point prediction, and correcting the first prediction result of each time period according to the multiple decision information to obtain the second prediction result of each time period; in the present application, the multiple decision information is obtained based on the meteorological mean data, and the meteorological mean data has a lower prediction difficulty and higher accuracy than the meteorological forecast data of each time period. The use of relatively high-precision multiple decision information to adjust the first prediction result of each time period can effectively improve the prediction accuracy and reduce the prediction error. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a schematic flow chart of a photovoltaic power generation prediction method in one embodiment;

[0044] Figure 2 A photovoltaic output sequence diagram at the same time every day in a historical time period in an embodiment;

[0045] Figure 3 A scatter plot comparison of power generation in each time period before and after data cleaning in an embodiment;

[0046] Figure 4 A training process of a slice prediction model in another embodiment;

[0047] Figure 5 It is a schematic diagram of a flow chart of predicting a first prediction result in another embodiment;

[0048] Figure 6 A comparison diagram of predictions of normalized and non-normalized slice prediction models in one embodiment;

[0049] Figure 7 is a schematic diagram of a prediction error conditional probability distribution function in one embodiment;

[0050] Figure 8 A result comparison diagram of a first prediction result before and after being corrected by multiple decision information in an embodiment;

[0051] Fig. 9 It is a structural block diagram of a photovoltaic power generation prediction device in one embodiment. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0053] Embodiment 1:

[0054] The integration of photovoltaic power generation may have an impact on the stability of the power grid. Accurate prediction can help grid operators better plan and manage the power system, ensuring the reliable integration of photovoltaic power generation and the stable operation of the power grid.

[0055] The time period prediction of photovoltaic power generation is affected by the meteorological information of each time period and has low accuracy. This embodiment provides a photovoltaic power generation prediction method, which corrects the time period prediction results in combination with a higher accuracy point prediction, thereby improving the accuracy of the time period prediction.

[0056] like Figure 1 As shown, the prediction method includes:

[0057] S11: Obtaining forecast input data of the photovoltaic power station for each time period on the date to be forecasted, wherein the forecast input data includes forecast meteorological data corresponding to the time period and historical output reference data.

[0058] S12: Input the prediction input data into the slice prediction model corresponding to each time period to obtain the first prediction result of each time period.

[0059] S13: Based on the point prediction and / or interval prediction of the photovoltaic power generation on the previous day, multiple decision-making information with meteorological characteristics is obtained.

[0060] S14: According to the multiple decision information, the first prediction result of each time period is corrected to obtain the second prediction result of each time period.

[0061] The above steps are described in detail below.

[0062] In step S11, each day may be divided into a plurality of time periods at certain time intervals. For example, a day may be divided into 24 time periods at hourly intervals.

[0063] Each time period has corresponding forecast input data for that time period, and the forecast input data includes the forecast meteorological data corresponding to the time period and the historical output reference data.

[0064] The predicted meteorological data includes at least one of the predicted ambient temperature data, ambient humidity data, radiation intensity data, and scattered radiation data for the period.

[0065] The historical output reference data is data in the historical power generation data that can provide additional reference for the prediction, such as the power generation or power generation in the same period of the day before the predicted date, the power generation or power generation in multiple periods or multiple days before the predicted period. For example, for predicting the power generation from 12 to 13 noon on the second day, the power generation in the same period of the previous day, the power generation from 9 to 10, 10 to 11, and 11 to 12 in the past days can be used as the historical output reference data.

[0066] For step S12, each time period corresponds to a slice prediction model for power generation prediction, wherein the slice prediction model corresponding to the time period is trained based on the historical output data of the photovoltaic power station at the same time period every day in the historical time period and the corresponding historical meteorological data.

[0067] The historical time period in the embodiment of the present application may be any time period in the past, for example, the past two years, the past year, the past month, etc.

[0068] For each day in the historical time period, the influence of meteorological data such as light intensity and temperature changes at each moment. Slicing, aggregating and classifying the historical output data and historical meteorological data of the same period can effectively reduce the prediction error of moments with similar characteristics and improve the similarity of the internal laws of historical photovoltaic power data.

[0069] For example, according to the 24 time periods of a day, the historical output data and historical meteorological data in the historical time period are divided into a time series data set at the same time of each day, and a total of 24 time series data sets are sliced ​​and aggregated. Modeling training is performed based on the 24 time series data sets to obtain a slice prediction model for each time period.

[0070] like Figure 2 As shown, the photovoltaic output sequence diagram at the same time of the day in the historical time period is disclosed as an example. The horizontal axis of the figure is the number of days, and the vertical axis is the normalized power generation of each day. The figure shows the output sequences of 17 time periods in parallel. The first two and the last are horizontal lines, indicating that there is usually no power generation at night, and the night time is simplified. In the figure, the photovoltaic output in the morning and evening periods fluctuates significantly with the season, with small mean and amplitude, and is nonlinear overall; when the light irradiation intensity is high at noon, the photovoltaic output fluctuates less with the season, with large mean and amplitude, and is linear overall.

[0071] The historical output data of each time period may be the power generated by the photovoltaic power station in the corresponding time period; or may be the actual power generation of the photovoltaic power station in the time period.

[0072] The historical meteorological data for each time period may be the ambient temperature data, ambient humidity data, irradiation intensity data, and scattered radiation data of the location of the photovoltaic power station during the corresponding time period.

[0073] The first prediction result may be the predicted power generation in each time period, or the power generation power in each time period.

[0074] Optionally, the historical output data in the embodiment of the present application may be data after the 3σ criterion and local outlier factors are eliminated, so as to effectively eliminate noise. Figure 3 As shown in FIG. 1 , a scatter plot comparison of power generation in each time period versus total light irradiance intensity before and after data cleaning is disclosed. It can be seen that the scatter noise can be effectively removed after data cleaning.

[0075] In particular, it should be pointed out that for general photovoltaic forecasts, the existing technology uses unified normalization processing for historical output data and historical meteorological data. However, when modeling and predicting, normalizing all values ​​may not be the best method. The reason is that a part of the data with smaller values ​​will be normalized and reduced, which will reduce the accuracy and reliability of data analysis and modeling, and thus affect the prediction accuracy of this segment of data.

[0076] In order to improve the accuracy of the prediction, in the embodiment of the present application, the data of the same period are respectively normalized and non-normalized, and then a more accurate prediction result is selected. The process is exemplified as follows:

[0077] On the one hand, based on the predicted input data after normalization, the normalized predicted output data corresponding to the time period is predicted. In this embodiment, the predicted input data after normalization is input into the model for prediction, and the output is denormalized to obtain the normalized predicted output data (for example, the predicted power generation for a certain time period).

[0078] On the other hand, according to the non-normalized predicted input data, the non-normalized predicted output data corresponding to the time period is predicted. In this embodiment, the predicted input data is input into the model for prediction, and its output is the non-normalized predicted output data;

[0079] When the non-normalized predicted output data and the normalized predicted output data corresponding to a time period are both smaller than the segmentation threshold of the time period, the non-normalized predicted output data is determined to be the first prediction result of the time period; otherwise, the normalized predicted output data is determined to be the first prediction result of the time period.

[0080] The following describes a slice prediction model training process and prediction process.

[0081] Exemplarily, this embodiment provides the following training process of the slice prediction model to reduce the impact of the prediction accuracy caused by the unified normalization process. The training process is as follows:

[0082] Step S21, according to the 24 time periods of a day, the historical output data and historical meteorological data in the historical time period are divided into a time series data set at the same time of each day, and a total of 24 time series data sets are sliced, aggregated and classified.

[0083] In step S22, the 24 time series data sets sliced, aggregated and classified in step S21 are normalized respectively to obtain the normalized historical output data and the normalized historical meteorological data of each time period, which are hereinafter referred to as normalized training data; on the other hand, the original 24 time series data sets are used as non-normalized training data.

[0084] Step S23, use the initial neural network model to model and learn the normalized training data and the non-normalized training data respectively to obtain 48 prediction models, thereby establishing a prediction model library, the 48 prediction models include non-normalized slice prediction models corresponding to 24 time periods obtained by training based on the non-normalized historical output data of each time period and the non-normalized historical meteorological data, and normalized slice prediction models corresponding to 24 time periods obtained by training based on the normalized historical output data of each time period and the normalized historical meteorological data.

[0085] The initial neural network model is, for example, an LSTM long short-term memory neural network. The training of the LSTM long short-term memory neural network refers to the existing method and will not be described in detail here.

[0086] like Figure 4 As shown, an exemplary embodiment discloses the training process of a slice prediction model in an embodiment, firstly, data cleaning, slicing by time period, 3σ criterion processing and local outlier factor removal are performed, and the data of the same period are respectively input into the model in two paths, normalization and non-normalization, and the model training is started one by one. The initial neural network model adopts LSTM long short time memory neural network, and judges whether model i is the last one according to whether model i is less than or equal to the number of slices n. After the i-th model training is completed, it can be saved and used to predict the photovoltaic output of the i-th time period.

[0087] In the same time period, two prediction results can be obtained based on the non-normalized slice prediction model and the normalized slice prediction model, and a more accurate prediction can be obtained based on the selection of the two prediction results.

[0088] Based on the non-normalized slice prediction model and the normalized slice prediction model, for step S12, the first prediction result of each time period is predicted in the following steps:

[0089] Sub-step S121: taking the predicted input data of each time period as non-normalized input data, inputting it into the non-normalized slice prediction model corresponding to each time period, and obtaining the non-normalized predicted output data of each time period.

[0090] Sub-step S122: performing normalization processing on the predicted input data corresponding to each time period to obtain normalized input data for each time period, inputting the normalized input data into the normalized slice prediction model corresponding to each time period to obtain normalized predicted output data for each time period.

[0091] Sub-step S123: When the non-normalized predicted output data and the normalized predicted output data corresponding to the time period are both smaller than the segmentation threshold of the time period, the non-normalized predicted output data is determined to be the first prediction result of the time period; otherwise, the normalized predicted output data is determined to be the first prediction result of the time period.

[0092] For sub-step S121, the predicted input data of each time period is not normalized and can be used as the input of the non-normalized slice prediction model to obtain the non-normalized predicted output data of each time period, such as the predicted power generation of each time period.

[0093] For sub-step S122, normalization processing such as maximum and minimum normalization is performed, and after normalization, the normalized slice prediction model is input to obtain a normalized output, and the output is denormalized to obtain the normalized predicted output data for each time period, such as the predicted power generation for each time period.

[0094] For sub-step S123, the prediction results of the two models are compared with the segmentation threshold. The segmentation threshold is set based on the typical interval of smaller values ​​in the output data that are easily affected by normalization. It can be flexibly set according to needs during actual implementation. By way of example, the segmentation threshold is defined as 1 (in kw). When the prediction results of the two prediction models are both less than 1, the prediction results adopt the prediction results of the non-normalized slice prediction model; when the prediction results of the two prediction models are not both less than 1, the prediction results adopt the prediction results of the normalized slice prediction model.

[0095] like Figure 5 As shown, an exemplary process of determining a first prediction result in an embodiment is disclosed, in which the prediction input data is cleaned, sliced ​​by time periods, processed by the 3σ criterion, and local outlier factors are eliminated. The data of the same time period are input into the model in two paths, normalized and non-normalized, respectively, and the model prediction is started one by one. A judgment is made based on the two prediction results to obtain the final first prediction result.

[0096] like Figure 6In the comparison diagram of time period prediction using normalized and non-normalized slice prediction models, the horizontal and vertical axes represent the actual results and the predicted results, respectively. As shown in the figure, for data less than the segmentation threshold 1, the predicted results are closer to the actual results without normalization, and for data greater than the segmentation threshold 1, the predicted results are closer to the actual results after normalization. Therefore, based on the segmentation threshold, the normalized prediction results and the non-normalized prediction results are comprehensively judged to obtain the first prediction result, which is more accurate than the prediction method that uses unified normalization processing.

[0097] Based on the above process, the first prediction result for each time period can be obtained. However, the first prediction result is affected by the predicted meteorological data of each time period. The prediction of meteorological data for each time period is difficult and has low accuracy, resulting in the accuracy of the first prediction result needs to be further improved.

[0098] For step S13, the multiple decision information is photovoltaic power generation prediction data obtained based on more accurate weather forecast information. In this embodiment, the multiple decision information includes the total power generation predicted on the day before the prediction date, which can be obtained based on the point prediction of the photovoltaic power generation on the day before. Alternatively, the multiple decision information includes the interval of the total power generation predicted on the day before the prediction date, and the interval of the total power generation predicted on the day before the prediction date can be obtained based on the interval prediction of the photovoltaic power generation on the day before the prediction.

[0099] More accurate weather forecast information, such as the average meteorological data for the date to be predicted, is easier to predict and the prediction results are more accurate than the predicted meteorological data for each period of the day. The average meteorological data for the day include average ambient temperature, average ambient humidity, average radiation intensity, and average scattered radiation.

[0100] Point forecasting predicts the total power generation on the day ahead based on meteorological mean data. Compared with time period forecasting, it is less difficult and more accurate.

[0101] In this embodiment, point prediction is implemented based on a point prediction model for power generation. The meteorological mean data of the date to be predicted is input into the point prediction model for power generation to obtain the total power generation predicted on the day before. The point prediction model for power generation is trained based on the meteorological mean data and actual output data of each day in a historical time period.

[0102] The interval forecast is realized based on the forecast error analysis of the point forecast. By analyzing the forecast error of the point forecast, the upper and lower limits of the total power generation forecast on the day before are obtained based on the total power generation forecast on the day before.

[0103] For step S14, the total power generation predicted on the day before with lower prediction difficulty and higher accuracy or the interval of the total power generation predicted on the day before is used as multi-decision information to correct the first prediction result and obtain a more accurate second prediction result.

[0104] The second prediction result may be data of the same type as the first prediction result. For example, if the first prediction result is the power generation in each time period, the second prediction result is the corrected power generation in each time period.

[0105] The correction method of the first prediction result can be implemented based on a neural network. According to the obtained multi-decision information, the multi-decision information and the first prediction result of each time period are input into a pre-trained correction model to obtain the second prediction result of each time period. The correction model is, for example, a BP neural network model.

[0106] In the above embodiment, the power generation point prediction model is obtained by training based on the daily meteorological mean data and actual output data in the historical time period. The following is an exemplary explanation of the training process of the power generation point prediction model:

[0107] Step S31: Obtain actual daily output data of the photovoltaic power station in a historical period and corresponding meteorological mean data.

[0108] Step S32: using the rain flow counting method to perform feature extraction on the first actual output value sequence determined by the actual output data to obtain corresponding counting feature data; and / or performing differential processing on the second actual output value sequence determined by the actual output data to obtain corresponding output trend feature data;

[0109] Step S33: Based on the counting characteristic data and / or the output trend characteristic data, the actual output data belonging to the same target weather type and the meteorological mean data corresponding to the actual output data are divided into the same data set to obtain the first data set corresponding to each target weather type.

[0110] Step S34: training is performed according to a first data set corresponding to a target weather type in a historical time period to obtain a power generation point prediction model corresponding to the target weather type.

[0111] The above process divides the historical power generation data belonging to the same target weather type into the same data set based on the counting feature data, thereby ensuring the consistency and regularity of each historical power generation data in the target data set corresponding to the same target weather type. The historical power generation data belonging to the same target weather type is divided into the same data set based on the output trend feature data, thereby eliminating the seasonal influence and ensuring the accuracy of the obtained target data set, thereby improving the accuracy of the training results and thus improving the accuracy of the prediction results.

[0112] The above steps are described in detail below.

[0113] Each actual output data in the embodiment of the present application may include the first actual output value corresponding to each first unit time of the photovoltaic power station on the corresponding historical date, and / or the second actual output value corresponding to each second unit time. The first actual output value and the second actual output value may be the actual power generation of the photovoltaic power station in the corresponding unit time, for example, the actual power generation of the photovoltaic power station every five minutes every day; or the actual power generation of the photovoltaic power station in the unit time, for example, the actual power generation of the photovoltaic power station every five minutes every day.

[0114] It should be noted that the duration corresponding to the first unit time and the duration corresponding to the second unit time in the embodiment of the present application can be flexibly set by the developer. Based on the first actual output value corresponding to the first unit time, rain flow counting learning can ensure the accuracy of the rain flow learning results, accurately distinguish the actual output data corresponding to each target weather type, and at the same time ensure the consistency and regularity of each actual output data under the same target weather type. Differential processing is performed based on the second actual output value corresponding to the second unit time, fully considering the balance between the efficiency and accuracy of data processing.

[0115] For step S32, in the embodiment of the present application, a three-point rain flow counting method can be used to obtain counting feature data. Of course, other types of rain flow counting methods can also be used to obtain counting feature data.

[0116] The process of extracting features from the first actual output value sequence composed of the first actual output values ​​by using the rain flow counting method to obtain corresponding counting feature data may be as follows:

[0117] Sub-step S321: extracting peak-valley value data from the first actual output value sequence to obtain a corresponding peak-valley value data sequence.

[0118] Sub-step S322: placing the maximum peak value or the minimum valley value in the peak-to-valley value data sequence at the starting position in the peak-to-valley value data sequence and performing cyclic counting to obtain the frequency, mean value and amplitude difference corresponding to each actual output value interval.

[0119] Sub-step S323: taking the three-dimensional data consisting of the frequency, mean value and amplitude difference corresponding to each actual output value interval as the counting feature data corresponding to the actual output data.

[0120] In the embodiment of the present application, for each actual output data, all the first actual output values ​​corresponding thereto may be sequentially composed of a first actual output value sequence in chronological order. In the above sub-step S321, the first actual output value sequence may be filtered so that it only contains peak data and valley data, thereby obtaining a corresponding peak-valley value data sequence. In some embodiments, after filtering the first actual output value sequence to obtain peak data and valley data, invalid amplitudes may be removed to obtain a more accurate peak-valley value data sequence.

[0121] In the above sub-step S322, any one of the maximum peak value and the minimum valley value in the peak-valley value data sequence may be placed at the starting position in the peak-valley value data sequence and then cycle counting may be performed.

[0122] In the above sub-step S323, the three-dimensional data consisting of the frequency, mean and amplitude difference corresponding to each actual output value interval can be used as the counting feature data corresponding to the actual output data, that is, for each actual output data, it corresponds to multiple groups of counting feature data.

[0123] In the embodiment of the present application, the rain flow counting method is used to perform feature value learning on the actual daily output data in the historical time period. After peak and valley value detection and invalid amplitude removal, the waveform frequency, amplitude difference and mean changes caused by weather changes can be effectively captured.

[0124] Step S33 may include the following sub-steps:

[0125] Sub-step S331: Obtain daily meteorological mean data for the location of the photovoltaic power station in a historical time period.

[0126] Sub-step S332: Obtain the daily professional weather types at the location of the photovoltaic power station in the historical time period to obtain M professional weather types.

[0127] Sub-step S333: Based on the meteorological mean data, the M professional weather types at the location of the photovoltaic power station are divided into N target weather types to obtain a mapping relationship between the professional weather types and the target weather types; wherein M>N.

[0128] Sub-step S334: Based on the mapping relationship and the daily professional weather type in the historical time period, the actual output data of each day in the historical time period is initialized for weather, and the target weather type corresponding to each actual output data is obtained.

[0129] Sub-step S335: According to the target weather type corresponding to each actual output data, the actual output data belonging to the same target weather type are divided into the same data set to obtain an initial data set corresponding to each target weather type.

[0130] Sub-step S336: determining the geometric center of each initial data set according to the first actual output value or the second actual output value of each actual output data in each initial data set.

[0131] Sub-step S337: Aggregate and classify each actual output data based on the counting feature data and / or output trend feature data corresponding to each actual output data, and each geometric center, and divide each meteorological mean data into the data set where the corresponding actual output data is located, to obtain a first data set corresponding to each target weather type.

[0132] The meteorological mean data in the embodiment of the present application includes but is not limited to at least one of the average ambient temperature, the average ambient humidity, the average irradiance intensity, and the average scattered irradiance. For example, the irradiance intensity may include the attenuation degree of the surface irradiance relative to the external irradiance. In the above sub-step S331, the attenuation degree of the daily average surface irradiance relative to the external irradiance at the location of the power station in the historical time period may be obtained.

[0133] Generally speaking, there are 33 broad professional weather types. In one embodiment, the M professional weather types at the location of the photovoltaic power station can be divided into N target weather types to meet the preset first division target and second division target, wherein the first division target is: the distribution quantity of each actual output data on each target weather type is balanced, and the second division target is: multiple professional weather types with similar attenuation degrees are divided into the same target weather type. In this embodiment, the distribution of each actual output data on different target weather types can be more balanced, so that the sample data corresponding to each target weather type can be sufficient. In some other embodiments, the M professional weather types at the location of the photovoltaic power station can be divided into N target weather types, as long as it is ensured to meet the preset first division target or the second division target.

[0134] For the first classification target, balanced distribution data means that the difference between the total number of actual output data corresponding to each target weather type is within a preset range. For the second classification target, when the difference between the attenuation degrees corresponding to any two professional weather types is within a preset range, it can be considered that the attenuation degrees of the two professional weather types are similar.

[0135] Through sub-step S333, M kinds of professional weather types can be divided into N kinds of target weather types, that is, a mapping relationship between professional weather types and target weather types can be obtained.

[0136] In sub-step S334, based on the mapping relationship and the daily professional weather type in the historical time period, the weather of the daily power generation historical data in the historical time period is initialized to obtain the target weather type corresponding to each actual output data. Exemplarily, 33 generalized professional weather types can be divided into 4 target weather types, for example, they can be divided into four target weather types A, B, C, and D. Assume that the four generalized professional weather types of "sunny", "sunny and cloudy", "cloudy and sunny" and "cloudy" have similar attenuation degrees and are classified under the same target weather type A. And assuming that the professional weather type under the historical date corresponding to a certain actual output data in sub-step S332 is "sunny and cloudy", then in sub-step S333, according to the mapping relationship between "sunny and cloudy" and target weather type A, the weather type corresponding to the actual output data can be initialized to target weather type A. Therefore, under the premise of the assumption of this example, target weather type A can be used as the target weather type corresponding to the actual output data in sub-step S337.

[0137] Through the above sub-step S335, the actual output data corresponding to each target weather type can be obtained, and the set of actual output data corresponding to the same target weather type is the initial data set corresponding to the target weather type.

[0138] In the above sub-step S336, the geometric center of each initial data set may be determined according to the first actual output value or the second actual output value of each actual output data in each initial data set.

[0139] In the above sub-step S337, the counting feature data and / or the output trend feature data can be used as the classification dimension, and the geometric center can be used as the center of each category to aggregate and classify the actual output data. When performing aggregate classification, the Euclidean distance calculation formula can be used to calculate the Euclidean distance between each actual output data and each geometric center, and the actual output data can be classified based on the Euclidean distance. Exemplarily, the actual output data can be aggregated and classified based on the K-means unsupervised learning algorithm, so that a reclassified data set can be obtained for each target weather type.

[0140] In an embodiment of the present application, the geometric center of each initial data set is determined according to the first output value or the second output value of each actual output data in each initial data set, and clustering is performed based on the geometric center. This can avoid the problem of high randomness of clustering results and inaccurate clustering caused by randomly selecting the geometric center for clustering.

[0141] After the historical data are classified according to the target weather types, the power generation point prediction model corresponding to each target weather type can be trained based on the first training set corresponding to each target weather type.

[0142] Based on the power generation point prediction model corresponding to each target weather type, the point prediction of the predicted date is completed. The process is as follows:

[0143] Determine the target weather type for the location of the PV power station on the date to be predicted.

[0144] The meteorological mean data is input into a power generation point prediction model corresponding to the target weather type of the date to be predicted to obtain the total power generation predicted on the day before.

[0145] In some embodiments, based on the prediction error of the point prediction, obtaining the interval of the total power generation predicted the day before includes obtaining a lower error limit and an upper error limit under a preset confidence level according to the conditional probability distribution of the prediction error, and obtaining the interval of the total power generation predicted the day before based on the total power generation predicted the day before according to the lower error limit and the upper error limit.

[0146] The prediction error conditional probability distribution is obtained based on the actual output data of each day in the historical time period and the corresponding predicted output data. The process is exemplified as follows:

[0147] Step S41: determining a corresponding Copula function for the target weather type of the date to be predicted; the variable of the Copula function is a second data set corresponding to the same target weather type, and the second data set includes actual output data and predicted output data.

[0148] Step S42: Calculate a correlation coefficient based on the actual output data corresponding to the target weather type and the corresponding predicted output data, and calculate the Copula parameter of the Copula function corresponding to the target weather type based on the correlation coefficient.

[0149] Step S43: Based on the actual output data corresponding to the target weather type, calculate the actual output edge density function and the actual output edge distribution function corresponding to the target weather type, and based on the predicted output data corresponding to the target weather type, calculate the predicted output edge distribution function corresponding to the target weather type.

[0150] Step S44: Calculate the prediction error conditional probability distribution under the target predicted output value according to the Copula function corresponding to the target weather type, the Copula parameters, the actual output marginal density function, the actual output marginal distribution function and the predicted output marginal distribution function.

[0151] For step S41, the daily output of the photovoltaic power station in the historical time period can be predicted according to a preset output prediction model (such as the aforementioned power generation point prediction model) to obtain corresponding predicted output data. In this way, each actual output data has its corresponding predicted output data.

[0152] The division process of the second data set is similar to that of the first data set. First, the rain flow counting method is used to extract features of the first actual output value sequence determined by the actual output data to obtain corresponding counting feature data; and / or the second actual output value sequence determined by the actual output data is differentially processed to obtain corresponding output trend feature data; then, based on the counting feature data and / or the output trend feature data, the actual output data belonging to the same target weather type and the predicted output data corresponding to the actual output data are divided into the same data set to obtain the second data set corresponding to each target weather type.

[0153] In the above step S41, each target weather type has its corresponding Copula function. Exemplarily, the target weather type includes a first target weather type and a second target weather type, and the output value of the photovoltaic power station per unit time under the first target weather type is greater than the output value of the photovoltaic power station per unit time under the second target weather type, and the Copula function corresponding to the first target weather type is a function that is sensitive to changes at the lower tail, and the Copula function corresponding to the second target weather type is a function that is sensitive to changes at the upper tail.

[0154] Take the target weather types of "sunny", "cloudy", "moderate rain" and "heavy rain" as examples for illustrative explanation. The photovoltaic output level of "sunny" and "cloudy" is relatively high. When it changes to "moderate rain" or "heavy rain", the Copula function should be able to describe the photovoltaic output decreasing with the weather changes. Therefore, for "sunny" and "cloudy", a Copula function that is sensitive to changes in the lower tail can be selected for it. Similarly, the photovoltaic output level of "moderate rain" and "heavy rain" is relatively low. When it changes to "sunny" or "cloudy", the Copula function should be able to describe the photovoltaic output increasing with the weather changes. Therefore, for "moderate rain" and "heavy rain", a Copula function that is sensitive to changes in the upper tail can be selected for it.

[0155] Exemplarily, the Copula function corresponding to the first target weather type is a Clayton Copula function, and the Copula function corresponding to the second target weather type is a Gumbel Copula function.

[0156] For step S42, the correlation coefficient refers to a coefficient that can reflect the correlation between the two variables, the actual output data and the predicted output data. Any coefficient that can characterize the correlation between the two variables can be used as the correlation coefficient here. Exemplarily, the Kendall correlation coefficient can be calculated based on the actual output data corresponding to the target weather type and the corresponding predicted output data. The method for calculating the Kendall correlation coefficient can refer to the existing method, which will not be repeated here.

[0157] After the correlation coefficient is obtained, the Copula parameter of the corresponding Copula function can be calculated according to the correlation coefficient.

[0158] The correlation coefficient is the Kendall correlation coefficient, and the Copula functions are the Clayton Copula function and the Gumbel Copula function. After calculating the Kendall correlation coefficient τ according to the actual output data corresponding to the target weather type and the corresponding predicted output data, the Kendall correlation coefficient τ is substituted into the following formula (1) and formula (2) to obtain the Copula parameter θ1 of the Clayton Copula function and the Copula parameter θ2 of the Gumbel Copula function.

[0159] The Clayton Copula function is:

[0160]

[0161] Among them, θ1 represents the Copula parameter of the Clayton Copula function, θ1∈[0,∞], and the relationship between the Kendall correlation coefficient and the Copula parameter θ1 of the Clayton Copula function is as follows:

[0162]

[0163] The Gumbel Copula function is:

[0164]

[0165] Among them, θ2 represents the Copula parameter of the Gumbel Copula function, θ2∈[0,∞], and the relationship between the Kendall correlation coefficient and the Copula parameter θ2 of the Gumbel Copula function is as follows:

[0166]

[0167] In the above formula, u represents the actual output marginal distribution function mentioned below, and v represents the predicted output marginal distribution function mentioned below. The specific method for obtaining the function is described in detail below.

[0168] Step S43 includes the following sub-steps:

[0169] S431: Determine the actual output value interval range to be subjected to probability statistics according to the actual output data corresponding to the target weather type, and determine the predicted output value interval range to be subjected to probability statistics according to the predicted output data corresponding to the target weather type.

[0170] S432: According to the actual output data corresponding to the target weather type, a first number of actual output data falling within each actual output value interval is counted, and according to the predicted output data corresponding to the target weather type, a second number of predicted output data falling within each predicted output value interval is counted.

[0171] S433: Obtaining an actual output edge density function according to each first quantity and each actual output value interval range, and obtaining a predicted output edge density function according to each second quantity and each predicted output value interval range.

[0172] S434: Integrate the actual output edge density function to obtain the actual output edge distribution function, and integrate the predicted output edge density function to obtain the predicted output edge distribution function.

[0173] It should be noted that the duration corresponding to the first time unit and the duration corresponding to the second time unit mentioned above in the embodiment of the present application can be less than 24 hours, and the time unit is made small enough to ensure the accuracy of the data set division. At this time, the actual output data can also include the daily actual output value, which represents the actual total output value of the photovoltaic power station on the corresponding historical date. Correspondingly, the predicted output data may include the daily predicted output value, which represents the predicted total output value of the photovoltaic power station on the corresponding historical date. At this time, when performing calculations related to the marginal density function, the marginal distribution function and the prediction error conditional probability distribution, the daily actual output value and the daily predicted output value can be calculated as two related variables.

[0174] Assume that the actual daily output values ​​corresponding to a certain target weather type are X1, X2, …, X n ∈[a,b), the interval [a,b) can be divided as follows:

[0175] a=a0<a1<a2<…<a k = b;

[0176] Define any small interval as I i =[a i-1 ,ai ), i=1,2,…k,[a i-1 ,a i ) is the actual output value range, obviously:

[0177] And I i ∩I j = Ф, i≠j;

[0178] Definition i ={X i ∈I i} is within the actual output value range I i The first number of daily actual output values ​​in , so the histogram density estimate can be defined as:

[0179]

[0180] In actual operation, the same interval is often taken, that is, I i The width of is h, then:

[0181]

[0182] The above functions (3) and (4) represent the actual output edge density function f X , which can also be written as f X (x), X represents the actual daily output value as a variable, where h is the normalization parameter and represents the group distance of each group, which is called window width or bandwidth.

[0183] The above describes the calculation of the actual output edge density function by the histogram density method. It is understandable that the predicted output edge density function f can also be calculated in a similar way as above. Y , which can also be written as f Y (y), which will not be described here.

[0184] It should be noted that, in some other embodiments, a first total number of actual output data corresponding to the target weather type can be calculated, and a first ratio between the first number and the first total number can be calculated. The corresponding actual output edge density function can be obtained from the first ratio and the corresponding actual output value interval. Similarly, a second total number of predicted output data corresponding to the target weather type can be calculated. Generally speaking, the first total number and the second total number are equal. The second ratio between the second number and the second total number is calculated. The corresponding predicted output edge density function can be obtained from the second ratio and the corresponding predicted output value interval.

[0185] In step S434, the actual output edge density function f X Integrate to get the actual output marginal distribution function F X, which can also be written as F X (x). And the predicted output edge density function f Y Integrate to obtain the predicted output marginal distribution function F Y , which can also be written as F Y (y). Actual output marginal density function f X and the actual output marginal distribution function F X The independent variable is the actual output data x, and the predicted output marginal density function f Y And the predicted output marginal distribution function F Y The independent variable is the predicted output data y.

[0186] Next, the function of the prediction error conditional probability distribution is derived based on the Copula function, the actual output data of the photovoltaic power station, and the predicted output data of the photovoltaic power station:

[0187] The joint distribution function of actual output data and predicted output data is:

[0188] F X,Y (x,y)=C(F X (x),F Y (y));

[0189] Where C is the Copula function, F X (x) is the actual output marginal distribution function, F Y (y) is the predicted output marginal distribution function.

[0190] The joint density function of actual output data and predicted output data is:

[0191]

[0192] Where c is the derivative of the Copula function C, f X (x) is the actual output edge density function, f Y (y) is the predicted output marginal density function.

[0193] When the predicted output data takes the target predicted output value, that is, when y=p, the probability density of the actual output data is:

[0194]

[0195] Let e ​​= xy, then the conditional probability distribution of the prediction error is:

[0196]

[0197] Therefore, in step S44, according to formula f E|Y (e|y=p)=c(F X(p+e),F Y (p))·f X (p+e) calculates the conditional probability distribution of the prediction error under the target predicted output value, where f E|Y represents the prediction error conditional probability distribution, y represents the predicted output data, p represents the target predicted output value, y=p represents the predicted output data taking the target predicted output value, and e represents the error between the actual output data and the corresponding predicted output data. For each target weather type, the corresponding prediction error conditional probability distribution can be obtained by the above-mentioned method. In some embodiments, in order to control the error, after obtaining the prediction error conditional probability distribution, the lower limit and upper limit of the error under the target predicted output value can also be calculated according to the preset confidence corresponding to the target weather type.

[0198] The prediction error conditional probability distribution function can be found in Figure 7 As shown, the shaded area in the curve represents the probability that the error falls outside the confidence interval.

[0199] Given a confidence level, we can use Calculate the lower limit of error under the target predicted output value, and Calculate the upper limit of the error under the target predicted output value; where 1-α represents the preset confidence corresponding to the target weather type, e1 represents the lower limit of the error, and e2 represents the upper limit of the error. Of course, in other embodiments, other methods can also be used to obtain the upper and lower limits of the error. The value of α can be flexibly set by the developer, for example, it can be set to 0.05.

[0200] It should be noted that the specific size of the target predicted output value in the embodiment of the present application can be set arbitrarily, thereby enabling the estimation of the conditional probability distribution of the prediction error of the actual photovoltaic output corresponding to any point and the estimation of the upper and lower limit errors of the photovoltaic output to be achieved under a given confidence level.

[0201] After obtaining the upper and lower error limits, the lower limit of the total power generation predicted the day before is the sum of the lower error limit and the total power generation predicted the day before, the upper limit of the total power generation predicted the day before is the sum of the upper error limit and the total power generation predicted the day before, and the interval of the total power generation predicted the day before is the range between the lower and upper limits of the total power generation predicted the day before.

[0202] The above process, for the second data set corresponding to different target weather types, can effectively use the Copula function that is sensitive to changes at the upper tail and the Copula function that is sensitive to changes at the lower tail according to the data change law. When estimating the conditional probability distribution of the prediction error of the actual photovoltaic output corresponding to the prediction at any point, it effectively improves the estimation accuracy and can provide more decision-making information.

[0203] In another embodiment, the multiple decision information further includes a forecast error corrected total power generation amount, wherein the forecast error corrected total power generation amount is obtained according to the integral of the forecast error conditional probability distribution between the error lower limit and the error upper limit.

[0204] The total power generation predicted the day before, the interval of the total power generation predicted the day before, and the total power generation corrected by the prediction error are taken together as multi-decision information, and the BP neural network model is used to model and correct the photovoltaic power generation prediction results in each period, which can effectively improve the accuracy of photovoltaic power generation prediction in each period 24 hours a day and reduce the prediction error.

[0205] like Figure 8 In the comparison diagram of the results before and after the correction of multiple decision information in an embodiment shown, it can be clearly seen that the corrected prediction result is closer to the actual data than the uncorrected prediction result.

[0206] Embodiment 2:

[0207] Based on the same inventive concept, the embodiment of the present application provides a photovoltaic power generation prediction device 500. Fig. 9 The photovoltaic power generation prediction device 500 includes a processor 501 and a memory 502. The memory 502 stores a computer program. The processor 501 and the memory 502 communicate through a communication bus. The processor 501 executes the computer program to implement the steps of the method in the above-mentioned embodiment 1, which will not be repeated here.

[0208] Understandably, Fig. 9 The structure shown is for illustration only. The photovoltaic power generation prediction device may also include Fig. 9 More or fewer components as shown, or with Fig. 9 Different configurations are shown.

[0209] The processor 501 may be an integrated circuit chip with signal processing capabilities. The processor 501 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. It may implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0210] The memory 502 may include, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable read-only memory (EPROM), electrically erasable read-only memory (EEPROM), etc.

[0211] This embodiment also provides a computer-readable storage medium, such as a floppy disk, a CD, a hard disk, a flash memory, a USB flash disk, an SD card, an MMC card, etc., in which one or more programs for implementing the above steps are stored. The one or more programs can be executed by one or more processors 501 to implement the steps of the method in the above embodiment 1, which will not be repeated here.

[0212] It should be understood that, although the steps in the above flowchart are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows.

[0213] It should be noted that the diagram provided in the present embodiment only illustrates the basic concept of the present invention in a schematic manner, so the diagram only shows the components related to the present invention rather than drawing according to the number, shape and size of the components during actual implementation. The type, quantity and ratio of each component during actual implementation can be a random change, and the component layout type may also be more complicated. The structure, ratio, size, etc. illustrated in the drawings of the present specification are only used to match the content disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the limiting conditions that the present invention can implement, so they have no technical substantive significance. Any modification of the structure, change of the proportional relationship or adjustment of the size should still fall within the scope of the technical content disclosed by the present invention without affecting the effect that the present invention can produce and the purpose that can be achieved. At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" quoted in this specification are only for the convenience of narration, and are not used to limit the scope of the present invention. The change or adjustment of its relative relationship should also be regarded as the scope of the present invention without substantially changing the technical content.

[0214] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0215] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A photovoltaic power generation prediction method, characterized in that: include: Obtaining forecast input data of the photovoltaic power station for each time period on the date to be forecasted, wherein the forecast input data includes forecast meteorological data corresponding to the time period and historical output reference data; Input the prediction input data into the slice prediction model corresponding to each time period to obtain the first prediction result of each time period, wherein the slice prediction model corresponding to the time period is trained based on the historical output data of the photovoltaic power station at the same time period every day in the historical time period and the corresponding historical meteorological data; Based on the point prediction and / or interval prediction of the photovoltaic power generation on the previous day, multiple decision information with meteorological characteristics is obtained, wherein the multiple decision information includes the total power generation predicted on the previous day and / or the interval of the total power generation predicted on the previous day on the date to be predicted, wherein the point prediction includes inputting the meteorological mean data of the date to be predicted into the power generation point prediction model to obtain the total power generation predicted on the previous day, wherein the power generation point prediction model is trained based on the meteorological mean data and actual output data of each day in a historical time period; the interval prediction includes obtaining the interval of the total power generation predicted on the previous day based on the prediction error of the point prediction; The first prediction result of each time period is corrected according to the multiple decision information to obtain the second prediction result of each time period.

2. The photovoltaic power generation prediction method according to claim 1, characterized in that: The interval for obtaining the total power generation predicted on the day ahead based on the prediction error of the point prediction includes: According to the prediction error conditional probability distribution, a lower error limit and an upper error limit under a preset confidence level are obtained, wherein the prediction error conditional probability distribution is obtained according to the actual output data and the corresponding predicted output data of each day in a historical time period; According to the lower error limit and the upper error limit, on the basis of the total power generation predicted on the day before, an interval of the total power generation predicted on the day before is obtained.

3. The photovoltaic power generation prediction method according to claim 2, characterized in that: The multiple decision information also includes the total amount of power generation corrected by the forecast error; The obtaining of multiple decision information with meteorological characteristics also includes: Obtaining the forecast error corrected total power generation according to the integral of the forecast error conditional probability distribution between the error lower limit and the error upper limit; The first prediction result of each time period is corrected by taking the total power generation predicted on the day before, the interval of the total power generation predicted on the day before, and the total power generation corrected by the prediction error as multi-decision information.

4. The photovoltaic power generation prediction method according to claim 1, characterized in that: The step of correcting the first prediction result of each time period according to the multiple decision information to obtain the second prediction result of each time period includes: The multiple decision information and the first prediction result of each time period are input into a pre-trained correction model to obtain the second prediction result of each time period.

5. The photovoltaic power generation prediction method according to claim 1, characterized in that: The step of inputting the prediction input data into the slice prediction model corresponding to each time period to obtain the first prediction result of each time period includes: According to the normalized forecast input data, forecast the normalized forecast output data corresponding to the time period; According to the non-normalized predicted input data, predict and obtain non-normalized predicted output data corresponding to the time period; When the non-normalized predicted output data and the normalized predicted output data corresponding to a time period are both smaller than the segmentation threshold of the time period, the non-normalized predicted output data is determined to be the first prediction result of the time period; otherwise, the normalized predicted output data is determined to be the first prediction result of the time period.

6. The photovoltaic power generation prediction method according to claim 5, characterized in that: The step of inputting the prediction input data into the slice prediction model corresponding to each time period to obtain the first prediction result of each time period includes: The predicted input data of each time period is used as non-normalized input data, and input into the non-normalized slice prediction model corresponding to each time period to obtain the non-normalized predicted output data of each time period; Performing normalization processing on the predicted input data corresponding to each time period to obtain normalized input data for each time period; Inputting the normalized input data into the normalized slice prediction model corresponding to each time period to obtain the normalized predicted output data of each time period; When the non-normalized predicted output data and the normalized predicted output data corresponding to a time period are both less than the segmentation threshold of the time period, the non-normalized predicted output data is determined to be the first prediction result of the time period; otherwise, the normalized predicted output data is determined to be the first prediction result of the time period; Among them, the non-normalized slice prediction model corresponding to the time period is trained based on the non-normalized historical output data and the non-normalized historical meteorological data of the time period; the normalized slice prediction model corresponding to the time period is trained based on the normalized historical output data and the normalized historical meteorological data of the time period.

7. The photovoltaic power generation prediction method according to claim 1, characterized in that: The step of inputting the meteorological mean data of the date to be predicted into the power generation point prediction model to obtain the total power generation predicted on the day before comprises: Determine the target weather type for the location of the photovoltaic power station on the date to be predicted; The meteorological mean data is input into the power generation point prediction model corresponding to the target weather type of the date to be predicted to obtain the total power generation predicted on the day before, wherein the power generation point prediction model corresponding to the target weather type is trained based on a first data set corresponding to the target weather type in a historical time period, and the first data set includes the meteorological mean data corresponding to the target weather type in a historical time period, and the actual output data corresponding to the meteorological mean data.

8. The photovoltaic power generation prediction method according to claim 2, characterized in that: The prediction error conditional probability distribution is obtained according to the actual output data and the corresponding predicted output data of each day in the historical time period, including: For each target weather type, determine a corresponding Copula function; the variable of the Copula function is a second data set corresponding to the target weather type, and the second data set includes the actual output data and the predicted output data; Calculating a correlation coefficient according to the actual output data corresponding to the target weather type and the corresponding predicted output data, and calculating a Copula parameter of the Copula function corresponding to the target weather type according to the correlation coefficient; According to the actual output data corresponding to the target weather type, an actual output marginal density function and an actual output marginal distribution function corresponding to the target weather type are calculated, and according to the predicted output data corresponding to the target weather type, a predicted output marginal distribution function corresponding to the target weather type is calculated; According to the Copula function corresponding to the target weather type, the Copula parameters, the actual output marginal density function, the actual output marginal distribution function and the predicted output marginal distribution function, the prediction error conditional probability distribution under the target predicted output value is calculated.

9. The photovoltaic power generation prediction method according to any one of claims 7 or 8, characterized in that: The first data set or the second data set corresponding to the target weather type in the historical time period is determined according to the following steps: Using a rain flow counting method to extract features from a first actual output value sequence determined by the actual output data to obtain corresponding counting feature data; and / or performing differential processing on a second actual output value sequence determined by the actual output data to obtain corresponding output trend feature data; Based on the counting feature data and / or the output trend feature data, the actual output data belonging to the same target weather type and the meteorological mean data corresponding to the actual output data are divided into the same data set to obtain the first data set corresponding to each target weather type; Based on the counting characteristic data and / or the output trend characteristic data, the actual output data belonging to the same target weather type and the predicted output data corresponding to the actual output data are divided into the same data set to obtain the second data set corresponding to each target weather type.

10. A photovoltaic power generation prediction device, characterized in that: The method comprises a processor and a memory, wherein a computer program is stored in the memory, and the processor executes the computer program to implement the method according to any one of claims 1 to 9.

Images