A day-ahead photovoltaic power prediction method and related device independent of meteorological data

Abstract

A day-ahead photovoltaic power prediction method and related device provided by the present invention that do not rely on meteorological data, including obtaining the weather categories of several days before the day to be predicted; inputting the weather categories of several days before the day into a weather type probability prediction model to obtain the weather type probability distribution of the day to be predicted; performing multiple random samplings on the weather type probability distribution; selecting the corresponding photovoltaic power prediction model according to the weather type obtained by sampling, and using each photovoltaic power prediction model for prediction; performing weighted fusion on the weather type probabilities obtained by multiple samplings and the corresponding multiple photovoltaic power prediction results to obtain the final prediction result. The present invention fully mines the hidden meteorological information in historical photovoltaic output data through a clustering method, and can effectively consider the weather changes in long-term operation without the input of weather forecasts by combining the weather type probability prediction model and the photovoltaic power prediction model, making the application conditions more relaxed.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power prediction, and particularly relates to a day-ahead photovoltaic power prediction method and related device that do not rely on meteorological data. Background Art

[0002] Photovoltaic power generation is the renewable energy with the fastest growth rate in the new power system. Since photovoltaic power is greatly affected by weather changes, in a power system with a high proportion of photovoltaics, the grid dispatching operation has put forward high requirements for the accurate prediction of day-ahead photovoltaic power.

[0003] The power generation of new energy is strongly related to the microenvironment and micro-meteorology of the location. Therefore, the grid's prediction of new energy power generation often relies on each new energy manufacturer to upload their own day-ahead power prediction data and sum them up. However, a large number of photovoltaic power generations are widely distributed in the distribution system in the form of distributed power sources, with small single capacities, large differences in their respective installation environments and operating conditions, and many do not have the conditions to obtain real-time weather prediction data for power generation prediction. In this regard, establishing a data-driven day-ahead photovoltaic power prediction model, training the model using the historical power generation records of each photovoltaic power source, and then using the trained model to achieve customized day-ahead photovoltaic power prediction is an effective method with low cost. The technical difficulty lies in that in order to ensure the actual usability of the prediction model, the prediction model should not require accurate weather forecast data as the prediction input, otherwise it lacks application conditions in many distributed power generation scenarios.

[0004] Existing photovoltaic power prediction methods generally require weather forecast data. Some models that do not require weather forecast input do not consider the impact of weather changes on photovoltaic power and are difficult to adapt to different weather conditions. Summary of the Invention

[0005] In view of this, the present invention aims to provide a photovoltaic power prediction method and related device that do not require weather forecast data but can predict weather changes. Through fully mining the meteorological information hidden in historical photovoltaic output data, the method and related device can effectively consider weather changes in long-term operation without weather forecast input, more accurately predict the photovoltaic output in the next 24 hours, and have more relaxed application conditions.

[0006] To achieve the above object, the technical solution provided by the present invention is as follows:

[0007] In the first aspect, the present invention provides a day-ahead photovoltaic power prediction method that does not rely on meteorological data, including the following steps:

[0008] Obtain the weather categories for several days before the day to be predicted. The weather categories are obtained by clustering historical photovoltaic power generation data, and each clustering category corresponds to a specific weather type;

[0009] Input the weather categories for several days before the day to be predicted into the weather type probability prediction model to obtain the probability distribution of the weather type for the day to be predicted. The weather type probability prediction model is trained based on all specific weather types obtained by clustering historical data;

[0010] Perform multiple random samplings on the probability distribution of the weather type;

[0011] Select the corresponding photovoltaic power prediction model according to the weather type obtained by sampling, and input the photovoltaic power generation data for several days before the day to be predicted into each photovoltaic power prediction model respectively to obtain multiple photovoltaic power prediction results for the day to be predicted;

[0012] Perform weighted fusion on the weather type probabilities obtained by multiple samplings and the corresponding multiple photovoltaic power prediction results to obtain the final photovoltaic power prediction result for the day to be predicted.

[0013] Furthermore, the training process of the weather type probability prediction model includes:

[0014] Number all specific weather types obtained by clustering historical photovoltaic power generation data and form a training sample set. Each sample in the training sample set includes a prediction day and the weather type numbers for several days before the prediction day;

[0015] Use the weather category numbers for several days before the day to be predicted as the input and the weather category number for the prediction day as the output, and use the training sample set to train the neural network model. The neural network model adopts a backpropagation network structure, and the output of the normalization layer is the probability of each specific weather category.

[0016] Furthermore, the training process of the photovoltaic power prediction model includes:

[0017] Preprocess the historical photovoltaic power generation data so that the daily photovoltaic power generation data meets the standardization requirements and carries time information;

[0018] For each sample, form the sample data of the sample with the preprocessed photovoltaic power generation data for several days before the date to which the sample belongs, and label the sample according to the specific weather type corresponding to the date;

[0019] Group all samples according to the labeled weather categories to obtain several groups of training sample sets;

[0020] Use each group of training sample sets to train the neural network model to obtain the photovoltaic power prediction model corresponding to each specific weather type.

[0021] Further, the preprocessing process includes:

[0022] Dividing the daily photovoltaic power generation power time series by the rated photovoltaic capacity for normalization to obtain the normalized daily photovoltaic power generation power time series;

[0023] Determining the time sequence information of each photovoltaic power generation power data in the time series by calculating the positional encoding;

[0024] Using the algebraic sum of each data in the daily photovoltaic power generation power time series and the corresponding time sequence information to form the preprocessed daily photovoltaic power generation power data.

[0025] Further, the photovoltaic power prediction model adopts an encoder-decoder structure.

[0026] Further, the final photovoltaic power prediction result is calculated according to the following formula:

[0027]

[0028] In the formula, is the final photovoltaic power prediction result, is the number of sampling times, is the weather type probability obtained after normalization corresponding to the th sampling result, is the predicted power sequence of the photovoltaic power prediction model corresponding to the

[0029] Further, clustering the historical photovoltaic power generation power data includes:

[0030] Taking days as units, obtaining the daily photovoltaic power generation power data sequence from the historical power generation power data of the photovoltaic power station;

[0031] Calculating the selected characteristic indexes for the daily photovoltaic power generation power data sequence;

[0032] Clustering the daily photovoltaic power generation power data sequence based on the characteristic indexes, and each clustering category corresponds to a specific weather category respectively.

[0033] In a second aspect, the present invention provides a day-ahead photovoltaic power prediction device that does not rely on meteorological data, including:

[0034] A data acquisition module for acquiring the weather categories of several days before the day to be predicted, where the weather categories are obtained by clustering the historical photovoltaic power generation power data, and each clustering category corresponds to a specific weather type;

[0035] A weather type prediction module, which is used to input the weather categories of several days before the current day into a weather type probability prediction model to obtain the weather type probability distribution of the day to be predicted. The weather type probability prediction model is trained based on all specific weather types obtained by clustering historical data;

[0036] A sampling module, which is used to perform multiple random samplings on the weather type probability distribution;

[0037] A prediction module, which is used to select the corresponding photovoltaic power prediction model according to the weather type obtained by sampling, and input the photovoltaic power generation data of several days before the current day into each photovoltaic power prediction model respectively to obtain multiple photovoltaic power prediction results of the day to be predicted;

[0038] A result output module, which is used to perform weighted fusion on the weather type probabilities obtained by multiple samplings and the corresponding multiple photovoltaic power prediction results to obtain the final photovoltaic power prediction result of the day to be predicted.

[0039] In a third aspect, the present invention also provides a computer device, which includes a processor and a memory:

[0040] The memory is used to store a computer program and send the instructions of the computer program to the processor;

[0041] The processor executes a day-ahead photovoltaic power prediction method independent of meteorological data as described in the first aspect according to the instructions of the computer program.

[0042] In a fourth aspect, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements a day-ahead photovoltaic power prediction method independent of meteorological data as described in the first aspect.

[0043] In summary, a day-ahead photovoltaic power prediction method and related device provided by the present invention include obtaining weather categories of several days before the day to be predicted. The weather categories are obtained by clustering historical photovoltaic power generation data, and each clustering category corresponds to a specific weather type. Inputting the weather categories of several days before the day into a weather type probability prediction model to obtain the weather type probability distribution of the day to be predicted. Conducting multiple random samplings on the weather type probability distribution. Selecting the corresponding photovoltaic power prediction model according to the weather type obtained by sampling, and inputting the photovoltaic power generation data of several days before the day into each photovoltaic power prediction model respectively to obtain multiple photovoltaic power prediction results of the day to be predicted. Weightedly fusing the weather type probabilities obtained by multiple samplings with the corresponding multiple photovoltaic power prediction results to obtain the final photovoltaic power prediction result of the day to be predicted. The present invention fully excavates the meteorological information hidden in historical photovoltaic output data through clustering, and combines the weather type probability prediction model and the photovoltaic power prediction model to effectively consider the weather changes in long-term operation without the input of weather forecasts, and more accurately realize the photovoltaic output prediction for the next 24 hours, and the application conditions are more relaxed. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0045] Figure 1 It is a flowchart of a day-ahead photovoltaic power prediction method provided by an embodiment of the present invention;

[0046] Figure 2 It is a structure diagram of a photovoltaic power prediction model provided by an embodiment of the present invention;

[0047] Figure 3 It is a structure diagram of a backpropagation network provided by an embodiment of the present invention;

[0048] Figure 4 It is a prediction flowchart of 2 random samplings provided by an embodiment of the present invention;

[0049] Figure 5 It is a DB index - K relationship diagram provided by an embodiment of the present invention;

[0050] Figure 6 It is a schematic diagram of the visualization result of clustering data provided by an embodiment of the present invention;

[0051] Figure 7aIt is a diagram showing the change process of the loss function of the deep learning prediction model 0 provided by the embodiment of the present invention;

[0052] Figure 7b It is a diagram showing the change process of the loss function of the deep learning prediction model 1 provided by the embodiment of the present invention;

[0053] Figure 7c It is a diagram showing the change process of the loss function of the deep learning prediction model 2 provided by the embodiment of the present invention;

[0054] Figure 8 It is a diagram showing the output curves of two prediction models provided by the embodiment of the present invention;

[0055] Figure 9 It is a diagram for comparing the prediction result with the actual value provided by the embodiment of the present invention;

[0056] Figure 10 It is a block diagram of a day-ahead photovoltaic power prediction device that does not rely on meteorological data provided by the embodiment of the present invention;

[0057] Figure 11 It is a block diagram of a computer device provided by the embodiment of the present invention. Specific embodiments

[0058] To make the objectives, features, and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the embodiments described below are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0059] Please refer to Figure 1 , this embodiment provides a day-ahead photovoltaic power prediction method that does not rely on meteorological data, including the following steps:

[0060] S11: Obtain the weather categories of several days before the day to be predicted. The weather categories are obtained by clustering historical photovoltaic power generation data, and each clustering category corresponds to a specific weather type.

[0061] It should be noted that in this step, through clustering analysis, it is not necessary to rely on real-time or historical meteorological data, but directly use the inherent characteristics of historical photovoltaic power generation data to identify weather types.

[0062] S12: Input the weather categories of several days before the day to be predicted into the weather type probability prediction model to obtain the weather type probability distribution of the day to be predicted. The weather type probability prediction model is trained based on all specific weather types obtained by clustering historical data.

[0063] It should be noted that the model trained with historical data in this step can predict the weather type without meteorological data, reducing the dependence on meteorological data.

[0064] S13: Conduct multiple random samplings on the probability distribution of weather types.

[0065] It should be noted that in this step, through random sampling, considering the uncertainty of weather types, the robustness of the prediction is enhanced.

[0066] S14: Select the corresponding photovoltaic power prediction model according to the weather type obtained by sampling, and input the photovoltaic power generation data of several days before the day to be predicted into each photovoltaic power prediction model respectively to obtain multiple photovoltaic power prediction results for the day to be predicted.

[0067] It should be noted that in this step, by selecting a dedicated photovoltaic power prediction model for each weather type, the pertinence and accuracy of the prediction are improved.

[0068] S15: Perform weighted fusion on the weather type probabilities obtained from multiple samplings and the corresponding multiple photovoltaic power prediction results to obtain the final photovoltaic power prediction result for the day to be predicted.

[0069] It should be noted that in this step, through the weighted fusion technology, comprehensively considering the probability distribution of weather types and the photovoltaic power prediction results, the overall accuracy and reliability of the prediction results are improved.

[0070] This embodiment provides a method for predicting the photovoltaic power of the day ahead without relying on meteorological data. Through multiple steps such as cluster analysis, weather type probability prediction, random sampling, selection of photovoltaic power prediction models, and weighted fusion, the problems existing in the traditional photovoltaic power prediction method are effectively solved, and the prediction accuracy and robustness are improved.

[0071] In some embodiments, clustering the historical photovoltaic power generation data includes:

[0072] S21: Taking the day as the unit, obtain the daily photovoltaic power generation data sequence from the historical power generation data of the photovoltaic power station.

[0073] It should be noted that in this step, by directly using the historical power generation data of the photovoltaic power station instead of relying on external meteorological data, the dependence on external data sources is reduced.

[0074] S22: Calculate the selected characteristic indexes for the daily photovoltaic power generation data sequence.

[0075] It should be noted that in this step, by calculating the characteristic indexes, useful information can be extracted from the photovoltaic power generation data, providing more meaningful basic data for cluster analysis and reducing the dependence on meteorological data.

[0076] S23: Cluster the daily photovoltaic power generation data sequence based on the characteristic indexes, and each cluster category corresponds to a specific weather category.

[0077] It should be noted that in this step, through cluster analysis, specific weather types can be directly identified from the photovoltaic power generation data without relying on meteorological data, improving the independence and robustness of the prediction method.

[0078] The following introduces this embodiment with a numerical example.

[0079] In the first step, taking the day as the unit, the characteristic statistics of the historical power generation of the photovoltaic power station are recorded.

[0080] Let the photovoltaic power generation data sequence on the t-th day be :

[0081] (1)

[0082] where N represents the number of sampling points. The power sequence values are collected every 15 minutes from 6:00 to 18:00, so N = 48; represents the power value at the i-th sampling point on the t-th day.

[0083] Calculate the following characteristic indexes:

[0084] (1) (Daily maximum value):

[0085] (2)

[0086] (2) (Total daily power)

[0087] (3)

[0088] (3) (Daily average power)

[0089] (4)

[0090] (4) (Variance)

[0091] (5)

[0092] (5) (Coefficient of variation)

[0093] (6)

[0094] (6) (Skewness)

[0095] (7)

[0096] (7) (Kurtosis)

[0097] (8)

[0098] Among them, (1)-(3) are used as basic statistical features, (4)-(5) are used as discrete statistical features, and (6)-(7) are used as distribution statistical features.

[0099] In the second step, this example selects a clustering algorithm and uses the above 7 indicators as features to perform sample clustering on each historical day.

[0100] The feature vector of the t-th day composed of 7 indicators is denoted as :

[0101] This example adopts the K-means clustering algorithm. It should be noted that for the determination of the number of clusters (K value) in clustering, it is recommended to use the Davies-Bouldin (DB) index to evaluate the clustering quality to find the optimal number of cluster centers.

[0102] The calculation formula of the DB index is as follows:

[0103] (9)

[0104] Among them represents the within-cluster distance of cluster . Assume that cluster contains samples, and each sample is a 7-dimensional feature vector

[0105] (10)

[0106] The center point of cluster is denoted as . Then the calculation formula of the within-cluster distance is

[0107] (11)

[0108] represents the distance between the -th sample and the center , and the calculation formula is:

[0109] (12)

[0110] Wherein and respectively represent the sample and the center point at the -th feature value.

[0111] In formula (12), represents the calculated distance between clusters. For clusters and cluster center points and The formula for calculating the inter-cluster distance is as follows:

[0112] (13)

[0113] The above calculation is the ratio of the sum of the average intra-cluster compactness between each cluster and other clusters to the separation degree between cluster and , and then take the average value. A lower DB index indicates better clustering quality, which reflects the ratio of the distance between samples within a cluster to the distance between samples between clusters.

[0114] The third step is to record the clustering categories as the weather categories. For example, if the photovoltaic power sequence on the -th day is clustered and assigned to the -th cluster, then record the weather category of the -th day as .

[0115] In some embodiments, the training process of the photovoltaic power prediction model includes:

[0116] S31: Preprocess the historical photovoltaic power generation data so that the daily photovoltaic power generation data meets the standardization requirements and has time information;

[0117] S32: For each sample, form the sample data of the sample with the preprocessed photovoltaic power generation data of several days before the date to which the sample belongs, and label the sample according to the specific weather type corresponding to the date;

[0118] S33: Group all samples according to the labeled weather categories to obtain several groups of training sample sets;

[0119] S34: Use each group of training sample sets to train the neural network model to obtain the photovoltaic power prediction model corresponding to each specific weather type.

[0120] In a further embodiment, the preprocessing process includes:

[0121] S41: Divide the daily photovoltaic power generation power time series by the rated photovoltaic capacity for normalization to obtain the daily photovoltaic power generation power time series after normalization;

[0122] S42: Determine the time sequence information of each photovoltaic power generation power data in the time series by calculating the position encoding;

[0123] S43: Use the algebraic sum of each data in the daily photovoltaic power generation power time series and the corresponding time sequence information to form the daily photovoltaic power generation power data after preprocessing.

[0124] The introduction of the photovoltaic power prediction model will continue with the foregoing numerical example.

[0125] (1) Training sample input design and sample organization

[0126] First, when predicting the photovoltaic power time series of the t-th day, the photovoltaic power records of the (t - 1)-th, (t - 2)-th, and (t - 3)-th days are used as the input. Taking the time period from 6:00 to 18:00 every day as an example, the input is the photovoltaic power at 48 moments. The photovoltaic power is normalized by dividing it by the rated photovoltaic capacity, denoted as:

[0127] (14)

[0128] where represents the normalized photovoltaic power on the t-th day, represents the input photovoltaic power on the t-th day, represents the rated photovoltaic capacity.

[0129] The position encoding is denoted as: PE. After calculating the position encoding, an algebraic summation operation is performed, and the result after summation is used as the input data of the network ( ):

[0130] (15)

[0131] The formed samples are labeled according to the weather category of the day t to be predicted. After all samples are generated and labeled, all samples are grouped according to the labeled categories to form K groups of training sample sets.

[0132] (2) Photovoltaic power prediction deep learning model structure

[0133] The deep learning prediction model designed in this numerical example is as Figure 2As shown in the figure. The model adopts an encoder-decoder structure. The normalized power data is algebraically added to the positional encoding and then input into the encoder. The encoder consists of a fully connected layer, a multi-head attention network, and a feed-forward network. The fully connected layer is used to increase the dimension of the data, and the multi-head attention and feed-forward networks are used to find information in the high-dimensional data space. To prevent the problem of gradient disappearance during training, residual connections and layer normalization operations are added at the output ends of the two networks. After the encoder extracts the data information, it is input into the decoder. The decoder consists of an LSTM network and a fully connected layer. After the LSTM compiles the high-dimensional information, the fully connected layer adjusts the dimension of the data output and outputs the final power prediction data. There are no future known variables in the prediction process of this model, which is a network for predicting power with power.

[0134] In the deep learning model, the feature dimensions of each link are set as follows:

[0135] Table 1 Explanation table of the feature dimensions of the deep learning model

[0136]

[0137] (3)Training of the prediction model

[0138] For each set of training sample sets, the Adam optimization algorithm is used to train a set of prediction model parameters. A total of K prediction models are obtained.

[0139] In some embodiments, the training process of the weather type probability prediction model includes:

[0140] S31: Number all specific weather types obtained by clustering historical photovoltaic power generation data to form a training sample set. Each sample in the training sample set includes a prediction day and the weather type numbers of several days before the prediction day;

[0141] S32: Using the weather category numbers of several days before as the input and the weather category number of the prediction day as the output, use the training sample set to train the neural network model. The neural network model adopts a backpropagation network structure, and the output of the normalization layer is the probability of each specific weather category.

[0142] The weather type probability prediction model will be further introduced below in combination with the above-mentioned example.

[0143] The weather types of each day in history obtained in the previous steps are used to form a training sample set. The input of the t-th sample is the weather type numbers of the t-3, t-2, t-1, and t days, and the output is the weather type number of the t+1 day.

[0144] Use the obtained sample set to train a neural network model with the structure as shown in Figure 3 the figure.

[0145] The input data dimensions of each layer in the network are as follows:

[0146] Table 2 Explanation Table of BP Network Data Dimensions

[0147]

[0148] The output of the normalization layer is denoted as :

[0149] (16)

[0150] Since

[0151] (17)

[0152] (18)

[0153] Therefore, is regarded as the probability that the th weather category may occur

[0154] The following continues to introduce the day-ahead photovoltaic power prediction process in combination with the above example.

[0155] In this example, random sampling is carried out 2 times for introduction.

[0156] Record the prediction day as the t-th day, and the photovoltaic power sequence to be predicted is for the (t + 1)-th day.

[0157] (The first step) Cluster the photovoltaic power sequences of the (t - 3)-th, (t - 2)-th, (t - 1)-th, and t-th days according to the aforementioned clustering model respectively to obtain the weather category of each day.

[0158] (The second step) Input the weather categories of the (t - 3)-th, (t - 2)-th, (t - 1)-th, and t-th days into the weather type probability prediction model to obtain the weather type probability distribution for the (t + 1)-th day .

[0159] (The third step) Conduct 2 random samplings according to the probability distribution in the second step

[0160] (The fourth step) If the results of the two samplings are the same, select the corresponding photovoltaic power prediction model according to the weather type obtained by sampling. Organize the input features according to the requirements of the day-ahead photovoltaic power prediction deep learning model, input them into the prediction model, and obtain the prediction result of the next-day photovoltaic power.

[0161] If the results of the two samplings are different, record the result of the first sampling as , and the result of the second sampling as , then first perform normalization processing on the sampling results to obtain :

[0162] (19)

[0163] Wherein

[0164] (20)

[0165] Using the corresponding power prediction model And the power prediction model Generate a predicted power sequence And , the final predicted power

[0166] (21)

[0167] The flow chart is as Figure 4 shown

[0168] The above is an example of the process of realizing the day-ahead photovoltaic power prediction by randomly sampling and selecting the power prediction model twice. It can be understood that this step can be extended to the number of times of random sampling several times to complete a more accurate day-ahead photovoltaic power prediction process. Based on this, in some embodiments, the final photovoltaic power prediction result is calculated according to the following formula:

[0169] (22)

[0170] In the formula, is the final photovoltaic power prediction result, is the number of sampling times, is the probability of the weather type obtained after normalization corresponding to the th sampling result, is the th predicted power sequence of the photovoltaic power prediction model corresponding to the sampling result

[0171] The following introduces the day-ahead photovoltaic power prediction method without relying on meteorological data proposed in the above embodiments in combination with an application example

[0172] This application example uses the measured data of a photovoltaic power station in a certain area of the country to organize training samples. The data sampling interval is 15 minutes, there are 96 points per day, and 35040 points in a year. Since the output of photovoltaic is almost 0 at night, considering the local sunrise and sunset times, only the 48 points of the recorded data during the period from 6:00 to 18:00 are taken as the data for subsequent analysis and calculation

[0173] Calculate 7 eigenvalues for the 48 - point sequence of each day. According to the clustering method described above, set the number of clusters K = 3, 4, 5, 6, 7, 8 respectively for clustering. Calculate the DB index of each clustering result. Draw the relationship graph of the DB index - K.

[0174] As Figure 5 , when the clustering center is 3, the value of the DB index is the lowest. Therefore, take K = 3.

[0175] The visualization results of the daily power curves of different categories are as Figure 6 shown.

[0176] From Figure 6 it can be seen that the classification results are strongly correlated with the weather types. The three types respectively correspond to the sunny - turning - cloudy mode, the sunny mode and the rainy mode. After obtaining the clustering center of each category, each time a new daily power sequence is input, the nearest - neighbor weather type label can be obtained.

[0177] In the previous step, 3 sample sets of categories were obtained through the clustering method. Use the sample set of each category to train the deep - learning model for photovoltaic power prediction to obtain the model parameters of this category. A total of 3 models are obtained.

[0178] The number of training rounds of the three models is 100, 150, and 100 rounds in sequence. The changes of the loss functions of the three models during the training

[0179] process are shown in Figure 7, where Figure 7a is the change process of the loss function of model 0; Figure 7b is the change process of the loss function of model 1; Figure 7c is the change process of the loss function of model 2.

[0180] Use the well - clustered sample set to train the next - day weather category prediction model in four quarters to obtain the model parameters of each quarter. A total of 4 models are obtained. Use the accuracy rate as the evaluation index of the model prediction performance. The evaluation results are as follows in the table:

[0181] Table 3 Prediction accuracies of each quarter

[0182]

[0183] Take predicting the photovoltaic power of a certain day in this area as an example. The trained clustering model can judge the weather category of the current day. Among them, the sunny - turning - cloudy is recorded as mode 0, the sunny day is recorded as mode 1, and the rainy day is recorded as mode 2. Input the power sequences of the 4 days before the prediction day, and the clustering model gives the weather category results of these 4 days in chronological order:

[0184]

[0185] Input this segment into the weather category prediction model, and the probabilities of each type of weather occurring output by the model are:

[0186]

[0187] According to the probabilities Perform sampling. The sampling result of the first time is mode 0, and the sampling result of the second time is mode 2.

[0188] Since the results of the two samplings are different, the result of the secondary sampling is denoted as :

[0189]

[0190] Among them

[0191]

[0192] represents the probability of mode 1 after the sampling result is normalized. Similarly.

[0193] After normalizing the power sequences of the previous 3 days before the prediction date, input them into the trained power sequence models 1 and 2 respectively, and obtain two power prediction sequences P1 and P2, as Figure 8 .

[0194] According to the probability weighted summation method, obtain the final power prediction result P:

[0195]

[0196] Compare the final prediction result with the actual value as Figure 9 shown. It can be seen that without weather information, the model correctly predicts the photovoltaic output level on a rainy day.

[0197] Based on the same inventive concept, the embodiment of the present application also provides a weather-data-independent day-ahead photovoltaic power prediction device for implementing the above-mentioned weather-data-independent day-ahead photovoltaic power prediction method. The implementation solution provided by this device to solve the problem is similar to the implementation solution described in the above method. Therefore, the specific limitations in the embodiment of the weather-data-independent day-ahead photovoltaic power prediction device provided below can refer to the limitations on the weather-data-independent day-ahead photovoltaic power prediction method in the above text, and will not be repeated here.

[0198] Please refer to Figure 10 , this embodiment provides a weather-data-independent day-ahead photovoltaic power prediction device, including:

[0199] A data acquisition module, configured to acquire the weather categories of several days before the day to be predicted. The weather categories are obtained by clustering historical photovoltaic power generation data, and each clustering category corresponds to a specific weather type;

[0200] A weather type prediction module, configured to input the weather categories of several days before the day into a weather type probability prediction model to obtain the weather type probability distribution of the day to be predicted. The weather type probability prediction model is trained based on all specific weather types obtained by clustering historical data;

[0201] A sampling module, configured to perform multiple random samplings on the weather type probability distribution;

[0202] A prediction module, configured to select the corresponding photovoltaic power prediction model according to the weather type obtained by sampling, and input the photovoltaic power generation data of several days before the day into each photovoltaic power prediction model respectively to obtain multiple photovoltaic power prediction results of the day to be predicted;

[0203] A result output module, configured to perform weighted fusion on the weather type probabilities obtained by multiple samplings and the corresponding multiple photovoltaic power prediction results to obtain the final photovoltaic power prediction result of the day to be predicted.

[0204] Those skilled in the art can clearly understand that, for the convenience and conciseness of description, only the above-mentioned division of each functional unit and module is used as an example for illustration. In practical applications, the above-mentioned functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of the present application. The specific working processes of the units and modules in the above-mentioned system can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0205] Refer to Figure 11 , an embodiment of the present invention further provides a computer device, including: a memory, a processor, and a computer program stored on the memory. When the computer program is executed on the processor, it implements the method for predicting the photovoltaic power before the day without relying on meteorological data as described in any one of the above methods.

[0206] The computer device may be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The computer device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that Figure 11 merely examples of computer devices, which do not constitute a limitation on computer devices, may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, it may also include input / output devices, network access devices, etc.

[0207] The so-called processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0208] The memory may be an internal storage unit of the computer device in some embodiments, such as the hard disk or memory of the computer device. The memory may also be an external storage device of the computer device in other embodiments, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device. Further, the memory may also include both the internal storage unit and the external storage device of the computer device. The memory is used to store an operating system, application programs, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program. The memory may also be used to temporarily store data that has been output or will be output.

[0209] The embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, it implements the day-ahead photovoltaic power prediction method independent of meteorological data as described in any one of the above methods.

[0210] In this embodiment, if the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, to implement all or part of the processes in the above-described embodiment methods of this application, a computer program can be used to instruct relevant hardware to complete. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the photographing device / terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk, or an optical disc, etc. In some jurisdictions, according to legislation and patent practice, the computer-readable medium cannot be an electrical carrier signal and a telecommunication signal.

[0211] In the above embodiments, the descriptions of the respective embodiments have their own focuses. For parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0212] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0213] In the embodiments disclosed in this application, it should be understood that the disclosed device / terminal device and method can be implemented in other ways. For example, the device / terminal device embodiments described above are merely illustrative. For example, the division of the modules or units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0214] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A day-ahead photovoltaic power prediction method independent of meteorological data, characterized in that, Including the following steps: Obtain the weather categories of several days before the day to be predicted. The weather categories are obtained by clustering historical photovoltaic power generation data, and each clustering category corresponds to a weather type; Input the weather categories of several days before the day into the weather type probability prediction model to obtain the weather type probability distribution of the day to be predicted. The weather type probability prediction model is trained based on all the weather types obtained by clustering historical photovoltaic power generation data; Perform multiple random samplings on the weather type probability distribution; Select the corresponding photovoltaic power prediction model according to the weather type obtained by sampling, and input the photovoltaic power generation data of several days before the day into each of the photovoltaic power prediction models to obtain multiple photovoltaic power prediction results of the day to be predicted; Perform weighted fusion on the weather type probabilities obtained by multiple samplings and the corresponding multiple photovoltaic power prediction results to obtain the final photovoltaic power prediction result of the day to be predicted.

2. The day-ahead photovoltaic power prediction method independent of meteorological data according to claim 1, wherein The training process of the weather type probability prediction model includes: Number all the weather types obtained by clustering historical photovoltaic power generation data and form a training sample set. Each sample in the training sample set includes a prediction day and the weather type numbers of several days before the prediction day; Using the weather type numbers of several days before the day as the input and the weather type number of the prediction day as the output, train the neural network model using the training sample set. The neural network model adopts a backpropagation network structure, and the output of the normalization layer is the probability of each weather category.

3. The day-ahead photovoltaic power prediction method independent of meteorological data according to claim 1, characterized in that, The training process of the photovoltaic power prediction model includes: Preprocess the historical photovoltaic power generation data so that the daily photovoltaic power generation data meets the standardization requirements and has time information; For each sample, form the sample data of the sample with the preprocessed photovoltaic power generation data of several days before the date of the sample, and label the sample according to the weather type corresponding to the date; Group all samples according to the labeled weather categories to obtain several groups of training sample sets; Use each group of the training sample sets to train the neural network model to obtain the photovoltaic power prediction model corresponding to each weather type.

4. The day-ahead photovoltaic power prediction method independent of meteorological data according to claim 3, characterized in that, The preprocessing process includes: Divide the daily photovoltaic power generation time series by the rated photovoltaic capacity for normalization processing to obtain the normalized daily photovoltaic power generation time series; Determine the time order information of each photovoltaic power generation data in the time series by calculating the position encoding; Form the preprocessed daily photovoltaic power generation data using the algebraic sum of each data in the daily photovoltaic power generation time series and the corresponding time order information.

5. The day-ahead photovoltaic power prediction method independent of meteorological data according to claim 1, characterized in that The photovoltaic power prediction model adopts an encoder-decoder structure.

6. The day-ahead photovoltaic power prediction method independent of meteorological data according to claim 1, characterized in that, The final photovoltaic power prediction result is calculated according to the following formula: Wherein, P is the final photovoltaic power prediction result, n is the number of sampling times, and p i is the probability of the weather type obtained after normalization corresponding to the i-th sampling result, and P i is the predicted power sequence of the photovoltaic power prediction model corresponding to the i-th sampling result.

7. The day-ahead photovoltaic power prediction method independent of meteorological data according to claim 1, wherein Clustering the historical photovoltaic power generation data includes: Taking days as the unit, obtain the daily photovoltaic power generation data sequence from the historical power generation data of the photovoltaic power station; For the daily photovoltaic power generation data sequence, calculate the selected characteristic indicators; Cluster the daily photovoltaic power generation data sequence based on the feature indicators, and each clustering category corresponds to a weather category respectively.

8. A day-ahead photovoltaic power prediction device that does not rely on meteorological data, characterized in that, It includes: A data acquisition module, configured to acquire the weather categories of several days before the day to be predicted. The weather categories are obtained by clustering historical photovoltaic power generation data, and each clustering category corresponds to a weather type respectively; A weather type prediction module, configured to input the weather categories of several days before the day into a weather type probability prediction model to obtain the weather type probability distribution of the day to be predicted. The weather type probability prediction model is trained based on all the weather types obtained by clustering historical photovoltaic power generation data; A sampling module, configured to perform multiple random samplings on the weather type probability distribution; A prediction module, configured to select the corresponding photovoltaic power prediction model according to the weather type obtained by sampling, and input the photovoltaic power generation data of several days before the day into each of the photovoltaic power prediction models to obtain multiple photovoltaic power prediction results of the day to be predicted; A result output module, configured to perform weighted fusion on the weather type probabilities obtained by multiple samplings and the corresponding multiple photovoltaic power prediction results to obtain the final photovoltaic power prediction result of the day to be predicted.

9. A computer device, characterized in that, The device includes a processor and a memory: The memory is used to store a computer program and send the instructions of the computer program to the processor; The processor executes a day-ahead photovoltaic power prediction method independent of meteorological data according to any one of claims 1-7 based on the instructions of the computer program.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by the processor, it implements a day-ahead photovoltaic power prediction method independent of meteorological data according to any one of claims 1-7.

Images