Optimization cascade-based distributed photovoltaic power prediction method and system

Abstract

The application discloses a kind of distributed photovoltaic power prediction method and system based on optimization cascade, which comprises: obtaining the historical photovoltaic power of centralized photovoltaic power station and distributed photovoltaic power station;Based on historical photovoltaic power, the Pearson correlation coefficient between centralized and distributed photovoltaic power station is calculated, and a linear correlation model of centralized photovoltaic power and distributed photovoltaic power is established;The meteorological data of the prediction day is input into the centralized photovoltaic power prediction model based on KDE-CatBoost, and the centralized photovoltaic power prediction value is output;The centralized photovoltaic power prediction model is trained using Adam optimization algorithm, the input meteorological data is estimated by kernel density estimation method, the photovoltaic power distribution probability prediction result is output, and then the CatBoost algorithm is used to output the photovoltaic power prediction value;According to the centralized photovoltaic power prediction value and the linear correlation model, the distributed photovoltaic power prediction value is obtained, and the accuracy of power prediction is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic power prediction, and particularly relates to a distributed photovoltaic power prediction method and system based on optimized cascade. BACKGROUND

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute the prior art.

[0003] Under the background of energy restructuring and transformation, it is crucial to improve the effective utilization rate of renewable energy, especially solar energy. Centralized solar photovoltaic power generation is one of the most important application ways of solar energy. However, the natural volatility and randomness of solar energy lead to unstable photovoltaic output power, which in turn affects the utilization rate of solar energy. Therefore, it is particularly important to study the probabilistic distribution law of photovoltaic power and high-precision prediction algorithm.

[0004] In the existing photovoltaic power prediction technology, a single prediction algorithm still cannot meet the requirements in terms of prediction accuracy and reliability, and therefore a hybrid model with higher accuracy and better reliability needs to be obtained by combining multiple algorithms to predict the photovoltaic output. The traditional prediction method is mainly based on physical models and statistical models, and uses weather data, system parameters and model assumptions for prediction. However, this method often requires a large amount of system parameters and weather data, and the prediction effect is limited under complex weather conditions, and the requirements for data and parameters are high. With the rapid development of machine learning technology, photovoltaic power prediction methods based on machine learning have gradually become a research hotspot. Machine learning methods can learn historical data and establish nonlinear models to capture the complex relationship between photovoltaic power and weather conditions. However, due to the influence of multiple external factors on photovoltaic power prediction, the data volume is large, and in the existing photovoltaic power prediction method based on machine learning, it is difficult to capture subtle changes in a large amount of data through a simplified model, and due to the influence of geographical location, time change and other factors, the existing model cannot consider multiple time and space input variables, resulting in poor prediction performance of the existing model. SUMMARY

[0005] To solve the above problems of the prior art, the present application provides a distributed photovoltaic power prediction method and system based on optimized cascade, which combines kernel density estimation, CatBoost algorithm and optimized cascade, predicts the distribution probability of centralized photovoltaic power through kernel density estimation, predicts photovoltaic power using the CatBoost algorithm, and obtains the distributed photovoltaic power prediction value according to the centralized photovoltaic power prediction value by combining the optimized cascade, thereby improving the accuracy of photovoltaic power prediction.

[0006] In a first aspect, the present disclosure provides a distributed photovoltaic power prediction method based on optimized cascade.

[0007] A distributed photovoltaic power prediction method based on optimized cascade, comprising:

[0008] Obtaining historical photovoltaic power of centralized photovoltaic power stations and distributed photovoltaic power stations;

[0009] Based on the historical photovoltaic power, calculating the Pearson correlation coefficient between the centralized and distributed photovoltaic power stations, and establishing a linear correlation model of the centralized photovoltaic power and the distributed photovoltaic power;

[0010] Inputting meteorological data of a prediction day into a centralized photovoltaic power prediction model based on KDE-CatBoost, and outputting a centralized photovoltaic power prediction value; the centralized photovoltaic power prediction model is trained by using an Adam optimization algorithm, the input meteorological data is processed by a kernel density estimation method, and a photovoltaic power distribution probability prediction result is output; based on the photovoltaic power distribution probability prediction result, a CatBoost algorithm is used to output a photovoltaic power prediction value;

[0011] According to the centralized photovoltaic power prediction value and the linear correlation model, a distributed photovoltaic power prediction value is obtained.

[0012] In a second aspect, the present disclosure provides a distributed photovoltaic power prediction system based on optimized cascade, comprising:

[0013] A data acquisition module is configured to obtain historical photovoltaic power of centralized photovoltaic power stations and distributed photovoltaic power stations;

[0014] A linear correlation model construction module is configured to calculate the Pearson correlation coefficient between the centralized and distributed photovoltaic power stations based on the historical photovoltaic power, and establish a linear correlation model of the centralized photovoltaic power and the distributed photovoltaic power;

[0015] A centralized photovoltaic power prediction model is configured to input meteorological data of a prediction day into a centralized photovoltaic power prediction model based on KDE-CatBoost, and output a centralized photovoltaic power prediction value; the centralized photovoltaic power prediction model is trained by using an Adam optimization algorithm, the input meteorological data is processed by a kernel density estimation method, and a photovoltaic power distribution probability prediction result is output; based on the photovoltaic power distribution probability prediction result, a CatBoost algorithm is used to output a photovoltaic power prediction value;

[0016] A distributed photovoltaic power prediction model is configured to obtain a distributed photovoltaic power prediction value according to the centralized photovoltaic power prediction value and the linear correlation model.

[0017] In a third aspect, the present disclosure further provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and running on the processor, when the computer instructions are run by the processor, the steps of the method of the first aspect are completed.

[0018] In a fourth aspect, the present disclosure also provides a computer readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the method of the first aspect.

[0019] The above one or more technical solutions have the following beneficial effects:

[0020] 1. The present application provides a distributed photovoltaic power prediction method and system based on optimized cascade, which uses Adam optimizer to optimize CatBoost algorithm and kernel density estimation algorithm, calculates the gradient of each parameter and updates the parameters using Adam algorithm to accelerate the convergence of the model and improve the prediction performance; through the adaptive adjustment of learning rate by Adam algorithm, different scales and changes in photovoltaic power data can be better handled, so as to more accurately predict photovoltaic power.

[0021] 2. In the present application, the advantages of combining CatBoost algorithm and kernel density estimation method are used to construct a photovoltaic power prediction model based on KDE-CatBoost, and the optimal prediction model is trained to achieve more accurate photovoltaic power prediction results.

[0022] 3. The present application establishes a cascade between centralized photovoltaic power and distributed photovoltaic power through Pearson correlation coefficient and lasso regression algorithm, and realizes accurate prediction of distributed photovoltaic power through accurate prediction value of centralized photovoltaic power. BRIEF DESCRIPTION OF DRAWINGS

[0023] The drawings accompanying the specification of the present application form a part thereof and serve to provide further understanding of the present application, the illustrative embodiments of the present application and their description serve to explain the present application and do not constitute an improper limitation of the present application.

[0024] Figure 1 A flowchart of the centralized photovoltaic power prediction method in the embodiment of the present disclosure;

[0025] Figure 2 A distribution comparison chart of kernel density estimation prediction value and true value in the embodiment of the present disclosure;

[0026] Figure 3 A distribution comparison chart of kernel density estimation prediction value and true value in the embodiment of the present disclosure;

[0027] Figure 4 A correlation coefficient chart between the output power of the photovoltaic power generation system and its related meteorological factors in the embodiment of the present disclosure;

[0028] Figure 5 A comparison chart of the predicted power output by the model and the actual photovoltaic output power in the embodiment of the present disclosure;

[0029] Figure 6 A comparison chart of the predicted power output by the model described in the embodiments of the present disclosure, the predicted power output by the existing method, and the actual photovoltaic output power;

[0030] Figure 7 A comparison chart of the predicted power output by the model described in the embodiments of the present disclosure, the predicted power output by the five comparison models, and the scheme, and the actual photovoltaic output power;

[0031] Figure 8 A linear relationship model chart of the centralized photovoltaic power station and three different distributed photovoltaic power stations in the embodiments of the present disclosure;

[0032] Figure 9 A comparison chart of the photovoltaic prediction value and the actual power value of the three different distributed photovoltaic power stations in the embodiments of the present disclosure. DETAILED DESCRIPTION

[0033] It should be noted that the following detailed description is exemplary only, is for the purpose of describing specific embodiments, and is intended to provide further explanation of the present application, and is not intended to limit the exemplary embodiments according to the present application. Unless otherwise indicated, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the present application belongs. In addition, it should be understood that when the terms "comprise" and / or "include" are used in the present specification, they refer to the presence of a feature, step, operation, device, component, and / or combination thereof.

[0034] Embodiment One

[0035] In order to solve the problem of poor prediction accuracy of existing photovoltaic power prediction models, the present embodiment proposes a distributed photovoltaic power prediction method based on optimized cascade, which takes into account that photovoltaic power prediction is affected by multiple external factors, and through a large amount of historical data, learns the complex nonlinear relationship and corresponding rules, so as to more accurately predict the photovoltaic power and the photovoltaic power change; in addition, the present embodiment considers multiple spatio-temporal correlation input variables at the same time to better predict the power change.

[0036] The present embodiment provides a distributed photovoltaic power prediction method based on optimized cascade, which comprises the following steps:

[0037] Obtain the historical photovoltaic power of the centralized photovoltaic power station and the distributed photovoltaic power station;

[0038] Based on the historical photovoltaic power, calculate the Pearson correlation coefficient between the centralized and distributed photovoltaic power stations, and establish a linear correlation model of the centralized photovoltaic power and the distributed photovoltaic power;

[0039] The meteorological data of the predicted day is input into the centralized photovoltaic power prediction model based on KDE-CatBoost, and a centralized photovoltaic power prediction value is output; the centralized photovoltaic power prediction model is trained by using an Adam optimization algorithm, the input meteorological data is processed by using a kernel density estimation method, a photovoltaic power distribution probability prediction result is output, and based on the photovoltaic power distribution probability prediction result, a photovoltaic power prediction value is output by using a CatBoost algorithm.

[0040] According to the centralized photovoltaic power prediction value and a linear correlation model, a distributed photovoltaic power prediction value is obtained.

[0041] In this embodiment, a certain photovoltaic power station in northern China is taken as an example to introduce the above-mentioned method in more detail. The above-mentioned method in this embodiment proposes a centralized photovoltaic power prediction model based on KDE-CatBoost, which realizes accurate prediction of photovoltaic power generation of centralized photovoltaic power stations; and proposes a cascading method of distributed photovoltaic power and centralized photovoltaic power, which realizes accurate prediction of photovoltaic power generation of multiple distributed photovoltaic power stations through optimization of cascading.

[0042] Firstly, the framework of the centralized photovoltaic power prediction model based on KDE-CatBoost and its training process are introduced. That is, the actual data of the centralized photovoltaic power station is obtained, the obtained data is processed, the required data is selected, the abnormal data is deleted, and the missing data is filled; then, the centralized photovoltaic power prediction model based on KDE-CatBoost is used for photovoltaic power prediction. The power prediction process of the above-mentioned centralized photovoltaic power prediction model based on KDE-CatBoost is shown in FIG. 1, which includes: Figure 1

[0043] Step S1, the historical photovoltaic power and historical meteorological data of the centralized photovoltaic power station are obtained, and the obtained data is preprocessed. For the centralized photovoltaic power station, the historical photovoltaic power and the historical meteorological data related thereto are obtained, and the obtained data is cleaned, denoised and normalized to eliminate the interference of abnormal values and noise on the prediction result.

[0044] Step S2, the preprocessed data is analyzed and encoded by using the CatBoost algorithm, the effective meteorological features related to the photovoltaic power are extracted, and a training set including the historical photovoltaic power and the effective meteorological features is constructed. The preprocessed data is analyzed, including feature selection, feature transformation and feature construction. In this embodiment, the automatic feature encoding function of the CatBoost algorithm is used to process the category type features, and the effective meteorological features related to the photovoltaic power are extracted, and the training data set is constructed by using the effective meteorological features and the corresponding historical photovoltaic power, as shown in FIG. 2.​Figure 4 as shown.

[0045] The feature selection refers to selecting a set of features with the most information and the most relevance from various possible features for the construction of the prediction model. In this embodiment, the feature selection is performed by calculating the correlation coefficients of different meteorological factors and power. Specifically, the Pearson correlation coefficient is used to calculate the correlation coefficients (correlations) between different meteorological factors and power, and the most relevant meteorological features (i.e. meteorological factors) are selected according to the sizes of the correlation coefficients. The finally selected meteorological factors depend on the calculated correlation coefficients. The meteorological factors with higher correlation with power can have greater impact on the final power generation, ensuring the accuracy of the final prediction.

[0046] The feature transformation refers to mathematical transformation of the original feature data, so as to change the representation or distribution of the data. The purpose of this step is to reduce the dimension of the data, eliminate the correlation between the data, and improve the performance of the model.

[0047] The feature construction refers to constructing new features or using different combinations of existing features to enhance the expression ability of the data. The purpose of the feature construction is to provide features with more information and more suitable for the model, so as to improve the performance of the model.

[0048] Considering that there may be a certain relationship between the meteorological factors and the categorical features, the seasonal conditions (i.e. categorical features) can affect the meteorological factors. For example, summer can be accompanied by higher temperature and stronger sunshine. Therefore, when extracting effective features, the combination or cross of the categorical features and the meteorological factors is considered to capture the correlation between them, so as to improve the accuracy and effect of the photovoltaic power prediction. In this embodiment, the extracted categorical feature is spring, and the effective meteorological features include total radiation, direct radiation, scattered radiation, temperature, ambient temperature, air pressure, ambient humidity, and photovoltaic power.

[0049] Step S3, a photovoltaic power distribution probability prediction model based on kernel density estimation is constructed, the constructed model is trained using the training set, and the photovoltaic power distribution probability prediction model is optimized using the Adam optimizer.

[0050] Step S4, a photovoltaic power prediction model based on CatBoost algorithm is constructed, and the constructed model is trained using the training set. In this embodiment, the data set processed by the above method is used to construct a photovoltaic power prediction model using the CatBoost algorithm. Through the learning of the training set, the model can learn the complex relationship between the photovoltaic power and the meteorological factors, the historical data, and the final fitting result.

[0051] In this embodiment, taking the predicted wind power as an example, through the above steps S1 and S2, the photovoltaic power and the effective meteorological features corresponding to the power are obtained to constitute a training data set. Further, the data set is split, 80% of the data is used as a training set to train the following prediction model, and 20% of the data is used as a test set to test the trained prediction model. In steps S3 and S4, first, the kernel density estimation is performed on the feature-processed data set (including total radiation, direct radiation, scattered radiation, temperature, ambient temperature, air pressure, ambient humidity, and photovoltaic power), the photovoltaic power distribution probability characteristics are analyzed through the kernel density estimation method, and the interval prediction of the photovoltaic power is realized. By comparing and analyzing using different kernel functions and bandwidth parameters, the optimal solution is obtained, and the final probability distribution graph is obtained, as shown in Figure 2 and Figure 3 , wherein the kernel function includes KDE, NP-PWE (Nonparametric Weighted Estimation, a non-parametric estimation method, which is used to estimate an unknown probability distribution function), PF (Probability Estimation, which estimates a discrete probability distribution through sample data), etc. The final optimized kernel density estimation model is used to predict and analyze the power in the test set, and the prediction result is obtained. The obtained prediction result is input into the kernel density estimation model to obtain the best kernel function and the corresponding bandwidth value. Unsuitable kernel functions may cause function execution errors due to the deviation of some test points during system calculation, and the smaller the bandwidth is, the less smooth the probability density function obtained is, and the larger the bandwidth is, the more details the probability density function will lose.

[0052] The expression of the kernel density estimation is:

[0053]

[0054] Wherein, n is the sample capacity, h is the bandwidth, K(x) represents the kernel function, x represents the position or data point to be estimated, and X i represents the i-th data point in the sample.

[0055] The bandwidth selection of the above kernel density estimation method follows certain principles, and the expression of the index is:

[0056] ISE(x) = ∫ [f h (x) - f(x)] 2 dx

[0057] MIE(z) = ∫E[f h (x) - f(x)] 2 dx

[0058]

[0059] The above formulas are, in turn, the integrated squared error, representing the integral of the square of the error of the estimated result f h (x) and the true function f(x); the average integrated squared error, used to measure the average value of the error square of the estimated result f h (x) and the true function f(x), which can be regarded as the average value of the integrated squared error with respect to the integrated density, where E represents the mathematical expectation in probability theory; the asymptotic mean squared integrated error, representing the result obtained by omitting the minimum term of the Taylor expansion of the integrated mean squared error.

[0060] Based on the above three error indicators, the bandwidth and kernel function that minimizes the three error indicators are selected, so that the photovoltaic power distribution probability prediction model based on kernel density estimation can achieve better photovoltaic power distribution probability prediction.

[0061] Further, in order to determine whether the predicted photovoltaic power distribution conforms to the actual distribution, the photovoltaic power data is divided into m groups of non-repeating data, and the goodness of fit is used for testing, and the formula is:

[0062]

[0063] In the above formula, the χ2 test is a method for measuring the goodness of fit, which evaluates whether the fitting meets the standard by calculating the difference between the theoretical frequency and the observed frequency. Among them, A i represents the observed frequency of photovoltaic power in the ith interval, E i represents the expected frequency of photovoltaic power in the ith interval.

[0064] Among them, the CatBoost algorithm is used to explore the nonlinear coupling characteristics of photovoltaic power generation. The CatBoost prediction model is used to predict the trend of photovoltaic power generation, and further, on the basis of the above, a super short-term photovoltaic power prediction framework based on CatBoost and kernel density estimation is proposed and constructed, further improving the accuracy and reliability of photovoltaic power prediction. The expression of CatBoost is:

[0065]

[0066] In the above formula, p represents the added prior distribution term, which is taken as the average value of the target variable in all data; a (greater than 0) represents the weight coefficient, ranging from 0 to 1; represents k random feature vectors; represents the label value; represents if then otherwise

[0067] In the present embodiment, a neural network model is constructed and trained, an Adam optimization algorithm is used as an optimizer of the training process, a training data set is constructed with effective meteorological features and corresponding historical photovoltaic power in the training process, the following four loss function indicators are used to train the constructed model, the loss function is calculated by a back propagation algorithm, and the parameters of the model are updated by using the Adam optimization algorithm, and the parameters of the model are adjusted to minimize the prediction error.

[0068] Among them, the evaluation index (that is, the loss function index) of the prediction model is the mean absolute error MAE, the mean square error MSE, the root mean square error RMSE, the determination coefficient R 2 , and the calculation formula is respectively:

[0069]

[0070]

[0071]

[0072]

[0073] Among them, is the predicted power, y i is the actual power, and n represents the number of samples.

[0074] By using kernel density estimation, only the bandwidth value needs to be set, the algorithm result is stable, and sample initialization is not needed; the CatBoost algorithm analyzes and processes the category features in the training process, and uses a symmetric tree as a base model to avoid overfitting.

[0075] Further, an Adam optimization algorithm is used as an optimizer of the training process, the core idea of the Adam algorithm is to dynamically adjust the learning rate based on the first moment estimation and the second moment estimation of the gradient, which maintains the momentum variable (first moment estimation) of each parameter and the variance variable (second moment estimation) of the parameter. The optimization process of the Adam algorithm is:

[0076] (1) initialize the momentum variable m to 0 and the variance variable v to 0;

[0077] (2) in each iteration, calculate the gradient of the current batch;

[0078] (3) update the first moment estimation m t and the second moment estimation v t , and calculate the exponential moving average of the first moment estimation: m t = β1m t-1 +(1-β1)g tand the exponentially moving average of the second moment estimate: v t = β2v t-1 + (1- β2) g t 2 ;

[0079] (4) Parameter updating according to momentum variable and variance variable, including:

[0080] Learning rate correction:

[0081] Where t represents the current iteration step number;

[0082] Learning rate correction:

[0083] Parameter updating:

[0084] Where ε is a very small constant, used to avoid the denominator being 0.

[0085] By calculating the gradient of each parameter and using the Adam algorithm to update the parameters, the convergence of the model is accelerated and the prediction performance is improved; by adaptively adjusting the learning rate, the Adam algorithm can better handle different scales and changes in photovoltaic power data, thus more accurately predicting photovoltaic power.

[0086] Finally, the prediction algorithm proposed in this embodiment is applied to actual examples, and by comparing with existing commonly used prediction algorithms, it is further verified that the model proposed in this embodiment can effectively improve the prediction performance.

[0087] Specifically, the actual data of a photovoltaic power plant in northern China is selected, the required data is selected, and the collected data is quality inspected and preprocessed to process abnormal values and missing values; then feature processing is performed to extract and construct feature values related to power prediction; finally, the data set is divided into training set and test set for model training and evaluation.

[0088] Then, according to the data characteristics, a suitable prediction model is selected. According to the selected model, the model is trained and the parameters are adjusted, and the training set is used to train the model so that it can learn the relationship between photovoltaic power and meteorological data. Among them, the first 90% of the data is the training set, the last 10% of the data is the test set, the small batch sample number is selected as 10, and the dropout rate is set to 0.5. The comparative models trained in this embodiment include: a photovoltaic power prediction model based on SVR, a photovoltaic power prediction model based on DTR, a photovoltaic power prediction model based on KNN, a photovoltaic power prediction model based on LSTM, and a photovoltaic power prediction model based on LightGBM.

[0089] Based on the error index: mean absolute error SMAE Mean square error S MSE Root mean square error S RMSE Coefficient of determination S R2 The performance evaluation and error analysis of the above comparative models as deterministic prediction models can be performed to reflect the prediction accuracy of the models.

[0090] Finally, based on the obtained deterministic prediction results, the predictive performance of the model is evaluated. The superiority of the method described in this embodiment is verified by comparing the prediction results of different models. Based on the deterministic prediction results, the root mean square error of the prediction results of each different prediction model is calculated, showing reductions of 27.59%, 8.69%, 16.21%, 23.33%, and 12.56%, respectively. Combined with... Figure 5 and Figure 6 As shown, the prediction results of the scheme described in this embodiment are closer to the actual situation. Therefore, the prediction of the scheme described in this embodiment is more accurate and has less error.

[0091] In addition, five photovoltaic (PV) models were established to verify the robustness of the algorithm, including Model 1: a KDE-Catboost algorithm model considering the past 7 days of historical PV data; Model 2: a KDE-Catboost algorithm model considering an additional 30 days of historical PV data; Model 3: a KDE-Catboost algorithm model considering different seasonal characteristics; Model 4: a KDE-Catboost algorithm model considering data from different power plants; and Model 5: a KDE-Catboost model based on Model 4 with added weather features. The final prediction results are as follows: Figure 7 As shown, the solution described in this embodiment has superior robustness.

[0092] This embodiment establishes a prediction model based on CatBoost and kernel density estimation, and compares this model with SVR, DTR, KNN, LSTM, and LightGBM for predicting photovoltaic power. The prediction model optimized by CatBoost and kernel density estimation shortens the prediction time, has a better convergence speed, improves the prediction accuracy, and has a good training effect.

[0093] After training the centralized photovoltaic power prediction model based on KDE (Kernel Density Estimation)-CatBoost using the above methods, the distributed photovoltaic power is cascaded with the centralized photovoltaic power, and the distributed photovoltaic power prediction result is obtained based on the accurately predicted centralized photovoltaic power.

[0094] Specifically, firstly, the historical photovoltaic power of centralized and distributed photovoltaic power plants is obtained. In this embodiment, actual data from a photovoltaic power plant in Australia is selected, the required data is chosen, outlier data is deleted, and missing data is filled in. A dataset is constructed based on the preprocessed data. This dataset includes the target variable (the variable to be predicted, i.e., historical distributed photovoltaic power) and a set of feature variables (variables that may affect the target variable, i.e., historical centralized photovoltaic power).

[0095] Next, the Pearson correlation coefficient was used to analyze the linear correlation between each feature variable and the target variable. Based on the results of the correlation analysis, feature variables that are highly correlated with the target variable were selected. Highly correlated feature variables are more likely to be helpful in predicting the target variable; therefore, selecting these features can improve the predictive performance of the model.

[0096] The Pearson correlation coefficient ranges from -1 to 1, where a value of 1 indicates a perfect positive correlation, -1 indicates a perfect negative correlation, and 0 indicates no linear relationship. The relevant calculation formula is as follows:

[0097]

[0098] Where, x i and y i These represent the values ​​of the two variables respectively. and Let represent the averages of the two variables, and n represent the sample size.

[0099] In photovoltaic (PV) power generation forecasting, the correlation between distributed PV and centralized PV can be obtained by calculating their Pearson correlation coefficient. Specifically, assume there are n time points with distributed PV power generation values ​​x1, x2, ..., x... n The corresponding centralized photovoltaic power generation values ​​are y1, y2, ..., y n Then the Pearson correlation coefficient between them is:

[0100]

[0101] Where x and y represent the average values ​​of distributed photovoltaic and centralized photovoltaic, respectively.

[0102] By calculating the Pearson correlation coefficient between distributed photovoltaic (PV) and centralized PV, the linear correlation between their power output can be assessed. If the correlation is strong, the relationship between the two can be used to improve the accuracy of PV power generation prediction.

[0103] like Figure 8As shown in (a), (b), (c) are three different distributed photovoltaic power station and centralized photovoltaic power station linear relationship model, three Pearson correlation coefficient are all close to 1, can draw three distributed photovoltaic power station and its centralized photovoltaic power station linear correlation is extremely high, so that the correlation between data can be associated, and construct linear correlation model.

[0104] Finally, based on the centralized photovoltaic power prediction result, the prediction result is input into the linear correlation model, and the prediction value of the distributed photovoltaic power is output.

[0105] As shown in (a), (b), (c) are three different distributed photovoltaic power station and centralized photovoltaic power station linear relationship model, three Pearson correlation coefficient are all close to 1, can draw three distributed photovoltaic power station and its centralized photovoltaic power station linear correlation is extremely high, so that the correlation between data can be associated, and construct linear correlation model. Figure 9 As shown in (a), (b), (c) are three different distributed photovoltaic power station and centralized photovoltaic power station linear relationship model, three Pearson correlation coefficient are all close to 1, can draw three distributed photovoltaic power station and its centralized photovoltaic power station linear correlation is extremely high, so that the correlation between data can be associated, and construct linear correlation model.

[0106] Embodiment two

[0107] The embodiment provides a distributed photovoltaic power prediction system based on optimized cascade, which comprises:

[0108] A data acquisition module is configured to acquire historical photovoltaic power of centralized photovoltaic power stations and distributed photovoltaic power stations.

[0109] A linear correlation model construction module is configured to calculate Pearson correlation coefficients between centralized photovoltaic power stations and distributed photovoltaic power stations based on the historical photovoltaic power, and establish a linear correlation model between centralized photovoltaic power and distributed photovoltaic power.

[0110] A centralized photovoltaic power prediction model is configured to input meteorological data of a prediction day into the centralized photovoltaic power prediction model based on KDE-CatBoost, and output a centralized photovoltaic power prediction value; the centralized photovoltaic power prediction model is trained by using an Adam optimization algorithm, the input meteorological data is processed by using a kernel density estimation method, a photovoltaic power distribution probability prediction result is output, and a photovoltaic power prediction value is output by using a CatBoost algorithm based on the photovoltaic power distribution probability prediction result.

[0111] A distributed photovoltaic power prediction model is configured to obtain a distributed photovoltaic power prediction value according to the centralized photovoltaic power prediction value and the linear correlation model.

[0112] Embodiment three

[0113] The embodiment provides an electronic device, which comprises a memory, a processor, computer instructions stored in the memory and running on the processor, and steps in the distributed photovoltaic power prediction method based on optimized cascade are completed when the computer instructions are run by the processor.

[0114] Embodiment Four

[0115] The embodiments also provide a computer readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps in the method for distributed photovoltaic power prediction based on optimized cascade as described above.

[0116] The steps and methods involved in the above embodiments two to four correspond to the method embodiment one, and the specific implementation can refer to the relevant description part of the embodiment one. The term "computer readable storage medium" should be understood as including a single medium or multiple media of one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying instruction sets for execution by a processor and causing the processor to perform any of the methods in the present application.

[0117] Those skilled in the art should understand that each module or step of the present application described above can be realized by a general computer device, alternatively, they can be realized by program codes executable by a computing device, so that they can be stored in a storage device for execution by a computing device, or they can be respectively made into each integrated circuit module, or a plurality of modules or steps among them can be made into a single integrated circuit module to realize. The present application is not limited to any specific combination of hardware and software.

[0118] The above description is only the preferred embodiments of the present application, although the specific embodiments of the present application are described in combination with the drawings, but it is not a limitation on the protection scope of the present application, those skilled in the art should understand that various modifications or changes made on the basis of the technical solutions of the present application without creative labor are still within the protection scope of the present application.

Claims

1. A method for distributed photovoltaic power prediction based on optimized cascading, characterized in that, The application comprises the following steps: acquiring historical photovoltaic power of centralized photovoltaic power stations and distributed photovoltaic power stations; based on the historical photovoltaic power, calculating the Pearson correlation coefficient between the centralized and distributed photovoltaic power stations, and establishing a linear correlation model of the centralized photovoltaic power and the distributed photovoltaic power; inputting meteorological data of a predicted day into a centralized photovoltaic power prediction model based on KDE-CatBoost, and outputting a centralized photovoltaic power prediction value; the centralized photovoltaic power prediction model is trained by using an Adam optimization algorithm, the input meteorological data is processed by using a kernel density estimation method, a photovoltaic power distribution probability prediction result is output, based on the photovoltaic power distribution probability prediction result, a photovoltaic power prediction value is output by using a CatBoost algorithm; obtaining a distributed photovoltaic power prediction value according to the centralized photovoltaic power prediction value and the linear correlation model; the training process of the centralized photovoltaic power prediction model based on KDE-CatBoost comprises the following steps: acquiring historical photovoltaic power and historical meteorological data of centralized photovoltaic power stations, and preprocessing the acquired data; performing feature analysis and coding on the preprocessed data by using a CatBoost algorithm; the feature analysis comprises feature selection, feature transformation and feature construction; processing category type features by using the automatic feature coding function of the CatBoost algorithm, extracting effective meteorological features related to photovoltaic power, and constructing a training data set by using the category type features, the effective meteorological features and corresponding historical photovoltaic power; constructing a photovoltaic power distribution probability prediction model based on kernel density estimation, training the constructed model by using the training set, and performing optimization processing on the photovoltaic power distribution probability prediction model by using an Adam optimizer; constructing a photovoltaic power prediction model based on the CatBoost algorithm, and training the constructed model by using the training set; wherein the expression of the kernel density estimation is: ； in, n For sample size, h For bandwidth, K(x) Represents the kernel function. x This represents the location or data point where the probability density is to be estimated. Indicates the first in the sample i Data points.

2. The method of claim 1, wherein the optimization-based cascading is performed by a plurality of optimization-based cascading modules. the category type features are spring, and the effective meteorological features comprise total radiation, direct radiation, scattered radiation, temperature, ambient temperature, air pressure, ambient humidity and photovoltaic power.

3. The method of claim 1, wherein the optimization-based cascading is performed by a plurality of optimization-based cascading modules, each of which is configured to perform the optimization-based cascading on a subset of the plurality of PV power plants. performing optimization processing on the photovoltaic power distribution probability prediction model by using the Adam optimizer, which comprises the following steps: Initialize momentum variable to 0 and variance variable to 0; in each iteration, calculating the gradient of the current batch; updating the first moment estimate and the second moment estimate and computing an exponential moving average of the first moment estimate and computing an exponential moving average of the second moment estimate; updating parameters according to the momentum variable and the variance variable.

4. A distributed photovoltaic power prediction system based on optimized cascading, characterized in that, The application comprises the following steps: a data acquisition module is configured to acquire historical photovoltaic power of centralized photovoltaic power stations and distributed photovoltaic power stations; a linear correlation model construction module is configured to calculate the Pearson correlation coefficient between the centralized and distributed photovoltaic power stations based on the historical photovoltaic power, and establish a linear correlation model of the centralized photovoltaic power and the distributed photovoltaic power; a centralized photovoltaic power prediction module is configured to input meteorological data of a predicted day into a centralized photovoltaic power prediction model based on KDE-CatBoost, and output a centralized photovoltaic power prediction value; the centralized photovoltaic power prediction model is trained by using an Adam optimization algorithm, the input meteorological data is processed by using a kernel density estimation method, a photovoltaic power distribution probability prediction result is output, based on the photovoltaic power distribution probability prediction result, a photovoltaic power prediction value is output by using a CatBoost algorithm; The distributed photovoltaic power prediction module is used for obtaining a distributed photovoltaic power prediction value according to a centralized photovoltaic power prediction value and a linear correlation model; The training process of the KDE-CatBoost-based centralized photovoltaic power prediction model comprises the following steps: Obtaining historical photovoltaic power and historical meteorological data of a centralized photovoltaic power station, and preprocessing the obtained data; Performing feature analysis and coding on the preprocessed data by using a CatBoost algorithm, wherein the feature analysis comprises feature selection, feature transformation and feature construction; processing category type features by using the automatic feature coding function of the CatBoost algorithm, extracting effective meteorological features related to photovoltaic power, and constructing a training data set by using the category type features, the effective meteorological features and corresponding historical photovoltaic power; Constructing a photovoltaic power distribution probability prediction model based on kernel density estimation, training the constructed model by using the training set, and performing optimization processing on the photovoltaic power distribution probability prediction model by using an Adam optimizer; Constructing a photovoltaic power prediction model based on the CatBoost algorithm, and training the constructed model by using the training set; The expression of the kernel density estimation is as follows: ； wherein, n is the sample size, h is the bandwidth, K(x) denotes the kernel function, x denotes the position or data point for which the probability density is to be estimated, denotes the th data point in the sample, i denotes the th data point in the sample.

5. The distributed photovoltaic power forecasting system based on optimized cascade according to claim 4, characterized in that, The optimization processing on the photovoltaic power distribution probability prediction model by using the Adam optimizer comprises the following steps: Initialize momentum variable to 0 and variance variable to 0; In each iteration, the gradient of the current batch is calculated; updating the first moment estimate and the second moment estimate and computing an exponential moving average of the first moment estimate and computing an exponential moving average of the second moment estimate; According to the momentum variable and the variance variable, the parameters are updated.

6. An electronic device, characterized by comprising: Computer instructions stored in the memory and running on the processor, when the computer instructions are run by the processor, complete the steps of the distributed photovoltaic power prediction method based on the optimized cascade in any one of claims 1-3.

7. A computer readable storage medium characterized by, Computer instructions for storing computer instructions, when the computer instructions are executed by the processor, complete the steps of the distributed photovoltaic power prediction method based on the optimized cascade in any one of claims 1-3.

Images