Photovoltaic power generation power short-term prediction algorithm based on improved CEEMDAN multiple decomposition

Abstract

The invention relates to the technical field of photovoltaic power generation power short-term prediction, and particularly discloses a photovoltaic power generation power short-term prediction algorithm based on improved CEEMDAN multiple decomposition, and the algorithm comprises the following steps: S01, obtaining real-time photovoltaic power generation power data, and carrying out the preprocessing of a missing value and an outlier in the data, so as to obtain a feature serial number; s02, decomposing the feature serial number based on complete adaptive noise set empirical mode decomposition, adding a group of k-order imf in specific white noise in decomposition, and removing noise through iterative calculation to obtain a plurality of intrinsic mode functions; according to the method, through methods of environment factor feature selection decomposition dimensionality reduction, photovoltaic sequence decomposition, high-frequency reconstruction noise reduction and the like, the photovoltaic power generation power has better precision and lower errors in a strongly coupled data set or a training set with an unobvious trend.

Description

Technical Field

[0001] The invention relates to the technical field of photovoltaic power short-term prediction, and in particular to a photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition. Background Art

[0002] The contradiction between the continuous population growth and the increasing shortage of energy has always been a major problem facing modern society. In order to achieve sustainable development, it is necessary to give full play to the alternative role of renewable energy, reduce the use of fossil fuels, and achieve harmonious coexistence between man and nature. Solar energy is inexhaustible and has the advantages of being clean and pollution-free, which is worthy of our full development and utilization. The photovoltaic industry is not only an important part of the renewable energy industry, but also an important engine to promote my country's energy transformation. However, photovoltaic power generation has obvious periodicity, randomness and volatility. Large-scale access to photovoltaic power generation will bring certain impacts to the stable operation of the power grid. Therefore, it is necessary to predict the photovoltaic power generation power in advance, so that the photovoltaic power generation department can formulate a power dispatching plan in advance to improve the safety and reliability of the power grid. Therefore, it is crucial to develop an accurate photovoltaic power prediction system.

[0004] Common photovoltaic power generation prediction methods include the following aspects: ultra-short-term prediction method, short-term prediction method, medium-term prediction method and long-term prediction method. From the perspective of power grid operation, a shorter prediction period is more conducive to timely optimization of scheduling and prevention of emergencies, and a longer prediction period is more conducive to photovoltaic power stations to plan installed capacity in advance and estimate the power generation income of power stations. In terms of prediction models: photovoltaic power generation prediction models are mainly divided into two categories: one is a physical model based on photovoltaic components and local weather and meteorological factors. This type of model is greatly affected by environmental conditions, has many parameters, is difficult to model, and is rarely used; the other is a data analysis method based on the intrinsic relationship between historical power generation data and environmental factors such as solar irradiance, component temperature, air temperature, air pressure and humidity. The development of computer technology has prompted the application of a variety of machine learning methods in this field, such as artificial neural networks, support vector machines, extreme learning machines, etc. However, a single model is often difficult to meet the prediction accuracy requirements, so a variety of combined prediction methods have emerged.

[0005] The combination method integrates models with different characteristics, selects appropriate methods to combine models, gives full play to the advantages of different models, and achieves more accurate prediction accuracy. Prediction methods can be divided into two categories. The first is to use different types of prediction models to predict photovoltaic power generation, and after weighting their prediction values, the prediction results are recombined to produce the final prediction results. The second method is to divide the initial data set into subsets, use different prediction models to predict these subsets, and finally superimpose the prediction values ​​of the subsets to obtain the final prediction results. Through the combination prediction method, the advantages of different prediction models can be effectively utilized, avoiding the high prediction error generated by a single model algorithm, and increasing the prediction accuracy of the model.

[0006] With the rapid development of artificial intelligence, the decomposition algorithm is increasingly used in photovoltaic power prediction. Since environmental data and photovoltaic power history series have strong non-stationarity, it is difficult to meet the demand by simply using neural networks to increase the prediction accuracy of the model. Therefore, the importance of frequency analysis of time series data has gradually become prominent. The more widely used signal time domain and frequency domain decomposition techniques include empirical mode decomposition method and adaptive noise complete set empirical mode decomposition method. Using signal decomposition technology to reduce the non-stationarity of photovoltaic power history series enables the neural network model to better extract data features, reduces the limitations of model application, and can improve the accuracy of photovoltaic power prediction model to a certain extent.

[0007] However, most of the existing artificial intelligence prediction algorithms have the following problems

[0008] 1. Usually only a single power sequence is targeted, the impact of environmental factors is ignored, and there is a lack of effective use of multi-environmental sequence information.

[0009] 2. No secondary processing is performed on the decomposed sequence, resulting in a large amount of information hidden in the high-frequency sequence being lost.

[0010] 3. Due to the volatility, randomness, nonlinearity and seasonality of photovoltaic power generation, the prediction effect will decrease for some time periods with unclear trends or sudden changes, and it has a greater dependence on the stability and trend of the data.

[0011] Therefore, in response to the above problems, this paper proposes a short-term prediction algorithm for photovoltaic power generation based on improved CEEMDAN multiple decomposition. Through environmental factor feature selection decomposition and dimensionality reduction, photovoltaic sequence decomposition and high-frequency reconstruction denoising methods, the photovoltaic power generation power has better accuracy and lower error in both strongly coupled data sets and training sets with unclear trends. Summary of the invention

[0012] The purpose of the present invention is to provide a photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition, which can effectively deal with the volatility, randomness, nonlinearity and seasonality of photovoltaic power, especially its high-frequency characteristics, so as to improve the prediction accuracy.

[0013] In order to achieve the above object, the present invention provides the following technical solutions:

[0014] A photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition includes the following steps:

[0015] S01, obtaining photovoltaic power generation and environmental factor information, and preprocessing photovoltaic power generation information;

[0016] S02, performing fully adaptive noise set empirical mode decomposition and high-frequency sequence denoising on the photovoltaic power generation sequence, performing variational mode decomposition and kernel principal component analysis dimensionality reduction on the environmental factors, and after reconstructing the data set, decomposing the subsequences into a training set, a validation set, and a test set;

[0017] S03, construct a BiGRU neural network, and input the test set into the trained BiGRU network to obtain the test results of each subsequence;

[0018] S04. Input the test set into the trained BiGRU network to obtain the test results of each subsequence, and obtain the final test result after reconstruction and denormalization.

[0019] Preferably, the photovoltaic power generation information is preprocessed in step S01, which is a process for processing missing values ​​and outliers in the Yuanshu data, including:

[0020] S11. For missing values ​​in the data, interpolation is used to fill them:

[0021]

[0022] Where x(t) is the estimated value of the missing value, x(t1) and x(t2) are two known data points adjacent to the missing value, and t1 and t2 are the times corresponding to the data points;

[0023] S12. Use the quartile method to handle outliers in the data:

[0024] S121, calculate the first quartile (Q1) and the third quartile (Q3) of the data, and then calculate the interquartile range (IQR): IQR = Q3-Q1;

[0025] S122. Define the outlier range based on the interquartile range, and replace the detected outliers with the median:

[0026] x new =median(X);

[0027] S123. After completing the data, normalize the data:

[0028]

[0029] Among them, x′ is the normalized eigenvalue, max(x) and min(x) are the minimum and maximum eigenvalues, respectively.

[0030] S13. Through normalization, the eigenvalues ​​are scaled to the interval [0,1] so that different features have the same dimension.

[0031] Preferably, the interquartile range in step S122 defines the outlier range as any data point smaller than Q1-1.5IQR or larger than Q3+1.5IQR.

[0032] Preferably, the processing step of decomposing the characteristic sequence number based on fully adaptive noise set empirical mode decomposition in step S02 to obtain a plurality of denoised intrinsic mode functions includes:

[0033] S21. Calculate the sample entropy of each intrinsic mode function:

[0034]

[0035] Among them, m is the embedding dimension, r is the similarity tolerance, N is the data length, and A m and B m are the similarity statistics when the embedding dimensions are m and m+1 respectively;

[0036] S22. Apply empirical wavelet transform to the identified high-frequency sequence to determine the maximum value in its spectrum, thereby dividing the signal into several sub-bands by the spectrum maximum value, and adaptively selecting and constructing wavelet basis functions.

[0037] Preferably, the step of processing the acquired environmental characteristic data of the photovoltaic power generation installation by combining the Pearson correlation coefficient, variational mode decomposition and kernel principal component analysis in step S02 includes:

[0038] S31. Use the Pearson correlation coefficient to select environmental feature data, and determine the feature columns with higher correlation according to the threshold value |r|≥0.3:

[0039]

[0040] Among them, r represents the Pearson correlation coefficient, x i and y are the eigenvalue and target value respectively, and is their mean;

[0041] S32. After feature selection, the environmental feature data is decomposed using the variational mode decomposition method:

[0042] S321, first select the kernel function and calculate the kernel matrix K,

[0043] S322, obtaining the principal component by eigenvalue decomposition,

[0044]

[0045] Among them, K(x,y) is the Gaussian kernel function and σ is the kernel width parameter.

[0046] Preferably, constructing a BiGRU neural network in S03 includes:

[0047] S41, determining the network structure and parameter selection based on the input processed feature data;

[0048] S42, using the data processed in step S41 as a training test set, converting it into a time series format suitable for BiLSTM model input, and performing model training on the training set to obtain a BiLSTM model;

[0049] S43, predicting the test set through the trained BiLSTM model, and outputting the predicted value of future photovoltaic power generation;

[0050] S44, denormalizing the predicted values ​​and then superimposing them to obtain an actual output power value.

[0051] Preferably, the network structure and parameter selection process in step S41 includes:

[0052] S31, input the processed feature data, the dimension of the input data includes a time step and a feature number, wherein each time step contains a multi-dimensional feature;

[0053] S32, extracting sequence features by stacking a first layer of BiLSTM, where the first layer of BiLSTM includes 128 hidden units;

[0054] S33, add a Dropout layer after the first BiLSTM layer, with a dropout rate of 0.1;

[0055] S34, extracting sequence features by stacking a second layer of BiLSTM, where the second layer of BiLSTM includes 64 hidden units;

[0056] S35, add a Dropout layer after the second BiLSTM layer, with a dropout rate of 0.1;

[0057] S36. Connect the output of the BiLSTM layer to a fully connected layer, which contains 16 neurons and uses a sigmoid activation function to extract features and adjust the data scale output.

[0058] Preferably, the Adam optimizer is used for parameter optimization in the network structure and parameter selection process, the mean square error is selected as the loss function, the initial learning rate is set to 0.001, and the training is automatically stopped when the error of the validation set fails to decrease within several training cycles.

[0059] In the above technical solution, the present invention provides a photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition, which has the following beneficial effects:

[0060] 1. The improved fully adaptive noise ensemble empirical mode decomposition is used to decompose the photovoltaic power time series into several intrinsic mode functions by adding white noise and decomposing it, which effectively separates the different frequency components of the signal, reduces the modal aliasing phenomenon, and improves the reliability and accuracy of the decomposition results. The sample entropy calculation for each intrinsic mode function can quantify the dynamic characteristics and information content of the sequence, identify high-frequency sequences, and evaluate the complexity and uncertainty of the sequence. It is suitable for processing random fluctuations and complex changes in photovoltaic power generation. The empirical wavelet transform is applied to the high-frequency sequence for denoising. By adaptively selecting the waveform, the noise can be effectively removed, the hidden useful information can be extracted, and the signal quality can be enhanced.

[0061] 2. For environmental feature selection, by calculating the Pearson coefficient, select environmental features with high correlation with photovoltaic power generation, reduce the complexity and redundant information of the model, and improve the training efficiency and prediction accuracy of the model. Subsequently, the selected Polygonatum sibiricum features are decomposed into multiple intrinsic modal functions using variational mode decomposition, extract information of different frequency components, avoid modal aliasing, and provide a high-quality data foundation. Then, kernel principal component analysis dimensionality reduction is performed based on the variational mode decomposition results to maintain the main features, reduce volatility and variable dimensions, and improve the correlation between feature columns and target columns. Finally, a bidirectional long short-term memory network is used for sequence prediction to process long-term dependencies and complex time series features in photovoltaic power generation, capture contextual information, and significantly improve prediction accuracy and stability. The combination of these technologies enables the present invention to effectively handle the volatility, randomness, nonlinearity and seasonality of photovoltaic power generation, thereby significantly improving the accuracy of short-term predictions.

[0062] 3. It can effectively handle the volatility, randomness, nonlinearity and seasonality of photovoltaic power generation, especially extract information on its high-frequency characteristics and environmental variables, thereby improving prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0064] Figure 1 A flowchart provided for an embodiment of the present invention;

[0065] Figure 2 A data processing framework diagram provided for an embodiment of the present invention;

[0066] Figure 3 Photovoltaic sequence decomposition provided by embodiments of the present invention;

[0067] Figure 4 An empirical wavelet transform denoising flow chart provided by an embodiment of the present invention;

[0068] Figure 5 This is an empirical wavelet change denoising effect diagram provided by an embodiment of the present invention;

[0069] Figure 6 A BiLSTM model diagram provided for an embodiment of the present invention;

[0070] Figure 7 A test set prediction result diagram provided by an embodiment of the present invention;

[0071] Figure 8 This is a graph of prediction results for one day in the test set provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0073] like Figure 1-8 As shown, a photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition includes the following steps:

[0074] S01, obtaining photovoltaic power generation and environmental factor information, and preprocessing the photovoltaic power generation information, that is, obtaining real-time photovoltaic power generation data, and preprocessing missing values ​​and outliers in the data to obtain characteristic sequence numbers;

[0075] The processing steps include:

[0076] include:

[0077] S11. For missing values ​​in the data, interpolation is used to fill them:

[0078]

[0079] Where x(t) is the estimated value of the missing value, x(t1) and x(t2) are two known data points adjacent to the missing value, and t1 and t2 are the times corresponding to the data points;

[0080] S12. Use the quartile method to handle outliers in the data:

[0081] S121, calculate the first quartile (Q1) and the third quartile (Q3) of the data, and then calculate the interquartile range (IQR): IQR = Q3-Q1;

[0082] S122. Define the outlier range based on the interquartile range, and replace the detected outliers with the median:

[0083] x new =median(X);

[0084] S123. After completing the data, normalize the data:

[0085]

[0086] Among them, x is the normalized eigenvalue, max(x) and min(x) are the minimum and maximum eigenvalues, respectively.

[0087] S13. Through normalization, the eigenvalues ​​are scaled to the interval [0,1] so that different features have the same dimension. Making different features have the same dimension helps improve the prediction performance of the model.

[0088] It should be noted that the interquartile range in step S122 defines the outlier range as any data point smaller than Q1-1.5IQR or larger than Q3+1.5IQR.

[0089] The above-mentioned method for processing missing values ​​and outliers can significantly improve the quality of photovoltaic power data, while data normalization makes the features have the same dimension, ensuring the accuracy and reliability of subsequent analysis and prediction models.

[0090] S02, performing fully adaptive noise set empirical mode decomposition and high-frequency sequence denoising on the photovoltaic power generation sequence, performing variational mode decomposition and kernel principal component analysis dimensionality reduction on the environmental factors, and after reconstructing the data set, decomposing the subsequences into a training set, a validation set, and a test set;

[0091] The decomposition and denoising tasks of photovoltaic power generation series are carried out by combining the three techniques of improved fully adaptive noise ensemble empirical mode decomposition, sample entropy calculation and high-frequency sequence denoising.

[0092] Firstly, the improved fully adaptive noise ensemble empirical mode decomposition method is used to decompose the time series data of photovoltaic power generation. The specific steps are as follows: adaptively adjusted white noise is added in each decomposition process, and the influence of noise is removed by iterative calculation, and finally several intrinsic mode functions are obtained. This method improves the decomposition effect of the traditional EMD method by adding white noise, effectively reduces the mode aliasing phenomenon, and improves the stability and accuracy of the decomposition.

[0093] Secondly, the sample entropy of each intrinsic mode function is calculated. Sample entropy is an unsupervised complexity measurement method. By calculating the sample entropy of each IMF, its dynamic characteristics and information content can be quantified.

[0094] First, the sample entropy of each intrinsic mode function is calculated:

[0095]

[0096] Among them, m is the embedding dimension, r is the similarity tolerance, N is the data length, and A m and B m They are the similarity statistics when the embedding dimension is m and m+1 respectively. By calculating the sample entropy, high-frequency sequences can be identified and determined, providing a basis for subsequent denoising processing.

[0097] Then, the empirical wavelet transform is applied to the identified high-frequency sequence to determine the maximum value in its spectrum, so as to divide the signal into several sub-bands by the maximum value of the spectrum, and adaptively select and construct the wavelet basis function. This step denoises the high-frequency sequence, effectively removes the noise and extracts the useful information hidden in the high-frequency signal, thereby enhancing the quality and reliability of the signal. It can fully extract the information in the decomposed high-frequency sequence, which is suitable for complex and highly volatile photovoltaic power generation systems.

[0098] Based on the Pearson correlation coefficient, variational mode decomposition and kernel principal component analysis, the acquired environmental characteristic data of the installed photovoltaic power generation are processed to obtain the characteristic data after dimensionality reduction;

[0099] First, the Pearson correlation coefficient is used to select environmental feature data, and the feature columns with higher correlation are determined according to the threshold value |r|≥0.3:

[0100]

[0101] Among them, r represents the Pearson correlation coefficient, x i and y are the eigenvalue and target value respectively, and is their mean.

[0102] Here, the Pearson correlation coefficient can be used to effectively screen out key features that have a significant impact on photovoltaic power generation, reduce the complexity and redundant information of the model, and thus improve the training efficiency and prediction accuracy of the model.

[0103] Secondly, the variational mode decomposition method is used to decompose the environmental feature data after feature selection. Variational mode decomposition decomposes the input signal into several intrinsic mode functions by performing spectral decomposition. Each intrinsic mode function represents the signal component in a different frequency range. Finally, based on the intrinsic mode function obtained by variational mode decomposition, kernel principal component analysis is used for dimensionality reduction. Kernel principal component analysis uses kernel functions to map the original data to a high-dimensional space, and performs principal component analysis in the high-dimensional space to capture nonlinear features. The specific steps include:

[0104] The environmental feature data after feature selection is decomposed using the variational mode decomposition method:

[0105] First, select the kernel function and calculate the kernel matrix K.

[0106] Then the principal components are obtained through eigenvalue decomposition.

[0107]

[0108] Among them, K(x,y) is the Gaussian kernel function and σ is the kernel width parameter.

[0109] The key features are screened out by the Pearson correlation coefficient, and VMD accurately separates different frequency components to avoid modal aliasing, while the kernel principal component analysis retains important features during the dimensionality reduction process, reduces data noise and redundancy, and improves the training efficiency and prediction accuracy of the model. The combination of these technologies enables the present invention to better handle complex and changeable environmental data and provide more reliable short-term predictions of photovoltaic power generation.

[0110] S03, construct a BiGRU neural network, and input the test set into the trained BiGRU network to obtain the test results of each subsequence;

[0111] For the feature data after dimensionality reduction, a bidirectional long short-term memory network is used for sequence prediction to capture the complex temporal characteristics and long-term dependencies of the time series. The specific network structure and parameter selection are as follows:

[0112] S31, input the processed feature data, the dimension of the input data includes a time step and a feature number, wherein each time step contains a multi-dimensional feature;

[0113] S32, extract sequence features by stacking the first layer of BiLSTM, the first layer of BiLSTM contains 128 hidden units;

[0114] S33, add a Dropout layer after the first BiLSTM layer, with a dropout rate of 0.1;

[0115] S34, extract sequence features by stacking the second layer of BiLSTM, the second layer of BiLSTM contains 64 hidden units;

[0116] S35, add a Dropout layer after the second BiLSTM layer, with a dropout rate of 0.1;

[0117] S36. Connect the output of the BiLSTM layer to the fully connected layer, which contains 16 neurons and uses the sigmoid activation function to extract features and adjust the data scale output.

[0118] During the network training process, the Adam optimizer was used for parameter optimization in the network structure and parameter selection process. The mean square error was selected as the loss function, and the initial learning rate was set to 0.001. When the error of the validation set failed to decrease within several training cycles, the training was automatically stopped.

[0119] S04: Based on the obtained feature data after dimensionality reduction, the test set is input into the trained BiGRU network to obtain the test results of each subsequence. The steps include:

[0120] S41, determining the network structure and parameter selection based on the input processed feature data;

[0121] S42, using the data processed in step S41 as a training test set, converting it into a time series format suitable for BiLSTM model input, and performing model training on the training set to obtain a BiLSTM model;

[0122] S43, predicting the test set through the trained BiLSTM model, and outputting the predicted value of future photovoltaic power generation;

[0123] S44, performing inverse normalization on the predicted values ​​and then superimposing them to obtain the actual output power value.

[0124] After the input feature data is processed as described above and divided into training and test sets, it is converted into a time series format suitable for BiLSTM model input. The training set is used for model training, and the test set is predicted by the trained BiLSTM model to output the predicted value of future photovoltaic power generation. The results are denormalized and superimposed, and the actual power value is finally output.

[0125] Through the design and training prediction of the above-mentioned BiLSTM network structure, the present invention can fully utilize the advantages of the bidirectional long short-term memory network, capture the complex timing characteristics and long-term dependencies in the photovoltaic power generation time series, and significantly improve the prediction accuracy and robustness of the model.

[0126] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0127] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0128] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0129] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0130] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0131] The embodiments of the present application also provide a specific implementation of an electronic device capable of implementing all steps of the method in the above embodiments, and the electronic device specifically includes the following contents:

[0132] Processor, memory, communications interface and bus;

[0133] Wherein, the processor, memory, and communication interface communicate with each other via the bus;

[0134] The processor is used to call the computer program in the memory, and when the processor executes the computer program, all the steps in the method in the above embodiment are implemented.

[0135] The embodiments of the present application also provide a computer-readable storage medium capable of implementing all the steps of the method in the above embodiments, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, all the steps of the method in the above embodiments are implemented.

[0136] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the hardware + program embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. Although the embodiment of this specification provides the method operation steps described in the embodiment or flow chart, more or less operation steps can be included based on conventional or non-creative means. The order of steps listed in the embodiment is only one way of executing the order of many steps, and does not represent the only execution order. When the device or terminal product in practice is executed, it can be executed in sequence or in parallel according to the method shown in the embodiment or the accompanying drawings (for example, a parallel processor or a multi-threaded processing environment, or even a distributed data processing environment). The term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, product or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such a process, method, product or device. In the absence of more restrictions, it is not excluded that there are other identical or equivalent elements in the process, method, product or device including the elements. For the convenience of description, the above device is described by dividing it into various modules according to its functions. Of course, when implementing the embodiments of this specification, the functions of each module can be implemented in the same one or more software and / or hardware, or the module implementing the same function can be implemented by a combination of multiple sub-modules or sub-units. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as 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 coupling or direct coupling or communication connection between each other shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms. The present invention is described with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the process and / or box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0137] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, the embodiments of this specification may be in the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the embodiments of this specification may be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts refer to the partial description of the method embodiment. In the description of this specification, the description of the reference term "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of this specification.

[0138] In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradicting each other. The above is only an embodiment of the embodiment of this specification and is not intended to limit the embodiment of this specification. For those skilled in the art, the embodiment of this specification may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiment of this specification shall be included in the scope of the claims of the embodiment of this specification.

Claims

1. A photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition, characterized in that: The following steps are involved: S01, obtaining photovoltaic power generation and environmental factor information, and preprocessing photovoltaic power generation information; S02, performing fully adaptive noise set empirical mode decomposition and high-frequency sequence denoising on the photovoltaic power generation sequence, performing variational mode decomposition and kernel principal component analysis dimensionality reduction on the environmental factors, and after reconstructing the data set, decomposing the subsequences into a training set, a validation set, and a test set; S03, construct a BiGRU neural network, and input the test set into the trained BiGRU network to obtain the test results of each subsequence; S04. Input the test set into the trained BiGRU network to obtain the test results of each subsequence, and obtain the final test result after reconstruction and denormalization.

2. The photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition according to claim 1 is characterized in that: The photovoltaic power generation information is preprocessed in step S01, which is a process for processing missing values ​​and outliers in the data, including: S11. For missing values ​​in the data, interpolation is used to fill them: Where x(t) is the estimated value of the missing value, x(t1) and x(t2) are two known data points adjacent to the missing value, and t1 and t2 are the times corresponding to the data points; S12. Use the quartile method to handle outliers in the data: S121, calculate the first quartile (Q1) and the third quartile (Q3) of the data, and then calculate the interquartile range (IQR): IQR = Q3-Q1; S122. Define the outlier range based on the interquartile range, and replace the detected outliers with the median: x new =median(X); S123. After completing the data, normalize the data: Among them, x′ is the normalized eigenvalue, max(x) and min(x) are the minimum and maximum eigenvalues, respectively. S13. Through normalization, the eigenvalues ​​are scaled to the interval [0,1] so that different features have the same dimension.

3. The photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition according to claim 2 is characterized in that: The interquartile range in step S122 defines the outlier range as any data point smaller than Q1-1.5IQR or larger than Q3+1.5IQR.

4. The photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition according to claim 1 is characterized in that: The processing steps of decomposing the characteristic sequence number based on fully adaptive noise set empirical mode decomposition in step S02 to obtain multiple denoised intrinsic mode functions include: S21. Calculate the sample entropy of each intrinsic mode function: Among them, m is the embedding dimension, r is the similarity tolerance, N is the data length, and A m and B m are the similarity statistics when the embedding dimensions are m and m+1 respectively; S22. Apply empirical wavelet transform to the identified high-frequency sequence to determine the maximum value in its spectrum, thereby dividing the signal into several sub-bands by the spectrum maximum value, and adaptively selecting and constructing wavelet basis functions.

5. The photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition according to claim 1 is characterized in that: In step S02, the steps of processing the acquired environmental characteristic data of photovoltaic power generation installation by combining Pearson correlation coefficient, variational mode decomposition and kernel principal component analysis include: S31. Use the Pearson correlation coefficient to select environmental feature data, and determine the feature columns with higher correlation according to the threshold value |r|≥0.3: Among them, r represents the Pearson correlation coefficient, x i and y are the eigenvalue and target value respectively, and is their mean; S32. After feature selection, the environmental feature data is decomposed using the variational mode decomposition method: S321, first select the kernel function and calculate the kernel matrix K, S322, obtaining the principal component by eigenvalue decomposition, Among them, K(x,y) is the Gaussian kernel function and σ is the kernel width parameter.

6. The photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition according to claim 1 is characterized in that: In step S04, the test set is input into the trained BiGRU network to obtain the test results of each subsequence, including: S41, determining the network structure and parameter selection based on the input processed feature data; S42, using the data processed in step S41 as a training test set, converting it into a time series format suitable for BiLSTM model input, and performing model training on the training set to obtain a BiLSTM model; S43, predicting the test set through the trained BiLSTM model, and outputting the predicted value of future photovoltaic power generation; S44, denormalizing the predicted values ​​and then superimposing them to obtain an actual output power value.

7. The photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition according to claim 6 is characterized in that: The BiGRU neural network is constructed in S03, including: S31, input the processed feature data, the dimension of the input data includes a time step and a feature number, wherein each time step contains a multi-dimensional feature; S32, extracting sequence features by stacking a first layer of BiLSTM, where the first layer of BiLSTM includes 128 hidden units; S33, add a Dropout layer after the first BiLSTM layer, with a dropout rate of 0.1; S34, extracting sequence features by stacking a second layer of BiLSTM, where the second layer of BiLSTM includes 64 hidden units; S35, add a Dropout layer after the second BiLSTM layer, with a dropout rate of 0.1; S36. Connect the output of the BiLSTM layer to a fully connected layer, which contains 16 neurons and uses a sigmoid activation function to extract features and adjust the data scale output.

8. The photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition according to claim 6 is characterized in that: The Adam optimizer is used for parameter optimization in the network structure and parameter selection process, the mean square error is selected as the loss function, the initial learning rate is set to 0.001, and the training is automatically stopped when the error of the verification set fails to decrease within several training cycles.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition as claimed in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the photovoltaic power short-term prediction algorithm based on improved CEEMDAN multiple decomposition as claimed in any one of claims 1 to 8 are implemented.

Images