Photovoltaic output prediction method and system considering illumination radiation flux calculation model

Abstract

The invention discloses a photovoltaic output prediction method and system considering an illumination radiation flux calculation model, and relates to the technical field of photovoltaic power generation prediction, and the method comprises the steps: building a sun operation model, analyzing the parameters of a photovoltaic power station, and collecting and preprocessing photovoltaic power generation data; hyper-parameters are intelligently set through a Bayesian method, space and time features are extracted, a Kupman-like operator and a cost function are constructed, and iterative solution is carried out; and taking the total radiation flux predicted value as input, and performing photovoltaic output prediction through a convolutional neural network. The photovoltaic output prediction method considering the illumination radiation flux calculation model provided by the invention adapts to photovoltaic output changes in different regions and under different climate conditions, the generalization ability and prediction precision of the model are enhanced, a physically constrained space-time neural network is constructed, the interpretability is enhanced, and the prediction accuracy is improved. The method has better effects in the aspects of generalization ability, prediction precision and interpretability.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation prediction, and in particular to a photovoltaic output prediction method considering a light radiation flux calculation model. Background Art

[0002] With the transformation of the global energy structure and the challenges of climate change, renewable energy, especially solar photovoltaic power generation, plays an increasingly important role in the energy field. Distributed photovoltaic systems have been widely used in residential buildings, commercial buildings and industrial facilities due to their flexibility and adaptability. However, the output of photovoltaic power generation is affected by many factors, especially the dynamic changes in the total radiation flux of light, which plays a decisive role in the energy output of photovoltaic power stations. Therefore, accurate prediction of the output of photovoltaic power stations is of great significance for grid dispatching, energy management and improving the efficiency of photovoltaic systems. Traditional prediction methods often rely on simplified models and assumptions, and are difficult to adapt to complex and changeable actual environments. In recent years, the development of artificial intelligence and neural network learning technology has provided new possibilities for improving prediction accuracy.

[0003] The shortcomings of current research are mainly reflected in the following aspects: many existing photovoltaic output forecasting methods perform well under specific conditions, but when faced with complex and changeable meteorological conditions and highly random and highly volatile time series data, the prediction accuracy often drops significantly and lacks sufficient generalization ability. Many existing prediction models, especially those based on deep learning, can provide higher prediction accuracy, but their internal mechanisms are relatively complex, and it is difficult to explain the rationality of the prediction process and results, resulting in low transparency and credibility of the model. Some studies are not comprehensive enough in feature extraction and fail to fully utilize the spatial and temporal characteristics of time series data, resulting in insufficient robustness and accuracy of the prediction model. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: the existing photovoltaic output prediction method has insufficient generalization ability and poor model interpretability, and how to solve the optimization problem of incomplete feature extraction.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a photovoltaic output prediction method considering a light radiation flux calculation model, including constructing a solar operation model and analyzing photovoltaic power station parameters, collecting photovoltaic power generation data and preprocessing it; intelligently setting hyperparameters through a Bayesian method, extracting spatial and temporal features, constructing a Koopman-like operator and cost function and iteratively solving them; taking the total radiation flux prediction value as input, and predicting photovoltaic output through a convolutional neural network.

[0007] As a preferred solution of the photovoltaic output prediction method considering the light radiation flux calculation model described in the present invention, wherein: the construction of the solar operation model and the analysis of the photovoltaic power station parameters include the calculation model of the solar azimuth and altitude angle, collecting the layout array, position, orientation, and solar panel function parameters of the photovoltaic power station, combining the astronomical factors of the earth's rotation, revolution, and inclination, and calculating the change law of solar radiation intensity and solar angle within a day cycle. The solar azimuth angle is expressed as:

[0008]

[0009] Among them, h and m are the hour and minute of the 24-hour Beijing time system, θ t0 is the difference between local time and Beijing time, θ t is the solar azimuth angle, and the solar altitude angle is expressed as:

[0010] sin(θ h )=sin(φ)×sin(δ)+cos(φ)×cos(δ)×cos(θ t )

[0011] δ=0.006918-0.399912cos(b)+0.0070257sin(b)-0.006758cos(2b)+0.000907sin(2b)-0.002697cos(3b)+0.00148sin(3b)

[0012]

[0013] Among them, φ is the latitude of the photovoltaic power station, δ is the declination, N is the cumulative day, b is the angle parameter, θ h The solar azimuth and altitude angles are used as the features of the prediction model, and the data are preprocessed by the ICEEMDAN method.

[0014] As a preferred solution of the photovoltaic output prediction method considering the light radiation flux calculation model described in the present invention, wherein: the photovoltaic power generation data collection and preprocessing includes decomposition characteristics based on ICEEMDAN, through the differential effects of solar illumination at different times, ICEEMDAN with white noise added is selected to decompose the load data, and the original data is decomposed into a series of intrinsic mode function components with different frequency characteristics, each intrinsic mode function represents different frequency and amplitude characteristics, and the nonlinear characteristics of the multi-energy microgrid load are reduced. ICEEMDAN improves the original time series data and adds white noise to the initial signal. The addition of white noise and the iterative decomposition of ICEEMDAN generate intrinsic mode functions, each intrinsic mode function represents the characteristics of the original sequence at different frequencies, and the decomposition of ICEEMDAN is expressed as:

[0015]

[0016] Among them, y(t) is the original load sequence, y j (t) is the noise signal, r n (t) and IMF n (t) are the nth order residual component and modal component, E n (·) is the nth mode obtained by EMD algorithm, M(·) is the envelope calculation, η j (t) is the jth white noise added, α n-1 is the standard deviation of the noise removed.

[0017] As a preferred solution of the photovoltaic output prediction method considering the light radiation flux calculation model described in the present invention, wherein: the extraction of spatial and temporal features includes RNN and BiGRU, RNN is responsible for spatial feature extraction, and is used to process sequence data. There is a sequential relationship between data points. In photovoltaic data, spatial features are regarded as sequence data, and photovoltaic array data at different geographical locations are sequences arranged along a certain dimension. RNN memorizes past inputs through hidden states, connects the nodes of the hidden layer, and makes the output of the hidden layer at the previous moment superimposed on the output of the input layer as new input information, and predicts and adjusts the load in a real-time changing environment. BiGRU is responsible for temporal feature extraction, and the information flow of BiGRU is divided into forward propagation and backward propagation. The two GRU models in opposite directions of the BiGRU model extract the positive and negative bidirectional time series features of the power load to obtain the dependency relationship. The forward GRU calculation is expressed as:

[0018]

[0019] Among them, x t is the input at time t, h t is the output at time t, n t is the candidate hidden state at the current moment, r t is the reset gate, z t is the update gate, W xr , W hr , W xz , W hz , W hn , W xn is the weight coefficient matrix, b xr , b hr , b xz , b hz , b hn , b xn is the bias matrix, σ is the sigmoid function, is the dot product of two matrices.

[0020] As a preferred solution of the photovoltaic output prediction method considering the light radiation flux calculation model described in the present invention, wherein: the construction of the Koopman-like operator and the cost function and the iterative solution include approximating the Koopman-like operator based on the PCL framework and the autoencoder, describing the spatiotemporal state evolution trend of the photovoltaic data, and making predictions, the autoencoder deep neural network is composed of an encoder and a decoder, the encoder is composed of a one-dimensional convolutional neural network and a layer of gated recurrent unit neural network, the decoder is composed of a layer of gated recurrent unit neural network and a deconvolutional neural network, and the reconstruction loss of KAE is expressed as:

[0021]

[0022] Among them, l is the prediction step size for multi-step prediction, n is the total number of steps of l, and λ l is the original nonlinear data, k l For the reconstructed version, the Koopman-like operators A and B are introduced, and the reverse and forward dynamics are expressed as:

[0023]

[0024] In the formula, λ(τ) is the original data or feature of the system at time τ, k(τ-1) and k(τ+1) are the variable states of (τ-1) and (τ+1) respectively, and the action transformation χ between operator A and operator B is dec With the transformation χ enc In the multi-step prediction mode, after the operation of operators A and B, the cost function of forward prediction is expressed as:

[0025]

[0026] Among them, l+r is the forward prediction step size for multi-step prediction, and m is the total number of steps of r.

[0027] As a preferred solution of the photovoltaic output prediction method considering the light radiation flux calculation model described in the present invention, the construction of the Koopman-like operator and the cost function and the iterative solution also includes constraining operators A and B in the training process, performing high-order vector singular value decomposition on operators A and B, expressed as:

[0028] A=Q1A'P1

[0029] B=Q2B'P2

[0030] Among them, Q and P are reversible matrices. During the training process of operators A and B, correlation constraints are added to operators A and B. The constraint function when operators A and B are trained simultaneously is expressed as:

[0031]

[0032] in, is the first s rows of the matrix A', A' *s is the left s columns of the matrix A', is the first s rows of the matrix B', B' *s is the left s columns of matrix B', E s is a unit matrix of size s×s, which is the same size as A'. When predicting multiple variables at the same time, the cost function of the multi-step multivariate forward prediction task is expressed as:

[0033] ε4=aε1+bε2

[0034] When predicting a single variable from multiple variables, the cost function of the multi-step many-to-one prediction task is expressed as:

[0035]

[0036] Among them, α and b are the weight coefficients of the loss term, κ *i* is the i-th column of κ, λ *i* is the i-th column of λ, is the i-th column of κ at time lr, is the i-th column of λ at time lr, and the cost function of the single-step multivariate to univariate forward prediction is expressed as:

[0037]

[0038] In the multivariate versus univariate prediction scenario, when multivariate data is fed into the model, the model will arbitrarily extract multivariate features and focus on the features to be predicted.

[0039] As a preferred solution of the photovoltaic output prediction method considering the light radiation flux calculation model described in the present invention, wherein: the photovoltaic output prediction through the convolutional neural network includes taking the total radiation flux prediction value as input, and performing photovoltaic output prediction through the convolutional neural network. The distribution pattern of solar radiation has spatial characteristics in the image, and the convolutional neural network captures the spatial characteristics. In the convolution layer of the convolutional neural network, the input is convolved by a filter to obtain a feature map, and the maximum dimension reduction is performed through the pooling layer. By applying the convolution kernel to the input sequence, the information between different types of data is captured. With the sliding of the time window, the convolutional neural network model captures the dynamic changes of the sequence. Each weight w in W judges the weight of the influence on the final recognition result for different features, and multiplies it with the pixel x in X, and then sums it. The convolution operation is expressed as:

[0040]

[0041] Among them, k is the number of convolution kernels, represents the number of filters in the convolution layer, and W is an m×n matrix.

[0042] Another object of the present invention is to provide a photovoltaic output prediction system that takes into account the light radiation flux calculation model, which can extract the spatial and temporal characteristics of photovoltaic output data respectively using RNN and BiGRU through the extraction construction module, thereby solving the current problem of insufficient extraction of spatial and temporal characteristics of photovoltaic output data.

[0043] As a preferred solution of the photovoltaic output prediction system considering the light radiation flux calculation model described in the present invention, it includes: a collection preprocessing module, an extraction construction module, and a photovoltaic output prediction module; the collection preprocessing module is used to build a solar operation model and analyze the parameters of the photovoltaic power station, collect photovoltaic power generation data and preprocess it; the extraction construction module is used to intelligently set hyperparameters through the Bayesian method, extract spatial and temporal characteristics, construct a Koopman-like operator and cost function and iteratively solve them; the photovoltaic output prediction module is used to use the total radiation flux prediction value as input to predict the photovoltaic output through a convolutional neural network.

[0044] A computer device includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that the processor executes the computer program to implement a step of a photovoltaic output prediction method considering a light radiation flux calculation model.

[0045] 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 a photovoltaic output prediction method considering a light radiation flux calculation model are implemented.

[0046] Beneficial effects of the present invention: The photovoltaic output prediction method considering the light radiation flux calculation model provided by the present invention is adaptable to the changes in photovoltaic output in different regions and under different climatic conditions, and enhances the generalization ability and prediction accuracy of the model. By extracting key features from complex data, the adaptability of the model to new data is improved. By capturing the dynamic trend of photovoltaic output changes, the accuracy and stability of the prediction are improved. A physically constrained spatiotemporal neural network is constructed to enhance the interpretability. The spatial and temporal features are extracted by RNN and BiGRU respectively, and the robustness and accuracy of the prediction model are improved. The present invention achieves better results in generalization ability, prediction accuracy and interpretability. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0048] Figure 1An overall flow chart of a photovoltaic output prediction method considering a light radiation flux calculation model provided in the first embodiment of the present invention.

[0049] Figure 2 A diagram showing changes in the solar azimuth and altitude angle during a day for a photovoltaic output prediction method taking into account a light radiation flux calculation model provided in the first embodiment of the present invention.

[0050] Figure 3 A flow chart of a solution to a photovoltaic output prediction method taking into account a light radiation flux calculation model provided in the first embodiment of the present invention.

[0051] Figure 4 An overall module diagram of a photovoltaic output prediction system taking into account a light radiation flux calculation model provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0053] Example 1

[0054] Reference Figure 1-Figure 3 , which is an embodiment of the present invention, provides a photovoltaic output prediction method considering a light radiation flux calculation model, including:

[0055] S1: Build a solar operation model and analyze the parameters of the photovoltaic power station, collect photovoltaic power generation data and pre-process it.

[0056] Furthermore, a solar operation model is constructed and the photovoltaic power station parameters are analyzed, including calculation models of the solar azimuth and altitude angles.

[0057] It should be noted that the array layout, location, orientation, and panel function parameters of the photovoltaic power station are collected, and combined with the astronomical factors of the earth's rotation, revolution, and inclination, the changing pattern of solar radiation intensity and solar angle in a day cycle is calculated. The solar azimuth angle is expressed as:

[0058]

[0059] Among them, h and m are the hour and minute of the 24-hour Beijing time system, θ t0 is the difference between local time and Beijing time, θ t is the solar azimuth angle, and the solar altitude angle is expressed as:

[0060] sin(θ h )=sin(φ)×sin(δ)+cos(φ)×cos(δ)×cos(θ t )

[0061] δ=0.006918-0.399912cos(b)+0.0070257sin(b)-0.006758cos(2b)+0.000907sin(2b)-0.002697cos(3b)+0.00148sin(3b)

[0062]

[0063] Among them, φ is the latitude of the photovoltaic power station, δ is the declination, N is the cumulative day, b is the angle parameter, θ h The solar azimuth and altitude angles are used as the features of the prediction model, and the data are preprocessed by the ICEEMDAN method.

[0064] It should also be noted that the luminous power of photovoltaic panels is expressed as:

[0065] P=H·A·η·K

[0066] Among them, P is the luminous power of the photovoltaic panel, H is the total radiation flux of light, A is the area of ​​the photovoltaic panel, η is the efficiency of the photovoltaic panel, and K is the correction factor. The total radiation flux of light refers to the total amount of radiation energy passing through a certain cross section or irradiating to a certain surface in unit time, and the unit is kilowatt-hour per square meter. The total radiation flux of light directly affects the ability of photovoltaic cells to generate current. The higher the irradiance, the greater the power generation power of the photovoltaic panel. The area of ​​the photovoltaic panel refers to the effective area of ​​the solar panel, which is used to receive sunlight and convert it into electrical energy. The unit is square meter. The efficiency of the photovoltaic panel refers to the ratio of the maximum output power of the photovoltaic panel to the incident light power under standard test conditions. The standard test conditions usually refer to the ambient temperature of 25°C and the irradiance of 1000W / m 2 , which is related to the quality of the product. The correction coefficient is a comprehensive coefficient used to compensate for various losses and undesirable factors in actual operation, such as dust shielding, temperature rise, line loss, inverter efficiency, etc. In the CNN convolution process, the accuracy of power generation prediction is improved mainly by compensating for the correction coefficient. The correction coefficient can more truly reflect the power generation efficiency of photovoltaic panels in actual applications. Common correction coefficients include the attenuation coefficient of long-term operation of components, power loss correction coefficient caused by dust shielding and temperature rise, line correction coefficient, inverter efficiency, and orientation and tilt angle correction coefficient.

[0067] It should also be noted that the power generation of photovoltaic panels is mainly determined by the total radiation flux of the photovoltaic panels and the angle of the sun. The azimuth and altitude angle of the sun change over time in 24 hours. Figure 2 As shown, the azimuth is the angle between the projection of sunlight on the horizontal plane and the south direction, and the altitude is the angle between sunlight and the horizontal plane. The incident angle of sunlight can be described by the azimuth and altitude. The azimuth is 0° at 12:00, and increases or decreases by 15° for every increase or decrease of 1 hour. It is -90° at 6:00 and +90° at 18:00.

[0068] It should also be noted that for a specific power station or power station group, its latitude and time zone difference are fixed values. It can be proved that even if there is a certain error in this value or when an arbitrary assumed value is used, its fixed error can also be regarded as an implicit parameter of the prediction model and still does not affect the final prediction accuracy. The solar declination can be directly calculated from the sampling time.

[0069] Furthermore, PV power generation data are collected and preprocessed including decomposition characteristics based on ICEEMDAN.

[0070] It should be noted that, through the different effects of sunlight at different times, ICEEMDAN with white noise is selected to decompose the load data, and the original data is decomposed into a series of intrinsic mode function components with different frequency characteristics. Each intrinsic mode function represents different frequency and amplitude characteristics, and the nonlinear characteristics of the multi-energy microgrid load are eliminated. ICEEMDAN improves the original time series data and adds white noise to the initial signal. The addition of white noise and the iterative decomposition of ICEEMDAN generate intrinsic mode functions. Each intrinsic mode function represents the characteristics of the original sequence at different frequencies. The decomposition of ICEEMDAN is expressed as:

[0071]

[0072] Among them, y(t) is the original load sequence, y j (t) is the noise signal, r n (t) and IMF n (t) are the nth order residual component and modal component, E n (·) is the nth mode obtained by EMD algorithm, M(·) is the envelope calculation, η j (t) is the jth white noise added, α n-1 is the standard deviation of the noise removed.

[0073] It should also be noted that the choice of using ICEEMDAN with white noise added to decompose the load data not only reduces the interference caused by frequency aliasing and pseudo-modes of the decomposed data, ensures the independence of light radiation at different times, but also retains the correlation and continuity between adjacent data.

[0074] S2: Intelligently set hyperparameters through Bayesian methods, extract spatial and temporal features, construct Koopman-like operators and cost functions, and iteratively solve them.

[0075] Furthermore, extracting spatial and temporal features involves RNN and BiGRU.

[0076] It should be noted that RNN is responsible for spatial feature extraction and is used to process sequence data. There is a sequential relationship between data points. In photovoltaic data, spatial features are regarded as sequence data. Photovoltaic array data at different geographical locations are sequences arranged along a certain dimension. RNN memorizes past inputs through hidden states, connects the nodes of the hidden layer, and makes the output of the hidden layer at the previous moment superimposed on the output of the input layer as new input information to predict and adjust the load in a real-time changing environment. BiGRU is responsible for temporal feature extraction. The information flow of BiGRU is divided into forward propagation and backward propagation. The two GRU models in opposite directions of the BiGRU model extract the positive and negative bidirectional time series features of the power load and obtain the dependency relationship. The forward GRU calculation is expressed as:

[0077]

[0078] Among them, x t is the input at time t, h t is the output at time t, n t is the candidate hidden state at the current moment, r t is the reset gate, z t is the update gate, W xr , W hr , W xz , W hz , W hn , W xn is the weight coefficient matrix, b xr 、b hr 、b xz 、b hz 、b hn 、b xn is the bias matrix, σ is the sigmoid function, is the dot product of two matrices.

[0079] It should also be noted that photovoltaic arrays in adjacent geographical locations may have similar output powers. RNN can capture this spatial correlation through its state transfer mechanism. The time series characteristics of photovoltaic data are very obvious, such as changes in light intensity during the day, seasonal changes, etc. BiGRU can effectively capture these time dependencies, thereby extracting time features more accurately.

[0080] Furthermore, a Koopman-like operator and a cost function are constructed and iteratively solved, including approximating the Koopman-like operator based on a PCL framework and an autoencoder.

[0081] It should be noted that the autoencoder deep neural network describes the temporal and spatial state evolution trend of photovoltaic data and makes predictions. The encoder consists of an encoder and a decoder. The encoder consists of a one-dimensional convolutional neural network and a layer of gated recurrent unit neural network. The decoder consists of a layer of gated recurrent unit neural network and a deconvolutional neural network. The reconstruction loss of KAE is expressed as:

[0082]

[0083] Among them, l is the prediction step size for multi-step prediction, n is the total number of steps of l, and λ l is the original nonlinear data, k l For the reconstructed version, the Koopman-like operators A and B are introduced, and the reverse and forward dynamics are expressed as:

[0084]

[0085] In the formula, λ(τ) is the original data or feature of the system at time τ, k(τ-1) and k(τ+1) are the variable states of (τ-1) and (τ+1) respectively, and the action transformation χ between operator A and operator B is dec With the transformation χ enc In the multi-step prediction mode, after the operation of operators A and B, the cost function of forward prediction is expressed as:

[0086]

[0087] Among them, l+r is the forward prediction step size for multi-step prediction, and m is the total number of steps of r.

[0088] Furthermore, constructing a Koopman-like operator and a cost function and solving them iteratively also includes constraining operators A and B during the training process.

[0089] It should be noted that the singular value decomposition of the high-order vectors of operators A and B is expressed as:

[0090] A=Q1A'P1

[0091] B=Q2B'P2

[0092] Among them, Q and P are reversible matrices. During the training process of operators A and B, correlation constraints are added to operators A and B. The constraint function when operators A and B are trained simultaneously is expressed as:

[0093]

[0094] in, is the first s rows of the matrix A', A'*s is the left s columns of the matrix A', is the first s rows of the matrix B', B' *s is the left s columns of matrix B', E s is a unit matrix of size s×s, which is the same size as A'. When predicting multiple variables at the same time, the cost function of the multi-step multivariate forward prediction task is expressed as:

[0095] ε4=aε1+bε2

[0096] When predicting a single variable from multiple variables, the cost function of the multi-step many-to-one prediction task is expressed as:

[0097]

[0098] Among them, a and b are the weight coefficients of the loss term, is the i-th column of κ, is the i-th column of λ, is the i-th column of κ at time lr, is the i-th column of λ at time lr, and the cost function of the single-step multivariate to univariate forward prediction is expressed as:

[0099]

[0100] In the multivariate versus univariate prediction scenario, when multivariate data is fed into the model, the model will arbitrarily extract multivariate features and focus on the features to be predicted.

[0101] It should also be noted that multi-step predictions are usually prone to the superposition of step-by-step errors and are difficult to achieve optimal accuracy. Single-step predictions are often used in practical applications.

[0102] It should also be noted that Bayesian optimization is an algorithm that estimates the maximum value of a function based on existing sampling points when the function equation is unknown. The Bayesian algorithm optimizes multiple hyperparameters such as RNN and BiGRU, thereby improving the effectiveness and accuracy of photovoltaic power generation prediction. The Bayesian optimization algorithm is for several existing sample points. It calculates the posterior probability distribution of the previous n points through Gaussian process regression to obtain the expected mean and variance of each hyperparameter at each value point. The larger the mean, the greater the impact of this point will make the final indicator of the model. The variance represents the uncertainty of the effect of this point. The larger the variance, the less certain it is that this point may obtain a value that meets the conditions. Let X = x1, x2, ..., xn be a set of hyperparameter combinations, and use f(x) to represent the objective function of the hyperparameter x. Bayesian optimization finds x∈X such that:

[0103] x * = arg max f(x), x∈X

[0104] Among them, x * is the global optimal value of the constraint domain of f(x), including real numbers, integers, or categorical feature values.

[0105] S3: Taking the total radiation flux prediction value as input, the photovoltaic output is predicted through the convolutional neural network.

[0106] Furthermore, PV output prediction via convolutional neural networks includes taking the predicted total radiation flux as input.

[0107] It should be noted that the photovoltaic output is predicted by convolutional neural network. The distribution pattern of solar radiation has spatial characteristics in the image. The convolutional neural network captures the spatial characteristics. In the convolutional layer of the convolutional neural network, the input is convolved by the filter to obtain the feature map. The maximum dimension reduction is performed through the pooling layer. By applying the convolution kernel to the input sequence, the information between different types of data is captured. As the time window slides, the convolutional neural network model captures the dynamic changes of the sequence. Each weight w in W is used for different features. The influence of the final recognition result is judged by the weight, and multiplied by the pixel x in X, and then summed. The convolution operation is expressed as:

[0108]

[0109] Among them, k is the number of convolution kernels, represents the number of filters in the convolution layer, and W is an m×n matrix.

[0110] Example 2

[0111] An embodiment of the present invention provides a photovoltaic output prediction method considering a light radiation flux calculation model. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0112] After screening outliers for the photovoltaic power generation and related data actually collected, the above data sets are respectively input into the algorithm of the present invention and the classical artificial intelligence algorithm for training, and the prediction error is evaluated using the test set data after the training is completed.

[0113] As shown in Table 1, in the single-step prediction mode, the prediction performance of the KAE model is better than that of most traditional models. The table shows the comparison of prediction errors of various models. Among them, the prediction accuracy of the KAE model is better than that of most traditional models. The IMSE value of the KAE model is 0.0638, while the IMSE values ​​of other models such as CNN, LSTM, GRU and Simple-RNN are all higher than 0.0751. This shows that in the single-step prediction mode, the KAE model has a more ideal error control while maintaining a high prediction accuracy. The IRMSE values ​​of each model are generally lower than its IMSE value. This is reflected in the fact that when considering the relative error, the KAE model also shows a high prediction accuracy. The IRMSE value of the KAE model is 0.1483, while the IMSE value is 0.0638, showing that it has a strong control ability in relative error. Considering the two indicators of IMSE and IRMSE, the KAE model performs well in both dimensions, which means that the model not only has an advantage in absolute error, but also has a good control ability in relative error. This comprehensive performance improvement helps to improve the accuracy and reliability of the overall prediction.

[0114] Table 1 Prediction error table of various models

[0115] Model IMSE IRMSE IMAE KAE 0.0638 0.0671 0.0471 CNN 0.0797 0.0786 0.0794 LSTM 0.0775 0.0967 0.0785 GRU 0.0726 0.0958 0.0794 Simple-RNN 0.0751 0.1402 0.1215 KNN 0.1082 0.1483 0.1352

[0116] Example 3

[0117] Reference Figure 4 , which is an embodiment of the present invention, provides a photovoltaic output prediction system considering a light radiation flux calculation model, including: an acquisition preprocessing module, an extraction construction module, and a photovoltaic output prediction module.

[0118] The acquisition and preprocessing module is used to build a solar operation model and analyze the parameters of the photovoltaic power station, collect photovoltaic power generation data and preprocess it; the extraction and construction module is used to intelligently set hyperparameters through the Bayesian method, extract spatial and temporal characteristics, construct a Koopman-like operator and cost function and iteratively solve them; the photovoltaic output prediction module is used to take the total radiation flux prediction value as input and predict the photovoltaic output through a convolutional neural network.

[0119] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0120] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0121] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0122] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0123] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A photovoltaic output prediction method considering a light radiation flux calculation model, characterized in that: include: Construct solar operation model and analyze photovoltaic power station parameters, collect photovoltaic power generation data and pre-process; The Bayesian method is used to intelligently set hyperparameters, extract spatial and temporal features, construct Koopman-like operators and cost functions, and solve them iteratively. The total radiation flux prediction value is taken as input and photovoltaic output is predicted through convolutional neural network.

2. The photovoltaic output prediction method considering the light radiation flux calculation model as claimed in claim 1, characterized in that: The solar operation model is constructed and the parameters of the photovoltaic power station are analyzed, including the calculation model of the solar azimuth and altitude angle, collecting the layout array, position, orientation, and functional parameters of the photovoltaic power station, and combining the astronomical factors of the earth's rotation, revolution, and inclination to calculate the variation law of solar radiation intensity and solar angle within a day cycle. The solar azimuth angle is expressed as: Among them, h and m are the hour and minute of the 24-hour Beijing time system, θ t0 is the difference between local time and Beijing time, θ t is the solar azimuth angle, and the solar altitude angle is expressed as: sin(θ h )=sin(φ)×sin(δ)+cos(φ)×cos(δ)×cos(θ t ) δ=0.006918-0.399912cos(b)+0.0070257sin(b)-0.006758cos(2b)+0.000907sin(2b)-0.002697cos(3b)+0.00148sin(3b) Among them, φ is the latitude of the photovoltaic power station, δ is the declination, N is the cumulative day, b is the angle parameter, θ h The solar azimuth and altitude angles are used as the features of the prediction model, and the data are preprocessed by the ICEEMDAN method.

3. The photovoltaic output prediction method considering the light radiation flux calculation model as claimed in claim 2, characterized in that: The photovoltaic power generation data collection and preprocessing includes decomposing the load data based on the decomposition characteristics of ICEEMDAN, through the difference of sunlight at different times, and selecting ICEEMDAN with white noise added to decompose the load data, decomposing the original data into a series of intrinsic mode function components with different frequency characteristics, each intrinsic mode function represents different frequency and amplitude characteristics, and reduces the nonlinear characteristics of the multi-energy microgrid load. ICEEMDAN improves the original time series data and adds white noise to the initial signal. The addition of white noise and the iterative decomposition of ICEEMDAN generate intrinsic mode functions. Each intrinsic mode function represents the characteristics of the original sequence at different frequencies. The decomposition of ICEEMDAN is expressed as: Among them, y(t) is the original load sequence, y j (t) is the noise signal, r n (t) and IMF n (t) are the nth order residual component and modal component, E n (·) is the nth mode obtained by EMD algorithm, M(·) is the envelope calculation, η j (t) is the jth white noise added, α n-1 is the standard deviation of the noise removed.

4. The photovoltaic output prediction method considering the light radiation flux calculation model as claimed in claim 3 is characterized by: The extraction of spatial and temporal features includes RNN and BiGRU. RNN is responsible for spatial feature extraction and is used to process sequence data. There is a sequential relationship between data points. In photovoltaic data, spatial features are regarded as sequence data. Photovoltaic array data at different geographical locations are sequences arranged along a certain dimension. RNN memorizes past inputs through hidden states, connects the nodes of the hidden layer, and makes the output of the hidden layer at the previous moment superimposed on the output of the input layer as new input information, so as to predict and adjust the load in a real-time changing environment. BiGRU is responsible for temporal feature extraction. The information flow of BiGRU is divided into forward propagation and backward propagation. The two GRU models in opposite directions of the BiGRU model extract the positive and negative bidirectional time series features of the power load to obtain the dependency relationship. The forward GRU calculation is expressed as: Among them, x t is the input at time t, h t is the output at time t, n t is the candidate hidden state at the current moment, r t is the reset gate, z t is the update gate, W xr , W hr , W xz , W hz , W hn , W xn is the weight coefficient matrix, b xr , b hr , b xz , b hz , b hn , b xn is the bias matrix, σ is the sigmoid function, and ° is the dot product of the two matrices.

5. The photovoltaic output prediction method considering the light radiation flux calculation model as claimed in claim 4, characterized in that: The construction of the Koopman-like operator and the cost function and the iterative solution include approximating the Koopman-like operator based on the PCL framework and the autoencoder, describing the spatiotemporal state evolution trend of the photovoltaic data, and making predictions. The autoencoder deep neural network is composed of an encoder and a decoder. The encoder is composed of a one-dimensional convolutional neural network and a layer of gated recurrent unit neural network. The decoder is composed of a layer of gated recurrent unit neural network and a deconvolutional neural network. The reconstruction loss of KAE is expressed as: Among them, l is the prediction step size for multi-step prediction, n is the total number of steps of k, and λ l is the original nonlinear data, k l For the reconstructed version, the Koopman-like operators A and B are introduced, and the reverse and forward dynamics are expressed as: In the formula, λ(τ) is the original data or feature of the system at time τ, k(τ-1) and k(τ+1) are the variable states of (τ-1) and (τ+1) respectively, and the transformation χ between operator A and operator B is: dec With the transformation χ enc In the multi-step prediction mode, after the operation of operators A and B, the cost function of forward prediction is expressed as: Among them, l+r is the forward prediction step size for multi-step prediction, and m is the total number of steps of r.

6. The photovoltaic output prediction method considering the light radiation flux calculation model as claimed in claim 5, characterized in that: The method of constructing a Koopman-like operator and a cost function and iteratively solving the same also includes constraining operators A and B in the training process and performing singular value decomposition of high-order vectors of operators A and B, which is expressed as: A=Q1A'P1 B=Q2B'P2 Among them, Q and P are reversible matrices. During the training process of operators A and B, correlation constraints are added to operators A and B. The constraint function when operators A and B are trained simultaneously is expressed as: Among them, A” s* is the first s rows of the matrix A', A' *s is the left s columns of matrix A', B' s* is the first s rows of the matrix B', B' *s is the left s columns of matrix B', E s is a unit matrix of size s×s, which is the same size as A'. When predicting multiple variables at the same time, the cost function of the multi-step multivariate forward prediction task is expressed as: ε4=aε1+bε2 When predicting a single variable from multiple variables, the cost function of the multi-step many-to-one prediction task is expressed as: Among them, a and b are the weight coefficients of the loss term, κ *i* is the i-th column of κ, λ *i* is the i-th column of λ, κ l-r|*i* is the i-th column of κ at time lr, λ l-r|*i* is the i-th column of λ at time lr, and the cost function of the single-step multivariate to univariate forward prediction is expressed as: In the multivariate versus univariate prediction scenario, when multivariate data is fed into the model, the model will arbitrarily extract multivariate features and focus on the features to be predicted.

7. The photovoltaic output prediction method considering the light radiation flux calculation model as claimed in claim 6, characterized in that: The photovoltaic output prediction by convolutional neural network includes taking the total radiation flux prediction value as input, and predicting the photovoltaic output by convolutional neural network. The distribution pattern of solar radiation has spatial characteristics in the image, and the convolutional neural network captures the spatial characteristics. In the convolution layer of the convolutional neural network, the input is convolved by the filter to obtain a feature map, and the maximum dimension reduction is performed through the pooling layer. By applying the convolution kernel to the input sequence, the information between different types of data is captured. With the sliding of the time window, the convolutional neural network model captures the dynamic changes of the sequence. Each weight w in W judges the weight of the influence on the final recognition result for different features, and multiplies it with the pixel x in X, and then sums it. The convolution operation is expressed as: Among them, k is the number of convolution kernels, represents the number of filters in the convolution layer, and W is an m×n matrix.

8. A system using a photovoltaic output prediction method taking into account a light radiation flux calculation model as described in any one of claims 1 to 7, characterized in that: It includes acquisition preprocessing module, extraction construction module and photovoltaic output prediction module; The acquisition and preprocessing module is used to build a solar operation model and analyze photovoltaic power station parameters, collect photovoltaic power generation data and preprocess; The extraction building module is used to intelligently set hyperparameters through the Bayesian method, extract spatial and temporal features, construct a Koopman-like operator and a cost function, and iteratively solve them; The photovoltaic output prediction module is used to take the total radiation flux prediction value as input and perform photovoltaic output prediction through a convolutional neural network.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the photovoltaic output prediction method considering the light radiation flux calculation model described in any one of claims 1 to 7 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 output prediction method considering the light radiation flux calculation model described in any one of claims 1 to 7 are implemented.

Images