Photovoltaic power prediction method, device, equipment, medium and product

Abstract

The invention discloses a photovoltaic power prediction method, device and equipment, a medium and a product. The method comprises the following steps: acquiring photovoltaic power data of a photovoltaic power generation system, and meteorological data and satellite remote sensing data of an area where the photovoltaic power generation system is located; determining multi-dimensional feature data according to the photovoltaic power data, the meteorological data and the satellite remote sensing data; processing the multi-dimensional feature data based on a multi-level dynamic system and the dynamic feature weight, and determining weighted feature data; and generating photovoltaic power prediction data based on the weighted feature data and a pre-constructed multi-channel nested neural network model. Data from different sources are combined to be analyzed through a multi-level dynamic system, deep mining and accurate prediction of photovoltaic power dynamic characteristics are achieved, and the problem that in the prior art, signal extraction is not sufficient is solved. Through the multi-path nested neural network, fusion of cross-level features is realized, the precision of the prediction model under variable weather conditions is effectively improved, and the prediction precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power prediction, and in particular to a photovoltaic power prediction method, device, equipment, medium and product. Background Art

[0002] Ultra-short-term prediction of photovoltaic power is crucial for optimizing the dispatching and management of photovoltaic power generation and improving the stability of the power system. As the installed capacity of photovoltaic power generation increases year by year, its intermittent and volatile nature poses a challenge to the smooth operation of the power system. In order to ensure the maximum utilization of photovoltaic power generation and the balance of the power grid, accurate ultra-short-term photovoltaic power prediction has become an indispensable tool. This prediction usually focuses on the power output in the next 0 to 4 hours, and its accuracy directly affects the operating efficiency and economy of the power grid.

[0003] Existing ultra-short-term photovoltaic power prediction technologies usually use multi-mode decomposition and multi-branch input methods. For example, historical photovoltaic power data and related meteorological factor data are preprocessed, and then the data is converted from the time domain to the frequency domain using fast Fourier transform, and the data is classified in combination with the adaptive neighbor propagation clustering algorithm. After that, the multi-mode decomposition algorithm is used to process the data and a multi-branch input network prediction model is constructed.

[0004] However, these methods have several defects: although multi-mode decomposition can reduce the complexity of data, its extraction of complex signals is still not comprehensive enough and cannot fully capture the multi-dimensional characteristics in the data; the existing technology has low prediction accuracy when dealing with the highly random and chaotic nature of meteorological factors, especially under extreme weather conditions. Summary of the invention

[0005] The present invention provides a photovoltaic power prediction method, device, equipment, medium and product to achieve accurate prediction of photovoltaic power.

[0006] According to a first aspect of the present invention, there is provided a photovoltaic power prediction method, comprising:

[0007] Acquiring photovoltaic power data of the photovoltaic power generation system, meteorological data of the area where the photovoltaic power generation system is located, and satellite remote sensing data;

[0008] Determining multidimensional feature data according to the photovoltaic power data, the meteorological data and the satellite remote sensing data;

[0009] Processing the multi-dimensional feature data based on a multi-level dynamic system and dynamic feature weights to determine weighted feature data;

[0010] Photovoltaic power prediction data is generated based on the weighted feature data and the pre-built multi-way nested neural network model.

[0011] According to a second aspect of the present invention, there is provided a photovoltaic power prediction device, comprising:

[0012] A data acquisition module, used to obtain photovoltaic power data of the photovoltaic power generation system, meteorological data of the area where the photovoltaic power generation system is located, and satellite remote sensing data;

[0013] A data determination module, used to determine multi-dimensional feature data according to the photovoltaic power data, the meteorological data and the satellite remote sensing data;

[0014] A feature mapping module, for processing the multi-dimensional feature data based on a multi-level dynamic system and dynamic feature weights to determine weighted feature data;

[0015] The power prediction module is used to generate photovoltaic power prediction data based on the weighted feature data and a pre-built multi-way nested neural network model.

[0016] According to a third aspect of the present invention, there is provided an electronic device, the electronic device comprising:

[0017] at least one processor; and

[0018] a memory communicatively connected to the at least one processor; wherein,

[0019] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the photovoltaic power prediction method described in any embodiment of the present invention.

[0020] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the photovoltaic power prediction method described in any embodiment of the present invention when executed.

[0021] According to a fifth aspect of the present invention, an embodiment of the present invention further provides a computer program product, the computer program product comprising a computer program, and when the computer program is executed by a processor, the photovoltaic power prediction method of any embodiment of the present invention is implemented.

[0022] The technical solution of the embodiment of the present invention obtains the photovoltaic power data of the photovoltaic power generation system, the meteorological data of the area where the photovoltaic power generation system is located, and the satellite remote sensing data; determines the multidimensional feature data based on the photovoltaic power data, meteorological data, and satellite remote sensing data; processes the multidimensional feature data based on the multi-level dynamic system and the dynamic feature weights to determine the weighted feature data; generates photovoltaic power prediction data based on the weighted feature data and the pre-built multi-path nested neural network model. By combining data from different sources for analysis with a multi-level dynamic system, deep mining and accurate prediction of the dynamic characteristics of photovoltaic power are achieved, solving the problem of insufficient signal extraction in the prior art. Through the multi-path nested neural network, the fusion of cross-level features is achieved, which effectively improves the accuracy of the prediction model under changeable weather conditions and improves the prediction accuracy.

[0023] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of 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 work.

[0025] Figure 1 is a flow chart of a photovoltaic power prediction method provided according to Embodiment 1 of the present invention;

[0026] Figure 2 is a flow chart of a photovoltaic power prediction method provided according to Embodiment 2 of the present invention;

[0027] Figure 3 is an example diagram of a multi-level dynamic system in a photovoltaic power prediction method provided in Embodiment 2 of the present invention;

[0028] Figure 4 This is an example diagram of a multi-path nested neural network model in a photovoltaic power prediction method provided in Embodiment 2 of the present invention;

[0029] Figure 5 is a schematic diagram of the structure of a photovoltaic power prediction device provided according to Embodiment 3 of the present invention;

[0030] Figure 6 It is a schematic diagram of the structure of an electronic device implementing an embodiment of the present invention. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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 should fall within the scope of protection of the present invention.

[0032] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0033] Embodiment 1

[0034] Figure 1 A flowchart of a photovoltaic power prediction method is provided for the first embodiment of the present invention. This embodiment is applicable to the prediction of photovoltaic power. The method can be performed by a photovoltaic power prediction device. The photovoltaic power prediction device can be implemented in the form of hardware and / or software. The photovoltaic power prediction device can be configured in an electronic device. Figure 1 As shown, the method includes:

[0035] S110, obtaining photovoltaic power data of the photovoltaic power generation system, meteorological data of the area where the photovoltaic power generation system is located, and satellite remote sensing data.

[0036] In this embodiment, the photovoltaic power generation system can be understood as a system that generates electricity in the form of photovoltaics. Photovoltaic power data can be understood as the output power of the photovoltaic power generation system in a specific period of time. Photovoltaic power data directly reflects the output power of the photovoltaic power generation system in a specific period of time and is the key data for predicting future photovoltaic power generation. Meteorological data can be understood as environmental parameters of the area where the photovoltaic power generation system is located, such as solar radiation, temperature, humidity, wind speed and other environmental parameters, which have a significant impact on the output power of the photovoltaic power generation system. Satellite remote sensing data is used to provide a large range of continuous ground observation information, which can supplement the deficiencies of ground meteorological observations, especially in terms of spatial resolution and coverage. Satellite remote sensing data can provide more comprehensive meteorological and environmental information, thereby improving the spatial accuracy of the prediction.

[0037] Specifically, the processor can obtain photovoltaic power data collected by sensors of the photovoltaic power generation system by setting time periods, obtain meteorological data of the area where the photovoltaic power generation system is located through weather-related apps, and obtain satellite remote sensing data of the area where the photovoltaic power generation system is located through related platforms.

[0038] S120. Determine multi-dimensional feature data based on photovoltaic power data, meteorological data, and satellite remote sensing data.

[0039] In this embodiment, multi-dimensional feature data can be understood as feature data extracted from different dimensions.

[0040] Specifically, the processor can clean and preprocess photovoltaic power data, meteorological data and satellite remote sensing data to remove noise and outliers, and then extract core features of the processed data and standardize them to form standardized multi-dimensional feature data.

[0041] S130, processing the multi-dimensional feature data based on the multi-level dynamic system and the dynamic feature weights to determine the weighted feature data.

[0042] In this embodiment, the multi-level dynamic system can be understood as consisting of multiple subsystems with different characteristic behaviors. The dynamic feature weight can be understood as the dynamically updated weight used to weight the features of different dimensions. The weighted feature data can be understood as the feature data set after preliminary processing and weight adjustment, which contains the dynamic change characteristics of the photovoltaic power time series and related meteorological conditions.

[0043] Specifically, the multi-dimensional feature data includes multi-dimensional information obtained through preliminary processing and feature extraction, such as the time series characteristics of photovoltaic power and changes in meteorological conditions. The processor can map the multi-dimensional feature data to a multi-level dynamic system, analyze the dynamic behavior of photovoltaic power at different levels, and transform the time series data into trajectories in the phase space through phase space reconstruction. These trajectories reflect the dynamic characteristics of each dynamic system, obtain the reconstructed features, and weight the reconstructed features with dynamic feature weights to determine the weighted feature data.

[0044] S140, generating photovoltaic power prediction data based on weighted feature data and a pre-built multi-way nested neural network model.

[0045] In this embodiment, the multi-path nested neural network model can be understood as a model formed by nesting neural networks of multiple parallel paths. The photovoltaic power prediction data can be understood as the predicted photovoltaic power data.

[0046] Specifically, the processor can input the weighted feature data into a multi-path nested neural network model according to different dimensions. The weighted feature data of each dimension is processed through multiple parallel paths and nested neural networks to determine the high-order feature data in each dimension. The high-order feature data in each dimension is then fused to obtain fused data and generate photovoltaic power prediction data through adaptive learning models and other methods.

[0047] The technical solution of the embodiment of the present invention obtains the photovoltaic power data of the photovoltaic power generation system, the meteorological data of the area where the photovoltaic power generation system is located, and the satellite remote sensing data; determines the multidimensional feature data based on the photovoltaic power data, meteorological data, and satellite remote sensing data; processes the multidimensional feature data based on the multi-level dynamic system and the dynamic feature weights to determine the weighted feature data; generates photovoltaic power prediction data based on the weighted feature data and the pre-built multi-path nested neural network model. By combining data from different sources for analysis with a multi-level dynamic system, deep mining and accurate prediction of the dynamic characteristics of photovoltaic power are achieved, solving the problem of insufficient signal extraction in the prior art. Through the multi-path nested neural network, the fusion of cross-level features is achieved, which effectively improves the accuracy of the prediction model under changeable weather conditions and improves the prediction accuracy.

[0048] In the prediction process, the processor first generates photovoltaic power prediction data through a multi-way nested neural network model and an adaptive ensemble learning model. The photovoltaic power prediction data represents the model's estimate of future photovoltaic power. However, due to the complexity and unpredictability of environmental conditions (such as weather and temperature), the prediction results will inevitably differ from the actual observed data. Therefore, in order to improve the prediction accuracy, the processor must effectively manage this difference.

[0049] As a first optional embodiment of the first embodiment, after generating photovoltaic power prediction data based on weighted feature data and a pre-built multi-way nested neural network model, the method further includes:

[0050] The actual observation data is obtained, and the optimization parameters are determined through the meta-learning algorithm based on the actual observation data and photovoltaic power prediction data.

[0051] In this embodiment, the actual observation data can be understood as the actual photovoltaic power data collected by the photovoltaic power generation system corresponding to the photovoltaic power prediction data time period. The meta-learning algorithm can be understood as a method for optimizing the algorithm itself to better adapt to new tasks. The optimization parameters can be understood as a machine learning technique for optimizing model parameters, which can regulate and optimize between learning models at different levels. The optimization parameters can include parameters included in a multi-way nested neural network model, an adaptive ensemble learning model, and a multi-level dynamic system model.

[0052] Specifically, the processor can obtain the actual observation data collected by the photovoltaic power generation system after experiencing the time corresponding to the prediction data. The processor can compare the integrated prediction data with the actual observation data to generate error feedback data. For example, the error feedback data can be calculated by comparing the current photovoltaic power prediction data with the actual photovoltaic power to calculate the error value (for example, mean square error or absolute error). This error value is the error feedback data, which not only reflects the current prediction accuracy of the model, but also provides a key basis for subsequent model optimization. The parameters of the multi-path nested neural network model, the adaptive ensemble learning model, and the multi-level dynamic system mapping and nonlinear adaptive regression model are optimized by the meta-learning method, and the optimized parameters are applied to the above models and steps to form a dynamic feedback closed loop, and the photovoltaic power ultra-short-term prediction process is continuously optimized. Among them, the step of generating the error feedback data includes comparing the integrated prediction data with the actual observation data, calculating the error between the two, and generating the error feedback data, and the error feedback data is used to optimize the parameters of the multi-path nested neural network model, the adaptive ensemble learning model, and the multi-level dynamic system mapping and nonlinear adaptive regression model by the meta-learning method. Through the meta-learning algorithm, the processor can optimize and adjust the parameters of multiple core models (including multi-path nested neural network models, adaptive ensemble learning models, and multi-level dynamic system mapping and nonlinear adaptive regression models) according to error feedback data.

[0053] For example, if the processor identifies that a certain feature contributes more to the prediction error under the current environmental conditions, the meta-learning method will increase the weight of the feature in the corresponding model so that it plays a greater role in the next prediction. On the contrary, if a certain feature has a small impact, its weight may be reduced or even temporarily ignored. Through such adjustments, the model can better adapt to changes in the actual environment, thereby gradually reducing the prediction error and improving the overall prediction effect. For example, the system predicts the photovoltaic power on a sunny morning, but the actual photovoltaic power is affected by sudden cloud cover, resulting in a large error between the predicted value and the actual value. The system calculates the error value by comparing the integrated prediction data with the actual observation data, and generates error feedback data. Subsequently, the meta-learning method identifies that the fluctuation of light intensity caused by cloud changes is the main source of error. The system then optimizes the feature weights related to the cloud layer and adjusts the parameter settings of the adaptive ensemble learning model and the multi-path nested neural network model. In the next round of prediction, the optimized model more accurately reflects the impact of cloud cover on photovoltaic power, significantly reducing the prediction error.

[0054] In the first optional embodiment of the first embodiment, the model parameters are optimized through a meta-learning algorithm, and a dynamic feedback closed loop of the prediction process is realized through a continuous cycle of optimization process. Whenever new error feedback data is generated, the parameter optimization process is automatically started, the parameter settings of each model are adjusted, and these optimized parameters are applied to the next prediction step. This feedback mechanism ensures that the model will not stagnate in a specific state, but can be adjusted in real time as the environment changes, so as to always maintain a high prediction accuracy.

[0055] Embodiment 2

[0056] Figure 2 This is a flow chart of a photovoltaic power prediction method provided by Embodiment 2 of the present invention. This embodiment is a further refinement of the above embodiment. Figure 2 As shown, the method includes:

[0057] S201. Acquire photovoltaic power data of the photovoltaic power generation system, meteorological data of the area where the photovoltaic power generation system is located, and satellite remote sensing data.

[0058] S202: Preprocess photovoltaic power data, meteorological data and satellite remote sensing data to generate fused feature data.

[0059] In this embodiment, the fused feature data can be understood as data obtained by fusing data from multiple sources in multiple dimensions.

[0060] Specifically, after obtaining these data, since photovoltaic power prediction is highly dependent on the accuracy and completeness of the data, the processor can clean and preprocess them to remove noise and outliers to ensure data quality. The cleaning process can include anomaly detection and data smoothing. Anomaly detection is to identify and remove abnormal data points that may be caused by sensor failure or data transmission errors; data smoothing is used to reduce random noise in the data and maintain the trend characteristics of the data. The processor can perform preliminary feature extraction on the cleaned data. For example, the processor can extract core features related to photovoltaic power generation from the cleaned meteorological data, such as solar radiation intensity and temperature change rate. These features are important factors affecting photovoltaic power generation. By extracting these features, the dimension of the original data can be reduced and the most useful information for prediction can be extracted. The photovoltaic power data, meteorological data and satellite remote sensing data that have been cleaned and preprocessed with feature extraction are fused to obtain fused feature data. The purpose of data fusion is to combine data from different sources and different types to obtain more comprehensive and accurate input data. The fusion process takes into account the temporal synchronization, spatial consistency and physical correlation of the data to ensure that the information from different data sources can be organically combined to generate fused feature data with higher prediction capabilities. Through data fusion, not only can the shortcomings of a single data source be made up, but the robustness and accuracy of the prediction model can also be improved through complementary information.

[0061] For example, the processor obtains local solar radiation data from a ground-based meteorological station, but due to limited coverage, it may not accurately reflect the lighting conditions of the entire area. At this time, through satellite remote sensing data, the processor can supplement the missing data from the ground-based meteorological station and provide a more comprehensive radiation distribution map. Combined with historical photovoltaic power data, the processor can use data fusion technology to integrate this information into a complete fused feature data set. This fused feature data set not only reflects the current environmental status, but also covers the spatial variation characteristics within the geographic area, providing a more reliable basis for subsequent ultra-short-term photovoltaic power forecasts.

[0062] S203, performing random slice Fourier transform on the fused feature data to obtain intermediate feature data.

[0063] In this embodiment, random slice Fourier transform is used to analyze the frequency characteristics of the fused feature data. The intermediate feature data can be understood as.

[0064] It can be understood that the Fourier transform essentially converts the signal in the time domain into a representation in the frequency domain, so that the periodic components in the data can be fully displayed. Different frequency components may correspond to different periodic changes in photovoltaic power generation output, such as daily changes in sunshine intensity and short-period disturbances of weather factors. Traditional Fourier transforms have limitations when processing multidimensional data. Therefore, the present invention adopts the random slicing Fourier transform method to randomly slice the multidimensional data and then perform Fourier transforms separately. This method can effectively capture the frequency characteristics in multidimensional data, especially when the data is unevenly distributed or there are complex correlations. The random slicing Fourier transform can more comprehensively reveal the spectral information in the data.

[0065] Specifically, the processor obtains the fused feature data after fusion, which contains multi-dimensional information such as sunshine intensity, temperature and wind speed. By processing the fused feature data through random slice Fourier transform, the processor can identify the periodic change patterns in these data, such as the daily cycle of sunshine intensity and the diurnal fluctuation of temperature.

[0066] S204: Perform fractal dimension analysis on the intermediate feature data to generate multi-dimensional feature data.

[0067] In this embodiment, fractal dimension analysis is used to extract the complexity characteristics of data. Fractal dimension is a measure used to describe the self-similarity and complexity of data.

[0068] Photovoltaic power data and meteorological data are usually highly complex and nonlinear, especially in short-term changes. Therefore, fractal dimension analysis can better capture the complexity and dynamic changes in the data. The higher the fractal dimension, the more complex the data structure is, and the more factors the model needs to consider when predicting. Therefore, through fractal dimension analysis, the extracted features are not limited to simple frequency information, but also include the complexity and dynamic characteristics of the data.

[0069] Specifically, fractal dimension analysis further reveals the complexity of these data in the short term, such as the dramatic fluctuations in wind speed due to sudden weather changes. Ultimately, the processor can fuse these frequency features and complexity features to generate standardized multi-dimensional feature data. This data set can not only reflect the short-term fluctuation trend of photovoltaic power, but also reveal the complex impact of meteorological conditions on power generation, providing richer and more accurate input for subsequent prediction models. Through random slice Fourier transform and fractal dimension analysis, standardized multi-dimensional feature data are generated. These feature data are standardized to ensure that features of different dimensions have a unified scale and distribution in the subsequent modeling process, thereby avoiding model training bias caused by differences in feature scales. This multi-dimensional feature data not only enriches the input information of the model, but also improves the model's ability to cope with complex data situations, thereby improving the accuracy and stability of ultra-short-term prediction of photovoltaic power.

[0070] S205, mapping the multi-dimensional feature data to a multi-level dynamic system, and generating phase space feature data through phase space reconstruction.

[0071] In this embodiment, the basic principle of phase space reconstruction is to embed the time series data into a high-dimensional space, and by constructing the phase space, the dynamic behavior in the data can be fully displayed in the space. The phase space feature data can be understood as a spatial concept that describes all possible states. In the phase space, the evolution trajectory of the time series reveals the inherent laws and dynamic characteristics of photovoltaic power at different time scales.

[0072] It is known that phase space reconstruction is particularly critical for ultra-short-term prediction of photovoltaic power, because the output of photovoltaic power generation systems is affected by many factors, such as weather changes, fluctuations in solar radiation intensity, etc., which often manifest as nonlinear dynamic changes. If we only rely on traditional linear prediction methods, it is difficult to accurately capture these complex dynamic characteristics, resulting in large prediction errors. Through phase space reconstruction, the system can construct the dynamic trajectory of photovoltaic power data in high-dimensional space, so as to better consider these nonlinear factors in the prediction and improve the prediction accuracy.

[0073] Specifically, the multidimensional feature data contains multidimensional information obtained through preliminary processing and feature extraction, such as the time series characteristics of photovoltaic power and changes in meteorological conditions. The processor can map this data to a multi-level dynamic system and analyze the dynamic behavior of photovoltaic power at different time scales provided by the multi-level dynamic system. Dynamical system theory is very effective in dealing with complex dynamic systems. It can convert time series data into trajectories in phase space, which reflect the dynamic characteristics of the system. In this way, complex nonlinear time series relationships are mapped to geometric structures in phase space, allowing the model to understand the dynamic behavior of the data more intuitively. Then, phase space feature data is generated through phase space reconstruction.

[0074] Exemplarily, the phase space reconstruction step includes implementing phase space reconstruction by delayed embedding of the time series, wherein the calculation formula for phase space reconstruction is:

[0075] X(t)={x(t),x(t+τ),…,x(t+(m-1)τ)}

[0076] Among them, X(t) represents the feature data vector after phase space reconstruction; x(t) is the time series data in the multidimensional feature data; τ is the time delay, which represents the delay step of the time series data; m is the embedding dimension, which represents the dimension size of the embedding space. Through the above reconstruction process, the phase space feature data at different time scales are extracted to capture the potential nonlinear dynamic behavior in the time series. The core principle of phase space reconstruction is to map the one-dimensional time series data into a high-dimensional space through delay embedding to form a geometric structure that can reveal the dynamic behavior of the system. Specifically, given a time series data x(t), it is mapped into a high-dimensional vector X(t) through delay embedding, as shown in the above formula. Through this reconstruction process, the original one-dimensional time series data is expanded into a vector sequence in an m-dimensional space. The vector sequence forms a trajectory in the phase space, which reflects the dynamic behavior characteristics of the system. For example, in photovoltaic power data, the power fluctuations originally caused by cloud cover may appear as a series of complex fluctuations in the one-dimensional time series, but through phase space reconstruction, these fluctuations will form specific trajectories, thereby revealing the dynamic laws of the system at different time scales. These phase space characteristic data can not only capture short-term fluctuation patterns, but also reveal trend changes on longer time scales, thus providing a multi-level predictive basis for the model.

[0077] Exemplarily, the processor can extract the time series x(t) from the photovoltaic power data of the past 12 hours. By setting the time delay τ to 10 minutes and the embedding dimension m to 3, the system maps the time series into a three-dimensional vector X(t). In the phase space, these three-dimensional vectors form a complex trajectory. Analysis shows that the trajectory clearly reflects the short-term power fluctuations caused by cloud movement and the trend changes caused by changes in sunlight angles. Based on this phase space feature data, the system can more accurately predict the next photovoltaic power changes, especially the drastic fluctuations that may occur in a short period of time. In the end, the prediction results were proven to significantly reduce the prediction error in actual operation, providing more reliable data support for the scheduling of photovoltaic power generation.

[0078] Phase space reconstruction through delayed embedding of time series can effectively extract nonlinear dynamic features in photovoltaic power data. This phase space feature data not only improves the input quality of the prediction model, but also significantly enhances the model's ability to capture complex dynamic behaviors, thereby achieving higher prediction accuracy and reliability.

[0079] S206. Perform regression analysis on the phase space feature data through kernel function mapping, minimum absolute shrinkage and selection operator regression in the nonlinear adaptive regression model to obtain target feature data.

[0080] In this embodiment, the nonlinear adaptive regression model is a flexible regression method that can capture nonlinear relationships in data. Kernel function mapping is a tool for mapping low-dimensional data to high-dimensional space. It can separate nonlinear features that are originally difficult to separate in a linear manner in high-dimensional space, thereby simplifying the complexity of subsequent regression analysis. Least Absolute Shrinkage and Selection Operator (Lasso) regression is a regression method for feature selection and model optimization. It adds an L1 regularization term to ordinary regression, so that the model can automatically select the most explanatory features and reduce the weights of unimportant features to zero. This feature is particularly suitable for processing high-dimensional feature data, because in high-dimensional space, the data may contain a large number of redundant features. Lasso regression ensures the simplicity and predictive power of the model by penalizing unimportant features.

[0081] It is understandable that when processing photovoltaic power data, nonlinear relationships often originate from the complex influence of environmental factors (such as light intensity, temperature changes, etc.), and these nonlinear factors make it difficult for traditional linear regression methods to accurately model. In order to solve this problem, the present invention introduces the concept of kernel function mapping. Specifically, after the phase space feature data is generated, these feature data are further analyzed and predicted by nonlinear adaptive regression. Nonlinear adaptive regression is a flexible regression method that can capture nonlinear relationships in data and automatically adjust model parameters according to changes in input data to improve the accuracy of prediction. In photovoltaic power prediction, nonlinear adaptive regression is particularly suitable for processing those irregular and nonlinear dynamic changes, such as photovoltaic power fluctuations caused by sudden weather changes.

[0082] Specifically, the nonlinear adaptive regression step includes regression analysis of phase space feature data through kernel function mapping and Lasso regression. Among them, the kernel function is a radial basis function, which is used to map the phase space feature data to a high-dimensional feature space. The radial basis function (RBF) is a common and effective kernel function. The core feature of the radial basis function is that it can calculate the mapping result according to the distance between the data points, that is, the mapping results of close data points in the high-dimensional space will also be very close, while the data points far away will be mapped farther. Through this mapping method, the radial basis function can convert the phase space feature data that originally has a complex nonlinear relationship into a form that is easy to handle in the high-dimensional space, providing a more suitable input for Lasso regression. The processor can map the phase space feature data to a high-dimensional feature space through the radial basis function, and the model can better capture the nonlinear relationship between photovoltaic power and environmental factors. Subsequently, Lasso regression analyzes these high-dimensional feature data, automatically selects the features that have the greatest impact on the photovoltaic power change, and thus generates an optimized prediction model. This model can not only accurately predict the short-term fluctuations of photovoltaic power, but also has strong robustness and can effectively cope with changes in environmental conditions.

[0083] Exemplarily, the processor can obtain high-dimensional feature data from the phase space reconstruction step, which contains the dynamic characteristics of photovoltaic power at different time scales. However, due to the complex nonlinear relationship between these feature data, it is difficult for the processor to make effective predictions through a simple linear model. To solve this problem, the processor can first use the radial basis function to map these feature data to a higher-dimensional feature space. In this high-dimensional space, the originally complex nonlinear relationship is linearized, and the processor can more easily identify the correlation between different features. Then, the processor can perform regression analysis on these high-dimensional feature data through Lasso regression. The L1 regularization term of Lasso regression enables the model to automatically filter out the features that are most sensitive to photovoltaic power changes and ignore those features that contribute less to the prediction. Finally, a concise but efficient prediction model is generated. This model can not only accurately predict the photovoltaic power changes in the next hour, but also has strong generalization ability and can work stably under different weather conditions. The implementation results show that through this nonlinear adaptive regression method, the accuracy of photovoltaic power prediction is significantly improved, especially in the face of complex and nonlinear environments, the prediction error is greatly reduced.

[0084] S207: Weighting the target feature data according to the dynamic feature weight to obtain weighted feature data.

[0085] Specifically, the processor may adjust the target feature data through dynamic feature weights to obtain weighted feature data.

[0086] Exemplarily, the processor acquires standardized photovoltaic power and meteorological data, which exhibit complex nonlinear dynamic characteristics. Through phase space reconstruction, the processor maps these data into a high-dimensional phase space. In this space, the processor can clearly see the trajectory of photovoltaic power changes over time, and predict future trends through these trajectories. Subsequently, the processor performs regression analysis on the phase space feature data through a nonlinear adaptive regression model, and dynamically adjusts the feature weights in combination with the real-time feedback mechanism to adapt to the weather changes of the day. During a sudden cloudy day, the real-time feedback mechanism detected an increase in the prediction error and immediately adjusted the feature weights related to sunshine, enabling the model to more accurately predict the decline in photovoltaic power that afternoon. Finally, the processor generates dynamically weighted feature data to ensure the high accuracy and timeliness of the prediction results.

[0087] Furthermore, the processor can adjust the model in time when the data changes through the real-time feedback mechanism, and the generated dynamic weighted feature data greatly improves the accuracy and response speed of ultra-short-term prediction of photovoltaic power. In addition, the introduction of the real-time feedback mechanism enables the feature weights to be dynamically adjusted according to the actual prediction error. Specifically, the real-time feedback mechanism monitors the error between each prediction result and the actual observation value, and dynamically adjusts the feature weights in the model according to the size and direction of the error. This process ensures that the model can adapt to changes in data in a timely manner, thereby maintaining efficient prediction performance. Finally, through nonlinear adaptive regression combined with real-time feedback, the system generates dynamically weighted feature data, which contains feature weights dynamically adjusted according to the current environment and historical behavior, and is the core input for further processing of the prediction model.

[0088] Exemplarily, the dynamic feature weight of nonlinear adaptive regression is dynamically adjusted by the following formula:

[0089] w i (t+1)=w i (t)+α·e(t)

[0090] Among them, w i (t) represents the weight of the i-th feature at time step t; w i (t+1) represents the updated weight of the tth feature at time step t+1; α is the learning rate, which is used to control the step size of the weight update; e(t) is the error feedback value at the current time step t, which represents the difference between the predicted value and the actual value; the above formula is used to realize dynamic adjustment of feature weights, so that the regression model can adapt to data changes in real time.

[0091] In photovoltaic power prediction, due to the complexity and dynamics of environmental conditions (such as weather changes), a single static regression model is difficult to continuously provide high-precision prediction results. In order to cope with this dynamic change, the present invention introduces a dynamic adjustment mechanism based on feedback, so that the regression model can self-adjust and optimize according to the actual situation. The core of this adjustment mechanism is real-time feedback. After each time step t, the system compares the prediction result with the actual observation value and calculates the error feedback value e(t). The error feedback value reflects the prediction accuracy of the current model: when the prediction error is large, the model needs to be adjusted more; when the prediction error is small, the adjustment range of the model is relatively small. The learning rate α determines the response speed of the model to the error feedback, which controls the step size of each adjustment. If α is too large, the model may be over-adjusted, resulting in unstable prediction; if α is too small, the model may be under-adjusted and difficult to adapt to changes in time.

[0092] Through the above formula, the system can dynamically adjust the weight of each feature at each time step. For example, if a feature has a greater impact on the prediction error in the current time step (i.e., the α·e(t) value is larger), the system will increase the weight of the feature so that it contributes more to the prediction result in the next time step. Conversely, if a feature has a smaller impact, the system will reduce its weight. This adjustment mechanism based on error feedback enables the regression model to adapt to changes in the environment and data in real time, thereby improving the accuracy and robustness of the prediction.

[0093] For example, the processor is making an ultra-short-term prediction of the output power of a photovoltaic power station. During the prediction process, an unexpected cloud cover suddenly appeared, which caused a large error in the current photovoltaic power prediction. At this time step t, the processor calculated a large error feedback value e(t) and adjusted the weights w of related features (such as cloud density, sunshine intensity, etc.) according to the error value through the formula i (t). By increasing the weights of these key features, the processor can more accurately predict the change in PV power in the next time step t+1 and quickly adapt to the current meteorological conditions. Over time, if the weather conditions gradually return to normal, the processor will continue to adjust the weights based on real-time feedback to reduce the impact of features that are no longer important and maintain the simplicity and accuracy of the prediction model.

[0094] Furthermore, the multi-level dynamic system includes a short-term time scale dynamic system, a medium-term time scale dynamic system and a long-term time scale dynamic system. Accordingly, the multi-dimensional feature data is mapped to the multi-level dynamic system, and the phase space feature data is generated through phase space reconstruction, including:

[0095] The multidimensional characteristic data are divided into time scales to obtain the data to be input under different time scale types, including short-term expected input sub-data, medium-term expected input sub-data and long-term expected input sub-data; the short-term expected input sub-data are input into the short-term time scale dynamic system and the phase space is reconstructed to obtain short-term sub-phase space data; the medium-term expected input sub-data are input into the medium-term time scale dynamic system and the phase space is reconstructed to obtain medium-term sub-phase space data; the long-term expected input sub-data are input into the long-term time scale dynamic system and the phase space is reconstructed to obtain long-term sub-phase space data; the short-term sub-phase space data, the medium-term sub-phase space data and the long-term sub-phase space data are used as the phase space characteristic data.

[0096] It is understandable that the dynamic behavior of photovoltaic power data is not only affected by short-term environmental factors (such as instantaneous cloud changes), but also by medium-term meteorological conditions (such as daytime temperature changes) and long-term seasonal factors (such as seasonal changes). Therefore, data analysis that relies solely on a single time scale may not be able to fully capture these diverse dynamic behaviors. In order to solve this problem, the present invention introduces a mapping method for a multi-level dynamic system, that is, decomposing the data into different time scales of short-term, medium-term and long-term, so as to independently analyze the dynamic behaviors on these time scales.

[0097] Specifically, the processor can input the multidimensional feature data into multiple dynamic systems of different time scales, each dynamic system processes the short-term, medium-term and long-term time scale data respectively, and reconstructs the multidimensional feature data in phase space to generate phase space feature data corresponding to each time scale to capture the dynamic characteristics of photovoltaic power data at different time scales. Among them, the short-term time scale dynamic system mainly captures the rapid fluctuations and short-term disturbances in the data, such as the photovoltaic power fluctuations caused by instantaneous changes in illumination. The medium-term time scale dynamic system focuses on daily trends and changes, such as the gradual changes in daytime sunshine intensity. The long-term time scale dynamic system focuses on seasonal trends and long-term meteorological pattern changes, such as the difference in illumination between winter and summer. Once each dynamic system receives the standardized multidimensional feature data, they will reconstruct the data in phase space. For example, on a short-term time scale, the system may see the instantaneous fluctuation trajectory of photovoltaic power, while on a long-term time scale, the system can capture the trend line of seasonal changes. Phase space reconstruction provides a complete perspective for subsequent dynamic behavior analysis, enabling the system to better understand and predict the changes in photovoltaic power at different time scales.

[0098] Through this multi-level dynamic system mapping, the dynamic characteristics of photovoltaic power data at different time scales are effectively separated and captured. This not only improves the accuracy of the model, but also enables the model to adapt to environmental changes more flexibly, thereby showing excellent performance in short-term, medium-term and long-term forecasts.

[0099] Exemplary, using a specific example as a demonstration, Figure 3 This is an example diagram of a multi-level dynamic system in a photovoltaic power prediction method provided in the second embodiment of the present invention, such as Figure 3The following are shown: short-term, medium-term and long-term time scale dynamic systems. The processor collects photovoltaic power data from multiple sensors over the past 24 hours. The processor can first input this data into the short-term, medium-term and long-term time scale dynamic systems. The short-term time scale dynamic system processes the instantaneous data changes in the past hour. Through phase space reconstruction, the processor can capture the trajectory of photovoltaic power fluctuations caused by cloud changes. The medium-term time scale dynamic system processes the data of the past 12 hours. Through phase space reconstruction, the processor can obtain the trend line of photovoltaic power changes with the angle of the sun. The long-term time scale dynamic system processes the data of the past 24 hours. Through phase space reconstruction, the processor can identify the photovoltaic power pattern that alternates between day and night. Finally, these phase space feature data on different time scales are combined to provide the system with a comprehensive dynamic feature view, making the subsequent ultra-short-term prediction of photovoltaic power more accurate and stable.

[0100] The dynamic characteristics of photovoltaic power data on short-term, medium-term and long-term time scales are effectively captured. This process not only enables the system to better understand the dynamic behavior of the data, but also provides more accurate and comprehensive input features for subsequent prediction models, thereby significantly improving the accuracy and reliability of ultra-short-term photovoltaic power prediction.

[0101] S208, inputting the weighted feature data into the neural networks of multiple parallel paths in the pre-constructed multi-path nested neural network model to obtain the high-order features output by each neural network; wherein a nested structure is adopted between each neural network.

[0102] In order to make full use of these multi-dimensional and multi-level feature information, a multi-path nested neural network model was constructed. The core of this model is to process feature data of different dimensions through multiple parallel neural network paths. The neural network level of each path is a nested structure, and the connection between layers reflects the hierarchical relationship of the data. The advantage of this nested structure is that it can extract high-order features from the data layer by layer, allowing the model to capture more complex and deep relationships.

[0103] Specifically, the processor can input the weighted feature data into the neural networks of multiple parallel paths in the pre-built multi-path nested neural network model. In different paths of the model, each path processes different dimensions of feature data. For example, one path may focus on features in the time dimension, and another path may focus on features in the space dimension. This multi-path design enables the model to analyze and process feature data from different perspectives and fully tap the value of features in each dimension. The optimized feature data set not only has higher expressive power, but also contains complex association information between different levels and dimensions, providing a solid foundation for the final prediction. Among them, within each path, the neural network adopts a nested structure, which means that the network level is progressive layer by layer, and the output of each layer will be used as the input of the next layer for further feature extraction and processing. The advantage of the nested structure is that it can gradually extract high-order features in the data, thereby revealing deeper associations and patterns. For example, in the path of the time dimension, the first few layers of the network may only capture simple time series features, and as the network level progresses, the subsequent layers can identify more complex trends and periodic changes. This progressive feature extraction method ensures that the model can extract the most predictive information from the original data. Each path generates its own path feature data, which represents the high-order features of the dimension processed by the path.

[0104] Exemplarily, the processor can identify three main dimensions from the weighted feature data: time dimension (including historical photovoltaic power data), space dimension (including meteorological data in different geographical locations), and environmental factor dimension (such as temperature, humidity, etc.). The processor constructs an independent neural network path for each dimension. In the time dimension path, the processor gradually extracts short-term fluctuations and long-term trends in the time series through nested neural network layers; in the space dimension path, the network processes the data of each geographical location layer by layer to capture the correlation between different regions; in the environmental factor dimension path, the processor identifies the complex effects of temperature, humidity, etc. on photovoltaic power through layer-by-layer analysis. The path feature data finally generated by each path reflects the high-order features of the dimension, and these feature data will be further integrated for the final prediction. The test results show that this multi-path nested neural network model significantly improves the accuracy of ultra-short-term prediction of photovoltaic power, especially when facing multi-dimensional complex data, the model shows strong adaptability and accuracy.

[0105] S209, based on the feature exchange layer in the multi-way nested neural network model, the high-order features are fused to generate optimized fused feature data.

[0106] In this embodiment, the feature exchange layer is used to allow feature data in different paths to interact and fuse. The optimized fused feature data can be understood as data obtained after fusing high-order features.

[0107] It is understandable that each path of the multi-path nested neural network model focuses on processing feature data of a specific dimension, such as time, space, and environmental factors. In these paths, the network extracts and processes high-order features of the data in a layer-by-layer progressive manner. Although each path can generate useful feature data individually, these features are often limited to the expression range of their respective dimensions. If only relying on the features of a single path, the prediction model may not be able to fully utilize the potential correlation and complementarity between features of different dimensions. Therefore, in order to organically combine the feature data in different paths, a feature exchange layer is introduced.

[0108] Specifically, the processor can perform cross-level feature fusion of each high-order feature based on the feature exchange layer in the multi-path nested neural network model. This cross-level fusion helps the model to integrate the feature information extracted from each path, thereby generating an optimized feature data set, and generating optimized fused feature data as the final prediction model input. Through this multi-path parallel processing and layer-by-layer progressive feature extraction method, the model can fully explore the deep-level information in the multi-dimensional features, thereby greatly improving the accuracy of ultra-short-term photovoltaic power prediction. Among them, the cross-level feature fusion step includes setting a feature exchange layer in the multi-path nested neural network model to allow the feature data in different paths to interact and fuse in this layer, and generate fused feature data, and finally input the fused feature data into the adaptive ensemble learning model.

[0109] The feature exchange layer allows the feature data in different paths to interact and fuse at this layer. In the feature exchange layer, the system analyzes the correlation between the features of different paths, combines these features with each other, and generates a unified fused feature data set. This process is similar to multi-view fusion of data, that is, combining information from different viewpoints (paths) to form a more comprehensive and accurate representation. Through this feature fusion, the model is able to consider feature information in multiple dimensions at the same time, thereby better capturing the complex dynamic relationship of photovoltaic power data.

[0110] The fused feature data has stronger expressiveness and generalization capabilities, because it not only contains the high-order features extracted from each path, but also reflects the correlation between the features of different paths. This fusion can effectively reduce the model's dependence on single-dimensional features, avoid prediction bias caused by incomplete or distorted information in a certain dimension, and thus improve the model's robustness and prediction accuracy.

[0111] For example, in order to facilitate understanding of the construction of the multi-way nested neural network model in the present invention, a specific example is used for demonstration. Figure 4 This is an example diagram of a multi-path nested neural network model in a photovoltaic power prediction method provided in the second embodiment of the present invention, such as Figure 4 As shown, the construction of the multi-path nested neural network model includes inputting weighted feature data into neural networks of multiple parallel paths, each path processing different dimensional features in the dynamic weighted feature data respectively; the neural network of each path adopts a nested structure, including multiple progressive neural network layers, each layer performs further feature extraction and processing based on the output of the previous layer, and finally generates path feature data.

[0112] Photovoltaic power data and its related meteorological data usually have multi-dimensional characteristics, including time dimension, space dimension and dimensions of other environmental factors. The effects of features in different dimensions on photovoltaic power may vary. Therefore, in order to improve the accuracy of prediction, these features need to be processed independently and in depth. The multi-path nested neural network model is designed based on this requirement. The model inputs dynamically weighted feature data into neural networks with multiple parallel paths, each of which specializes in processing feature data of one dimension. This multi-path design ensures that the features of each dimension can be fully analyzed and utilized.

[0113] Exemplarily, the processor obtains high-order features of photovoltaic power time series, high-order features of space weather data, and high-order features of environmental factors from different paths. By setting a feature exchange layer in the multi-path nested neural network model, the processor can interactively fuse the feature data in these different paths. The feature exchange layer identifies the potential correlation between time series features and space weather features, and fuses these features to generate a unified feature data set. This fused feature data not only contains the core information of each path, but also reflects the complex correlation between them. Next, the processor inputs the fused feature data into the adaptive ensemble learning model, and by integrating the prediction results of multiple weak learners, the processor finally generates a highly accurate photovoltaic power prediction data. Tests show that this model based on cross-level feature fusion can significantly improve the stability and accuracy of predictions when facing changing meteorological conditions.

[0114] S210, inputting the optimized fusion feature data into a preset adaptive ensemble learning model to obtain photovoltaic power prediction data.

[0115] In this embodiment, the preset adaptive ensemble learning model can be understood as a technology that weightedly combines the output results of multiple learners. The core idea is to predict the input features through multiple weak learners (i.e., relatively simple models with certain predictive capabilities), and weighted average these prediction results to generate a more accurate and robust final prediction.

[0116] Specifically, these fused feature data are input into the adaptive ensemble learning model. The adaptive ensemble learning model further optimizes the accuracy of the prediction by combining the output results of multiple weak learners. Since the input fused feature data has integrated the information of multiple paths, the adaptive ensemble learning model can better integrate and utilize this information to generate more accurate and robust ensemble prediction results. By processing these optimized feature data through the preset adaptive ensemble learning method, the weights of each learner can be dynamically adjusted according to their performance, thereby adaptively optimizing the results of the ensemble prediction. This method can not only improve the overall prediction accuracy of the model, but also effectively reduce the overfitting problem that may arise from a single model.

[0117] Exemplarily, the calculation of adaptive ensemble learning is performed by the following formula:

[0118]

[0119] Among them, y is the final integrated prediction result; α i is the weight of the i-th weak learner; h i (x) is the predicted output of the i-th weak learner for the input feature x; n is the number of weak learners, which represents the total number of different weak learners used in ensemble learning.

[0120] Through the above formula, the prediction results of multiple weak learners are combined to generate the final integrated prediction data. In photovoltaic power prediction, different weak learners may be more sensitive to different types of input features or have stronger prediction capabilities. For example, some learners may be good at capturing rapid changes in the short term, while others are better at identifying long-term trends. Through adaptive ensemble learning, the prediction results of these weak learners are reasonably weighted to generate a more robust final prediction. Weight α i The setting is not fixed, but is dynamically adjusted according to the performance of each weak learner in different situations. In this way, the system can adaptively optimize the integrated model and enhance its responsiveness to different environmental changes. This weighted average approach can not only fully utilize the advantages of each weak learner, but also effectively reduce the bias of a single model. Since each weak learner has its own specific preferences and areas of expertise, by combining the outputs of these learners, the system can obtain a more comprehensive perspective, thereby improving the overall accuracy and robustness of the prediction.

[0121] The technical solution of the embodiment of the present invention generates multidimensional feature data by preprocessing the data from each source, performing random slice Fourier transform and fractal dimension analysis, so as to more comprehensively reveal the spectrum information in the data, and make the extracted features not limited to simple frequency information, but also include the complexity and dynamic characteristics of the data. The multidimensional feature data is processed by a multi-level dynamic system, phase space reconstruction and nonlinear adaptive regression model to obtain weighted feature data, which can effectively reveal the nonlinear dynamic behavior in the photovoltaic power time series, and effectively separate and capture the dynamic characteristics of photovoltaic power data on different time scales. This not only improves the accuracy of the model, but also enables the model to adapt to environmental changes more flexibly, so that it can show excellent performance in short-term, medium-term and long-term predictions. Through the combination of kernel function mapping and Lasso regression, the nonlinear relationship in the phase space feature data is effectively processed. The use of radial basis functions enables the model to simplify complex nonlinear features in high-dimensional space, while Lasso regression optimizes the performance of the model through feature selection. This method not only improves the accuracy of the prediction, but also enhances the robustness of the model, so that it can maintain a stable and accurate prediction effect under various complex environments. Through multiple parallel path neural networks in the multi-path nested neural network model, high-order features of weighted feature data in different dimensions are extracted, and fused through the feature exchange layer to generate optimized fused feature data. The model can simultaneously consider feature information in multiple dimensions, thereby better capturing the complex dynamic relationship of photovoltaic power data, and inputting the optimized fused feature data into the preset adaptive ensemble learning model to obtain photovoltaic power prediction data, thereby enhancing the ability to respond to different environmental changes. This model based on cross-level feature fusion can significantly improve the stability and accuracy of predictions when facing changing meteorological conditions.

[0122] Embodiment 3

[0123] Figure 5 This is a schematic diagram of the structure of a photovoltaic power prediction device provided by Embodiment 3 of the present invention. Figure 5 As shown, the device comprises:

[0124] The data acquisition module 51 is used to obtain the photovoltaic power data of the photovoltaic power generation system, the meteorological data of the area where the photovoltaic power generation system is located, and the satellite remote sensing data;

[0125] A data determination module 52, configured to determine multi-dimensional feature data according to the photovoltaic power data, the meteorological data and the satellite remote sensing data;

[0126] A feature mapping module 53, for processing the multi-dimensional feature data based on a multi-level dynamic system and dynamic feature weights to determine weighted feature data;

[0127] The power prediction module 54 is used to generate photovoltaic power prediction data based on the weighted feature data and the pre-built multi-way nested neural network model.

[0128] The technical solution of the embodiment of the present invention obtains the photovoltaic power data of the photovoltaic power generation system, the meteorological data of the area where the photovoltaic power generation system is located, and the satellite remote sensing data; determines the multidimensional feature data based on the photovoltaic power data, meteorological data, and satellite remote sensing data; processes the multidimensional feature data based on the multi-level dynamic system and the dynamic feature weights to determine the weighted feature data; generates photovoltaic power prediction data based on the weighted feature data and the pre-built multi-path nested neural network model. By combining data from different sources for analysis with a multi-level dynamic system, deep mining and accurate prediction of the dynamic characteristics of photovoltaic power are achieved, solving the problem of insufficient signal extraction in the prior art. Through the multi-path nested neural network, the fusion of cross-level features is achieved, which effectively improves the accuracy of the prediction model under changeable weather conditions and improves the prediction accuracy.

[0129] Furthermore, the data determination module 52 is specifically used for:

[0130] Preprocessing the photovoltaic power data, the meteorological data and the satellite remote sensing data to generate fusion feature data;

[0131] Performing random slice Fourier transform on the fused feature data to obtain intermediate feature data;

[0132] Perform fractal dimension analysis on the intermediate feature data to generate multi-dimensional feature data.

[0133] Furthermore, the feature mapping module includes:

[0134] A data reconstruction unit, used for mapping the multi-dimensional feature data to a multi-level dynamic system, and generating phase space feature data through phase space reconstruction;

[0135] A first determination unit is used to perform regression analysis on the phase space feature data through kernel function mapping and minimum absolute shrinkage and selection operator regression in a nonlinear adaptive regression model to obtain target feature data;

[0136] The second determining unit is used to weight the target feature data according to the dynamic feature weight to obtain weighted feature data.

[0137] The multi-level dynamic system includes a short-term time scale dynamic system, a medium-term time scale dynamic system and a long-term time scale dynamic system. Accordingly, the data reconstruction unit is specifically used for:

[0138] Dividing the multidimensional feature data by time scale to obtain data to be input under different time scale types, wherein the data to be input includes short-term expected input sub-data, medium-term expected input sub-data and long-term expected input sub-data;

[0139] Inputting the short-term expected input sub-data into a short-term time scale dynamic system and performing phase space reconstruction to obtain short-term sub-phase space data;

[0140] Inputting the intermediate expected input sub-data into the intermediate time scale dynamic system and reconstructing the phase space to obtain intermediate sub-phase space data;

[0141] Inputting the long-term waiting input sub-data into the long-term time scale dynamic system and reconstructing the phase space to obtain long-term sub-phase space data;

[0142] The short-term sub-phase space data, the medium-term sub-phase space data and the long-term sub-phase space data are used as the phase space characteristic data.

[0143] Furthermore, the power prediction module 54 is specifically used for:

[0144] Inputting the weighted feature data into neural networks of multiple parallel paths in a pre-constructed multi-path nested neural network model to obtain high-order features output by each of the neural networks; wherein a nested structure is adopted between each of the neural networks;

[0145] Based on the feature exchange layer in the multi-path nested neural network model, each of the high-order features is fused to generate optimized fused feature data;

[0146] The optimized fusion feature data is input into a preset adaptive ensemble learning model to obtain photovoltaic power prediction data.

[0147] Optionally, the device further comprises:

[0148] A parameter optimization module is used to obtain actual observation data after generating photovoltaic power prediction data based on the weighted feature data and the pre-built multi-way nested neural network model, and determine optimization parameters through a meta-learning algorithm based on the actual observation data and the photovoltaic power prediction data.

[0149] The photovoltaic power prediction device provided in the embodiment of the present invention can execute the photovoltaic power prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0150] Embodiment 4

[0151] Figure 6A schematic diagram of the structure of an electronic device 60 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0152] like Figure 6 As shown, the electronic device 60 includes at least one processor 61, and a memory connected to the at least one processor 61, such as a read-only memory (ROM) 62, a random access memory (RAM) 63, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 61 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 62 or the computer program loaded from the storage unit 68 to the random access memory (RAM) 63. In the RAM 63, various programs and data required for the operation of the electronic device 60 can also be stored. The processor 61, the ROM 62, and the RAM 63 are connected to each other via a bus 64. An input / output (I / O) interface 65 is also connected to the bus 64.

[0153] A number of components in the electronic device 60 are connected to the I / O interface 65, including: an input unit 66, such as a keyboard, a mouse, etc.; an output unit 67, such as various types of displays, speakers, etc.; a storage unit 68, such as a disk, an optical disk, etc.; and a communication unit 69, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 69 allows the electronic device 60 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0154] The processor 61 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 61 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 61 executes the various methods and processes described above, such as a photovoltaic power prediction method.

[0155] In some embodiments, the photovoltaic power prediction method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 68. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 60 via the ROM 62 and / or the communication unit 69. When the computer program is loaded into the RAM 63 and executed by the processor 61, one or more steps of the photovoltaic power prediction method described above may be performed. Alternatively, in other embodiments, the processor 61 may be configured to execute the photovoltaic power prediction method in any other appropriate manner (e.g., by means of firmware).

[0156] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0157] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0158] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0159] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0160] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0161] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0162] In one embodiment, the embodiment of the present invention further includes a computer program product, the computer program product includes a computer program, and the computer program implements the photovoltaic power prediction method of any embodiment of the present invention when executed by a processor.

[0163] In the process of implementation, the computer program product can be written in one or more programming languages ​​or a combination thereof to perform the computer program code of the present invention, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0164] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0165] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A photovoltaic power prediction method, characterized in that: include: Acquiring photovoltaic power data of the photovoltaic power generation system, meteorological data of the area where the photovoltaic power generation system is located, and satellite remote sensing data; Determining multidimensional feature data according to the photovoltaic power data, the meteorological data and the satellite remote sensing data; Processing the multi-dimensional feature data based on a multi-level dynamic system and dynamic feature weights to determine weighted feature data; Photovoltaic power prediction data is generated based on the weighted feature data and the pre-built multi-way nested neural network model.

2. The method according to claim 1, characterized in that Determining multi-dimensional feature data according to the photovoltaic power data, the meteorological data and the satellite remote sensing data includes: Preprocessing the photovoltaic power data, the meteorological data and the satellite remote sensing data to generate fusion feature data; Performing random slice Fourier transform on the fused feature data to obtain intermediate feature data; Perform fractal dimension analysis on the intermediate feature data to generate multi-dimensional feature data.

3. The method according to claim 1, characterized in that The multi-dimensional feature data is processed based on the multi-level dynamic system and the dynamic feature weights to determine the weighted feature data, including: Mapping the multidimensional feature data to a multi-level dynamic system, and generating phase space feature data through phase space reconstruction; Through kernel function mapping and minimum absolute shrinkage and selection operator regression in a nonlinear adaptive regression model, regression analysis is performed on the phase space feature data to obtain target feature data; The target feature data is weighted according to the dynamic feature weight to obtain weighted feature data.

4. The method according to claim 3, characterized in that The multi-level dynamic system includes a short-term time scale dynamic system, a medium-term time scale dynamic system and a long-term time scale dynamic system. Accordingly, the multi-dimensional feature data is mapped to the multi-level dynamic system, and the phase space feature data is generated by phase space reconstruction, including: Dividing the multidimensional feature data by time scale to obtain data to be input under different time scale types, wherein the data to be input includes short-term expected input sub-data, medium-term expected input sub-data and long-term expected input sub-data; Inputting the short-term expected input sub-data into a short-term time scale dynamic system and performing phase space reconstruction to obtain short-term sub-phase space data; Inputting the intermediate expected input sub-data into the intermediate time scale dynamic system and reconstructing the phase space to obtain intermediate sub-phase space data; Inputting the long-term waiting input sub-data into the long-term time scale dynamic system and reconstructing the phase space to obtain long-term sub-phase space data; The short-term sub-phase space data, the medium-term sub-phase space data and the long-term sub-phase space data are used as the phase space characteristic data.

5. The method according to claim 1, characterized in that The generating photovoltaic power prediction data based on the weighted feature data and the pre-built multi-path nested neural network model includes: Inputting the weighted feature data into neural networks of multiple parallel paths in a pre-constructed multi-path nested neural network model to obtain high-order features output by each of the neural networks; wherein a nested structure is adopted between each of the neural networks; Based on the feature exchange layer in the multi-path nested neural network model, each of the high-order features is fused to generate optimized fused feature data; The optimized fusion feature data is input into a preset adaptive ensemble learning model to obtain photovoltaic power prediction data.

6. The method according to claim 1, characterized in that After generating photovoltaic power prediction data based on the weighted feature data and the pre-built multi-way nested neural network model, the method further includes: Actual observation data is obtained, and optimization parameters are determined through a meta-learning algorithm based on the actual observation data and the photovoltaic power prediction data.

7. A photovoltaic power prediction device, characterized in that: include: A data acquisition module, used to obtain photovoltaic power data of the photovoltaic power generation system, meteorological data of the area where the photovoltaic power generation system is located, and satellite remote sensing data; A data determination module, used to determine multi-dimensional feature data according to the photovoltaic power data, the meteorological data and the satellite remote sensing data; A feature mapping module, for processing the multi-dimensional feature data based on a multi-level dynamic system and dynamic feature weights to determine weighted feature data; The power prediction module is used to generate photovoltaic power prediction data based on the weighted feature data and a pre-built multi-way nested neural network model.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the photovoltaic power prediction method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the photovoltaic power prediction method according to any one of claims 1 to 6 when executed.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the computer program implements the photovoltaic power prediction method according to any one of claims 1 to 6.

Images