Distribution area distributed photovoltaic output prediction method and system based on data driving

Abstract

The invention relates to the technical field of photovoltaic output prediction, and particularly discloses a transformer area distributed photovoltaic output prediction method and system based on data driving. According to the method, a data processing and analysis algorithm based on artificial intelligence and deep learning can be used for carrying out variational mode decomposition-based time sequence correlation analysis on environment factor time sequence data sets, so that environment factor multivariate time sequence implicit correlation feature information is extracted, and time sequence features of photovoltaic output power are captured at the same time. And furthermore, interactive responsiveness characteristic information between multivariable time sequence association semantics and photovoltaic output time sequence semantics of the environmental factors is utilized to carry out transformer area distributed photovoltaic output prediction, and a final photovoltaic output power short-time prediction value is obtained. Therefore, not only is the influence of environmental factors on photovoltaic output considered, but also the accuracy and reliability of photovoltaic output prediction are further improved through the feature engineering and the deep learning model, and powerful support is provided for energy management and scheduling of a smart power grid.

Description

Technical Field

[0001] The present application relates to the technical field of photovoltaic output prediction, and more specifically, to a data-driven distributed photovoltaic output prediction method and system. Background Art

[0002] As the global demand for renewable energy continues to increase, distributed photovoltaic power generation systems have been widely used and developed as one of the clean and environmentally friendly energy solutions. However, due to the intermittent and random nature of solar energy resources, the output power of photovoltaic systems also has corresponding volatility and uncertainty, which poses a challenge to the stable operation and energy management of the power grid.

[0003] In order to improve the adaptability and stability of the power grid and optimize the operating efficiency of distributed photovoltaic power generation systems, it is essential to accurately predict photovoltaic output. Traditional photovoltaic output prediction methods mostly rely on physical models, that is, theoretical calculations based on parameters such as the position of the sun and weather conditions. However, these methods usually require detailed equipment parameters and may not be accurate enough in complex and changeable actual environments. Specifically, traditional models often fail to fully consider key environmental factors that affect photovoltaic output, such as solar irradiance, component temperature, and atmospheric pressure. These factors have a direct impact on photovoltaic output, and ignoring them will lead to large deviations in prediction results. At the same time, there is a nonlinear relationship between photovoltaic power and multivariate environmental information. Traditional models are difficult to effectively capture this complex dynamic change, especially when faced with an increase in input variables, and are prone to overfitting problems, which limits the improvement of prediction accuracy. In addition, due to the instability and obvious intermittent fluctuations of photovoltaic power generation, when seasonal changes or day and night changes cause climate change, traditional models cannot be flexibly adjusted and cannot reflect the regular trend of photovoltaic output in different seasons and different time periods, resulting in traditional photovoltaic output prediction methods. It is difficult to effectively predict.

[0004] Therefore, an optimized distributed photovoltaic output prediction scheme is desired. Summary of the invention

[0005] The present application provides a data-driven distributed photovoltaic output prediction method and system for a substation, which not only takes into account the impact of environmental factors on photovoltaic output, but also further improves the accuracy and reliability of photovoltaic output prediction through feature engineering and deep learning models, providing strong support for energy management and scheduling of smart grids.

[0006] In the first aspect, a data-driven distributed photovoltaic output prediction method is provided, including:

[0007] Acquire a time series data set of environmental factors collected by the sensor assembly, the environmental factors including solar irradiance, assembly temperature, air temperature, relative humidity, and atmospheric pressure;

[0008] Get the time series of photovoltaic output power;

[0009] After the time series data set of the environmental factors is sorted into a time series of solar irradiance, a time series of component temperature, a time series of air temperature, a time series of relative humidity, and a time series of atmospheric pressure according to the parameter sample dimension, variational mode decomposition and sequence encoding are performed on them to obtain a multivariate time series encoding feature vector of the environmental factors;

[0010] Performing sequence encoding on the time series of the photovoltaic output power to obtain a photovoltaic output power time series encoding feature vector;

[0011] The environmental factor multivariate time-series coding feature vector and the photovoltaic output power time-series coding feature vector are subjected to feature interaction processing based on semantic flow field modulation to obtain a photovoltaic output power-environmental factor fine-grained aligned feature response representation, including: constructing a photovoltaic output power-environmental factor semantic flow field between the environmental factor multivariate time-series coding feature vector and the photovoltaic output power time-series coding feature vector; based on the photovoltaic output power-environmental factor semantic flow field, performing feature alignment-type fine-grained response encoding on the environmental factor multivariate time-series coding feature vector and the photovoltaic output power time-series coding feature vector to obtain the photovoltaic output power-environmental factor fine-grained aligned feature response representation;

[0012] Photovoltaic output prediction is performed based on the photovoltaic output power-environmental factor fine-grained alignment feature response representation to determine a short-term prediction value of photovoltaic output power.

[0013] Secondly, a data-driven distributed photovoltaic output prediction system is provided, including:

[0014] An environmental factor data set acquisition module is used to acquire a time series data set of environmental factors collected by the sensor component, wherein the environmental factors include solar irradiance, component temperature, air temperature, relative humidity and atmospheric pressure;

[0015] Photovoltaic output power acquisition module, used to obtain the time series of photovoltaic output power;

[0016] An environmental factor multivariate time series encoding module is used to organize the time series data set of the environmental factor into a time series of solar irradiance, a time series of component temperature, a time series of air temperature, a time series of relative humidity, and a time series of atmospheric pressure according to the parameter sample dimension, and then perform variational mode decomposition and sequence encoding on them to obtain a multivariate time series encoding feature vector of the environmental factor;

[0017] A photovoltaic output power time series encoding module, used for performing sequence encoding on the time series of the photovoltaic output power to obtain a photovoltaic output power time series encoding feature vector;

[0018] A photovoltaic output power-environmental factor feature interaction processing module is used to perform feature interaction processing based on semantic flow field modulation on the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector to obtain a photovoltaic output power-environmental factor fine-grained aligned feature response representation, including: a photovoltaic output power-environmental factor semantic flow field construction unit, which is used to construct a photovoltaic output power-environmental factor semantic flow field between the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector; a photovoltaic output power-environmental factor fine-grained alignment processing unit, which is used to perform feature alignment-type fine-grained response encoding on the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector based on the photovoltaic output power-environmental factor semantic flow field to obtain the photovoltaic output power-environmental factor fine-grained aligned feature response representation;

[0019] The photovoltaic output prediction module is used to perform photovoltaic output prediction based on the photovoltaic output power-environmental factor fine-grained alignment feature response representation to determine a short-term prediction value of photovoltaic output power.

[0020] The present application provides a data-driven distributed photovoltaic output prediction method and system for the substation, which can collect time series data sets of environmental factors through sensor components, including time series data of solar irradiance, component temperature, air temperature, relative humidity and atmospheric pressure, and collect historical and real-time data of photovoltaic output power, and then use data processing and analysis algorithms based on artificial intelligence and deep learning to perform time series correlation analysis based on variational mode decomposition on these environmental factor time series data sets, so as to extract the implicit correlation feature information of the multivariate time series of environmental factors, and capture the time series characteristics of photovoltaic output power. Further, the interactive responsiveness feature information between the multivariate time series correlation semantics of environmental factors and the photovoltaic output time series semantics is used to predict the distributed photovoltaic output of the substation, and the final short-term prediction value of photovoltaic output power is obtained. Compared with traditional solutions, this method not only takes into account the impact of environmental factors on photovoltaic output, but also further improves the accuracy and reliability of photovoltaic output prediction through feature engineering and deep learning models, providing strong support for energy management and scheduling of smart grids. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0022] Figure 1 This is a schematic flow chart of a data-driven distributed photovoltaic output prediction method for an area according to an embodiment of the present application.

[0023] Figure 2 Schematic diagram of data flow of a data-driven distributed photovoltaic output prediction method for an embodiment of the present application.

[0024] Figure 3 This is a schematic flowchart of step S5 in the data-driven distributed photovoltaic output prediction method in an embodiment of the present application.

[0025] Figure 4 This is a schematic flowchart of step S52 in the data-driven distributed photovoltaic output prediction method in an embodiment of the present application.

[0026] Figure 5 It is a schematic block diagram of a data-driven distributed photovoltaic output prediction system in an embodiment of the present application. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0028] In recent years, with the development of big data and artificial intelligence technology, data-driven methods have gradually become a new trend in photovoltaic output forecasting. Data-driven methods mainly use historical data and real-time monitoring data to establish prediction models through machine learning or deep learning algorithms, thereby achieving more accurate short-term or ultra-short-term photovoltaic output forecasts. The advantage of this method is that it can automatically learn complex nonlinear relationships from a large amount of data without having to deeply understand the specific physical process.

[0029] Based on this, in the technical solution of the present application, a data-driven distributed photovoltaic output prediction method for the substation is proposed, which can collect time series data sets of environmental factors through sensor components, including time series data of solar irradiance, component temperature, air temperature, relative humidity and atmospheric pressure, and collect historical and real-time data of photovoltaic output power, and then use data processing and analysis algorithms based on artificial intelligence and deep learning to perform time series correlation analysis based on variational mode decomposition on these environmental factor time series data sets, so as to extract the implicit correlation feature information of the multivariate time series of environmental factors, and capture the time series characteristics of photovoltaic output power. Further, the interactive responsiveness feature information between the multivariate time series correlation semantics of environmental factors and the photovoltaic output time series semantics is used to predict the distributed photovoltaic output of the substation, and the final short-term prediction value of photovoltaic output power is obtained. Compared with the traditional solution, this method not only takes into account the impact of environmental factors on photovoltaic output, but also further improves the accuracy and reliability of photovoltaic output prediction through feature engineering and deep learning models, providing strong support for energy management and scheduling of smart grids.

[0030] Specifically, Figure 1 and Figure 2 As shown, the data-driven distributed photovoltaic output prediction method for the substation includes: S1, obtaining a time series data set of environmental factors collected by a sensor component, wherein the environmental factors include solar irradiance, component temperature, air temperature, relative humidity and atmospheric pressure; S2, obtaining a time series of photovoltaic output power; S3, after the time series data set of the environmental factors is sorted into a time series of solar irradiance, a time series of component temperature, a time series of air temperature, a time series of relative humidity and a time series of atmospheric pressure according to the parameter sample dimension, variational mode decomposition and sequence encoding are performed to obtain a multivariate time series encoding feature vector of the environmental factor; S4, sequence encoding is performed on the time series of the photovoltaic output power to obtain a photovoltaic output power time series encoding feature vector; S5, feature interaction processing based on semantic flow field modulation is performed on the multivariate time series encoding feature vector of the environmental factor and the photovoltaic output power time series encoding feature vector to obtain a photovoltaic output power-environmental factor fine-grained aligned feature response representation; S6, photovoltaic output prediction is performed based on the photovoltaic output power-environmental factor fine-grained aligned feature response representation to determine a short-term prediction value of photovoltaic output power.

[0031] Exemplarily, in step S1, a time series data set of environmental factors collected by the sensor assembly is obtained, and the environmental factors include solar irradiance, component temperature, air temperature, relative humidity and atmospheric pressure. It should be understood that the characteristic analysis of distributed photovoltaic output shows that the output power of the photovoltaic system is not only affected by the day and night cycle, but also by seasonal changes and other environmental factors such as solar irradiance, component temperature, air temperature, relative humidity and atmospheric pressure. These factors work together to lead to the uncertainty and volatility of photovoltaic output. In order to improve the accuracy of distributed photovoltaic output prediction and build an effective probabilistic prediction model, it is necessary to deeply study and analyze these influencing factors and obtain the time series data set of environmental factors collected by the sensor assembly.

[0032] Obtaining a time series data set of environmental factors is crucial to understanding the variation pattern of photovoltaic output. It helps to identify the key factors affecting distributed photovoltaic output and provide data support for establishing more accurate prediction models, thereby optimizing energy management strategies and promoting the effective use of clean energy. Specifically, since distributed photovoltaic output has obvious diurnal and seasonal characteristics and is affected by climate change, it is necessary to obtain a time series data set of environmental factors to capture these dynamic changes. For example, photovoltaic output changes with the position of the sun during the day, gradually increasing in the morning, reaching a peak at noon, and then gradually decreasing in the afternoon until it stops at night; in different seasons, the length of sunshine time varies, and the solar altitude angle also varies, which directly affects the amount of solar radiation received by the photovoltaic system, and thus affects the photovoltaic output level. In addition, solar irradiance directly determines the amount of photovoltaic power generation, the increase in component temperature will lead to a decrease in conversion efficiency, the air temperature and relative humidity will affect the performance of the component, and the atmospheric pressure can indirectly reflect the weather conditions and affect the lighting conditions. All these factors increase the complexity and unpredictability of photovoltaic output.

[0033] In one embodiment, obtaining a time series data set of environmental factors collected by the sensor assembly includes: first deploying a series of high-precision sensors to monitor and record relevant parameters. Various types of sensors are reasonably arranged around the photovoltaic power station to ensure that the required environmental variables can be fully covered and accurately measured, such as using a radiometer or photoresistor to measure solar irradiance, a thermocouple or other temperature sensor to measure the component temperature, a temperature and humidity integrated sensor to simultaneously detect air temperature and humidity, and a barometer to measure atmospheric pressure. All sensors should be connected to a central control system or cloud platform to achieve real-time data transmission through a PLC (programmable logic controller), RTU (remote terminal unit) or other communication protocols. The system automatically collects data at a predetermined time interval (such as every 15 minutes) and uploads the data to the server for storage and preliminary processing. The collected data may contain noise or outliers, and must undergo preprocessing steps such as cleaning, screening and formatting to ensure data quality. Then, it is organized and stored in a database in time series for subsequent use in feature extraction, model training and verification processes.

[0034] Exemplarily, in step S2, the time series of photovoltaic output power is obtained. It should be understood that by obtaining the time series data of photovoltaic output power, the output characteristics of the photovoltaic system at different time points can be captured, which helps to understand the trend of photovoltaic output power changing over time and its relationship with various environmental factors.

[0035] In one embodiment, it is first necessary to deploy necessary sensors and monitoring devices in the photovoltaic power station, such as the built-in data acquisition module of the inverter, an independent power meter, a current and voltage transformer, etc. These devices are used to measure the output power of the photovoltaic array or a single photovoltaic panel in real time. Then, the power data collected on site is uploaded to the central control system or cloud server using PLC (programmable logic controller), RTU (remote terminal unit) or other communication protocols to ensure the security, integrity and timeliness of the data, and to facilitate subsequent data processing and analysis. The collected raw data may contain noise or outliers, so it is necessary to go through pre-processing steps such as cleaning, screening and formatting to ensure data quality. Next, it is summarized and stored in the database at certain time intervals (for example, every minute, every hour) to form a continuous time series.

[0036] Exemplarily, in step S3, after the time series data set of the environmental factors is sorted into the time series of solar irradiance, the time series of component temperature, the time series of air temperature, the time series of relative humidity, and the time series of atmospheric pressure according to the parameter sample dimension, variational mode decomposition and sequence encoding are performed to obtain the multivariate time series encoding feature vector of the environmental factors. It should be understood that since the environmental factors include solar irradiance, component temperature, air temperature, relative humidity and atmospheric pressure, these environmental factor data not only have dynamic change laws and trends in the time dimension, but also have mutual influence between different environmental factor time series data. Therefore, in order to be able to more deeply and comprehensively capture the time series implicit correlation feature information of multiple environmental factors in the time dimension, in the technical solution of the present application, the time series data set of the environmental factors is further sorted into the time series of solar irradiance, the time series of component temperature, the time series of air temperature, the time series of relative humidity, and the time series of atmospheric pressure according to the parameter sample dimension, and then variational mode decomposition and sequence encoding are performed to obtain the multivariate time series encoding feature vector of the environmental factors. In particular, variational mode decomposition (VMD) is an adaptive signal processing technique that can decompose complex original signals into several intrinsic mode components (IMFs). For each environmental variable (such as solar irradiance, component temperature, etc.), VMD can obtain a set of IMFs representing fluctuation patterns in different frequency ranges. This decomposition helps to separate the noise in the original signal and can capture time series features at different scales, thereby increasing the diversity of features. In other words, by performing VMD on the time series of each environmental factor separately, a series of purer and physically meaningful environmental factor subsequences (i.e., intrinsic mode components) can be obtained. These subsequences can better reflect the intrinsic connection between environmental factors and photovoltaic output than the original environmental factor time series, thereby increasing the possibility of the model learning effective information.

[0037] In one embodiment, after the time series data set of the environmental factors is sorted into a time series of solar irradiance, a time series of component temperature, a time series of air temperature, a time series of relative humidity, and a time series of atmospheric pressure according to the parameter sample dimension, variational mode decomposition and sequence encoding are performed on them to obtain a multivariate time series encoding feature vector of the environmental factors, including: performing variational mode decomposition on the time series of the solar irradiance, the time series of the component temperature, the time series of the air temperature, the time series of the relative humidity, and the time series of the atmospheric pressure to obtain a solar irradiance eigenmode. The method comprises the following steps of: collecting a set of eigenmodal components of solar irradiance, a set of eigenmodal components of component temperature, a set of eigenmodal components of air temperature, a set of eigenmodal components of relative humidity and a set of eigenmodal components of atmospheric pressure; arranging the set of eigenmodal components of solar irradiance, the set of eigenmodal components of component temperature, the set of eigenmodal components of air temperature, the set of eigenmodal components of relative humidity and the set of eigenmodal components of atmospheric pressure into an environmental multivariate intrinsic modal component aggregation matrix and then inputting the matrix into an environmental multivariate temporal association encoder based on a convolutional layer to obtain the environmental factor multivariate temporal coding feature vector.

[0038] Specifically, the steps of variational mode decomposition are as follows: First, construct a mathematical model of the original signal and prepare for decomposition. Then, the construction problem of the original signal is converted into a variational problem, and a form suitable for the VMD algorithm is established by introducing constraints and Lagrange multipliers. Finally, the alternating direction method of multipliers (ADMM) or other optimization algorithms are applied to solve the variational problem to obtain each modal component. In this step, the algorithm iteratively updates the modal components and their corresponding center frequencies until the convergence condition is met or the maximum number of iterations is reached.

[0039] Exemplarily, in step S4, the time series of the photovoltaic output power is sequence-encoded to obtain a photovoltaic output power time series encoding feature vector. In one embodiment, the time series of the photovoltaic output power is sequence-encoded to obtain a photovoltaic output power time series encoding feature vector, including: inputting the time series of the photovoltaic output power into a sequence encoder based on a forward LSTM model to obtain the photovoltaic output power time series encoding feature vector. It should be understood that since the photovoltaic output data has the dynamic characteristics and change rules of the time series in the time dimension, there is usually a strong autocorrelation. Directly using a simple statistical method or linear model may cause the model to be too simple to capture the true structure of the data (underfitting), or overfit the noise in the training data, and perform poorly on new data (overfitting). The LSTM model can balance the relationship between the length of memory by adjusting internal parameters, and will neither ignore important historical information nor be overly sensitive to short-term fluctuations, so these problems can be alleviated to a certain extent. Therefore, by using a sequence encoder based on a forward LSTM model to extract the time series features of the photovoltaic output power, the time series correlation feature information of the photovoltaic output power in the time dimension can be effectively captured.

[0040] Exemplarily, in step S5, the multivariate time-series encoding feature vector of the environmental factor and the photovoltaic output power time-series encoding feature vector are subjected to feature interaction processing based on semantic flow field modulation to obtain a photovoltaic output power-environmental factor fine-grained alignment feature response representation. It should be understood that since the multivariate time-series encoding feature vector of the environmental factor and the photovoltaic output power time-series encoding feature vector respectively contain multivariate time-series correlation feature information and photovoltaic output power time-series feature information about multiple environmental factors. In order to capture the complex time-series correlation relationship between environmental factors and photovoltaic output to perform short-term prediction of photovoltaic output, it is necessary to perform feature interaction response analysis and semantic information fusion on the multivariate time-series encoding feature vector of the environmental factor and the photovoltaic output power time-series encoding feature vector. However, when processing data from different sources or different modalities, directly fusing these features may result in information loss or inaccurate representation. Secondly, traditional feature fusion and interactive response analysis methods are difficult to cross the "semantic gap" from low-level perceptual features to high-level semantic concepts, which will cause the time-series feature response representation between photovoltaic output power and multivariate environmental factors to be inaccurate, thereby affecting the subsequent photovoltaic output prediction. Based on this, in the technical solution of the present application, the environmental factor multivariate time series encoding feature vector and the photovoltaic output power time series encoding feature vector are further subjected to feature interaction processing based on semantic flow field modulation to obtain photovoltaic output power-environmental factor fine-grained aligned feature response representation. In particular, the feature interaction processing based on semantic flow field modulation can realize fine-grained aligned modulation between multi-source features (i.e., environmental factor multivariate time series correlation features and photovoltaic output power time series features) by constructing and utilizing semantic flow fields, and utilizes converter structures to perform deep response interaction on the modulated feature vectors to generate highly comprehensive and semantically rich photovoltaic output power-environmental factor fine-grained aligned feature response representations.

[0041] In one embodiment, Figure 3 As shown, the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector are subjected to feature interaction processing based on semantic flow field modulation to obtain a photovoltaic output power-environmental factor fine-grained aligned feature response representation, including: S51, constructing a photovoltaic output power-environmental factor semantic flow field between the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector; S52, based on the photovoltaic output power-environmental factor semantic flow field, performing feature alignment-type fine-grained response encoding on the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector to obtain the photovoltaic output power-environmental factor fine-grained aligned feature response representation.

[0042] In one embodiment, in step S51, constructing the photovoltaic output power-environmental factor semantic flow field between the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector includes: constructing the photovoltaic output power-environmental factor semantic flow field between the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector to obtain the photovoltaic output power-environmental factor semantic flow field matrix as the photovoltaic output power-environmental factor semantic flow field. Specifically, the process can be expressed as follows:

[0043] v ′1 =Sigmoid[Conv 1×1 (v1)]

[0044] v ′2 =Sigmoid[Conv 1×1 (v2)]

[0045]

[0046] Wherein, v1 and v2 are respectively the multivariate time series encoding feature vector of the environmental factor and the time series encoding feature vector of the photovoltaic output power, Conv 1×1 (·) is the point convolution processing, Sigmoid (·) is the Sigmoid activation function, v ′1 and v ′2 are respectively the multivariate time series coding feature vector of the environmental factors after modulation and the time series coding feature vector of the photovoltaic output power after modulation, L is the length of the time series coding feature vector of the photovoltaic output power after modulation, Conv 5×5 (·) is a convolution layer with a convolution kernel of 5×5. is matrix multiplication, Conv 3×3 (·) is a convolution layer with a convolution kernel of 3×3, Upsampling(·) represents upsampling processing, and Ω is the photovoltaic output power-environmental factor semantic flow field matrix.

[0047] In one embodiment, Figure 4As shown, in step S52, based on the photovoltaic output power-environmental factor semantic flow field, the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector are subjected to feature alignment fine-grained response coding to obtain the photovoltaic output power-environmental factor fine-grained aligned feature response representation, including: S521, based on the photovoltaic output power-environmental factor semantic flow field, the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector are subjected to feature alignment to obtain the aligned environmental factor multivariate time series coding feature vector and the aligned photovoltaic output power multivariate time series coding feature vector. power time series encoding feature vector; S522, linearly transform the aligned environmental factor multivariate time series encoding feature vector to obtain an environmental factor multivariate time series query vector and an environmental factor multivariate time series value vector; S523, linearly transform the aligned photovoltaic output power time series encoding feature vector to obtain a photovoltaic output power time series key vector; S524, perform fine-grained response association encoding on the environmental factor multivariate time series query vector, the environmental factor multivariate time series value vector and the photovoltaic output power time series key vector to obtain the photovoltaic output power-environmental factor fine-grained alignment feature response representation.

[0048] In one embodiment, in step S521, based on the photovoltaic output power-environmental factor semantic flow field, the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector are respectively feature aligned to obtain the aligned environmental factor multivariate time series coding feature vector and the aligned photovoltaic output power time series coding feature vector, including: multiplying the photovoltaic output power-environmental factor semantic flow field matrix with the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector to obtain the aligned environmental factor multivariate time series coding feature vector and the aligned photovoltaic output power time series coding feature vector. Specifically, the process can be expressed by the formula:

[0049]

[0050] Among them, v ′1 and v ′2 are respectively the multivariate temporal coding feature vector of the environmental factors after modulation and the temporal coding feature vector of the photovoltaic output power after modulation, Ω is the semantic flow field matrix of photovoltaic output power-environmental factors, and v 1t and v 2t They are respectively the multivariate time series encoding feature vector of environmental factors after alignment and the time series encoding feature vector of photovoltaic output power after alignment.

[0051] Exemplarily, in step S522 and step S523, the aligned environmental factor multivariate time series encoding feature vector is linearly transformed to obtain an environmental factor multivariate time series query vector and an environmental factor multivariate time series value vector, and the aligned photovoltaic output power time series encoding feature vector is linearly transformed to obtain a photovoltaic output power time series key vector. Specifically, the process can be expressed as:

[0052]

[0053] Among them, v 1v and v 2v are the time series encoding feature vectors of the environmental factors after alignment and the time series encoding feature vectors of the photovoltaic output power after alignment, respectively. 1q and b 1q are the weight matrix and bias vector of multivariate time series query of environmental factors, respectively. 1v and b 1v are the weight matrix of the multivariate time series values ​​of environmental factors and the bias vector of the multivariate time series values ​​of environmental factors, respectively. 2k and b 2k are the photovoltaic output power timing key weight matrix and photovoltaic output power timing key bias vector, V 1q and V 1v are respectively the environmental factor multivariate time series query vector and the environmental factor multivariate time series value vector, V 2k is the photovoltaic output power timing key vector.

[0054] In one embodiment, in step S524, the environmental factor multivariate time series query vector, the environmental factor multivariate time series value vector and the photovoltaic output power time series key vector are fine-grained response associated encoded to obtain the photovoltaic output power-environmental factor fine-grained alignment feature response representation, including: inputting the environmental factor multivariate time series query vector, the environmental factor multivariate time series value vector and the photovoltaic output power time series key vector into a fine-grained response encoding module based on a converter structure to obtain a photovoltaic output power-environmental factor fine-grained alignment feature response encoding vector as the photovoltaic output power-environmental factor fine-grained alignment feature response representation. Specifically, the process can be expressed as follows:

[0055]

[0056] Among them, V 1q and V 1v are respectively the environmental factor multivariate time series query vector and the environmental factor multivariate time series value vector, V 2kis the photovoltaic output power time series key vector, T represents the transpose of the vector, d is the length of the photovoltaic output power time series key vector, softmax(·) is the Softmax function, V i A feature response encoding vector is generated for fine-grained alignment of the photovoltaic output power and environmental factors.

[0057] Specifically, the feature interaction processing based on semantic flow field modulation can capture the complex semantic relationship space between the environmental factor multivariate temporal coding feature vector and the photovoltaic output power temporal coding feature vector by constructing the semantic flow field between them, and then by performing feature alignment operations, it can more accurately identify and adjust the subtle differences between the environmental factor multivariate temporal correlation features and the photovoltaic output power temporal coding features, ensuring that the features extracted even under different conditions and contexts can accurately correspond, thereby improving the quality of feature interaction response analysis and the expressiveness of the photovoltaic output power-environmental factor fine-grained alignment feature response representation generated in the end. This step helps to solve the problem of photovoltaic output fluctuations caused by changes in environmental conditions, so that even if there are local deformations or scale changes, feature matching at a fine-grained level can be achieved. In addition, the feature interaction processing based on semantic flow field modulation can more deeply capture the subtle semantic relationship between the environmental factor multivariate temporal correlation features and the photovoltaic output power temporal coding features by introducing semantic flow fields and self-attention mechanisms, so that the model can better understand the true meaning and interactive influence of the data, and enhance the semantic richness of the photovoltaic output power-environmental factor fine-grained alignment feature response representation. This helps the distributed photovoltaic output prediction process in the substation area to more accurately simulate and predict the output of the photovoltaic system while taking into account a variety of environmental factors, providing a basis for subsequent short-term prediction of photovoltaic output power and providing more reliable data support for grid dispatching.

[0058] Exemplarily, in step S6, photovoltaic output prediction is performed based on the photovoltaic output power-environmental factor fine-grained alignment feature response representation to determine the photovoltaic output power short-term prediction value. In one embodiment, photovoltaic output prediction is performed based on the photovoltaic output power-environmental factor fine-grained alignment feature response representation to determine the photovoltaic output power short-term prediction value, including: inputting the photovoltaic output power-environmental factor fine-grained alignment feature response encoding vector into a decoder-based photovoltaic output prediction module to obtain the photovoltaic output power short-term prediction value. That is, the fine-grained alignment feature response representation between the photovoltaic output power time series characteristics and the environmental factor multivariate time series correlation characteristics is used to perform decoding regression, so as to perform distributed photovoltaic output prediction in the substation area and obtain the final photovoltaic output power short-term prediction value. Compared with the traditional scheme, this method not only takes into account the impact of environmental factors on photovoltaic output, but also further improves the accuracy and reliability of photovoltaic output prediction through feature engineering and deep learning models, providing strong support for energy management and scheduling of smart grids.

[0059] In a specific embodiment, the photovoltaic output power-environmental factor fine-grained alignment feature response encoding vector is input into a decoder-based photovoltaic output prediction module to obtain a photovoltaic output power short-term prediction value, including: first, defining an input layer to accept the photovoltaic output power-environmental factor fine-grained alignment feature response encoding vector. Next, adding several GRU layers after the input layer. The GRU layer is responsible for capturing long-term dependencies in time series data. The GRU unit contains two control structures, a reset gate and an update gate, which can effectively alleviate the gradient vanishing problem while keeping the information flow smooth. By stacking multiple GRU layers, the network can have stronger learning ability and better handle complex nonlinear mappings. Then, the output of the GRU layer is connected to a fully connected layer (also called a dense layer). The role of the fully connected layer is to convert the space output by the GRU layer into a form suitable for regression tasks. This usually means mapping high-dimensional features to low dimensions, such as directly predicting future photovoltaic output power values ​​or passing them to other components as intermediate representations. Finally, the output layer is responsible for generating the final prediction result, that is, the photovoltaic output power short-term prediction value.

[0060] Here, when the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector respectively represent the time series implicit correlation features of the environmental factor multivariate and the photovoltaic output power, after performing feature interaction responses based on semantic flow field modulation, the photovoltaic output power-environmental factor fine-grained aligned feature response coding vector will also have significant aligned feature response coding distribution spatial structure differences due to source time series distribution differences and data cross-modal differences, affecting the convergence consistency of the decoder, thereby affecting the accuracy of the photovoltaic output power short-term prediction value obtained by the photovoltaic output prediction module.

[0061] Therefore, in one example, the photovoltaic output power-environmental factor fine-grained alignment feature response encoding vector is optimized, and the optimization process includes:

[0062] The absolute value sum and the square root of the square sum of all eigenvalues ​​of the photovoltaic output power-environmental factor fine-grained alignment feature response encoding vector are calculated to obtain the first photovoltaic output power-environmental factor fine-grained alignment feature response encoding spatial structure value and the second photovoltaic output power-environmental factor fine-grained alignment feature response encoding spatial structure value, that is:

[0063]

[0064] Among them, v i represents the ith eigenvalue of the photovoltaic output power-environmental factor fine-grained alignment feature response encoding vector, w1 represents the first photovoltaic output power-environmental factor fine-grained alignment feature response encoding space structure value, and w2 represents the second photovoltaic output power-environmental factor fine-grained alignment feature response encoding space structure value;

[0065] Determine the number of eigenvalues ​​n of all eigenvalues ​​of the photovoltaic output power-environmental factor fine-grained alignment feature response coding vector, that is, the length of the photovoltaic output power-environmental factor fine-grained alignment feature response coding vector;

[0066] For each eigenvalue of the photovoltaic output power-environmental factor fine-grained aligned feature response encoding vector, the first photovoltaic output power-environmental factor fine-grained aligned feature response encoding spatial structure value minus the product of the eigenvalue and the number of eigenvalues ​​is calculated to obtain the first photovoltaic output power-environmental factor fine-grained aligned feature response encoding long-range dependency value x i =w1-v i ×n, where w1 represents the first photovoltaic output power-environmental factor fine-grained alignment feature response encoding spatial structure value, v i represents the ith eigenvalue of the photovoltaic output power-environmental factor fine-grained alignment feature response encoding vector, L represents the number of eigenvalues ​​of all eigenvalues ​​of the photovoltaic output power-environmental factor fine-grained alignment feature response encoding vector, x i Indicates the first PV output power-environmental factor fine-grained alignment feature response encoding long-range dependency value;

[0067] Calculate the product of the square root of the number of eigenvalues ​​multiplied by the eigenvalue minus the second photovoltaic output power-environmental factor fine-grained alignment feature response encoding spatial structure value to obtain the second photovoltaic output power-environmental factor fine-grained alignment feature response encoding long-range dependency value Where w2 represents the second photovoltaic output power-environmental factor fine-grained alignment feature response encoding spatial structure value, v i represents the ith eigenvalue of the PV output power-environmental factor fine-grained alignment feature response encoding vector, L represents the number of eigenvalues ​​of all eigenvalues ​​of the PV output power-environmental factor fine-grained alignment feature response encoding vector, y i Indicates the second PV output power-environmental factor fine-grained alignment feature response encoding long-range dependency value;

[0068] The index value calculated by taking the first photovoltaic output power-environmental factor fine-grained alignment feature response encoding long-range dependency value as the exponent of the natural constant and the inverse of the second photovoltaic output power-environmental factor fine-grained alignment feature response encoding long-range dependency value are weighted summed to obtain the optimized eigenvalue corresponding to each eigenvalue Among them, x i represents the first photovoltaic output power-environmental factor fine-grained alignment feature response encoding long-range dependency value, y i represents the second PV output power-environmental factor fine-grained alignment feature response encoding long-range dependency value, e represents a natural constant, β and γ represent weighted hyperparameters, and f ′i Indicates the optimized eigenvalue corresponding to each eigenvalue;

[0069] The optimized eigenvalues ​​are combined into an optimized photovoltaic output power-environmental factor fine-grained aligned feature response encoding vector.

[0070] That is, in view of the possible spatial structure loss of the feature set of the photovoltaic output power-environmental factor fine-grained aligned feature response coding vector in the high-dimensional space, which causes the decoder weight to implicitly infer spatial structure information based on the features, resulting in inconsistent convergence. A long-distance feature dependency relationship is established based on the spatial structure representation of the photovoltaic output power-environmental factor fine-grained aligned feature response coding vector relative to the overall feature scale of the photovoltaic output power-environmental factor fine-grained aligned feature response coding vector, so as to establish the feature local connectivity of the photovoltaic output power-environmental factor fine-grained aligned feature response coding vector, and to capture the spatial ambiguous information of the object feature value through the unstructured feature value point prediction of the photovoltaic output power-environmental factor fine-grained aligned feature response coding vector, thereby improving the spatial inductive bias perception ability of the feature set of the photovoltaic output power-environmental factor fine-grained aligned feature response coding vector, improving the convergence consistency of the decoder, and improving the accuracy of the photovoltaic output power short-term prediction value of the photovoltaic output power-environmental factor fine-grained aligned feature response coding vector obtained by the decoder-based photovoltaic output prediction module.

[0071] In summary, according to the data-driven distributed photovoltaic output prediction method of the embodiment of the present application, it is explained that it can collect time series data sets of environmental factors through sensor components, including time series data of solar irradiance, component temperature, air temperature, relative humidity and atmospheric pressure, and collect historical and real-time data of photovoltaic output power, and then use data processing and analysis algorithms based on artificial intelligence and deep learning to perform time series correlation analysis based on variational mode decomposition on these environmental factor time series data sets, so as to extract the implicit correlation feature information of the multivariate time series of environmental factors, and capture the time series characteristics of photovoltaic output power. Further, the interactive responsiveness feature information between the multivariate time series correlation semantics of environmental factors and the photovoltaic output time series semantics is used to predict the distributed photovoltaic output of the substation, and the final short-term prediction value of photovoltaic output power is obtained. Compared with traditional solutions, this method not only takes into account the impact of environmental factors on photovoltaic output, but also further improves the accuracy and reliability of photovoltaic output prediction through feature engineering and deep learning models, providing strong support for energy management and scheduling of smart grids.

[0072] Figure 5 Schematic block diagram of a data-driven distributed photovoltaic output prediction system for an embodiment of the present application. Figure 5 As shown, the data-driven distributed photovoltaic output prediction system 100 includes: an environmental factor data set acquisition module 110, which is used to acquire a time series data set of environmental factors collected by a sensor component, wherein the environmental factors include solar irradiance, component temperature, air temperature, relative humidity and atmospheric pressure; a photovoltaic output power acquisition module 120, which is used to acquire a time series of photovoltaic output power; an environmental factor multivariate time series encoding module 130, which is used to organize the time series data set of the environmental factors into a time series of solar irradiance, a time series of component temperature, a time series of air temperature, a time series of relative humidity and a time series of atmospheric pressure according to the parameter sample dimension, and then perform variational mode decomposition and Sequence encoding to obtain a multivariate time series encoding feature vector of environmental factors; a photovoltaic output power time series encoding module 140, which is used to sequence encode the time series of the photovoltaic output power to obtain a photovoltaic output power time series encoding feature vector; a photovoltaic output power-environmental factor feature interaction processing module 150, which is used to perform feature interaction processing based on semantic flow field modulation on the environmental factor multivariate time series encoding feature vector and the photovoltaic output power time series encoding feature vector to obtain a photovoltaic output power-environmental factor fine-grained aligned feature response representation; a photovoltaic output prediction module 160, which is used to perform photovoltaic output prediction based on the photovoltaic output power-environmental factor fine-grained aligned feature response representation to determine a short-term prediction value of the photovoltaic output power.

[0073] In one embodiment, the photovoltaic output power-environmental factor feature interaction processing module includes: a photovoltaic output power-environmental factor semantic flow field construction unit, which is used to construct a photovoltaic output power-environmental factor semantic flow field between the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector; a photovoltaic output power-environmental factor fine-grained alignment processing unit, which is used to perform feature-aligned fine-grained response encoding on the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector based on the photovoltaic output power-environmental factor semantic flow field to obtain the photovoltaic output power-environmental factor fine-grained alignment feature response representation.

[0074] In one embodiment, the environmental factor multivariate time series encoding module is used to: perform variational modal decomposition on the time series of the solar irradiance, the time series of the component temperature, the time series of the air temperature, the time series of the relative humidity, and the time series of the atmospheric pressure to obtain a set of solar irradiance intrinsic modal components, a set of component temperature intrinsic modal components, a set of air temperature intrinsic modal components, a set of relative humidity intrinsic modal components, and a set of atmospheric pressure intrinsic modal components; arrange the set of solar irradiance intrinsic modal components, the set of component temperature intrinsic modal components, the set of air temperature intrinsic modal components, the set of relative humidity intrinsic modal components, and the set of atmospheric pressure intrinsic modal components into an environmental multivariate intrinsic modal component aggregation matrix and then input the matrix into a convolutional layer-based environmental multivariate time series association encoder to obtain the environmental factor multivariate time series encoding feature vector.

[0075] Here, those skilled in the art can understand that the specific operations of each module and unit in the above-mentioned data-driven distributed photovoltaic output prediction system have been referred to above. Figures 1 to 4 It has been introduced in detail in the description of the data-driven distributed photovoltaic output prediction method in the substation, and therefore, its repeated description will be omitted.

[0076] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0077] It should be understood that the specific examples in this article are only intended to help those skilled in the art better understand the embodiments of the present application, rather than to limit the scope of the embodiments of the present application.

[0078] It should also be understood that in the various embodiments of the present application, the size of the serial number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0079] It should also be understood that the various implementation methods described in this specification can be implemented individually or in combination, and the embodiments of the present application are not limited to this.

[0080] Unless otherwise stated, all technical and scientific terms used in the embodiments of the present application are the same as the meanings generally understood by those skilled in the art of the technical field of the present application. The terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the scope of the present application. The term "and / or" used in the present application includes any and all combinations of one or more related listed items. The singular forms of "a kind of", "above" and "the" used in the embodiments of the present application and the appended claims are also intended to include majority forms, unless the context clearly indicates other meanings. In addition, the terms "first", "second" etc. are only used for description purposes and cannot be understood as indicating or suggesting relative importance.

[0081] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0082] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0083] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0084] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0085] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A data-driven distributed photovoltaic output prediction method for a substation, characterized in that: include: Acquire a time series data set of environmental factors collected by the sensor assembly, the environmental factors including solar irradiance, assembly temperature, air temperature, relative humidity, and atmospheric pressure; Get the time series of photovoltaic output power; After the time series data set of the environmental factors is sorted into a time series of solar irradiance, a time series of component temperature, a time series of air temperature, a time series of relative humidity, and a time series of atmospheric pressure according to the parameter sample dimension, variational mode decomposition and sequence encoding are performed on them to obtain a multivariate time series encoding feature vector of the environmental factors; Performing sequence encoding on the time series of the photovoltaic output power to obtain a photovoltaic output power time series encoding feature vector; The environmental factor multivariate time-series coding feature vector and the photovoltaic output power time-series coding feature vector are subjected to feature interaction processing based on semantic flow field modulation to obtain a photovoltaic output power-environmental factor fine-grained aligned feature response representation, including: constructing a photovoltaic output power-environmental factor semantic flow field between the environmental factor multivariate time-series coding feature vector and the photovoltaic output power time-series coding feature vector; based on the photovoltaic output power-environmental factor semantic flow field, performing feature alignment-type fine-grained response encoding on the environmental factor multivariate time-series coding feature vector and the photovoltaic output power time-series coding feature vector to obtain the photovoltaic output power-environmental factor fine-grained aligned feature response representation; Photovoltaic output prediction is performed based on the photovoltaic output power-environmental factor fine-grained alignment feature response representation to determine a short-term prediction value of photovoltaic output power.

2. The data-driven distributed photovoltaic output prediction method according to claim 1 is characterized in that: After the time series data set of the environmental factors is sorted into the time series of solar irradiance, the time series of component temperature, the time series of air temperature, the time series of relative humidity, and the time series of atmospheric pressure according to the parameter sample dimension, variational mode decomposition and sequence encoding are performed on them to obtain the multivariate time series encoding feature vector of the environmental factors, including: Performing variational modal decomposition on the time series of solar irradiance, the time series of component temperature, the time series of air temperature, the time series of relative humidity, and the time series of atmospheric pressure to obtain a set of solar irradiance intrinsic modal components, a set of component temperature eigenmodal components, a set of air temperature eigenmodal components, a set of relative humidity eigenmodal components, and a set of atmospheric pressure eigenmodal components; The set of solar irradiance intrinsic modal components, the set of component temperature intrinsic modal components, the set of air temperature intrinsic modal components, the set of relative humidity intrinsic modal components and the set of atmospheric pressure intrinsic modal components are arranged into an environmental multivariate intrinsic modal component aggregation matrix and then input into a convolutional layer-based environmental multivariate time series association encoder to obtain the environmental factor multivariate time series encoding feature vector.

3. The data-driven distributed photovoltaic output prediction method according to claim 2 is characterized in that: The time series of the photovoltaic output power is sequence-encoded to obtain a photovoltaic output power time series encoding feature vector, including: inputting the time series of the photovoltaic output power into a sequence encoder based on a forward LSTM model to obtain the photovoltaic output power time series encoding feature vector.

4. The data-driven distributed photovoltaic output prediction method according to claim 3 is characterized in that: Constructing a photovoltaic output power-environmental factor semantic flow field between the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector, including: constructing a photovoltaic output power-environmental factor semantic flow field between the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector to obtain a photovoltaic output power-environmental factor semantic flow field matrix as the photovoltaic output power-environmental factor semantic flow field.

5. The data-driven distributed photovoltaic output prediction method according to claim 4 is characterized in that: Based on the photovoltaic output power-environmental factor semantic flow field, the environmental factor multivariate time series encoding feature vector and the photovoltaic output power time series encoding feature vector are subjected to feature alignment fine-grained response encoding to obtain the photovoltaic output power-environmental factor fine-grained aligned feature response representation, including: Based on the photovoltaic output power-environmental factor semantic flow field, feature alignment is performed on the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector to obtain an aligned environmental factor multivariate time series coding feature vector and an aligned photovoltaic output power time series coding feature vector; Performing a linear transformation on the aligned environmental factor multivariate time series encoding feature vector to obtain an environmental factor multivariate time series query vector and an environmental factor multivariate time series value vector; Performing a linear transformation on the aligned photovoltaic output power time series encoding feature vector to obtain a photovoltaic output power time series key vector; Fine-grained response association encoding is performed on the environmental factor multivariate time series query vector, the environmental factor multivariate time series value vector and the photovoltaic output power time series key vector to obtain the photovoltaic output power-environmental factor fine-grained alignment feature response representation.

6. The data-driven distributed photovoltaic output prediction method according to claim 5 is characterized in that: Based on the photovoltaic output power-environmental factor semantic flow field, feature alignment is performed on the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector to obtain an aligned environmental factor multivariate time series coding feature vector and an aligned photovoltaic output power time series coding feature vector, including: The photovoltaic output power-environmental factor semantic flow field matrix is ​​multiplied with the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector respectively to obtain the aligned environmental factor multivariate time series coding feature vector and the aligned photovoltaic output power time series coding feature vector.

7. The data-driven distributed photovoltaic output prediction method according to claim 6 is characterized in that: The environmental factor multivariate time series query vector, the environmental factor multivariate time series value vector and the photovoltaic output power time series key vector are subjected to fine-grained response association encoding to obtain the photovoltaic output power-environmental factor fine-grained alignment feature response representation, including: The environmental factor multivariate time series query vector, the environmental factor multivariate time series value vector and the photovoltaic output power time series key vector are input into a fine-grained response encoding module based on a converter structure to obtain a photovoltaic output power-environmental factor fine-grained aligned feature response encoding vector as the photovoltaic output power-environmental factor fine-grained aligned feature response representation.

8. The data-driven distributed photovoltaic output prediction method according to claim 7 is characterized in that: Photovoltaic output prediction is performed based on the photovoltaic output power-environmental factor fine-grained alignment feature response representation to determine a short-term prediction value of photovoltaic output power, including: inputting the photovoltaic output power-environmental factor fine-grained alignment feature response encoding vector into a decoder-based photovoltaic output prediction module to obtain a short-term prediction value of photovoltaic output power.

9. A data-driven distributed photovoltaic output prediction method for a substation, characterized in that: include: An environmental factor data set acquisition module is used to acquire a time series data set of environmental factors collected by the sensor component, wherein the environmental factors include solar irradiance, component temperature, air temperature, relative humidity and atmospheric pressure; Photovoltaic output power acquisition module, used to obtain the time series of photovoltaic output power; An environmental factor multivariate time series encoding module is used to organize the time series data set of the environmental factor into a time series of solar irradiance, a time series of component temperature, a time series of air temperature, a time series of relative humidity, and a time series of atmospheric pressure according to the parameter sample dimension, and then perform variational mode decomposition and sequence encoding on them to obtain a multivariate time series encoding feature vector of the environmental factor; A photovoltaic output power time series encoding module, used for performing sequence encoding on the time series of the photovoltaic output power to obtain a photovoltaic output power time series encoding feature vector; A photovoltaic output power-environmental factor feature interaction processing module is used to perform feature interaction processing based on semantic flow field modulation on the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector to obtain a photovoltaic output power-environmental factor fine-grained aligned feature response representation, including: a photovoltaic output power-environmental factor semantic flow field construction unit, which is used to construct a photovoltaic output power-environmental factor semantic flow field between the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector; a photovoltaic output power-environmental factor fine-grained alignment processing unit, which is used to perform feature alignment-type fine-grained response encoding on the environmental factor multivariate time series coding feature vector and the photovoltaic output power time series coding feature vector based on the photovoltaic output power-environmental factor semantic flow field to obtain the photovoltaic output power-environmental factor fine-grained aligned feature response representation; The photovoltaic output prediction module is used to perform photovoltaic output prediction based on the photovoltaic output power-environmental factor fine-grained alignment feature response representation to determine a short-term prediction value of photovoltaic output power.

10. The data-driven distributed photovoltaic output prediction method according to claim 9, characterized in that: The environmental factor multivariate temporal coding module is used to: Performing variational modal decomposition on the time series of solar irradiance, the time series of component temperature, the time series of air temperature, the time series of relative humidity, and the time series of atmospheric pressure to obtain a set of solar irradiance intrinsic modal components, a set of component temperature eigenmodal components, a set of air temperature eigenmodal components, a set of relative humidity eigenmodal components, and a set of atmospheric pressure eigenmodal components; The set of solar irradiance intrinsic modal components, the set of component temperature intrinsic modal components, the set of air temperature intrinsic modal components, the set of relative humidity intrinsic modal components and the set of atmospheric pressure intrinsic modal components are arranged into an environmental multivariate intrinsic modal component aggregation matrix and then input into a convolutional layer-based environmental multivariate time series association encoder to obtain the environmental factor multivariate time series encoding feature vector.

Images