Photovoltaic power prediction method fusing irradiance physical information
By adopting a two-stage photovoltaic power prediction method that combines irradiance physical information and multi-scale time series decomposition, the optimization difficulties and weak adaptability of existing models in photovoltaic power prediction are solved, and more accurate photovoltaic power prediction is achieved.
Patent Information
- Application Number
- CN202511821985.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-05
- Publication Date
- 2026-03-10
AI Technical Summary
Existing photovoltaic power prediction models face optimization difficulties when integrating physical models and data-driven models, and struggle to effectively decouple complex physical processes from multidimensional spatiotemporal feature learning, leading to inaccurate prediction results.
The photovoltaic power prediction method adopts a two-stage architecture. First, it predicts irradiance by embedding a time-series memory neural network with physical information of irradiance. Then, it uses a photovoltaic power prediction model with multi-scale time series decomposition and adaptive multiprocessor synthesis to predict power, combining physical mechanisms and data-driven methods.
It achieves more accurate photovoltaic power prediction, can adapt to different weather conditions, improves the physical consistency and generalization ability of the model, and enhances prediction accuracy.
Smart Images
Figure CN121638562A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of photovoltaic power station power prediction, and in particular to a photovoltaic power prediction method fusing irradiance physical information. BACKGROUND
[0002] Photovoltaic energy is a clean and sustainable form of energy. However, the power output of a photovoltaic power generation system has inherent volatility and intermittency, and its output efficiency is highly dependent on a series of uncontrollable and dynamically changing meteorological factors such as solar irradiance, ambient temperature, cloud coverage, humidity, and the like. The uncertainty of photovoltaic energy poses great challenges to the stable operation of the power system, the economic benefits of the power grid dispatching and the energy market. Therefore, developing a high-precision photovoltaic power prediction technology is of great importance to ensuring the safety of the power grid, optimizing energy dispatching, and improving the economic value of photovoltaic power stations.
[0003] At present, the related technology can realize the prediction of photovoltaic power through three types of models, namely, physical models, data-driven models, and hybrid models. The physical model is established based on the physical description of the energy conversion process of the photovoltaic system, and the modeling process is extremely complex, requiring accurate power station parameters. In addition, data-driven models based on machine learning and deep learning, such as long short-term memory network (LSTM), bidirectional long short-term memory network (Bi-LSTM), and hybrid model of convolutional neural network and LSTM (CNN-LSTM), can learn complex nonlinear relationships from historical data and achieve good prediction results under certain conditions. However, due to the "black box" nature of the model and the strong dependence on the distribution of historical data, this type of model has deep limitations. Specifically, pure data-driven models may produce prediction results that violate physical common sense. Moreover, photovoltaic power is simultaneously affected by minute-level weather fluctuations and hour-level daily trends. Traditional models are difficult to adaptively extract and fuse dynamic features at different time scales, and cannot accurately mine the complex synergistic relationships between meteorological variables, resulting in inaccurate prediction of photovoltaic power changes.
[0004] On the other hand, some related technologies adopt a way of fusing physical models and data-driven models. In order to fuse the physical model and the data-driven model, the physical information neural network (PINN) embeds the physical law as a constraint into the loss function of the neural network. However, this way usually uses PINN for end-to-end prediction, which makes the optimization process of the model extremely difficult.
[0005] Therefore, there is an urgent need for a photovoltaic power prediction method that can make more use of meteorological information, has a more concise and efficient model structure, and can effectively decouple complex physical process modeling and multi-dimensional spatiotemporal feature learning. SUMMARY
[0006] The application provides a photovoltaic power prediction method fusing irradiance physical information, to solve the problems of lack of physical consistency and insufficient generalization ability of a pure data-driven model, and complexity and weak adaptability of a traditional physical model in the related art, and the technical solution is as follows: In a first aspect, the application provides a photovoltaic power prediction method fusing irradiance physical information, comprising: obtaining a historical operation data sequence of a target photovoltaic power station in a historical period; preprocessing the historical operation data sequence; inputting the preprocessed historical operation data sequence into a trained irradiance prediction model, and obtaining a predicted irradiance sequence of a prediction period through processing of the irradiance prediction model; inputting the preprocessed historical operation data sequence and the predicted irradiance sequence as feature variables into a trained photovoltaic power prediction model, and obtaining a predicted power sequence of the prediction period through processing of the photovoltaic power prediction model.
[0007] In an optional solution of the first aspect, the preprocessing of the operation data sequence comprises: detecting abnormal values of each type of data in the historical operation data sequence respectively, and eliminating the abnormal values of each type of data respectively; identifying missing values of each type of data in the historical operation data sequence, and filling the missing values through interpolation calculation respectively; performing normalization processing on each numerical value type parameter in the historical operation data sequence respectively, and outputting the normalized data.
[0008] In an optional solution of the first aspect, the training process of the irradiance prediction model comprises: obtaining a sample operation data sequence of the target photovoltaic power station in a sample period, the sample operation data sequence comprising actual irradiance, irradiance parameters, physical parameters of the target photovoltaic power station at each sample time in the sample period, and theoretical irradiance calculated according to the irradiance parameters and the physical parameters; constructing a sample set for training based on the sample operation data sequence, to train the irradiance prediction model through the sample set; calculating the predicted irradiance of the next time of the sample time through the irradiance prediction model based on the actual irradiance and the irradiance parameters of the sample time; constructing a data loss term of the irradiance prediction model according to the difference between the predicted irradiance and the actual irradiance of the corresponding time, and constructing a physical loss term of the irradiance prediction model according to the difference between the predicted irradiance and the theoretical irradiance of the corresponding time; The loss function of the irradiance prediction model is obtained by weighted fusion of the data loss term and the physical loss term, and the weight parameters of the irradiance prediction model are optimized according to the loss function until convergence, so as to obtain a trained irradiance prediction model.
[0009] In an optional implementation of the first aspect, the irradiance parameters include direct normal irradiance, global horizontal irradiance, sky diffuse irradiance, and extraterrestrial normal irradiance, and the physical parameters of the target photovoltaic power station include an incidence angle of the sun relative to the photovoltaic panel, a ground albedo, a photovoltaic panel tilt angle, and a solar zenith angle. The step of calculating the theoretical irradiance specifically includes: A direct radiation component is calculated based on the incidence angle and the direct normal irradiance; A ground reflection component is calculated based on the global horizontal irradiance, the ground albedo, and the photovoltaic panel tilt angle; A sky diffuse component is calculated based on the sky diffuse irradiance, the direct normal irradiance, the extraterrestrial normal irradiance, the photovoltaic panel tilt angle, the incidence angle, and the solar zenith angle; The theoretical irradiance at the corresponding moment is calculated based on the direct radiation component, the ground reflection component, the sky diffuse component, and the incidence angle correction factor corresponding to each type of component.
[0010] In an optional implementation of the first aspect, the photovoltaic power prediction model includes a multi-scale time series decomposition module, a grouping star aggregation-redistribution module, and an adaptive multi-processor synthesis module. The predicted power sequence of the prediction period is obtained by processing the photovoltaic power prediction model, including: Each type of feature variable input is respectively subjected to down-sampling processing by the multi-scale time series decomposition module to obtain features of multiple different time scales, and a hybrid time scale feature vector is fused based on the features of multiple different time scales; The grouping star aggregation-redistribution module processes the system-level fusion feature vector obtained based on each type of feature variable input to obtain an updated fusion feature vector by splicing the system-level fusion feature vector with each type of feature variable input. The adaptive multi-processor synthesis module processes the hybrid time scale feature vector and the fusion feature vector to obtain the predicted power sequence of the prediction period.
[0011] In an optional implementation of the first aspect, the grouping star aggregation-redistribution module processes the system-level fusion feature vector obtained based on each type of feature variable input to obtain an updated fusion feature vector by splicing the system-level fusion feature vector with each type of feature variable input. The grouping star aggregation-redistribution module groups each type of feature variable according to the physical meaning of the inputted feature variable, and obtains a group-level feature vector of each group; The system-level fusion feature vector is spliced with each type of feature variable, and the spliced result is fused, and the fused result is connected with each type of feature variable through a residual connection to obtain an updated fusion feature vector.
[0012] In an optional implementation of the first aspect, the adaptive multi-processor synthesis module processes the hybrid time-scale feature vector and the fusion feature vector to obtain a predicted power sequence of a prediction period, and the processing includes: The hybrid time-scale feature vector is processed through a gating network to output a weight vector; The weight of each processor in the adaptive multi-processor synthesis module is determined based on the weight vector; Each processor processes the fusion feature vector to obtain a feature prediction sequence of a prediction period; The feature prediction sequences predicted by each processor are weighted and summed based on the weight of each processor to output a final predicted power sequence.
[0013] In a second aspect, the present application further provides a photovoltaic power prediction device fusing irradiance physical information, including: A data acquisition module is configured to acquire a historical operation data sequence of a target photovoltaic power station in a historical period; A data preprocessing module is configured to preprocess the historical operation data sequence; An irradiance prediction module is configured to input the preprocessed historical operation data sequence into a trained irradiance prediction model, and process the preprocessed historical operation data sequence through the irradiance prediction model to obtain a predicted irradiance sequence of a prediction period; A power prediction module is configured to input the preprocessed historical operation data sequence and the predicted irradiance sequence as feature variables into a trained photovoltaic power prediction model, and process the preprocessed historical operation data sequence and the predicted irradiance sequence through the photovoltaic power prediction model to obtain a predicted power sequence of a prediction period.
[0014] In a third aspect, the present application further provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method provided in the first aspect or any one of the implementation manners of the first aspect.
[0015] In a fourth aspect, the present application also provides a non-transitory computer-readable storage medium having stored thereon a computer program, which, when executed by a processor, implements the method provided by the first aspect of the present application or any of the implementation manners of the first aspect.
[0016] The technical scheme provided by some embodiments of the present application has at least the following beneficial effects: The present application realizes the combination of physical mechanism and data driving through a two-stage architecture. The irradiance prediction model in the first stage provides a prediction irradiance sequence with stronger physical consistency for the second-stage prediction of the power of the photovoltaic power station, and thus enables the photovoltaic power prediction model in the second stage to focus more on learning the complex nonlinear dynamics in the power conversion process.
[0017] In addition, in the prediction in the second stage, the present application sets a multi-scale time series decomposition module to enable the model to distinguish long-term trends and short-term fluctuations, and a grouping star aggregation-redistribution module to enable the model to efficiently capture the mutual influence between all meteorological variables through the grouping-aggregation-redistribution mechanism. The adaptive multi-processor synthesis module enables the model to dynamically adjust its prediction strategy according to the input, and through the prediction of each processor in the adaptive multi-processor synthesis module and the adjustment of the weight of each processor, the model can more finely cope with changing weather patterns to cope with the meteorological characteristics of different weather conditions, and more accurate power prediction can be achieved under different climate conditions. BRIEF DESCRIPTION OF DRAWINGS
[0018] In order to more clearly illustrate the technical solutions in the present application or related art, the following will briefly introduce the drawings needed to be used in the embodiments or related art descriptions. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.
[0019] Figure 1 is a flowchart of a photovoltaic power prediction method provided by an embodiment of the present application, which fuses irradiance physical information; Figure 2 is a comparison diagram of the effect of the irradiance prediction model provided by an embodiment of the present application and other models; Figure 3 is one of the comparison diagrams of the effect of the photovoltaic power prediction model provided by an embodiment of the present application and other models; Figure 4 is the second of the comparison diagrams of the effect of the photovoltaic power prediction model provided by an embodiment of the present application and other models; Figure 5 is a comparison diagram of the power prediction indicators of the model provided by an embodiment of the present application and other models. Figure 6 is a structural schematic diagram of a photovoltaic power prediction method device fusing irradiance physical information provided by an embodiment of the present application; Figure 7 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION
[0020] To make the objectives, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.
[0021] The terms “include” and “have” and any variations thereof in the specification and claims of the present application and the above-described drawings are intended to cover the non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or modules is not limited to the listed steps or modules, but can optionally further include steps or modules not listed or can optionally further include other steps or modules inherent to the process, method, product or device.
[0022] It should be noted that the terms “first” and “second” involved in the present application are only to distinguish similar objects, and do not represent a specific order of the objects. Understandably, “first” and “second” can be interchanged in a specific order or sequence as allowed. It should be understood that the objects distinguished by “first” and “second” can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those described or illustrated herein.
[0023] It should be noted that the physical process of the photovoltaic system, such as modeling the irradiance on the surface of the photovoltaic module, has a great influence on the generated power. Due to the focus on a single time scale and a single relationship between each meteorological factor and power, the related art is difficult to fully capture key information at different time scales and the synergistic relationship between each influencing feature, limiting the performance improvement of the model in handling complex dynamic changes.
[0024] The present application can cope with changing weather conditions by organically coupling photovoltaic system physical mechanism data and meteorological data. The present application will be described in detail in combination with specific embodiments.
[0025] Next, in combination with Figure 1 , a photovoltaic power prediction method fusing irradiance physical information provided by an embodiment of the present application is introduced. For details, please refer to Figure 1 , Figure 1A flowchart of a photovoltaic power prediction method provided by an embodiment of the present application is shown. As shown in Figure 1 The method comprises the following steps: S101, obtaining a historical operation data sequence of a target photovoltaic power station in a historical period; S102, preprocessing the historical operation data sequence; S103, inputting the preprocessed historical operation data sequence into a trained irradiance prediction model, and obtaining a predicted irradiance sequence of a prediction period by processing the irradiance prediction model; S104, inputting the preprocessed historical operation data sequence and the predicted irradiance sequence into a trained photovoltaic power prediction model as feature variables, and obtaining a predicted power sequence of the prediction period by processing the photovoltaic power prediction model.
[0026] It should be noted that the embodiment of the present application can realize photovoltaic power prediction of a prediction period based on historical operation data of a photovoltaic power station in a historical period according to the method of S101-S104, and obtain a predicted power sequence of the prediction period. For example, in S101, operation data in the period from August 24, 2021 to August 1, 2023 can be collected, and a prediction period after August 1, 2023 is predicted. The length of the prediction period can be adjusted according to actual needs, for example, set to a short period of 15 minutes, 30 minutes or 1 hour, and the predicted power at each time point in the short period is predicted to obtain a predicted power sequence of the short period.
[0027] It can be understood that the predicted power data can be used as new historical operation data to recursively predict subsequent prediction periods, so as to gradually predict a predicted power sequence of a subsequent long period.
[0028] Specifically, the historical operation data sequence in the historical period in S101 includes parameters generated and recorded by the target photovoltaic power station during operation in the corresponding period, including but not limited to irradiance-related parameters and environment-related parameters. The irradiance-related parameters include actual power, actual irradiance irradiance parameters, physical parameters of the target photovoltaic power station, and theoretical irradiance calculated according to the irradiance parameters and the physical parameters.
[0029] POA means Plane of Array, which represents a physical plane where a photovoltaic module or panel is located.
[0030] Specifically, the irradiance parameters include Global Horizontal Irradiance (GHI), Direct Normal Irradiance (DNI), Diffuse Horizontal Irradiance (DHI), and the physical parameters of the target photovoltaic power station include the incident angle of the sun relative to the photovoltaic panel, the ground albedo, the inclination angle of the photovoltaic panel, and the solar zenith angle. Specifically, the environment-related parameters include, but are not limited to, environmental temperature, relative humidity, wind direction, and rainfall, etc.
[0031] Further, S102 can preprocess each type of data in the acquired historical operation data sequence, including the steps of: respectively detecting outliers of each type of data in the historical operation data sequence, and respectively eliminating outliers of each type of data; identifying missing values of each type of data in the historical operation data sequence, and respectively filling the missing values by interpolation calculation; respectively performing normalization processing on each type of numerical parameter in the historical operation data sequence, and outputting the normalized data.
[0032] Specifically, the formula of the normalization process is: ; wherein, denotes the normalized data, denotes the original data, denotes the minimum value in the data, denotes the maximum value in the data.
[0033] Further, S103 and the subsequent steps are executed based on the normalized operation data sequence.
[0034] Specifically, the irradiance prediction model in S103 is a Physical informed temporal memory neural network (PITMNN) embedded with irradiance physical information, which can predict the predicted irradiance sequence of the future period based on the irradiance-related parameters in the historical operation data sequence.
[0035] In some embodiments, the training process of the irradiance prediction model specifically includes: S1031, acquire a sample operation data sequence of the target photovoltaic power station in a sample period, the sample operation data sequence including actual irradiance, irradiance parameters, physical parameters of the target photovoltaic power station at each sample time in the sample period, and theoretical irradiance calculated according to the irradiance parameters and the physical parameters.
[0036] S1032, construct a sample set for training based on the sample operation data sequence, so as to train the irradiance prediction model through the sample set.
[0037] Specifically, the sample period can be any historical period, and the sample operation data includes not only the irradiance parameters but also the physical information of the photovoltaic power station related to irradiance.
[0038] Specifically, the irradiance parameters include direct normal irradiance, global horizontal irradiance, sky diffuse irradiance, and extraterrestrial normal irradiance, and the physical parameters of the target photovoltaic power station include angle of incidence (AOI) of the sun relative to the photovoltaic panel, ground albedo , inclination of the photovoltaic panel , and solar zenith angle .
[0039] By embedding the physical information into the loss function of the model, the model can learn the influence of different physical information on irradiance in the training process, and thus predict more accurate irradiance prediction results.
[0040] In some embodiments, the step of calculating the theoretical irradiance specifically includes: calculating a direct radiation component based on the angle of incidence and the direct normal irradiance , applying the formula: ; calculating a ground reflection component based on the global horizontal irradiance, the ground albedo, and the inclination of the photovoltaic panel , applying the formula: ; calculating a sky diffuse component based on the sky diffuse irradiance, the direct normal irradiance, the extraterrestrial normal irradiance, the inclination of the photovoltaic panel, the angle of incidence, and the solar zenith angle ; the sky diffuse irradiance can be decomposed into an isotropic part and an anisotropic part around the sun, and the formula is: ; wherein, is an anisotropic index, is the extraterrestrial normal irradiance, For projection ratio.
[0041] Further, based on the direct radiation component, the ground reflection component, the sky scattering component and the incidence angle modifier corresponding to each component, the theoretical irradiance at the corresponding moment is calculated , the formula is applied: ; Among them, is the incidence angle modifier corresponding to the direct radiation component, is the incidence angle modifier corresponding to the sky scattering component, is the incidence angle modifier corresponding to the ground reflection component (Incidence Angle Modifier), which is used to quantify the optical loss of light when passing through the glass cover plate of the photovoltaic module.
[0042] Exemplarily, the incidence angle modifier can be calculated based on the physical model of Fresnel equation and Bouguer-Lambert-Beer law, and the formula is applied: ; ; ; ; ; Among them, is the refraction angle, n is the effective refractive index corresponding to the photovoltaic panel, is the reflection loss, is the absorption loss corresponding to the photovoltaic panel, is the transmittance at normal incidence, K is the extinction coefficient, and L is the panel thickness corresponding to the photovoltaic panel.
[0043] For the incidence angle modifier corresponding to the sky scattering component and the ground reflection component, Schlick approximation model can be used for calculation. Specifically, the Schlick approximation formula suitable for point light source is spatially integrated to obtain the equivalent modifier suitable for diffuse reflection light.
[0044] S1033, through the irradiance prediction model, based on the actual irradiance of the sample moment, the irradiance parameter is calculated to obtain the predicted irradiance of the next moment of the sample moment ; S1034, according to the difference between the predicted irradiance and the actual irradiance POA at the corresponding moment, the data loss term of the irradiance prediction model is constructed , according to the difference between the predicted irradiance and the theoretical irradiance The difference is used to construct the physical loss term of the irradiance prediction model. . Specifically, physical residual loss Defined as predicted irradiance Compared with theoretical irradiance The mean square error (MSE) between them is calculated using the formula: ; Where N represents the total number of samples in the sample set, and i is the sample number.
[0045] S1035, the loss function of the irradiance prediction model is obtained by weighted fusion of the data loss term and the physical loss term: ; in, These are the weight parameters.
[0046] S1036, Optimize the weight parameters of the irradiance prediction model according to the loss function until convergence, and obtain the trained irradiance prediction model.
[0047] Specifically, during training, the weight parameters can be adjusted based on the values of the data loss and physical loss terms. At the beginning of each training cycle, the moving average of the data loss and physical loss over the most recent cycles can be calculated. Then, based on the principle of "the greater the loss, the higher the weight", new weight parameters are calculated to make the model pay more attention to the parts that are currently learning worse. Finally, to prevent drastic fluctuations in weights, the exponential moving average (EMA) algorithm is used to smoothly update the weights and renormalize them to ensure that the sum of the weights is 1.
[0048] Specifically, this can be achieved through a custom callback function (Adaptive Loss Balancer), thereby balancing data fitting with physical constraints.
[0049] After training in the above manner, the irradiance prediction model (PITMNN model) can generate physically consistent predicted irradiance sequences.
[0050] For example, such as Figure 2 As shown, Figure 2 The example illustrates a comparison between the irradiance prediction model provided in this application and other existing technology models (including physical models, LSTM models, and PINN models) based on samples under different weather conditions. Figure 2 The PITMNN model is the irradiance prediction model provided in the embodiments of this application.
[0051] Further, a step of performing S104, inputting the pre-processed historical operation data sequence and the predicted irradiance sequence as feature variables into the trained photovoltaic power prediction model, and processing the photovoltaic power prediction model to obtain a predicted power sequence of a prediction period.
[0052] In some embodiments, the photovoltaic power prediction model is specifically a Grouped Star Aggregate-Redistribute Adaptive Multi-Processor Network (MSGStar-AMS), and specifically includes a multi-scale time series decomposition module (MDM), a grouped star aggregate-redistribute module (GSTAR), and an adaptive multi-processor synthesis module (AMS).
[0053] Specifically, the photovoltaic power prediction model can be trained based on a sample set constructed based on sample operation data in a sample period and predicted irradiance calculated by the irradiance prediction model based on the sample operation data, and the photovoltaic power prediction model can be trained until the model converges, with actual power in the sample operation data as a sample label, and all other parameters in the sample operation data and the predicted irradiance as sample inputs.
[0054] Specifically, processing the photovoltaic power prediction model to obtain a predicted power sequence of a prediction period includes: S1041, performing down-sampling processing on each type of input feature variable by the multi-scale time series decomposition module to obtain features of multiple different time scales, and fusing the features of multiple different time scales to obtain a mixed time scale feature vector.
[0055] Specifically, the mixed time scale feature vector is obtained by performing down-sampling processing on the originally fine-grained input features to obtain coarse-grained features, and fusing the coarse-grained features into fine-grained features by a feedforward network.
[0056] S1042, processing each type of input feature variable by the grouped star aggregate-redistribute module to obtain a system-level fused feature vector, and concatenating the system-level fused feature vector with each type of input feature variable to obtain an updated fused feature vector.
[0057] Specifically, the Group Star Aggregation-Redistribution module groups the inputted physical meanings of each type of feature variable to obtain a group-level feature vector (Group Core Representation) for each group, such as a main meteorological group (e.g., direct driving factors such as irradiance and temperature), a "context environment group" (e.g., secondary influencing factors such as wind direction and humidity), and a "historical power group" (i.e., the target variable); Further, all group-level representation vectors are spliced to obtain a system-level fusion feature vector (System Core Representation) through a Multilayer Perceptron (MLP); Next, the system-level fusion feature vector is spliced with the inputted each type of feature variable, and the spliced result is fused. The fused result is connected with the inputted each type of feature variable in residual connection to obtain an updated fusion feature vector.
[0058] In this way, the fusion feature vector enables the model to better learn the global feature variation trend.
[0059] S1043, the adaptive multi-processor synthesis module processes the mixed time scale feature vector and the fusion feature vector to obtain a predicted power sequence of a prediction period.
[0060] It should be noted that the adaptive multi-processor synthesis module includes a group of simple parallel MLP prediction networks, referred to as "processors", each of which is a dedicated predictor for a certain specific data pattern (e.g., sunny smooth pattern, cloudy high-frequency fluctuation pattern, etc.).
[0061] In some embodiments, the processing process of the adaptive multi-processor synthesis module includes: S10431, the mixed time scale feature vector is processed through a gating network to output a weight vector.
[0062] Specifically, the gating network is composed of an MLP. For an MLP with an output dimension of k, the obtained weight vector includes k weights, and the number of output dimensions is the same as the number of processors performing the calculation.
[0063] S10432, the weight of each processor in the adaptive multi-processor synthesis module is determined based on the weight vector, and the formula is: ; ; wherein m is the mixed time scale feature vector, k is the number of processors, is a standard Gaussian noise, a learnable weight noise matrix, a calculated weight vector, a weight, , , are all functions.
[0064] S10433, the feature dimension of the fusion feature vector can be mapped to the length of the predicted sequence through a linear layer, and each processor processes a feature prediction sequence of a prediction period based on the fusion feature vector.
[0065] It should be noted that when the processor processes based on the fusion feature vector, the prediction result of each feature variable is obtained, which includes the power prediction result of the target photovoltaic power station, and the results of each processor can be integrated by further weighted summation to extract the final predicted power sequence.
[0066] S10434, the feature prediction sequence predicted by each processor is weighted and summed based on the weight of each processor, and the final predicted power sequence is output, and the formula is: ; wherein, is the final predicted power sequence, is the weight of the i-th processor, is the feature prediction sequence predicted by the i-th processor. In some embodiments, the irradiance prediction model and the photovoltaic power prediction model (PITMNN-MSGStar-AMS) provided by the embodiments of the present application and other models (CNN-BiLSTM-Attention, Transformer, CGAformer) of related technologies are respectively based on given samples for prediction.
[0067] Based on weather-varying samples, for example, the change rate of the environmental temperature, humidity, wind direction, etc. corresponding to the sample is greater than a preset threshold, the comparison chart of the prediction results of the model provided by the embodiments of the present application and other models of related technologies is as shown in ; based on weather-stable samples, that is, the change rate of the environmental temperature, humidity, wind direction, etc. corresponding to the sample is less than or equal to the preset threshold, the comparison chart of the prediction results of the model provided by the embodiments of the present application and other models of related technologies is as shown in
[0068] . Figure 3 Figure 4 Based on , the embodiments of the present application have achieved better prediction effect than other models.
[0069] Figures 3-4
[0070] The evaluation indexes of the prediction results of the embodiments of the present application and other models are as shown in the following table. Figure 5 Figure 5 It is illustrated that the irradiance prediction model and the photovoltaic power prediction model provided by the embodiments of the present application can achieve good prediction effects under various weather conditions, can adapt to variable climate conditions, and are superior to other advanced models.
[0071] The following is a device embodiment of the present application, which can be used to execute the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, refer to the method embodiments of the present application.
[0072] Next, refer to Figure 6 The structure schematic diagram of the photovoltaic power prediction device fusing irradiance physical information provided by an exemplary embodiment of the present application is shown in the following figure, and the device comprises: A data acquisition module is configured to acquire a historical operation data sequence of a target photovoltaic power station in a historical period. A data preprocessing module is configured to preprocess the historical operation data sequence. An irradiance prediction module is configured to input the preprocessed historical operation data sequence into a trained irradiance prediction model, and obtain a predicted irradiance sequence of a prediction period through processing of the irradiance prediction model. A power prediction module is configured to input the preprocessed historical operation data sequence and the predicted irradiance sequence as feature variables into a trained photovoltaic power prediction model, and obtain a predicted power sequence of the prediction period through processing of the photovoltaic power prediction model.
[0073] It should be noted that the device provided by the above embodiments in executing the photovoltaic power prediction method fusing irradiance physical information is only exemplified by the division of the above functional modules, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device provided by the above embodiments and the photovoltaic power prediction method fusing irradiance physical information embodiments belong to the same concept, and the implementation process is detailed in the method embodiments, which will not be repeated here.
[0074] The embodiments of the present application also provide an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the method of any one of the above embodiments when executing the program.
[0075] Please refer to Figure 7 The structure block diagram of an electronic device provided by the embodiments of the present application is shown in the following figure.
[0076] As Figure 7 As shown, the electronic device includes a processor and a memory.
[0077] In the embodiments of the present application, the processor is the control center of the computer system, and can be a processor of a physical machine or a processor of a virtual machine. The processor can include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 1401 can be implemented in at least one of a hardware form of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), and a PLA (Programmable Logic Array).
[0078] The processor can also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also known as a CPU (Central Processing Unit). The coprocessor is a low-power processor for processing data in a standby state.
[0079] The memory can include one or more computer-readable storage media, which can be non-transitory. The memory can also include a high-speed random access memory and a nonvolatile memory such as one or more disk storage devices, flash storage devices. In some embodiments of the present application, the non-transitory computer-readable storage medium in the memory is used to store at least one instruction for being executed by the processor to implement the method in the embodiments of the present application.
[0080] In some embodiments, the electronic device further includes a peripheral device interface and at least one peripheral device. The processor, the memory, and the peripheral device interface can be connected through a bus or a signal line. Each peripheral device can be connected to the peripheral device interface through a bus, a signal line, or a circuit board. Specifically, the peripheral device interface in the display screen, the camera, and the audio circuit can be used to connect at least one peripheral device related to I / O (Input / Output) to the processor and the memory.
[0081] In some embodiments of the present application, the processor, the memory, and the peripheral device interface are integrated on the same chip or circuit board; in some other embodiments of the present application, any one or two of the processor, the memory, and the peripheral device interface can be implemented on a separate chip or circuit board. The embodiments of the present application do not make specific limitations in this regard.
[0082] The electronic device structure block diagram shown in the embodiments of the present application does not constitute a limitation on the electronic device, and the electronic device can include more or fewer components than shown, or combine certain components, or adopt a different component arrangement.
[0083] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the steps of the method of any of the preceding embodiments. The computer readable storage medium can include, but is not limited to, any type of disk, including a floppy disk, an optical disk, a DVD, a CD-ROM, a micro drive, and a magneto-optical disk, a ROM, a RAM, an EPROM, an EEPROM, a DRAM, a VRAM, a flash memory device, a magnetic card or an optical card, a nanosystem (including a molecular memory IC), or any type of medium or device suitable for storing instructions and / or data.
[0084] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software plus necessary universal hardware platforms, and of course can also be implemented by hardware. Based on such an understanding, the above technical solutions, essentially or in other words, the part that contributes to the related art, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some part of the embodiments.
[0085] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A method of fusing irradiance physical information for photovoltaic power prediction, characterized in that, The method comprises the following steps: obtaining a historical operation data sequence of a target photovoltaic power station in a historical period; preprocessing the historical operation data sequence; inputting the preprocessed historical operation data sequence into a trained irradiance prediction model, and obtaining a predicted irradiance sequence of a prediction period through processing of the irradiance prediction model; inputting the preprocessed historical operation data sequence and the predicted irradiance sequence as feature variables into a trained photovoltaic power prediction model, and obtaining a predicted power sequence of the prediction period through processing of the photovoltaic power prediction model.
2. The method of claim 1, wherein the method further comprises: The preprocessing of the operation data sequence comprises the following steps: detecting abnormal values of each type of data in the historical operation data sequence respectively, and eliminating the abnormal values of each type of data respectively; identifying missing values of each type of data in the historical operation data sequence, and filling the missing values through interpolation calculation respectively; performing normalization processing on each numerical type parameter in the historical operation data sequence respectively, and outputting the normalized data.
3. The method of claim 1, wherein the method further comprises: The training process of the irradiance prediction model comprises the following steps: obtaining a sample operation data sequence of the target photovoltaic power station in a sample period, wherein the sample operation data sequence comprises actual irradiance, irradiance parameters, physical parameters of the target photovoltaic power station at each sample time in the sample period, and theoretical irradiance calculated according to the irradiance parameters and the physical parameters; constructing a sample set for training based on the sample operation data sequence, so as to train the irradiance prediction model through the sample set; calculating the predicted irradiance of the next time of the sample time based on the actual irradiance and the irradiance parameters of the sample time through the irradiance prediction model; constructing a data loss term of the irradiance prediction model according to the difference between the predicted irradiance and the actual irradiance of the corresponding time, and constructing a physical loss term of the irradiance prediction model according to the difference between the predicted irradiance and the theoretical irradiance of the corresponding time; weighting and fusing the data loss term and the physical loss term to obtain a loss function of the irradiance prediction model, and optimizing the weight parameters of the irradiance prediction model according to the loss function until convergence, so as to obtain the trained irradiance prediction model.
4. The method of claim 3, wherein the method further comprises: The irradiance parameters comprise direct normal irradiance, global horizontal irradiance, sky diffuse irradiance and extraterrestrial normal irradiance, and the physical parameters of the target photovoltaic power station comprise the incidence angle of the sun relative to the photovoltaic panel, the ground albedo, the inclination of the photovoltaic panel and the solar zenith angle; The step of calculating the theoretical irradiance comprises the following steps: calculating a direct radiation component based on the incidence angle and the direct normal irradiance; calculating a ground reflection component based on the global horizontal irradiance, the ground albedo and the inclination of the photovoltaic panel; calculating a sky diffuse component based on the sky diffuse irradiance, the direct normal irradiance, the extraterrestrial normal irradiance, the inclination of the photovoltaic panel, the incidence angle and the solar zenith angle; calculating the theoretical irradiance of the corresponding time based on the direct radiation component, the ground reflection component, the sky diffuse component and the incidence angle correction factor corresponding to each component.
5. A method of photovoltaic power prediction fusing irradiance physical information according to any of claims 1-4, characterized in that, The photovoltaic power prediction model comprises a multi-scale time series decomposition module, a grouping star aggregation-redistribution module and an adaptive multi-processor synthesis module; The photovoltaic power prediction model comprises a multi-scale time series decomposition module, a grouping star aggregation-redistribution module and an adaptive multi-processor synthesis module; The multi-scale time series decomposition module is used for performing down-sampling processing on each type of input feature variable respectively to obtain a plurality of features of different time scales, and fusing the features of different time scales to obtain a hybrid time scale feature vector; The grouping star aggregation-redistribution module is used for processing the input feature variables to obtain a system-level fusion feature vector, splicing the system-level fusion feature vector with the input feature variables to obtain an updated fusion feature vector, and processing the updated fusion feature vector to obtain a feature prediction sequence of a prediction period. The adaptive multi-processor synthesis module is used for processing the hybrid time scale feature vector and the fusion feature vector to obtain a prediction power sequence of a prediction period.
6. The method of claim 5, wherein the method further comprises: The grouping star aggregation-redistribution module is used for processing the input feature variables to obtain a system-level fusion feature vector, splicing the system-level fusion feature vector with the input feature variables to obtain an updated fusion feature vector, and processing the updated fusion feature vector to obtain a feature prediction sequence of a prediction period. The grouping star aggregation-redistribution module is used for grouping the input feature variables according to physical meanings of the feature variables to obtain a group-level feature vector of each group. All group-level feature vectors are spliced to obtain a system-level fusion feature vector. The system-level fusion feature vector is spliced with the input feature variables, and the spliced result is fused with the input feature variables to obtain an updated fusion feature vector.
7. The method of claim 6, wherein the method further comprises: The adaptive multi-processor synthesis module is used for processing the hybrid time scale feature vector and the fusion feature vector to obtain a prediction power sequence of a prediction period. The gating network is used for processing the hybrid time scale feature vector to output a weight vector. The weight vector is used for determining weights of each processor in the adaptive multi-processor synthesis module. Each processor is used for processing the fusion feature vector to obtain a feature prediction sequence of a prediction period. The weights of each processor are used for weighting and summing the feature prediction sequences obtained by each processor to output a final prediction power sequence.
8. A photovoltaic power prediction device fusing irradiance physical information, characterized in that, The data acquisition module is used for acquiring a historical operation data sequence of a target photovoltaic power station in a historical period. The data preprocessing module is used for preprocessing the historical operation data sequence. The irradiance prediction module is used for inputting the preprocessed historical operation data sequence into a trained irradiance prediction model to obtain a prediction irradiance sequence of a prediction period through the irradiance prediction model. The power prediction module is used for inputting the preprocessed historical operation data sequence and the prediction irradiance sequence as feature variables into a trained photovoltaic power prediction model to obtain a prediction power sequence of a prediction period through the photovoltaic power prediction model. The processor implements the steps of the method according to any one of claims 1 to 7 when executing the program.
9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The computer program implements the steps of the method according to any one of claims 1 to 7 when executed by the processor.
10. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that,