Photovoltaic output prediction method based on combined prediction model and related device

Abstract

The invention relates to a photovoltaic output prediction method based on a combined prediction model and a related device. The method comprises the following steps: acquiring historical photovoltaic data; constructing a photovoltaic physical model based on the historical photovoltaic data, and converting the photovoltaic physical model into a photovoltaic network module; the method comprises the following steps of: constructing an Inform module; establishing a photovoltaic combined prediction model based on the photovoltaic network module and the Informer module, and training the photovoltaic combined prediction model by using the historical photovoltaic data; and predicting the photovoltaic output data of the photovoltaic cell based on the trained photovoltaic combined prediction model to obtain photovoltaic output prediction data. Compared with the prior art, the method can drive and optimize model parameters through a large amount of historical photovoltaic power generation power data and complex local meteorological data, so that the method is better suitable for power grid dispatching personnel to make decision support for photovoltaic output prediction in complex local weather.

Description

Technical Field

[0001] The present application relates to the technical field of photovoltaic output prediction of power grids, and in particular to a photovoltaic output prediction method and related devices based on a combined prediction model. Background Art

[0002] Under the dual pressures of global climate change and environmental protection, the dual carbon goals (i.e. carbon peak and carbon neutrality) have become a global consensus. According to the forecast of the International Energy Agency (IEA), by 2040, the proportion of new energy in the energy structure will reach 60%, of which photovoltaic and wind power will become dominant, accounting for more than 50%. At present, the development of new energy is particularly rapid, with an annual installed capacity of new energy reaching 127GW, among which the growth of distributed photovoltaic is particularly significant, and the newly installed capacity in 2023 will reach 96.29GW.

[0003] However, the output characteristics of photovoltaic energy have brought great challenges to the stable operation of the power system. Photovoltaic output is volatile and intermittent. This unstable output characteristic will have a huge impact on the safe operation of the power grid when a large proportion of it is connected to the grid. In addition, photovoltaic output is also significantly affected by complex local weather conditions, such as cloud cover and temperature changes. These factors further increase the volatility and uncertainty of photovoltaic power output, which brings great difficulties to the dispatch and operation of the power system. As the proportion of photovoltaic installed capacity increases year by year, the uncertainty of power system operation is also increasing. Therefore, accurate prediction of photovoltaic output can achieve efficient consumption of new energy while ensuring the safe operation of the power grid.

[0004] Among the existing photovoltaic output prediction methods, both traditional physical modeling and statistical models have certain limitations. Although the physical model can accurately reflect the physical mechanism of photovoltaic output, it requires a large amount of field data as support, and the calculation process is complicated, making it difficult to promote it on a large scale in practical applications. Traditional statistical models, such as time series analysis and regression analysis, have certain effects on photovoltaic output prediction in certain specific scenarios, but due to the complex nonlinear characteristics of photovoltaic power generation, especially the interaction between multiple variables, the prediction accuracy of these models in complex meteorological environments is greatly reduced. In addition, traditional statistical models often find it difficult to capture the complex relationship between photovoltaic output and various meteorological factors (such as irradiance, temperature, wind speed, etc.), and cannot effectively deal with the volatility and uncertainty of photovoltaic output on different time scales. Therefore, under complex meteorological conditions, the prediction results of these models often have large deviations from the actual values, which makes it difficult to meet the actual needs of the power system for photovoltaic output prediction accuracy.

[0005] In summary, the existing photovoltaic output prediction methods have certain limitations in practical applications and cannot meet the power system's demand for high-precision photovoltaic output prediction. Therefore, developing a more accurate and efficient photovoltaic output prediction method has become a hot and difficult issue in the current power industry research. Summary of the invention

[0006] Based on this, it is necessary to provide a photovoltaic output prediction method and related devices based on a combined prediction model that can improve the photovoltaic output prediction accuracy and prediction efficiency in the power system in response to the above technical problems.

[0007] In a first aspect, the present application provides a photovoltaic output prediction method based on a combined prediction model, the method comprising:

[0008] Obtain historical photovoltaic data;

[0009] Building a photovoltaic physical model based on the historical photovoltaic data, and converting the photovoltaic physical model into a photovoltaic network module;

[0010] Build the Informer module;

[0011] Establishing a photovoltaic combination prediction model based on the photovoltaic network module and the Informer module, and training the photovoltaic combination prediction model using the historical photovoltaic data;

[0012] The photovoltaic output data of the photovoltaic cells is predicted based on the trained photovoltaic combination prediction model to obtain photovoltaic output prediction data.

[0013] In one of the embodiments, the historical photovoltaic data includes historical meteorological data and historical photovoltaic output data of the photovoltaic cell under each historical meteorological data, and the historical meteorological data includes at least radiation intensity, temperature, humidity and / or wind speed.

[0014] In one embodiment, the step of constructing a photovoltaic physical model based on the historical photovoltaic data and converting the photovoltaic physical model into a photovoltaic network module includes:

[0015] Determine a first model parameter according to an equivalent circuit of a photovoltaic cell, and construct a photovoltaic physical model based on the first model parameter and the historical photovoltaic data;

[0016] The photovoltaic physical model is converted into a photovoltaic network module using a neural network algorithm.

[0017] In one embodiment, the constructing of the Informer module includes:

[0018] Building a coding submodule and a decoding submodule, and constructing an Informer module based on the coding submodule and the decoding submodule;

[0019] The encoding submodule is used to extract first correlation information between the historical meteorological data and the historical photovoltaic output data to form a first feature sequence, and to generate a hidden layer representation after performing redundant information distillation on the first feature sequence;

[0020] The decoding submodule is used to extract the second associated information of the historical photovoltaic parameters to form a second feature sequence, and fuse the second feature sequence with the hidden layer representation generated by the encoder submodule to output a prediction result.

[0021] In one embodiment, the photovoltaic combination prediction model is established based on the photovoltaic network module and the informer module, and the photovoltaic combination prediction model is trained using the historical photovoltaic data, including:

[0022] Establishing a photovoltaic combination prediction model based on the photovoltaic network module and the Informer module;

[0023] defining a second model parameter and a loss function of the photovoltaic combination prediction model;

[0024] Inputting the historical photovoltaic data into the photovoltaic combination prediction model, and outputting a first predicted power based on the forward transfer of the photovoltaic combination prediction model;

[0025] The first predicted power is transmitted back to the photovoltaic combination prediction model as the input of the photovoltaic combination prediction model, and the second model parameters are adjusted to iteratively train the photovoltaic combination prediction model until the loss function converges or reaches the maximum number of iterations, so as to obtain a trained photovoltaic combination prediction model.

[0026] In one embodiment, inputting the historical photovoltaic data into the photovoltaic combination prediction model, and outputting a first predicted power based on the forward transfer of the photovoltaic combination prediction model, comprises:

[0027] Inputting the historical meteorological data and the historical photovoltaic output data into the photovoltaic network module, and outputting a second predicted power based on the forward transfer of the photovoltaic network module;

[0028] Aligning the historical meteorological data and the historical photovoltaic output data in time to form photovoltaic time series data;

[0029] The photovoltaic time series data and the second predicted power are input into the encoding submodule, and the first association information between the data at different positions in the photovoltaic time series data is captured based on the multi-head probabilistic sparse self-attention layer of the encoding submodule, and the data at the corresponding position in the photovoltaic time series data is extracted according to the first association information to form a first feature sequence; and the data distillation layer of the encoding submodule performs redundant information distillation on the first feature sequence to generate a hidden layer representation;

[0030] The photovoltaic time series data and the hidden layer representation are input into the decoding submodule, and the second association relationship between the data before the current position in the photovoltaic time series data is captured based on the masked multi-head probabilistic sparse self-attention layer of the decoding submodule, and the data at the corresponding position in the photovoltaic time series data is extracted according to the second association relationship information to form a second feature sequence; then, the second feature sequence is fused with the hidden layer representation generated by the encoder submodule based on the multi-head self-attention layer of the decoding submodule, and the prediction result is output at one time to obtain the first prediction power.

[0031] In a second aspect, the present application also provides a photovoltaic output prediction device based on a combined prediction model, including:

[0032] Data acquisition module, used to obtain historical photovoltaic data;

[0033] A first model building module, used to build a photovoltaic physical model based on the historical photovoltaic data, and convert the photovoltaic physical model into a photovoltaic network module;

[0034] The second model building module is used to build the Informer module;

[0035] A model fusion and training module, used to establish a photovoltaic combination prediction model based on the photovoltaic network module and the Informer module, and train the photovoltaic combination prediction model using the historical photovoltaic data;

[0036] The data prediction module is used to predict the photovoltaic output data of the photovoltaic cell based on the trained photovoltaic combination prediction model to obtain photovoltaic output prediction data.

[0037] In a third aspect, the present application further provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the steps of the method described above when executing the computer program.

[0038] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method described above are implemented.

[0039] In a fifth aspect, the present application also provides a computer program product, including a computer program, which implements the steps of the method described above when executed by a processor.

[0040] The photovoltaic output prediction method and related device based on the combined prediction model obtain historical photovoltaic data; construct a photovoltaic network module and an Informer module to fuse to obtain a photovoltaic combination prediction model, and then use historical photovoltaic data to train the photovoltaic combination prediction model; finally, the photovoltaic output data of the photovoltaic cell is predicted based on the trained photovoltaic combination prediction model to obtain photovoltaic output prediction data; the photovoltaic combination prediction model has both the physical interpretability of the photovoltaic physical model and the optimizability of the model parameters of the Informer module, overcoming the shortcomings of the traditional photovoltaic physical modeling process that the physical parameters of the photovoltaic station are difficult to obtain and thus the practicality is not strong. The historical meteorological data and the historical photovoltaic output data are used as additional time series feature inputs of the photovoltaic combination prediction model, which can drive the optimization of the model parameters of the photovoltaic combination prediction model, and then effectively cope with the prediction of photovoltaic output data under complex local weather conditions, and improve the prediction accuracy of photovoltaic output data, so as to provide data support for the dispatching and planning of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0042] FIG1 is an application environment diagram of a photovoltaic output prediction method based on a combined prediction model in one embodiment;

[0043] Figure 2 A schematic diagram of a flow chart of a photovoltaic output prediction method based on a combined prediction model in one embodiment;

[0044] Figure 3 is a structural block diagram of the Informer module in step S203 in one embodiment;

[0045] Figure 4 is a flow chart of step S203 in one embodiment;

[0046] Figure 5 is a flow chart of step S205 in one embodiment;

[0047] Figure 6 It is a schematic diagram of the prediction results of a photovoltaic output prediction method based on a combined prediction model in an application example;

[0048] Figure 7 is a structural block diagram of a photovoltaic output prediction device based on a combined prediction model in one embodiment;

[0049] FIG. 8 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0051] The photovoltaic output prediction method based on the combined prediction model provided in the embodiment of the present application can be applied to the application environment shown in Figure 1. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The server 104 obtains historical photovoltaic data, constructs a photovoltaic network module and an Informer module respectively, and integrates the two modules to obtain a photovoltaic combination prediction model, and then uses the acquired historical photovoltaic data to train the photovoltaic combination prediction model. Finally, based on the trained photovoltaic combination prediction model, the output of the photovoltaic cells in the power system is predicted, so as to reasonably dispatch electricity and ensure the stable operation of the power system while realizing the efficient consumption of new energy. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, projection devices, etc. Portable wearable devices can be smart watches, smart bracelets, etc. The server 104 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0052] In an exemplary embodiment, Figure 2 As shown, a photovoltaic output prediction method based on a combined prediction model is provided, and the method is applied to the server in FIG. 1 as an example for explanation, including the following steps S201 to S205. Among them:

[0053] Step S201, obtaining historical photovoltaic data.

[0054] Among them, historical photovoltaic data includes historical meteorological data and historical photovoltaic output data of photovoltaic cells under various historical meteorological data. Both historical meteorological data and historical photovoltaic output data are presented in the form of time series to reflect the photovoltaic output data corresponding to photovoltaic cells under various meteorological conditions. Historical meteorological data include but are not limited to meteorological data such as radiation intensity, temperature, humidity and / or wind speed, and contain various meteorological data under complex weather conditions, which is conducive to predicting the output of photovoltaic cells when connected to the grid under extreme weather conditions to ensure the accuracy of the prediction. Historical photovoltaic output data mainly refers to the historical photovoltaic power data of photovoltaic cells.

[0055] Specifically, the server obtains historical photovoltaic output data of photovoltaic cells in various time periods and historical meteorological data corresponding to the time periods in a chronological manner through various channels and methods to obtain historical photovoltaic data.

[0056] The server may obtain historical meteorological data and historical photovoltaic output data through channels and methods such as from relevant departments such as the Meteorological Bureau and the Electric Power Bureau, or by crawling published meteorological data from the websites of the Meteorological Bureau and the Electric Power Bureau or other third-party websites.

[0057] In an optional embodiment, after acquiring the historical photovoltaic data, the server may pre-process the acquired historical photovoltaic data to facilitate the processing of the data in subsequent steps. The pre-processing method may include but is not limited to data cleaning, data partitioning and / or data normalization.

[0058] Step S202: construct a photovoltaic physical model based on historical photovoltaic data, and convert the photovoltaic physical model into a photovoltaic network module.

[0059] Among them, the photovoltaic physical model refers to a circuit model that treats the photovoltaic cell as a photocurrent source in parallel with a diode and its parasitic resistance (or equivalent to a series resistor and a parallel resistor). It can describe the photoelectric conversion process on the material surface of the photovoltaic cell under sunlight, so that the photovoltaic physical model has better explanatory power and physical constraints when considering physical factors.

[0060] Among them, the photovoltaic network module is used to capture the nonlinear relationships and hidden features in the photovoltaic physical model to overcome the defect of the photovoltaic physical model that photovoltaic site data is difficult to obtain under complex local weather conditions.

[0061] Specifically, the server constructs a photovoltaic physical model according to the acquired historical meteorological data and historical photovoltaic output data, and then constructs a corresponding photovoltaic network module based on the photovoltaic physical model.

[0062] The photovoltaic network module may be implemented using one or more algorithms including but not limited to neural network, regression network, vector machine, random forest, etc.

[0063] Step S203, constructing an Informer module.

[0064] Among them, the Informer module is a deep learning module used to predict time series, which can dynamically update and optimize model parameters based on the forward and reverse transmission of data.

[0065] Specifically, the server determines the structure and parameters of the Informer module according to the prediction requirements, and constructs the Informer module accordingly.

[0066] Step S204: establishing a photovoltaic combination prediction model based on the photovoltaic network module and the Informer module, and training the photovoltaic combination prediction model using historical photovoltaic data.

[0067] Specifically, the server establishes a photovoltaic combination prediction model based on the photovoltaic network module and the Informer module, and inputs the acquired historical meteorological data and historical photovoltaic output data into the photovoltaic combination prediction model. After obtaining a preliminary prediction data through the forward transmission of the photovoltaic combination prediction model, the preliminary prediction data is reversely transmitted to the photovoltaic combination prediction model to iteratively train the photovoltaic combination prediction model, ensuring that the trained photovoltaic combination prediction model has both physical interpretability and can optimize the parameters in the photovoltaic physical model in an end-to-end manner, thereby improving the practicality of photovoltaic predictions under complex weather conditions.

[0068] In an optional embodiment, when the photovoltaic combination prediction model is trained using historical meteorological data and historical photovoltaic output data, the historical meteorological data and historical photovoltaic output data can be randomly divided into a training set and a verification set to respectively train and perform performance verification on the photovoltaic combination prediction model to ensure the prediction accuracy of the photovoltaic combination prediction model.

[0069] Step S205 , predicting the photovoltaic output data of the photovoltaic cells based on the trained photovoltaic combination prediction model to obtain photovoltaic output prediction data.

[0070] Specifically, the acquired photovoltaic data (including meteorological data and photovoltaic output data) is input into the trained photovoltaic combination prediction model, and the prediction result is output based on the forward transfer of the photovoltaic combination prediction model to obtain the photovoltaic output prediction data.

[0071] In the above-mentioned photovoltaic output prediction method based on the combined prediction model, historical photovoltaic data is obtained; a photovoltaic network module and an Informer module are constructed to fuse to obtain a photovoltaic combination prediction model, and then the photovoltaic combination prediction model is trained using historical photovoltaic data; finally, the photovoltaic output data of the photovoltaic cell is predicted based on the trained photovoltaic combination prediction model to obtain photovoltaic output prediction data; the photovoltaic combination prediction model has both the physical interpretability of the photovoltaic physical model and the optimizability of the model parameters of the Informer module, overcoming the shortcomings of the traditional photovoltaic physical modeling process that the physical parameters of the photovoltaic station are difficult to obtain and thus the practicality is not strong, and historical meteorological data and historical photovoltaic output data are used as additional time series feature inputs of the photovoltaic combination prediction model, which can drive the optimization of the model parameters of the photovoltaic combination prediction model, thereby effectively coping with the prediction of photovoltaic output data under complex local weather conditions, and improving the prediction accuracy of photovoltaic output data, so as to provide data support for the dispatching and planning of the power system.

[0072] In one embodiment, the specific method of obtaining the photovoltaic network module in step S202 includes:

[0073] The first model parameters are determined according to the equivalent circuit of the photovoltaic cell, and a photovoltaic physical model is constructed based on the first model parameters and historical photovoltaic data; the photovoltaic physical model is converted into a photovoltaic network module using a neural network algorithm.

[0074] Among them, the equivalent circuit of the photovoltaic cell, as the basis for constructing the photovoltaic physical model, usually includes a photocurrent source, a diode and a parasitic resistance in parallel with it. The parasitic resistance includes a series resistor and a parallel resistor. The photocurrent source is used to represent the current generated by the photovoltaic cell under light, the diode simulates the PN junction characteristics inside the photovoltaic cell, and the parasitic resistor reflects the internal loss of the photovoltaic cell.

[0075] Among them, the magnitude of the photocurrent is proportional to the intensity of solar radiation, and it is the main source of electricity generated by photovoltaic cells. In the photovoltaic physics model, the photocurrent can be determined by parameters such as the intensity of solar radiation, the conversion efficiency of photovoltaic cells, and the area of ​​photovoltaic cells. The reverse saturation current of the diode is related to the temperature. In the photovoltaic physics model, the reverse saturation current of the diode can be determined by the temperature and the semiconductor parameters of the photovoltaic cell. The series resistance in the parasitic resistance mainly affects the maximum power of the photovoltaic cell, and the parallel resistance is mainly caused by manufacturing defects, causing power loss in the form of shunting. The values ​​of these resistances can be obtained through experimental measurements or according to the data manual of the photovoltaic cell.

[0076] Among them, the first model parameters refer to the parameters used to construct the photovoltaic physical model. According to the above, the first model parameters of this embodiment mainly include the photocurrent of the photovoltaic cell, the saturation current of the diode, the electron current, the series resistance value, the diode factor, the temperature coefficient of the short-circuit current, and the output current, output voltage and output power of the photovoltaic cell.

[0077] Among them, in the photovoltaic physical model, for historical meteorological data, the performance of photovoltaic cells under different radiation intensities is reflected by considering adjusting the coefficient of photocurrent; the influence of temperature on the performance of photovoltaic cells is reflected by considering the influence of temperature on the reverse saturation current, series resistance and parallel resistance of the diode; the influence of humidity on the surface temperature of photovoltaic cells is indirectly reflected to reflect its influence on the performance of photovoltaic cells; the influence of wind speed on the surface temperature of photovoltaic cells is reflected to reflect its influence on the performance of photovoltaic cells.

[0078] Specifically, the server constructs an equivalent circuit of the photovoltaic cell including a photocurrent source, a diode and its parasitic resistance by analyzing the performance of the photovoltaic cell, and determines the first model parameters and the mutual influence and relationship between the photovoltaic meteorological data and the first model parameters based on the equivalent circuit, and constructs a photovoltaic physical model that takes into account meteorological data such as radiation intensity, temperature, humidity, and wind speed based on the first model parameters and historical photovoltaic data.

[0079] In this embodiment, the photovoltaic physical model constructed according to the equivalent circuit of the photovoltaic cell is expressed as:

[0080] (1)

[0081] (2)

[0082] (3)

[0083] in, is the output current of the photovoltaic cell in the photovoltaic physical model, is the saturation current of the diode, represents the exponential function, is the electron charge, is the output voltage of the photovoltaic cell in the photovoltaic physical model, is the series resistance in the parasitic resistance, is the diode factor, is the Boltzmann constant, is the photogenerated current, The reference radiation intensity and temperature The photocurrent under is the radiation intensity, is the temperature, is the temperature coefficient of the short-circuit current, is the output power of the photovoltaic physical model.

[0084] The server then uses a neural network algorithm to convert the photovoltaic physical model into a photovoltaic network module. The photovoltaic network module uses historical meteorological data such as historical radiation intensity, temperature, humidity, wind speed, and historical photovoltaic power data under complex local weather as input to predict the performance of photovoltaic cells under different meteorological conditions. By inputting actual meteorological data such as radiation intensity, temperature, humidity, wind speed, etc., the output power of the photovoltaic cells under the corresponding meteorological conditions can be output.

[0085] Among them, the photocurrent in the above formulas (1), (2) and (3) is , Temperature coefficient of short-circuit current , diode factor and the series resistance in the parasitic resistance They can all be defined as trainable neural network parameters in the photovoltaic network module, so as to be optimized and updated together with the parameters of the Informer module in the subsequent photovoltaic combination prediction model, so as to improve the prediction accuracy of photovoltaic output through data-driven optimization of model parameters.

[0086] In this embodiment, by constructing a photovoltaic physical model, it has good interpretability when considering physical factors and can accurately reflect the working characteristics of photovoltaic cells; at the same time, by converting the constructed photovoltaic physical model into a photovoltaic network module using a neural network, and defining the parameters therein as parameters that can be optimized and updated, the model parameters can be optimized through data-driven in subsequent model training, which can effectively improve the prediction accuracy of photovoltaic processing conditions.

[0087] In one embodiment, the Informer module constructed in step S203 specifically includes:

[0088] Build the encoding submodule and the decoding submodule, and build the Informer module based on the encoding submodule and the decoding submodule.

[0089] Among them, Figure 3 As shown, the Informer module includes an encoding submodule and a decoding submodule. The encoding submodule is used to extract the first correlation information of historical meteorological data and historical photovoltaic output data to form a first feature sequence, and generate a hidden layer representation after performing redundant information distillation on the first feature sequence; the decoding submodule is used to extract the second correlation relationship of historical photovoltaic parameters to form a second feature sequence, and fuse the second feature sequence with the hidden layer representation generated by the encoder submodule to output the prediction result.

[0090] Among them, historical meteorological data and historical photovoltaic output data are extracted by the encoding submodule in the form of time series. The first association information refers to the association between historical meteorological data and historical photovoltaic output data at different positions and different time steps in the entire sequence, which can represent the impact of different meteorological data on photovoltaic output data, so as to determine the long-term dependence of historical meteorological data and historical photovoltaic output data in time series. The first feature sequence refers to the sequence formed by time series integration based on each first association relationship. The hidden layer representation refers to the data representation of high-level feature information in historical photovoltaic data obtained after extraction by the probabilistic sparse self-attention mechanism and data screening.

[0091] Among them, the encoding submodule has a multi-head probabilistic sparse self-attention layer and a data distillation layer.

[0092] Among them, the multi-head probabilistic sparse self-attention layer uses the probabilistic sparse self-attention mechanism to process the sequence of historical meteorological data and historical photovoltaic output data to capture the dependency between each data in the sequence of historical meteorological data and historical photovoltaic output data at different positions and time steps. In this process, the multi-head probabilistic sparse self-attention layer divides the input historical meteorological data and historical photovoltaic output data sequence into multiple heads, and each head uses the self-attention mechanism to calculate the attention weight of the corresponding position to other positions, and obtains multiple first associations to splice and generate the first feature sequence that only focuses on important information, thereby ignoring some irrelevant information and improving data processing efficiency.

[0093] The self-attention mechanism adopts the key-value-query mode, and assigns the corresponding value to the key by comparing the similarity between the key and the query. In this embodiment, the calculation formula of the self-attention mechanism of the Informer module is as follows:

[0094] (4)

[0095] in, is the calculation process of the self-attention mechanism, are matrices of query, key, and value, respectively. for The matrix OK, They are The first OK, For the prediction element Related elements The number of times the dot product result appears, Choose an asymmetric exponential function, that is, for , For the prediction element Related elements The total number of times the dot product result appears, For the The probability distribution of the key under the query, The final degree of association.

[0096] According to the above formula (4), let the key only focus on the front A query with high importance (Query) can be obtained:

[0097] (5)

[0098] in, is the calculation process of the probabilistic sparse self-attention mechanism, is the operation of the normalized exponential function, for The transpose of the matrix, To include only the front A sparse matrix of queries.

[0099] Based on the above, the dependency between the historical meteorological data and each data in the historical photovoltaic output data sequence at different positions and time steps can be determined, and the first characteristic sequence can be obtained.

[0100] Among them, the data distillation layer uses the self-attention distillation mechanism to process the first feature sequence to further reduce the redundant information in the first feature sequence, extract the most useful features for the subsequent prediction process, and form a hidden layer representation containing high-level feature information in the historical meteorological data and historical photovoltaic output data sequences.

[0101] Among them, the self-attention distillation mechanism can give higher weights to the advantageous features with important information and The layer matrix generates the self-attention feature map focused on the previous layer, namely:

[0102] (6)

[0103] in, For the Layer Matrix, For the Layer Matrix represents the maximum pooling operation, is the activation function, Indicates a one-dimensional convolution operation on the time series. It is the attention module, that is, the data distillation layer.

[0104] Based on the above, by assigning higher weights to dominant features with important information, the redundant information in the first feature sequence can be removed, forming a hidden layer representation containing high-level feature information in the historical meteorological data and historical photovoltaic output data sequences.

[0105] Among them, the historical meteorological data and the historical photovoltaic output data are extracted by the decoding submodule in the form of time series. The second association relationship refers to the association relationship between the historical meteorological data and the historical photovoltaic output data in the sequence before the current position or the current time step, which can represent the impact of the meteorological data before the current position or the current time step on the photovoltaic output data, so as to determine the long-term dependence of the historical meteorological data and the historical photovoltaic output data in time series. The second characteristic sequence refers to the sequence formed by time series integration based on each second association relationship.

[0106] Among them, the decoding submodule has a masked multi-head probabilistic sparse self-attention layer and a multi-head self-attention layer.

[0107] Among them, the masked multi-head probabilistic sparse self-attention layer uses a masking mechanism to determine the calculation range of the attention weight, that is, for the calculation of the attention weight of the current position or time step, only the previous data is used to ensure that only the previous information is considered when predicting future data; at the same time, the masked multi-head probabilistic sparse self-attention layer also uses a probabilistic sparse self-attention mechanism to process the historical meteorological data and historical photovoltaic output data within the determined sequence range, so as to capture the dependency relationship between the historical meteorological data and the historical photovoltaic output data sequence within the sequence range at different positions and time steps. In this process, the multi-head probabilistic sparse self-attention layer divides the historical meteorological data and historical photovoltaic output data sequence within the input sequence range into multiple heads, and each head uses a self-attention mechanism to calculate the attention weight of the corresponding position to other positions, and obtains multiple second associations, so as to splice and generate a second feature sequence that only focuses on important information, thereby ignoring some irrelevant information and improving data processing efficiency.

[0108] Among them, the specific process of extracting the second feature sequence by the masked multi-head probabilistic sparse self-attention layer can be found in the record of the above-mentioned encoding submodule and will not be repeated here.

[0109] Among them, the multi-head self-attention layer uses the self-attention mechanism to fuse the hidden layer representation output by the data distillation layer and the second feature sequence output by the masked multi-head probabilistic sparse self-attention layer to obtain the final prediction result. In this process, the multi-head self-attention layer uses the hidden layer representation of the encoder as one input of the multi-head self-attention layer and the second feature sequence of the masked multi-head probabilistic sparse self-attention layer as another input, applies the multi-head self-attention mechanism to the two inputs, and calculates the attention weight between the two to obtain the final prediction result.

[0110] Specifically, the server sets the parameters of the Informer module according to the corresponding instructions to construct the encoding submodule and the decoding submodule respectively, and then constructs the Informer module based on the encoding submodule and the decoding submodule.

[0111] Among them, the parameters of the Informer module include but are not limited to the input sequence length, target sequence length, predicted sequence length, hidden layer size, number of attention heads, etc.

[0112] In this embodiment, an Informer module including an encoding submodule and a decoding submodule is constructed, and the encoding submodule obtains the long-term dependency between the input historical meteorological data and the historical photovoltaic output data through a probabilistic sparse self-attention mechanism, and reduces the network dimension based on the self-attention distillation mechanism, so that the output hidden layer representation only focuses on high-level feature information, which can effectively reduce the calculation amount of the module to improve the calculation efficiency and prediction speed; while the decoding submodule fuses the hidden layer representation with the second feature information in a generative decoding manner, and outputs the prediction result at one time, which is conducive to further improving the prediction speed.

[0113] In one embodiment, Figure 4 As shown, step S204 includes the following steps S401 to S404. Among them:

[0114] Step S401: establishing a photovoltaic combination prediction model based on the photovoltaic network module and the Informer module.

[0115] Specifically, the server establishes a photovoltaic combination prediction model based on the photovoltaic network module and the Informer module; the photovoltaic combination prediction module takes historical meteorological data and historical photovoltaic output data as input.

[0116] Step S402, defining second model parameters and loss function of the photovoltaic combination prediction model.

[0117] The second model parameter refers to the parameter that can be dynamically optimized and updated in the photovoltaic combination prediction model, so as to adjust the second model parameter through iterative prediction of the photovoltaic combination model, thereby improving the prediction accuracy of the photovoltaic combination prediction model. The second model parameter includes the photocurrent defined in the photovoltaic network module , Temperature coefficient of short-circuit current , diode factor and the series resistance in the parasitic resistance As well as parameters such as input sequence length, target sequence length, predicted sequence length, hidden layer size, and number of attention heads in the Informer module.

[0118] In this embodiment, the loss function is expressed as:

[0119] (7)

[0120] (8)

[0121] (9)

[0122] in, is the loss function of the PV combination prediction model; is the loss weight coefficient; is the loss function of the photovoltaic network module, which is used to measure the predicted output power of the photovoltaic network module The actual power value The error, Output power The index of ; is the timing measurement error of the Informer module, is the final prediction output of the Informer module, i.e., the photovoltaic output prediction data. is the actual photovoltaic output data, The final prediction output The index of .

[0123] Step S403: inputting historical photovoltaic data into a photovoltaic combination prediction model, and outputting a first predicted power based on forward transfer of the photovoltaic combination prediction model.

[0124] Among them, the forward transfer of the photovoltaic combination prediction model refers to the input of historical photovoltaic data into the photovoltaic combination prediction model, which is processed and predicted by the photovoltaic network module and the informer module in turn. The first predicted power refers to the predicted power output by the photovoltaic combination prediction model during model training, that is, the photovoltaic output prediction data.

[0125] Specifically, the server inputs historical photovoltaic data into the photovoltaic combination prediction model, and after the photovoltaic network module and the Informer module perform data processing in sequence, a first predicted power is output.

[0126] More specifically, at the photovoltaic network module level, historical meteorological data and historical photovoltaic output data are input into the photovoltaic network module, and the output power of the photovoltaic network module is obtained through prediction by the photovoltaic network module (i.e. ); At the Informer module level, historical meteorological data and historical photovoltaic output data are embedded in the encoding submodule and decoding submodule as time series information respectively. At the same time, the output power of the photovoltaic network module is also used as the input of the encoding submodule. After being processed by the encoding submodule and the decoding submodule, the first predicted power is obtained.

[0127] Step S404, the first predicted power is transmitted back to the photovoltaic combination prediction model as the input of the photovoltaic combination prediction model, and the second model parameters are adjusted to iteratively train the photovoltaic combination prediction model until the loss function converges or reaches the maximum number of iterations, so as to obtain a trained photovoltaic combination prediction model.

[0128] Among them, when a first predicted power is obtained, a round of judgment is usually performed to determine whether the model training is completed in the current round. At this time, the loss function in the above formulas (7) to (9) is calculated based on the first predicted power to obtain the loss function value. If the loss function value tends to be stable or converged, the training is terminated. Otherwise, the first predicted power is continued to be used as the input of the photovoltaic combination prediction model for reverse transmission to perform the next round of training until the loss function value tends to be stable or converged, or the set maximum number of iterations is reached.

[0129] Specifically, the server determines the loss function according to the first predicted power and the corresponding actual photovoltaic output data using formula (9): The value of the photovoltaic network module is then input into the Informer module according to the output power of the photovoltaic network module in the current round. And the corresponding actual power and use formula (8) to determine the loss function After the value of , the loss function of the entire photovoltaic combination prediction model is determined using formula (7) To determine the value of the loss function The value of tends to be stable or converged or whether the maximum number of iterations is reached, so as to judge whether the first predicted power needs to be reversely transmitted to the photovoltaic combination prediction model. If necessary, the server transmits the first predicted power as the input of the photovoltaic combination prediction model to the photovoltaic combination prediction model, adjusts the second model parameters, and outputs a new first predicted power based on the forward transmission of the photovoltaic combination prediction model. Then, according to the above process, it is determined whether the loss function converges or whether the maximum number of iterations is reached, so as to realize iterative training of the photovoltaic combination prediction model, and then obtain the trained photovoltaic combination prediction model.

[0130] In this embodiment, the photovoltaic network module and the Informer module are integrated to obtain a photovoltaic combination prediction model, so that the photovoltaic combination prediction model can be driven by a large amount of historical photovoltaic power generation data and complex local meteorological data to optimize model parameters, thereby overcoming the shortcomings of traditional photovoltaic physical modeling methods that are difficult to obtain the physical parameters of photovoltaic sites and are not practical. As a result, the photovoltaic combination prediction model has better interpretability when considering physical factors, and can optimize the relevant parameters in the photovoltaic physical model in an end-to-end manner to adapt to the photovoltaic output prediction under complex local weather conditions, thereby improving the practicality of the model for photovoltaic prediction under complex local weather conditions.

[0131] In an exemplary embodiment, Figure 5 As shown, step S403 includes the following steps S501 to S504. Among them:

[0132] Step S501, inputting historical meteorological data and historical photovoltaic output data into a photovoltaic network module, and outputting a second predicted power based on forward transmission of the photovoltaic network module.

[0133] Among them, the forward transfer of the photovoltaic network module refers to the process in which historical meteorological data and historical photovoltaic data are input into the photovoltaic network module and flow from the input layer to the output layer of the photovoltaic network module. The second predicted power refers to the output power of the photovoltaic network module during model training (i.e. ).

[0134] Specifically, the server inputs historical meteorological data and historical photovoltaic output data into the photovoltaic network module, and outputs the second predicted power after being processed by the photovoltaic network module.

[0135] Step S502, aligning historical meteorological data and historical photovoltaic output data in time to form photovoltaic time series data.

[0136] Specifically, the server splices the input historical meteorological data and historical photovoltaic output data into an input sequence to ensure that the timestamps of each data are aligned, and then splices the data at the corresponding time points into a multi-dimensional matrix by column to form photovoltaic time series data.

[0137] Step S503, input the photovoltaic time series data and the second predicted power into the encoding submodule, capture the first correlation information between the data at different positions in the photovoltaic time series data based on the multi-head probabilistic sparse self-attention layer of the encoding submodule, and extract the data at the corresponding position in the photovoltaic time series data according to the first correlation information to form a first feature sequence; then, based on the data distillation layer of the encoding submodule, perform redundant information distillation on the first feature sequence to generate a hidden layer representation.

[0138] Specifically, the server inputs the photovoltaic time series data and the second predicted power into the multi-head probabilistic sparse self-attention layer of the encoding submodule, and processes the sequences of historical meteorological data and historical photovoltaic output data through the multi-head probabilistic sparse self-attention layer to capture the dependency between each data in the historical meteorological data and historical photovoltaic output data sequences at different positions and time steps, and obtains multiple first association relationships, which are spliced ​​to generate a first feature sequence that only focuses on important information; the first feature sequence is then input into the data distillation layer of the encoding submodule, and the first feature sequence is processed through the data distillation layer to further reduce redundant information in the first feature sequence, extract the most useful features for subsequent prediction processes, and form a hidden layer representation containing high-level feature information in the historical meteorological data and historical photovoltaic output data sequences.

[0139] Step S504, input the photovoltaic time series data and the hidden layer representation into the decoding submodule, capture the second correlation relationship between the data before the current position in the photovoltaic time series data based on the masked multi-head probabilistic sparse self-attention layer of the decoding submodule, and extract the data at the corresponding position in the photovoltaic time series data according to the second correlation relationship information to form a second feature sequence; then, based on the multi-head self-attention layer of the decoding submodule, fuse the second feature sequence with the hidden layer representation generated by the encoder submodule, output the prediction result at one time, and obtain the first predicted power.

[0140] Specifically, the server inputs the photovoltaic time series data into the masked multi-head probabilistic sparse self-attention layer of the decoding submodule, determines the calculation range of the attention weight through the masked multi-head probabilistic sparse self-attention layer, and processes the historical meteorological data and historical photovoltaic output data within the determined sequence range to capture the dependency between the historical meteorological data within the sequence range and each data in the historical photovoltaic output data sequence at different positions and time steps, and obtains multiple second association relationships to splice into a second feature sequence that only focuses on important information; then the second feature sequence and the hidden layer representation are synchronously input into the multi-head self-attention layer of the decoding submodule, and the hidden layer representation and the second feature sequence are fused through the multi-head attention layer to obtain the final prediction result.

[0141] In this embodiment, in the stage where the photovoltaic network module performs data processing, the accuracy of the second predicted power output can be ensured by processing massive data, and in the stage where the Informer module performs data processing, the dimensions of the photovoltaic time series data and the second predicted power can be effectively reduced by highlighting key information with main characteristics, which is conducive to better extracting important historical moment information, improving the efficiency and accuracy of data processing, and avoiding the problem of information overload; at the same time, since there is a complex and nonlinear interdependence relationship between photovoltaic output data in the prediction process, the prediction result is directly generated by decoding in a generative manner through the decoding submodule, which can effectively avoid the accumulation of errors in multi-step predictions, thereby further improving the efficiency and accuracy of data processing.

[0142] In an application example, in order to verify the effectiveness of the photovoltaic output prediction method based on the combined prediction model of this embodiment, the actual photovoltaic power generation data set of a photovoltaic power station in a certain place in the past three months is taken as an example, and the sampling frequency of the data set is 15 minutes. The photovoltaic power station data sample contains meteorological data such as irradiation intensity, temperature, humidity, wind speed (i.e. historical meteorological data) and the actual power generation power of the photovoltaic power station (i.e. historical photovoltaic output data). The hardware environment used in this example is CPU: i7-13700K, GPU: 3090, 24G video memory, memory: 128G; the software environment is python3.6, torch1.8.0 framework, Windows10 Professional Edition. The neuron data of the encoding submodule and the decoding submodule of the informer module are both set to 128.

[0143] The photovoltaic output prediction method based on the combined prediction model in this example is as follows:

[0144] Firstly, a photovoltaic physical model considering the meteorological data of radiation intensity, temperature, humidity and wind speed is established, and the photovoltaic physical model is converted into a photovoltaic network module.

[0145] Specifically, considering the meteorological data of radiation intensity, temperature, humidity, and wind speed under complex local weather conditions, the photovoltaic physical model established according to the photovoltaic cell equivalent circuit is:

[0146] (1')

[0147] (2')

[0148] (3')

[0149] in, is the output current of the photovoltaic cell in the photovoltaic physical model, is the saturation current of the diode, represents the exponential function, is the electron charge, is the output voltage of the photovoltaic cell in the photovoltaic physical model, is the series resistance in the parasitic resistance, is the diode factor, is the Boltzmann constant, is the photogenerated current, The reference radiation intensity and temperature The photocurrent under is the radiation intensity, is the temperature, is the temperature coefficient of the short-circuit current, is the output power of the photovoltaic physical model.

[0150] The established photovoltaic physical model is converted into a photovoltaic network module. The input data is the historical complex local weather radiation intensity, temperature, humidity, wind speed meteorological data and photovoltaic power data. The photocurrent in the photovoltaic physical model , Temperature coefficient of short-circuit current , diode factor and the series resistance in the parasitic resistance Parameters such as can be defined as trainable neural network weights, and thus optimized and updated together with the Informer model parameters during the subsequent training of the PV combination prediction model.

[0151] Next, the photovoltaic network module is combined with the Informer module to establish a photovoltaic combination prediction model that takes meteorological factors into consideration. Historical meteorological data and historical photovoltaic output data are input for multi-step prediction training to obtain the optimal parameters.

[0152] Since the use of a single traditional photovoltaic physical model neural network in the learning task needs to process a large amount of input information, but this may cause information overload problems, this embodiment introduces an attention mechanism, which can selectively focus on more important information and reduce attention to secondary information, thereby improving the efficiency and accuracy of task processing. In response to the problem of long-term series input of historical photovoltaic data, the Informer model reduces the dimension of the historical photovoltaic data input sequence by highlighting the key information with main features, so that important historical moment information can be better extracted. When dealing with photovoltaic output forecasting tasks, there are complex and nonlinear interdependencies, and the generative structure of the Informer model effectively avoids the problem of error accumulation in multi-step forecasts, showing unique advantages.

[0153] Specifically, the overall structure of the Informer module consists of two parts: the encoding submodule and the decoding submodule.

[0154] The encoding submodule is used to obtain the long-term dependency of the input historical meteorological data and historical photovoltaic output data time series, and the Informer module uses the ProbSparse Self-Attention mechanism to replace the traditional self-attention mechanism. In addition, the self-attention distillation method is used to reduce the dimension of the network. The self-attention mechanism usually adopts the key-value-query mode, and assigns the corresponding value to the key by comparing the similarity between the key and the query. The calculation formula of the Informer self-attention mechanism is as follows:

[0155] (4')

[0156] in, is the calculation process of the self-attention mechanism, are matrices of query, key, and value, respectively. for The matrix OK, They are The first OK, For the prediction element Related elements The number of times the dot product result appears, Choose an asymmetric exponential function, that is, for , For the prediction element Related elements The total number of times the dot product result appears, For the The probability distribution of the key under the query, The final degree of association.

[0157] Based on the above formula, we get the calculation formula of the probabilistic sparse self-attention mechanism, so that the key only pays attention to the previous A query of high importance, namely:

[0158] (5')

[0159] in, is the calculation process of the probabilistic sparse self-attention mechanism, is the operation of the normalized exponential function, for The transpose of the matrix, To include only the front A sparse matrix of queries.

[0160] In addition, self-attention distillation assigns higher weights to the dominant features with main information, and The layer generates the self-attention feature map focused on the previous layer. The formula is as follows:

[0161] (6')

[0162] in, For the Layer Matrix, For the Layer Matrix represents the maximum pooling operation, is the activation function, Indicates a one-dimensional convolution operation on the time series. It is the attention module, that is, the data distillation layer.

[0163] At the decoding sub-module level, the Informer module proposes a generative decoding method that can directly generate all prediction results, thereby significantly improving the prediction speed of the model and making the Informer module more efficient and accurate when processing photovoltaic output prediction tasks.

[0164] The input of the photovoltaic combination forecasting model established by combining the photovoltaic network module with the informer module is specifically the historical complex local weather radiation intensity, temperature, humidity, wind speed meteorological data and photovoltaic power data. The output of the photovoltaic network of the traditional photovoltaic physical model is used as one of the input features of the informer module. At the same time, the historical meteorological data under complex local weather is used as an additional time series feature input. The sparse attention mechanism is used in the encoding submodule of the informer module to extract the time series features of the input features and generate hidden layer representations. In the decoding submodule of the informer module, the hidden layer representation is mapped to the power prediction at the future moment through the multi-head attention mechanism.

[0165] When training the established photovoltaic combination prediction model, since the output of the traditional photovoltaic physical model has physical constraints and interpretability, the training goal of the aforementioned photovoltaic combination prediction model is to minimize the prediction error as much as possible. The loss function of the photovoltaic combination prediction model is defined as follows:

[0166] (7')

[0167] (8')

[0168] (9')

[0169] in, is the loss function of the PV combination prediction model; is the loss weight coefficient; is the loss function of the photovoltaic network module, which is used to measure the predicted output power of the photovoltaic network module The actual power value The error, Output power The index of ; is the timing measurement error of the Informer module, is the final prediction output of the Informer module, i.e., the photovoltaic output prediction data. is the actual photovoltaic output data, The final prediction output The index of .

[0170] Finally, the trained photovoltaic forecast combination model is used to obtain photovoltaic output forecast data under complex local weather conditions.

[0171] like Figure 6 FIG. 1 is a schematic diagram of obtaining the predicted daily result using the method of this embodiment. Figure 6 It can be seen that the shape of the actual power generation on the predicted day is similar to an arch, that is, the actual power generation of the photovoltaic power station is affected by meteorological conditions such as solar radiation intensity and temperature, which leads to volatility and uncertainty in photovoltaic output. Similarly, the prediction results based on the method of this embodiment also present a similar arch shape, which is basically consistent with the trend of the actual value. It can be seen that the method of this embodiment has high accuracy in predicting photovoltaic output data, and can further provide decision support for power system scheduling and planning.

[0172] This example overcomes the shortcomings of traditional photovoltaic physical modeling methods, which are difficult to obtain due to the difficulty in obtaining the physical parameters of photovoltaic sites and are not very practical. The traditional photovoltaic physical model is converted into a photovoltaic network module. After deep integration with the Informer module, the model parameters can be driven and optimized through a large amount of historical photovoltaic power generation data and complex local meteorological data, so as to better support grid dispatchers in making decisions on photovoltaic output forecasts under complex local weather conditions.

[0173] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0174] Based on the same inventive concept, the embodiment of the present application also provides a photovoltaic output prediction device based on a combined prediction model for implementing the photovoltaic output prediction method based on a combined prediction model involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the photovoltaic output prediction device based on a combined prediction model provided below can refer to the limitations of the photovoltaic output prediction method based on a combined prediction model above, and will not be repeated here.

[0175] In an exemplary embodiment, Figure 7 As shown, a photovoltaic output prediction device based on a combined prediction model is provided, comprising: a data acquisition module 701, a first model construction module 702, a second model construction module 703, a model fusion and training module 704 and a data prediction module 705, wherein:

[0176] The data acquisition module 701 is used to acquire historical photovoltaic data.

[0177] The first model building module 702 is used to build a photovoltaic physical model based on historical photovoltaic data and convert the photovoltaic physical model into a photovoltaic network module.

[0178] The second model building module 703 is used to build an Informer module.

[0179] The model fusion and training module 704 is used to establish a photovoltaic combination prediction model based on the photovoltaic network module and the Informer module, and train the photovoltaic combination prediction model using historical photovoltaic data.

[0180] The data prediction module 705 is used to predict the photovoltaic output data of the photovoltaic cells based on the trained photovoltaic combination prediction model to obtain photovoltaic output prediction data.

[0181] In an optional embodiment, the first model building module 702 is also used to determine the first model parameters according to the equivalent circuit of the photovoltaic cell, and build a photovoltaic physical model based on the first model parameters and the historical photovoltaic data; and use a neural network algorithm to convert the photovoltaic physical model into a photovoltaic network module.

[0182] In an optional embodiment, the second model construction module 703 is further used to construct a coding submodule and a decoding submodule, and to obtain an Informer module based on the coding submodule and the decoding submodule.

[0183] In an optional embodiment, the model fusion and training module 704 is also used to establish a photovoltaic combination prediction model based on the photovoltaic network module and the Informer module; define the second model parameters and loss function of the photovoltaic combination prediction model; input historical photovoltaic data into the photovoltaic combination prediction model, and output the first predicted power based on the forward transfer of the photovoltaic combination prediction model; reversely transfer the first predicted power to the photovoltaic combination prediction model as the input of the photovoltaic combination prediction model, and adjust the second model parameters to iteratively train the photovoltaic combination prediction model until the loss function converges or reaches the maximum number of iterations, so as to obtain a trained photovoltaic combination prediction model.

[0184] In an optional embodiment, the model fusion and training module 704 is also used to input historical meteorological data and historical photovoltaic output data into the photovoltaic network module, and output a second predicted power based on the forward transfer of the photovoltaic network module; align the historical meteorological data and the historical photovoltaic output data in time to form photovoltaic time series data; input the photovoltaic time series data and the second predicted power into the encoding submodule, and capture the first association information between the data at different positions in the photovoltaic time series data based on the multi-head probabilistic sparse self-attention layer of the encoding submodule, and extract the data at the corresponding position in the photovoltaic time series data according to the first association information to form a first feature sequence; then, based on the data distillation layer of the encoding submodule, perform redundant information distillation on the first feature sequence to generate a hidden layer representation; input the photovoltaic time series data and the hidden layer representation into the decoding submodule, and capture the second association relationship between the data before the current position in the photovoltaic time series data based on the masked multi-head probabilistic sparse self-attention layer of the decoding submodule, and extract the data at the corresponding position in the photovoltaic time series data according to the second association relationship information to form a second feature sequence; then, based on the multi-head self-attention layer of the decoding submodule, fuse the second feature sequence with the hidden layer representation generated by the encoder submodule, output the prediction result at one time, and obtain the first predicted power.

[0185] Each module in the photovoltaic output prediction device based on the combined prediction model can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0186] In an exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG8. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store historical meteorological data, historical photovoltaic output data, and various intermediate power data in the prediction process. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a photovoltaic output prediction method based on a combined prediction model is implemented.

[0187] Those skilled in the art will understand that the structure shown in FIG. 8 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0188] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the photovoltaic output prediction method based on the combined prediction model of the above embodiment is implemented.

[0189] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the photovoltaic output prediction method based on the combined prediction model of the above embodiment is implemented.

[0190] In one embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements the photovoltaic output prediction method based on the combined prediction model of the above embodiment.

[0191] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0192] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0193] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0194] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A photovoltaic output prediction method based on a combined prediction model, characterized in that: The method comprises: Obtain historical photovoltaic data; Building a photovoltaic physical model based on the historical photovoltaic data, and converting the photovoltaic physical model into a photovoltaic network module; Build the Informer module; Establishing a photovoltaic combination prediction model based on the photovoltaic network module and the Informer module, and training the photovoltaic combination prediction model using the historical photovoltaic data; The photovoltaic output data of the photovoltaic cells is predicted based on the trained photovoltaic combination prediction model to obtain photovoltaic output prediction data.

2. The method according to claim 1, characterized in that: The historical photovoltaic data includes historical meteorological data and historical photovoltaic output data of the photovoltaic cell under each historical meteorological data, and the historical meteorological data at least includes radiation intensity, temperature, humidity and / or wind speed.

3. The method according to claim 2, characterized in that The step of constructing a photovoltaic physical model based on the historical photovoltaic data and converting the photovoltaic physical model into a photovoltaic network module includes: Determine a first model parameter according to an equivalent circuit of a photovoltaic cell, and construct a photovoltaic physical model based on the first model parameter and the historical photovoltaic data; The photovoltaic physical model is converted into a photovoltaic network module using a neural network algorithm.

4. The method according to claim 3, characterized in that: The construction of the Informer module includes: Building a coding submodule and a decoding submodule, and constructing an Informer module based on the coding submodule and the decoding submodule; The encoding submodule is used to extract first correlation information between the historical meteorological data and the historical photovoltaic output data to form a first feature sequence, and to generate a hidden layer representation after performing redundant information distillation on the first feature sequence; The decoding submodule is used to extract the second associated information of the historical photovoltaic parameters to form a second feature sequence, and fuse the second feature sequence with the hidden layer representation generated by the encoder submodule to output a prediction result.

5. The method according to claim 4, characterized in that The photovoltaic combination prediction model is established based on the photovoltaic network module and the informer module, and the photovoltaic combination prediction model is trained by using the historical photovoltaic data, including: Establishing a photovoltaic combination prediction model based on the photovoltaic network module and the Informer module; defining a second model parameter and a loss function of the photovoltaic combination prediction model; Inputting the historical photovoltaic data into the photovoltaic combination prediction model, and outputting a first predicted power based on the forward transfer of the photovoltaic combination prediction model; The first predicted power is transmitted back to the photovoltaic combination prediction model as the input of the photovoltaic combination prediction model, and the second model parameters are adjusted to iteratively train the photovoltaic combination prediction model until the loss function converges or reaches the maximum number of iterations, so as to obtain a trained photovoltaic combination prediction model.

6. The method according to claim 5, characterized in that The step of inputting the historical photovoltaic data into the photovoltaic combination prediction model and outputting a first predicted power based on the forward transfer of the photovoltaic combination prediction model comprises: Inputting the historical meteorological data and the historical photovoltaic output data into the photovoltaic network module, and outputting a second predicted power based on the forward transfer of the photovoltaic network module; Aligning the historical meteorological data and the historical photovoltaic output data in time to form photovoltaic time series data; The photovoltaic time series data and the second predicted power are input into the encoding submodule, and the first association information between the data at different positions in the photovoltaic time series data is captured based on the multi-head probabilistic sparse self-attention layer of the encoding submodule, and the data at the corresponding position in the photovoltaic time series data is extracted according to the first association information to form a first feature sequence; and the data distillation layer of the encoding submodule performs redundant information distillation on the first feature sequence to generate a hidden layer representation; The photovoltaic time series data and the hidden layer representation are input into the decoding submodule, and the second association relationship between the data before the current position in the photovoltaic time series data is captured based on the masked multi-head probabilistic sparse self-attention layer of the decoding submodule, and the data at the corresponding position in the photovoltaic time series data is extracted according to the second association relationship information to form a second feature sequence; then, the second feature sequence is fused with the hidden layer representation generated by the encoder submodule based on the multi-head self-attention layer of the decoding submodule, and the prediction result is output at one time to obtain the first prediction power.

7. A photovoltaic output prediction device based on a combined prediction model, characterized in that: The device comprises: Data acquisition module, used to obtain historical photovoltaic data; A first model building module, used to build a photovoltaic physical model based on the historical photovoltaic data, and convert the photovoltaic physical model into a photovoltaic network module; The second model building module is used to build the Informer module; A model fusion and training module, used to establish a photovoltaic combination prediction model based on the photovoltaic network module and the Informer module, and train the photovoltaic combination prediction model using the historical photovoltaic data; The data prediction module is used to predict the photovoltaic output data of the photovoltaic cell based on the trained photovoltaic combination prediction model to obtain photovoltaic output prediction data.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images