Method for Improving the Prediction Accuracy of Photovoltaic Power Generation Capacity

Abstract

The present invention provides a method for improving the prediction accuracy of photovoltaic power generation capacity, which relates to the technical field of photovoltaic power generation prediction. The method includes: obtaining the photovoltaic power generation power within a historical time period and the set of meteorological characteristic parameters for the corresponding time period, and dividing the weather types of the historical time period according to the photovoltaic power generation power and the meteorological characteristic parameters; respectively using the photovoltaic power generation power and the set of meteorological characteristic parameters corresponding to each weather type as a training sample set to construct a photovoltaic power generation power prediction model for the corresponding weather type; determining the weather type of the day to be measured according to the set of meteorological characteristic parameters of the historical time period and the set of meteorological characteristic parameters obtained for the day to be measured, and inputting the set of meteorological characteristic parameters of the day to be measured into the photovoltaic power generation prediction model of the corresponding weather type to obtain the photovoltaic power generation power of the day to be measured. This solution can achieve rapid and accurate prediction of photovoltaic power generation power.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of photovoltaic power generation prediction, and particularly to a method for improving the prediction accuracy of photovoltaic power generation capacity. Background Art

[0002] In the global context of addressing climate change and promoting sustainable energy policies, the development and utilization of renewable energy have become strategic priorities. As an important form of renewable energy, photovoltaic power generation is widely used in power production due to its advantages such as pollution-free and renewable. However, its power generation is affected by multiple factors such as solar radiation, weather, and seasons, showing high volatility, which poses challenges to the stable operation of the power system. Therefore, accurate photovoltaic power generation prediction is crucial for optimizing power dispatching and improving system efficiency.

[0003] In the research field of photovoltaic power prediction, photovoltaic power prediction mainly includes three types of methods. Among them, one type of method is to establish a prediction model based on historical power generation data to predict the future power generation change trend; another type of method combines the data of numerical weather prediction (NWP), and improves the prediction accuracy by integrating short-term and long-term meteorological prediction information. This type of method uses the detailed forecast data provided by advanced meteorological models to better capture the key factors affecting photovoltaic power generation. Another type of method relies on satellite cloud observation data and multi-source data fusion technology, focusing on analyzing the time-series changes of atmospheric cloud maps or integrating different types of data resources to significantly enhance the accuracy and robustness of the prediction solution.

[0004] However, the above methods can only effectively predict stable weather. In the face of extreme meteorological conditions or rapidly changing weather conditions such as rapid cloud cover, the above methods cannot fully consider real-time weather changes, so it is difficult to accurately capture the impact of weather changes on photovoltaic output power, and thus it is impossible to accurately predict the power generation capacity of photovoltaic power generation systems.

[0005] Therefore, there is an urgent need to provide a method for improving the prediction accuracy of photovoltaic power generation capacity to meet the dispatching requirements in practical engineering. Summary of the Invention

[0006] In order to solve the problem that traditional photovoltaic power generation prediction methods are difficult to accurately capture the impact of weather changes on photovoltaic output power and cannot accurately predict the power generation capacity of photovoltaic power generation systems, the embodiments of the present invention provide a method for improving the prediction accuracy of photovoltaic power generation capacity.

[0007] In a first aspect, the embodiments of the present invention provide a method for improving the prediction accuracy of photovoltaic power generation capacity, the method comprising:

[0008] Obtain the photovoltaic power generation in a historical time period and the set of meteorological characteristic parameters for the corresponding time period, and classify the weather types in the historical time period according to the photovoltaic power generation and meteorological characteristic parameters; wherein, the weather types include simple weather and complex weather;

[0009] Respectively use the photovoltaic power generation and the set of meteorological characteristic parameters corresponding to each weather type as the training sample set to construct a photovoltaic power generation prediction model for the corresponding weather type; wherein, when training the photovoltaic power generation prediction model corresponding to the complex weather type, evaluate the correlation between each meteorological characteristic parameter and photovoltaic power generation at each preset time interval, and update the set of meteorological characteristic parameters input into the prediction model in real time according to this correlation;

[0010] Determine the weather type of the day to be measured according to the set of meteorological characteristic parameters in the historical time period and the set of meteorological characteristic parameters obtained on the day to be measured, and input the set of meteorological characteristic parameters of the day to be measured into the photovoltaic power generation prediction model of the corresponding weather type to obtain the photovoltaic power generation of the day to be measured.

[0011] In a second aspect, an embodiment of the present invention further provides a device for improving the prediction accuracy of photovoltaic power generation capacity, the device includes:

[0012] A division unit, configured to obtain the photovoltaic power generation in a historical time period and the set of meteorological characteristic parameters for the corresponding time period, and classify the weather types in the historical time period according to the photovoltaic power generation and meteorological characteristic parameters; wherein, the weather types include simple weather and complex weather;

[0013] A construction unit, configured to respectively use the photovoltaic power generation and the set of meteorological characteristic parameters corresponding to each weather type as the training sample set to construct a photovoltaic power generation prediction model for the corresponding weather type; wherein, when training the photovoltaic power generation prediction model corresponding to the complex weather type, evaluate the correlation between each meteorological characteristic parameter and photovoltaic power generation at each preset time interval, and update the set of meteorological characteristic parameters input into the prediction model in real time according to this correlation;

[0014] A prediction unit, configured to determine the weather type of the day to be measured according to the set of meteorological characteristic parameters in the historical time period and the set of meteorological characteristic parameters obtained on the day to be measured, and input the set of meteorological characteristic parameters of the day to be measured into the photovoltaic power generation prediction model of the corresponding weather type to obtain the photovoltaic power generation of the day to be measured.

[0015] In a third aspect, an embodiment of the present invention further provides a computing device, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the method described in any embodiment of this specification is implemented.

[0016] Fourthly, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed on a computer, the computer is made to execute the method described in any embodiment of this specification.

[0017] Fifthly, an embodiment of the present application further provides a computer program product. The computer program product includes a computer program. A processor of a computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the method described in any of the above embodiments.

[0018] An embodiment of the present invention provides a method for improving the prediction accuracy of photovoltaic power generation ability. First, according to the photovoltaic power generation power and the set of meteorological characteristic parameters in the historical time period, the weather types in the historical time period are divided into complex weather with dynamic changes and simple weather with single changes. Then, prediction models for different weather types are trained based on the historical data corresponding to each weather type. When training the prediction model for photovoltaic power generation in complex weather with dynamic changes, the influence of each meteorological characteristic parameter in the set of meteorological characteristic parameters on photovoltaic power generation is evaluated and the correlation is analyzed at regular intervals. The meteorological characteristic parameters affecting the prediction result are identified and adjusted in real time according to the real-time weather changes and historical data, so as to adaptively adjust the input feature set of the photovoltaic power generation prediction model, and the power generation power of the photovoltaic power generation system is predicted based on the constructed hybrid prediction model. In this way, not only the prediction accuracy of the photovoltaic power generation system is improved, but also the adaptability of the photovoltaic power generation system to different weather condition changes is enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0020] Figure 1 is a flowchart of a method for improving the prediction accuracy of photovoltaic power generation ability provided by an embodiment of the present invention;

[0021] Figure 2 is a hardware architecture diagram of a computing device provided by an embodiment of the present invention;

[0022] Figure 3 is a structural diagram of a device for improving the prediction accuracy of photovoltaic power generation ability provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0024] The following describes the specific implementation manners of the above concepts.

[0025] Please refer to Figure 1 , the embodiments of the present invention provide a method for improving the prediction accuracy of photovoltaic power generation ability. The method includes:

[0026] Step 100: Obtain the photovoltaic power generation power within a historical time period and the set of meteorological characteristic parameters for the corresponding time period, and classify the weather types of the historical time period according to the photovoltaic power generation power and the meteorological characteristic parameters; wherein, the weather types include simple weather and complex weather;

[0027] Step 102: Respectively use the photovoltaic power generation power and the set of meteorological characteristic parameters corresponding to each weather type as a training sample set to construct a photovoltaic power generation power prediction model for the corresponding weather type; wherein, when training the photovoltaic power generation prediction model corresponding to the complex weather type, evaluate the correlation between each meteorological characteristic parameter and photovoltaic power generation at each preset time interval, and update the set of meteorological characteristic parameters input into the prediction model in real time according to the correlation;

[0028] Step 104: Determine the weather type of the day to be measured according to the set of meteorological characteristic parameters of the historical time period and the set of meteorological characteristic parameters obtained for the day to be measured, and input the set of meteorological characteristic parameters of the day to be measured into the photovoltaic power generation prediction model of the corresponding weather type to obtain the photovoltaic power generation power of the day to be measured.

[0029] In the embodiments of the present invention, first, according to the photovoltaic power generation and the set of meteorological characteristic parameters in the historical time period, the weather types in the historical time period are divided into complex weather with dynamic changes and simple weather with single changes. Then, prediction models for different weather types are trained based on the historical data corresponding to each weather type. When training the prediction model for photovoltaic power generation in complex weather with dynamic changes, by regularly evaluating the influence of each meteorological characteristic parameter in the set of meteorological characteristic parameters on photovoltaic power generation and performing correlation analysis, and in real time identifying and adjusting the characteristics that affect the prediction result according to the real-time weather changes and historical data, the input characteristics of the prediction model are adaptively adjusted, and the power generation capacity of the photovoltaic power generation system is predicted based on the constructed new hybrid prediction model. In this way, not only the prediction accuracy of the photovoltaic power generation system is improved, but also the adaptability of the photovoltaic power generation system to different weather condition changes is enhanced.

[0030] Regarding step 100:

[0031] Considering that photovoltaic power generation is affected by meteorological factors such as cloud cover and irradiance changes, the power output by the photovoltaic power generation system in complex weather has obvious randomness and volatility. Under the simple weather type with single changes, the power output of the photovoltaic power generation system is relatively stable. Therefore, in the embodiments of the present invention, first, based on the historical power generation data of photovoltaic power generation and the changes in the meteorological characteristic parameters in the corresponding time period, the weather types are divided into complex weather and simple weather. Complex weather can be, for example, cloudy weather, and simple weather can be, for example, sunny, rainy or cloudy days. Different prediction models are constructed based on different weather types respectively, so as to accurately predict the power generation capacity of the photovoltaic power generation system under different meteorological conditions.

[0032] In some embodiments, the set of meteorological characteristic parameters includes solar radiation, cloud optical thickness, aerosol optical thickness, atmospheric temperature, atmospheric humidity, atmospheric transparency and air pressure;

[0033] In the embodiments of the present invention, by comprehensively considering the meteorological characteristic parameters affecting the photovoltaic power generation system and using multiple meteorological characteristic parameters to construct a multi-variable photovoltaic prediction model, it is beneficial to accurately predict the photovoltaic power generation capacity under complex and changeable meteorology.

[0034] In some embodiments, in step 100, the division of the weather type in the historical time period according to the photovoltaic power generation and the meteorological characteristic parameters includes:

[0035] Determine the first division threshold and the second division threshold according to the daily solar radiation and cloud optical thickness data in the historical time period;

[0036] Calculate the time series complexity of the daily photovoltaic power generation for a historical time period, and compare the time series complexity calculated daily with the first division threshold;

[0037] If the time series complexity of the current day is greater than the first division threshold, then classify the current day as a complex weather type;

[0038] If the time series complexity of the current day is less than the first division threshold, then perform differencing on the photovoltaic power generation at adjacent moments in the current day to obtain a number of time series difference values;

[0039] Compare each of the time series difference values with the second division threshold respectively. If there are a preset number of time series difference values greater than the second division threshold, then classify the current day as a complex weather type; if not, then classify the current day as a simple weather type.

[0040] In some specific embodiments, the first division threshold and the second division threshold are respectively determined by the following formulas:

[0041]

[0042]

[0043] In the formula, A 1 is the first division threshold, n is the total amount of daily solar radiation sample data in the historical time period, L i is the i-th sample data in the solar radiation sample data, is the average value of the daily cloud optical thickness sample data in the historical time period.

[0044] In the embodiments of the present invention, first, calculate the time series complexity of the daily photovoltaic power generation data in the historical time period to measure the complexity of the power generation sequence during this period, and classify its complexity according to the first division threshold. When the time series complexity is less than the first division threshold, further perform differencing on the power generation at adjacent moments, so as to further measure the magnitude of the power generation fluctuation at adjacent moments, and further classify the weather type of the current day based on this time series difference value. The two complement each other, so as to improve the accuracy of weather type classification on the basis of comprehensively considering power generation information in multiple dimensions. Moreover, this method is applicable to the meteorological condition division in different regions.

[0045] In some embodiments, after obtaining the photovoltaic power generation in the historical time period and the set of meteorological characteristic parameters for the corresponding time period, before classifying the weather type of the historical time period based on this, it further includes the step of respectively preprocessing the photovoltaic power generation data.

[0046] Considering that in the actual operation process of a photovoltaic power generation system, missing values or outliers will inevitably appear in the photovoltaic power generation data, and these abnormal conditions will affect the prediction accuracy of the prediction model. Therefore, in the embodiments of the present invention, the obtained photovoltaic power generation data is first preprocessed, and the preprocessing includes standardizing the data and filling in missing data or identifying abnormal data, etc., so as to improve the training efficiency and prediction accuracy of the prediction model.

[0047] In the embodiments of the present invention, a learning model combining a gated filling recurrent unit and a generative adversarial network is used to fill in the photovoltaic power generation data. Specifically, the learning model includes a generator and a discriminator, and both the generator and the discriminator are composed of a gated filling recurrent unit and a fully connected neural network layer. The gated filling recurrent unit is used to generate data, and the fully connected neural network layer is used to realize the conversion and control of data dimensions. During the training process, the photovoltaic power generation data is input into the generator, and the generator uses the gated filling recurrent unit to preliminarily fill in the missing values in the input data to generate a relatively reasonable first-round filling result. Based on the initial filling values provided by the gated filling recurrent unit of the generative adversarial network, more refined and actual distribution-compliant filling data is generated. The discriminator is used to evaluate whether the generated data is real and provide feedback to the generator for improvement. Moreover, during the model training process, the generative adversarial network model uses the Earth Mover's Distance as the loss function to measure the difference between the real data distribution and the generated data distribution. In this way, the difference between the real data and the generated data distribution can be more accurately reflected, so that the model training is more stable and the quality of the generated photovoltaic power generation data is improved.

[0048] In the embodiments of the present invention, the meteorological characteristic parameters are obtained by satellite remote sensing. The meteorological data obtained thereby can not only provide more extensive, more detailed and more real-time meteorological information, but also significantly improve the accuracy and robustness of the prediction model.

[0049] Regarding step 102:

[0050] In some embodiments, in step 102, the evaluating the correlation between each meteorological characteristic parameter and photovoltaic power generation at each preset time interval and updating the set of meteorological characteristic parameters input into the prediction model in real time according to the correlation includes:

[0051] Calculating the transfer information between each meteorological characteristic parameter in the set of meteorological characteristic parameters and the photovoltaic power generation power respectively, and adding the meteorological characteristic parameters corresponding to the transfer information exceeding the preset threshold to the initial training feature set;

[0052] Calculating the interaction correlation between each meteorological characteristic parameter in the set of meteorological characteristic parameters, and adding the feature with the maximum interaction correlation to the initial training feature set;

[0053] At each preset time interval, the transfer information and interaction correlation of each meteorological feature parameter in the initial training feature set are recalculated, and the meteorological feature parameter set in the training sample set is updated in real time based on the calculation results.

[0054] In some specific embodiments, the transfer information between each meteorological feature parameter in the meteorological feature parameter set and the photovoltaic power generation is calculated by the following formula:

[0055]

[0056] In the formula, is the transfer information between one of the meteorological feature parameters and the photovoltaic power generation data, is the joint probability distribution between one of the meteorological feature parameters and the photovoltaic power generation data, is the marginal probability distribution of one of the meteorological feature parameters, is the marginal probability distribution of the photovoltaic power generation data;

[0057] The interaction correlation between each meteorological feature parameter in the meteorological feature parameter set is calculated by the following formula:

[0058]

[0059] In the formula, is the interaction correlation between two meteorological feature parameters under the condition of known photovoltaic power generation data, is the marginal probability distribution of another meteorological feature parameter, is the joint probability distribution of two of the meteorological feature parameters under the condition of known photovoltaic power generation data, is the marginal probability distribution of one of the two meteorological feature parameters under the condition of known photovoltaic power generation data, is the marginal probability distribution of one of the two meteorological feature parameters under the condition of known photovoltaic power generation data, X and Y are different types of meteorological feature parameters, x and y are the values of the corresponding meteorological feature parameters, Z is the photovoltaic power, and z is the value of the photovoltaic power generation data.

[0060] Considering that the environmental and meteorological conditions are constantly changing, over time, the importance of the meteorological feature parameters affecting the photovoltaic power generation will change. For example, at noon, the solar radiation intensity is high, and at this time, cloud clusters and aerosols will significantly affect the penetration ability of solar radiation. Considering these two factors is crucial for accurately predicting the photovoltaic power generation at this time. In the afternoon, the cloud clusters change frequently and the atmospheric conditions are unstable, etc., which will make the weather conditions more complex. At this time, all relevant meteorological feature parameters must be considered to accurately predict the photovoltaic power generation.

[0061] Therefore, in the embodiments of the present invention, during the training process, the influence and correlation of the input meteorological characteristic parameters are evaluated regularly according to the real-time meteorological characteristic parameter data and photovoltaic power generation data, so as to evaluate the correlation between the meteorological characteristic parameters and the photovoltaic power generation regularly within a preset time interval. Thus, the meteorological characteristics with greater correlation are automatically selected according to the evaluation results, and the meteorological characteristics with smaller correlation are removed, and then the set of meteorological characteristic parameters input into the prediction model is dynamically adjusted. Further, in the embodiments of the present invention, the above method is used to evaluate the correlation between each meteorological characteristic parameter and the photovoltaic power generation and the interactive correlation between the meteorological characteristic parameters. In this way, the accuracy of the selection of the meteorological characteristic parameters in the input prediction model can be effectively guaranteed, which is not only beneficial to improving the efficiency and prediction accuracy of the prediction model, but also enables the prediction model to adapt to different weather changes.

[0062] In the embodiments of the present invention, when using the model for prediction, since the above feature selection method considers the temporal correlation between the meteorological characteristic parameters while selecting the features with high correlation with the photovoltaic power generation, therefore, by dynamically and adaptively selecting the set of meteorological characteristic parameters at a specific time, the set of meteorological characteristic parameters input into the prediction model can be adjusted according to the time sequence, so as to achieve accurate prediction of the photovoltaic power generation.

[0063] Regarding step 104:

[0064] In some embodiments, in step 104, determining the weather type of the day to be measured according to the set of meteorological characteristic parameters in the historical time period and the set of meteorological characteristic parameters obtained for the day to be measured includes:

[0065] Calculate the correlation coefficient between any one meteorological characteristic parameter in the set of meteorological characteristic parameters obtained for the day to be measured and the corresponding meteorological characteristic parameter in each historical day respectively, so as to determine the degree of association between each historical day and the day to be measured;

[0066] Calculate the Euclidean distance between any one meteorological characteristic parameter in the set of meteorological characteristic parameters obtained for the day to be measured and the corresponding meteorological characteristic parameter in each historical day respectively, so as to determine the similarity between each historical day and the day to be measured;

[0067] Select the photovoltaic power generation of the historical day corresponding to the maximum degree of association value and the photovoltaic power generation of the historical day corresponding to the maximum similarity value to construct a weather type evaluation function; wherein, the weather type evaluation function is a unary quadratic function about the photovoltaic power generation.

[0068] According to the meteorological parameters and photovoltaic power generation of any day corresponding to each weather type in the historical time period, determine the coefficient range corresponding to each weather type in the weather type evaluation function;

[0069] Iteratively adjust the coefficients in the weather type evaluation function until the deviation between the photovoltaic power generation calculated by the weather type evaluation function and the photovoltaic power generation value of the day to be measured is minimized, so as to determine the weather type of the day to be measured according to the range where the current coefficients are located.

[0070] In some specific embodiments, the correlation coefficient between any meteorological characteristic parameter in the set of meteorological characteristic parameters obtained for the day to be measured and the corresponding meteorological characteristic parameter in each historical day is calculated by the following formula:

[0071]

[0072] In the formula, is the correlation coefficient between the day to be measured and the d-th historical day for the k-th meteorological characteristic parameter, y(k) is the k-th meteorological characteristic parameter in the day to be measured, and x d (k) is the k-th meteorological characteristic parameter of the d-th historical day. is the discrimination coefficient, and its value ranges from 0 to 1.

[0073] The correlation degree between each historical day and the day to be measured is calculated by the following formula:

[0074]

[0075] In the formula, is the correlation degree between the d-th historical day and the day to be measured, m is the total number of meteorological characteristic parameters, and w k is the weight of the k-th meteorological characteristic parameter.

[0076] In some specific embodiments, the similarity between each historical day and the day to be measured is calculated by the following formula:

[0077]

[0078] In the formula, is the similarity between the d-th historical day and the day to be measured, is the k-th meteorological characteristic parameter in the d-th historical day, is the k-th meteorological characteristic parameter of the day to be measured, and λ k is the weight of the k-th meteorological characteristic parameter.

[0079] In some specific embodiments, the weather type evaluation function is constructed and calculated by the following formula:

[0080]

[0081] In the formula, F(P) is the photovoltaic power generation solved by the weather type evaluation function, is the coefficient of the weather type evaluation function. The photovoltaic power generation corresponding to the historical day with the highest correlation degree with the day to be measured The photovoltaic power generation corresponding to the historical day with the highest similarity with the day to be measured

[0082] In the embodiments of the present invention, the correlation between different meteorological characteristic variable parameters and similar days is considered, and the above two calculation methods are combined to determine the weather type of the day to be measured according to the data of historical days. Since this method not only considers the relative influence of each meteorological characteristic parameter on the weather type, but also can reduce the influence of outliers on the result through weighted adjustment. At the same time, it can dynamically adjust the weights according to historical data and real-time observations to adapt to different meteorological conditions and environmental changes, and further construct a weather type evaluation function by using the two, which can capture the small changes between variables, help to identify subtle but important meteorological characteristics, so as to more accurately measure the similarity of similar days under different weather types, improve the accuracy and efficiency of weather type determination, and further improve the accuracy and reliability of photovoltaic power generation prediction.

[0083] In some specific embodiments, when the weather type is complex weather, the photovoltaic power generation prediction model includes an encoding network and a decoding network connected in sequence. The encoding network is used to perform non-linear transformation, convolution operation and normalization processing on the input meteorological characteristic parameter set in sequence to identify the dependence relationship between features at different positions in the time series. The decoding network is used to obtain the photovoltaic power generation based on the dependence relationship between the feature information output by the encoding network.

[0084] In some specific embodiments, the encoding network includes a first feed-forward neural network layer, a multi-head attention layer, a temporal convolutional network layer, a second feed-forward neural network layer and a normalization layer connected in sequence; wherein, the first feed-forward neural network layer is used to perform preliminary non-linear transformation on the input meteorological characteristic parameter set, the multi-head attention layer is used to extract multi-level features output by the first feed-forward neural network layer, the temporal convolutional network layer is used to extract the features output by the multi-head attention layer, the second feed-forward neural network layer is used to perform non-linear transformation on the features output by the temporal convolutional network layer again, and the normalization layer is used to perform normalization operation on the features output by the second feed-forward neural network;

[0085] The decoding network includes a long short-term memory network layer and a fully connected layer. The long short-term memory network module is used to make predictions based on the medium and long-term dependence relationship of the time series features output by the normalization layer, and the fully connected layer is used to integrate the features output by the long short-term memory network layer to obtain the photovoltaic power generation.

[0086] In the embodiments of the present invention, a prediction model for photovoltaic power generation is constructed based on an encoder-decoder framework. A multi-level feature extraction mechanism is used to capture the complex dependencies of the input feature data. By setting two feed-forward neural network layers, the first feed-forward neural network layer performs a non-linear transformation on the input original feature data, and then the output features are passed to a multi-head attention layer with relative position encoding. This structure captures the dependencies between different positions in the feature data sequence and realizes cross-level feature fusion. The subsequent temporal convolutional network layer synchronously captures the local patterns and global trends of the features output by the multi-head attention layer through multi-scale convolutional kernels. Its output is further processed by the second feed-forward neural network layer, and finally, the normalization layer performs adaptive parameter calibration on the output features, thereby significantly improving the model training convergence efficiency and numerical stability. The encoder gradually enhances the feature representation ability, enabling the model to accurately analyze the complex interaction relationships hidden in the data, so that the prediction model can more accurately capture the complex features of the feature data.

[0087] Furthermore, since the long short-term memory network has the ability to capture long-term dependencies in time series data, the decoder part adopts a long short-term memory network, and the long short-term memory network layer receives the output of the encoder as input features for prediction processing, thereby further enhancing the ability of the prediction model to perceive temporal information and extract long-distance relevant information, and further enhancing the accuracy of photovoltaic power generation prediction.

[0088] In some specific embodiments, the prediction model for photovoltaic power generation is trained in the following manner: The training sample set is divided into a training set, a validation set, and a test set. The constructed deep learning network model is trained, and the hyperparameters and loss function (RMSE function) of the model are set. Subsequently, the historical meteorological feature parameter data output after correlation evaluation is used as input, and iterative training is performed through the model. During the training process, the loss and accuracy of the network model are calculated, and the network structure parameters are iterated until the model converges, obtaining the mapping relationship between the input historical meteorological feature parameters and the photovoltaic power generation, and completing the training.

[0089] It should be noted that since the changes in meteorological features in simple weather are relatively single, when the weather type is simple weather, the prediction model of photovoltaic power can be predicted using a generalized regression neural network or a method combining a neural network with other methods. The research on the prediction model for simple weather is relatively mature, and it will not be elaborated in the embodiments of the present invention.

[0090] In summary, in the embodiments of the present invention, first, the weather types within a historical time period are divided according to the photovoltaic power generation and the set of meteorological characteristic parameters within the historical time period. Then, the weather type of the day to be measured is determined based on the data of the historical time period and input into the corresponding prediction model for prediction. Moreover, when training the photovoltaic power generation prediction model for complex weather with dynamic changes, the input feature set of the photovoltaic power generation prediction model is dynamically and adaptively adjusted, and the power generation of the photovoltaic power generation system is predicted based on the constructed hybrid prediction model. In this way, not only the prediction accuracy of the short-term power of the photovoltaic power generation system is improved, but also the adaptability of the photovoltaic power generation system to different weather condition changes is enhanced.

[0091] As Figure 2 、 Figure 3 shown, the embodiments of the present invention provide a device for improving the prediction accuracy of photovoltaic power generation capacity. The device embodiments can be implemented by software, or by hardware, or by a combination of software and hardware. From the hardware level, as Figure 2 shown, it is a hardware architecture diagram of a computing device where the device for improving the prediction accuracy of photovoltaic power generation capacity provided by the embodiments of the present invention is located. In addition to Figure 2 the shown processor, memory, network interface, and non-volatile memory, the computing device where the device in the embodiments is located usually may also include other hardware, such as a forwarding chip responsible for processing packets, etc. Taking the software implementation as an example, as Figure 3 shown, as a logically meaningful device, it is formed by the CPU of its computing device reading the corresponding computer program in the non-volatile memory into the memory for running.

[0092] A device for improving the prediction accuracy of photovoltaic power generation capacity provided in this embodiment, the device includes:

[0093] A division unit 301, configured to obtain the photovoltaic power generation within a historical time period and the set of meteorological characteristic parameters within the corresponding time period, and divide the weather types within the historical time period according to the photovoltaic power generation and the meteorological characteristic parameters; wherein, the weather types include simple weather and complex weather;

[0094] A construction unit 302, configured to respectively use the photovoltaic power generation and the set of meteorological characteristic parameters corresponding to each weather type as a training sample set to construct a photovoltaic power generation prediction model for the corresponding weather type; wherein, when training the photovoltaic power generation prediction model corresponding to the complex weather type, evaluate the correlation between each meteorological characteristic parameter and photovoltaic power generation at each preset time interval, and update the set of meteorological characteristic parameters input into the prediction model in real time according to the correlation;

[0095] A prediction unit 303 is configured to determine the weather type of a day to be measured according to the set of meteorological characteristic parameters in a historical time period and the set of meteorological characteristic parameters obtained on the day to be measured, and input the set of meteorological characteristic parameters of the day to be measured into a photovoltaic power generation prediction model of the corresponding weather type to obtain the photovoltaic power generation power of the day to be measured.

[0096] In an embodiment of the present invention, the partitioning unit 301 may be configured to execute step 100 in the above method embodiment, the construction unit 302 may be configured to execute step 102 in the above method embodiment, and the prediction unit 303 may be configured to execute step 104 in the above method embodiment.

[0097] In an embodiment of the present invention, in the partitioning unit 301, the set of meteorological characteristic parameters includes solar radiation, cloud optical thickness, aerosol optical thickness, atmospheric temperature, atmospheric humidity, atmospheric transparency, and air pressure;

[0098] When the partitioning unit 301 partitions the weather types in a historical time period according to the photovoltaic power generation and meteorological characteristic parameters, it is configured to perform the following operations:

[0099] Determine a first partitioning threshold and a second partitioning threshold according to the solar radiation and cloud optical thickness data of each day in the historical time period;

[0100] Calculate the time series complexity of the photovoltaic power generation power of each day in the historical time period, and compare the time series complexity calculated for each day with the first partitioning threshold;

[0101] If the time series complexity of the current day is greater than the first partitioning threshold, then the current day is partitioned into a complex weather type;

[0102] If the time series complexity of the current day is less than the first partitioning threshold, then perform differencing on the photovoltaic power generation power at adjacent moments in the current day to obtain a number of time series difference values;

[0103] Compare each of the time series difference values with the second partitioning threshold respectively. If there are a preset number of time series difference values greater than the second partitioning threshold, then the current day is partitioned into a complex weather type. If not, then the current day is partitioned into a simple weather type.

[0104] In an embodiment of the present invention, when the construction unit 302 evaluates the correlation between each meteorological characteristic parameter and photovoltaic power generation at each preset time interval and updates the set of meteorological characteristic parameters input into the prediction model in real time according to the correlation, it is configured to perform the following operations:

[0105] Calculate the transfer information between each meteorological feature parameter in the meteorological feature parameter set and the photovoltaic power generation power respectively, and add the meteorological feature parameters corresponding to the transfer information exceeding the preset threshold to the initial training feature set;

[0106] Calculate the interaction correlation between each meteorological feature parameter in the meteorological feature parameter set, and add the feature with the largest amount of information to the initial training feature set;

[0107] At each preset time interval, recalculate the transfer information and interaction correlation of each meteorological feature parameter in the initial training feature set, and update the input training feature set in real time based on the calculation results.

[0108] In an embodiment of the present invention, when the prediction unit 303 determines the weather type of the day to be measured according to the meteorological feature parameter set in the historical time period and the meteorological feature parameter set obtained for the day to be measured, it is used to perform the following operations:

[0109] Calculate the correlation coefficient between any meteorological feature parameter in the meteorological feature parameter set obtained for the day to be measured and the corresponding meteorological feature parameter in each historical day respectively, so as to determine the correlation degree between each historical day and the day to be measured;

[0110] Calculate the Euclidean distance between any meteorological feature parameter in the meteorological feature parameter set obtained for the day to be measured and the corresponding meteorological feature parameter in each historical day respectively, so as to determine the similarity between each historical day and the day to be measured;

[0111] Select the photovoltaic power generation power of the historical day corresponding to the maximum correlation degree value and the photovoltaic power generation power of the historical day corresponding to the maximum similarity value to construct a weather type evaluation function; wherein, the weather type evaluation function is a unary quadratic function about the photovoltaic power generation power;

[0112] According to the meteorological parameters and photovoltaic power generation power of any day corresponding to each weather type in the historical time period, determine the coefficient range corresponding to each weather type in the weather type evaluation function;

[0113] Iteratively adjust the coefficients in the weather type evaluation function until the deviation between the photovoltaic power generation power of the weather type evaluation function and the photovoltaic power generation power value of the day to be measured is minimized, so as to determine the weather type of the day to be measured according to the range where the current coefficients are located.

[0114] In an embodiment of the present invention, in the construction unit 302, when the weather type is complex weather, the photovoltaic power prediction model includes an encoding network and a decoding network connected in sequence. The encoding network is used to perform non-linear transformation, convolution operation, and normalization processing on the input set of meteorological feature parameters in sequence to identify the dependencies between features at different positions in the time series. The decoding network is used to obtain the photovoltaic power based on the dependencies between the feature information output by the encoding network.

[0115] In an embodiment of the present invention, in the construction unit 302, the encoding network includes a first feed-forward neural network layer, a multi-head attention layer, a temporal convolutional network layer, a second feed-forward neural network layer, and a normalization layer connected in sequence. Among them, the first feed-forward neural network layer is used to perform preliminary non-linear transformation on the input set of meteorological feature parameters. The multi-head attention layer is used to extract multi-level features output by the first feed-forward neural network layer. The temporal convolutional network layer is used to extract features output by the multi-head attention layer. The second feed-forward neural network layer is used to perform non-linear transformation on the features output by the temporal convolutional network layer again. The normalization layer is used to perform normalization operation on the features output by the second feed-forward neural network.

[0116] The decoding network includes a long short-term memory network layer and a fully connected layer. The long short-term memory network module is used to make predictions based on the medium- and long-term dependencies of the time series features output by the normalization layer. The fully connected layer is used to integrate the features output by the long short-term memory network layer to obtain the photovoltaic power.

[0117] It can be understood that the structure schematically shown in the embodiments of the present invention does not constitute a specific limitation on a device for improving the prediction accuracy of photovoltaic power generation ability. In other embodiments of the present invention, a device for improving the prediction accuracy of photovoltaic power generation ability may include more or fewer components than shown in the figure, or combine certain components, or split certain components, or have different component arrangements. The components shown in the figure can be implemented in hardware, software, or a combination of software and hardware.

[0118] Regarding the information interaction, execution process, etc. between the various modules in the above device, since it is based on the same concept as the method embodiment of the present invention, the specific content can be referred to the description in the method embodiment of the present invention and will not be elaborated here.

[0119] The embodiment of the present invention also provides a computing device, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, it implements a method for improving the prediction accuracy of photovoltaic power generation ability in any embodiment of the present invention.

[0120] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the processor is caused to execute a method for improving the prediction accuracy of photovoltaic power generation capacity according to any one of the embodiments of the present invention.

[0121] Specifically, a system or device equipped with a storage medium can be provided, on which software program code for implementing the functions of any one of the above embodiments is stored, and the computer (or CPU or MPU) of the system or device is caused to read and execute the program code stored in the storage medium.

[0122] In this case, the program code read from the storage medium itself can implement the functions of any one of the above embodiments, so the program code and the storage medium storing the program code constitute a part of the present invention.

[0123] Examples of storage media for providing program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, the program code can be downloaded from a server computer via a communication network.

[0124] In addition, it should be clear that not only can the functions of any one of the above embodiments be realized by executing the program code read by the computer, but also by causing an operating system or the like operating on the computer based on the instructions of the program code to complete part or all of the actual operations.

[0125] In addition, it can be understood that the program code read from the storage medium is written into the memory provided in the expansion board inserted into the computer or the memory provided in the expansion module connected to the computer, and then based on the instructions of the program code, the CPU or the like installed on the expansion board or the expansion module is caused to execute part and all of the actual operations, thereby realizing the functions of any one of the above embodiments.

[0126] An embodiment of the present application further provides a computer-readable storage medium, on which at least one instruction, at least one segment of program, a code set or an instruction set is stored. The at least one instruction, at least one segment of program, a code set or an instruction set is loaded and executed by a processor to implement a method for improving the prediction accuracy of photovoltaic power generation capacity provided by the above method embodiments.

[0127] An embodiment of the present application further provides a computer program product. The computer program product includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the method for improving the prediction accuracy of photovoltaic power generation capacity described in any one of the foregoing embodiments.

[0128] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0129] Those of ordinary skill in the art can understand that all or part of the steps of implementing the foregoing method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the foregoing method embodiments; and the foregoing storage medium includes: various media such as ROM, RAM, magnetic disk or optical disc that can store program codes.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the 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 invention.

Claims

1. A method for improving the accuracy of photovoltaic power generation capacity prediction, characterized in that: include: Acquire the photovoltaic power generation power in a historical time period and a set of meteorological characteristic parameters of the corresponding time period, and classify the weather types of the historical time period according to the photovoltaic power generation power and the meteorological characteristic parameters; wherein the weather types include simple weather and complex weather; The photovoltaic power generation and meteorological characteristic parameter sets corresponding to each weather type are used as training sample sets to construct a photovoltaic power generation prediction model for the corresponding weather type. When training the photovoltaic power generation prediction model corresponding to complex weather types, the transfer information between each meteorological characteristic parameter in the meteorological characteristic parameter set and the photovoltaic power generation power is calculated respectively, and the meteorological characteristic parameters corresponding to the transfer information exceeding the preset threshold are added to the initial training feature set; wherein the transfer information is calculated by the following formula: In the formula, is the transfer information between one of the meteorological characteristic parameters and the photovoltaic power generation data. is the joint probability distribution of one of the meteorological characteristic parameters and the photovoltaic power generation data, is the marginal probability distribution of one of the meteorological characteristic parameters, is the marginal probability distribution of photovoltaic power data; The interactive correlation between each meteorological characteristic parameter in the meteorological characteristic parameter set is calculated, and the feature with the largest amount of information is added to the initial training feature set; wherein the interactive correlation is calculated by the following formula: In the formula, is the mutual correlation between two meteorological characteristic parameters when the photovoltaic power generation data is known, is the marginal probability distribution of another meteorological characteristic parameter, is the joint probability distribution of two meteorological characteristic parameters when the photovoltaic power generation data is known, is the marginal probability distribution of one of the two meteorological characteristic parameters when the photovoltaic power generation data is known, When the photovoltaic power generation data is known, the marginal probability distribution of one of the two meteorological characteristic parameters, X, Y are different types of meteorological characteristic parameters, x, y are the values ​​of the corresponding meteorological characteristic parameters, Z is the photovoltaic power generation, and z is the value of the photovoltaic power generation data; Recalculate the transfer information and interaction correlation of each meteorological characteristic parameter in the initial training feature set at each preset time interval, and update the meteorological characteristic parameter set in the input prediction model in real time based on the calculation results; The weather type of the day to be tested is determined based on the meteorological characteristic parameter set of the historical time period and the meteorological characteristic parameter set obtained on the day to be tested, and the meteorological characteristic parameter set of the day to be tested is input into the photovoltaic power generation prediction model of the corresponding weather type to obtain the photovoltaic power generation power of the day to be tested.

2. The method according to claim 1, characterized in that The meteorological characteristic parameter set includes solar radiation, cloud optical thickness, aerosol optical thickness, atmospheric temperature, atmospheric transparency, atmospheric humidity and air pressure; The classification of weather types in historical time periods according to the photovoltaic power generation power and meteorological characteristic parameters includes: Determine the first division threshold and the second division threshold according to the daily solar radiation and cloud optical thickness data in the historical time period; Calculating the time series complexity of photovoltaic power generation power every day in the historical time period, and comparing the time series complexity calculated every day with the first division threshold; If the time series complexity of the current day is greater than the first classification threshold, classifying the current day as a complex weather type; If the time series complexity of the current day is less than the first division threshold, then the photovoltaic power generation power at adjacent moments in the current day is differentiated to obtain a number of time series difference values; Each of the timing difference values ​​is compared with the second classification threshold respectively. If there are a preset number of timing difference values ​​greater than the second classification threshold, the current day is classified as a complex weather type. If not, the current day is classified as a simple weather type.

3. The method according to claim 1, characterized in that The step of determining the weather type of the day to be measured based on the meteorological characteristic parameter set of the historical time period and the meteorological characteristic parameter set obtained on the day to be measured includes: Calculate the correlation coefficient between any meteorological characteristic parameter in the meteorological characteristic parameter set obtained on the day to be measured and the corresponding meteorological characteristic parameter in each historical day to determine the correlation between each historical day and the day to be measured; Calculate the Euclidean distance between any meteorological characteristic parameter in the meteorological characteristic parameter set obtained on the day to be measured and the corresponding meteorological characteristic parameter in each historical day to determine the similarity between each historical day and the day to be measured; Select the photovoltaic power generation power of the historical day corresponding to the maximum correlation value and the photovoltaic power generation power of the historical day corresponding to the maximum similarity value to construct a weather type evaluation function; wherein the weather type evaluation function is a one-variable quadratic function about the photovoltaic power generation power; According to the meteorological parameters and photovoltaic power generation power of any day corresponding to each weather type in the historical time period, determine the coefficient range corresponding to each weather type in the weather type evaluation function; The coefficients in the weather type evaluation function are iteratively adjusted until the deviation between the photovoltaic power generation power of the weather type evaluation function and the photovoltaic power generation power value of the day to be tested is minimized, so as to determine the weather type of the day to be tested according to the range of the current coefficients.

4. The method according to claim 1, characterized in that: When the weather type is complex weather, the photovoltaic power generation prediction model includes an encoding network and a decoding network connected in sequence. The encoding network is used to perform nonlinear transformation, convolution operation and normalization processing on the input meteorological characteristic parameter set in sequence to identify the dependency relationship between features at different positions in the time series, and the decoding network is used to obtain the photovoltaic power generation based on the dependency relationship between the feature information output by the encoding network.

5. The method according to claim 4, characterized in that The encoding network includes a first feedforward neural network layer, a multi-head attention layer, a temporal convolutional network layer, a second feedforward neural network layer and a normalization layer connected in sequence; wherein the first feedforward neural network layer is used to perform a preliminary nonlinear transformation on the input meteorological feature parameter set, the multi-head attention layer is used to extract the multi-level features output by the first feedforward neural network layer, the temporal convolutional network layer is used to extract the features output by the multi-head attention layer, the second feedforward neural network layer is used to perform a nonlinear transformation on the features output by the temporal convolutional network layer again, and the normalization layer is used to perform a normalization operation on the features output by the second feedforward neural network; The decoding network includes a long short-term memory network layer and a fully connected layer. The long short-term memory network layer is used to predict based on the medium- and long-term dependencies of the time series features output by the normalization layer, and the fully connected layer is used to integrate the features output by the long short-term memory network layer to obtain the photovoltaic power generation power.

6. A device for improving the prediction accuracy of photovoltaic power generation capacity, characterized in that: include: A classification unit, used to obtain the photovoltaic power generation in a historical time period and a set of meteorological characteristic parameters of the corresponding time period, and classify the weather type of the historical time period according to the photovoltaic power generation and the meteorological characteristic parameters; wherein the weather type includes simple weather and complex weather; A construction unit, used to construct a photovoltaic power prediction model for the corresponding weather type by taking the photovoltaic power generation and meteorological characteristic parameter set corresponding to each weather type as a training sample set; When training the photovoltaic power generation prediction model corresponding to the complex weather type, the transfer information between each meteorological characteristic parameter in the meteorological characteristic parameter set and the photovoltaic power generation power is calculated respectively, and the meteorological characteristic parameters corresponding to the transfer information exceeding the preset threshold are added to the initial training feature set; wherein the transfer information is calculated by the following formula: In the formula, is the transfer information between one of the meteorological characteristic parameters and the photovoltaic power generation data. is the joint probability distribution of one of the meteorological characteristic parameters and the photovoltaic power generation data, is the marginal probability distribution of one of the meteorological characteristic parameters, is the marginal probability distribution of photovoltaic power data; The interactive correlation between each meteorological characteristic parameter in the meteorological characteristic parameter set is calculated, and the feature with the maximum amount of information is added to the initial training feature set; wherein the interactive correlation is calculated by the following formula: In the formula, is the mutual correlation between two meteorological characteristic parameters when the photovoltaic power generation data is known, is the marginal probability distribution of another meteorological characteristic parameter, is the joint probability distribution of two meteorological characteristic parameters when the photovoltaic power generation data is known, is the marginal probability distribution of one of the two meteorological characteristic parameters when the photovoltaic power generation data is known, When the photovoltaic power generation data is known, the marginal probability distribution of one of the two meteorological characteristic parameters, X, Y are different types of meteorological characteristic parameters, x, y are the values ​​of the corresponding meteorological characteristic parameters, Z is the photovoltaic power generation, and z is the value of the photovoltaic power generation data; Recalculate the transfer information and interaction correlation of each meteorological characteristic parameter in the initial training characteristic set at each preset time interval, and update the input meteorological characteristic parameter set in real time based on the calculation results; The prediction unit is used to determine the weather type of the day to be tested based on the meteorological characteristic parameter set of the historical time period and the meteorological characteristic parameter set obtained on the day to be tested, and input the meteorological characteristic parameter set of the day to be tested into the photovoltaic power generation prediction model of the corresponding weather type to obtain the photovoltaic power generation power of the day to be tested.

7. A computing device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the method according to any one of claims 1 to 5.

9. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 5 when being executed by a processor.

Images