Photovoltaic generating capacity prediction method and device based on dynamic decomposition and digital-analog fusion

Abstract

The invention provides a photovoltaic generating capacity prediction method and device based on dynamic decomposition and digital-analog fusion, and the method comprises the steps: carrying out the feature selection of a historical photovoltaic parameter sequence of a to-be-predicted photovoltaic system in a first historical period, and obtaining a target photovoltaic parameter sequence; performing dynamic decomposition on the target photovoltaic parameter sequence to obtain a historical moving average kernel component and a historical seasonal component in a first historical period; inputting the historical moving average kernel component and the historical seasonal component into a prediction module of a prediction model to obtain a moving average kernel component and a seasonal component in a to-be-predicted period; inputting the moving average kernel component and the seasonal component in the to-be-predicted period into a fusion module of the prediction model to obtain a power generation capacity prediction value of the to-be-predicted photovoltaic system in the to-be-predicted period; the prediction model is trained on the basis of a digital-analog fusion model fusing an unscented Kalman filter model and an extended long-short-term memory network model. According to the invention, the accuracy of photovoltaic generating capacity prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation, and in particular to a method and device for predicting photovoltaic power generation based on dynamic decomposition and digital-analog fusion. Background Art

[0002] As a clean and renewable energy, photovoltaics are crucial to meeting the growing demand for clean energy. Efficient and accurate photovoltaic power generation forecasting is crucial for the effective implementation of energy companies' pricing strategies, safety assessments, reserve management, and maintenance plans. Therefore, how to accurately achieve photovoltaic power generation forecasting is an important issue that the industry needs to solve urgently.

[0003] In related technologies, machine learning algorithms are usually used to train prediction models using a large amount of historical data, thereby realizing the prediction of photovoltaic power generation. However, the performance of this method depends largely on the quality and time span of the input data. The photovoltaic power generation process itself is affected by a variety of natural factors, resulting in a certain degree of randomness in the relevant parameters of photovoltaic power generation, making it difficult to guarantee the quality of the input data, and thus the accuracy of the photovoltaic power generation predicted by this method is also difficult to guarantee.

[0004] Therefore, there is an urgent need for a photovoltaic power generation prediction method and device based on dynamic decomposition and digital-analog fusion to solve the above technical problems. Summary of the invention

[0005] The present invention provides a photovoltaic power generation prediction method and device based on dynamic decomposition and digital-analog fusion, which is used to solve the defect in the prior art that the quality of input data is difficult to guarantee, resulting in difficulty in ensuring the accuracy of photovoltaic power generation prediction, and to improve the accuracy of photovoltaic power generation prediction.

[0006] The present invention provides a photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion, comprising: Perform feature selection on the historical photovoltaic parameter sequence of the photovoltaic system to be predicted in the first historical period to obtain the target photovoltaic parameter sequence; Dynamically decomposing the target photovoltaic parameter sequence to obtain a historical moving average kernel component in the first historical period and a historical seasonal component in the first historical period; Inputting the historical moving average kernel component and the historical seasonal component into a prediction module of a prediction model, and predicting the moving average kernel component under a period to be predicted and the seasonal component under the period to be predicted; Inputting the moving average kernel component under the period to be predicted and the seasonal component under the period to be predicted into the fusion module of the prediction model, and predicting the power generation forecast value of the photovoltaic system to be predicted in the period to be predicted; Among them, the prediction model is obtained by jointly training the sample photovoltaic parameter sequence of the sample photovoltaic system in the second historical period, and the moving average kernel component label, seasonal component label and power generation label in the third historical period on the basis of the digital-analog fusion model; the digital-analog fusion model is constructed based on the fusion of the unscented Kalman filter model and the extended long short-term memory network model.

[0007] According to a photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion provided by the present invention, the target photovoltaic parameter sequence is dynamically decomposed to obtain the historical moving average kernel component under the first historical period and the historical seasonal component under the first historical period, including: Based on the exponential decay model, calculating the weight coefficient of each parameter in the target photovoltaic parameter sequence; According to the weight coefficient, performing weighted moving average decomposition on the target photovoltaic parameter sequence to obtain a historical moving average kernel component in the first historical period; According to the target photovoltaic parameter sequence and the historical moving average kernel component in the first historical period, the historical seasonal component in the first historical period is decomposed to obtain.

[0008] According to a photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion provided by the present invention, the calculation formula of the exponential decay model is as follows: ; in, The timing position in the target photovoltaic parameter sequence is The weight coefficient of the parameter; is the attenuation coefficient; is the number of parameters in the target photovoltaic parameter sequence; and The position number of the parameter.

[0009] According to a photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion provided by the present invention, feature selection is performed on the historical photovoltaic parameter sequence of the photovoltaic system to be predicted in the first historical period to obtain the target photovoltaic parameter sequence, including: Performing outlier detection and outlier filling processing on the historical photovoltaic parameter sequence to obtain a preprocessed historical photovoltaic parameter sequence; Performing importance scoring on each dimension parameter in the sample photovoltaic parameter sequence to obtain an importance score value corresponding to each dimension parameter in the sample photovoltaic parameter sequence; Determining a target parameter from among the multi-dimensional parameters in the sample photovoltaic parameter sequence according to the importance score value; In the preprocessed historical photovoltaic parameter sequence, parameters matching the target parameters are selected to construct the target photovoltaic parameter sequence.

[0010] According to a photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion provided by the present invention, the importance of each dimensional parameter in the sample photovoltaic parameter sequence is scored to obtain the importance score value corresponding to each dimensional parameter in the sample photovoltaic parameter sequence, including: Calculating the correlation between every two dimensional parameters in the sample photovoltaic parameter sequence to obtain a first correlation value corresponding to each dimensional parameter in the sample photovoltaic parameter sequence; Calculating the correlation between each dimensional parameter in the sample photovoltaic parameter sequence and the power generation label to obtain a second correlation value corresponding to each dimensional parameter in the sample photovoltaic parameter sequence; The first correlation value and the second correlation value are combined to obtain the importance score value corresponding to each dimensional parameter in the sample photovoltaic parameter sequence.

[0011] According to a photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion provided by the present invention, the historical moving average kernel component and the historical seasonal component are input into a prediction module of a prediction model to predict the moving average kernel component under the period to be predicted and the seasonal component under the period to be predicted, including: Inputting the historical moving average kernel component into the first branch network of the prediction module to predict the moving average kernel component in the period to be predicted; Inputting the historical seasonal component into the second branch network of the prediction module to predict the seasonal component under the period to be predicted; Among them, the first branch network is constructed based on the unscented Kalman filter model; the second branch network is constructed based on the trained extended long short-term memory network model.

[0012] According to a photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion provided by the present invention, the historical photovoltaic parameter sequence is constructed based on historical irradiation information, historical environmental information and historical operating parameters of the photovoltaic system to be predicted in the first historical period.

[0013] The present invention also provides a photovoltaic power generation prediction device based on dynamic decomposition and digital-analog fusion, comprising: A feature selection unit is used to perform feature selection on a historical photovoltaic parameter sequence of a photovoltaic system to be predicted in a first historical period to obtain a target photovoltaic parameter sequence; A decomposition unit, used for dynamically decomposing the target photovoltaic parameter sequence to obtain a historical moving average kernel component in the first historical period and a historical seasonal component in the first historical period; A first prediction unit, used for inputting the historical moving average kernel component and the historical seasonal component into a prediction module of a prediction model, and predicting the moving average kernel component under a period to be predicted and the seasonal component under the period to be predicted; A second prediction unit is used to input the moving average kernel component of the period to be predicted and the seasonal component of the period to be predicted into the fusion module of the prediction model to predict the power generation forecast value of the photovoltaic system to be predicted in the period to be predicted; Among them, the prediction model is obtained by jointly training the sample photovoltaic parameter sequence of the sample photovoltaic system in the second historical period, and the moving average kernel component label, seasonal component label and power generation label in the third historical period on the basis of the digital-analog fusion model; the digital-analog fusion model is constructed based on the fusion of the unscented Kalman filter model and the extended long short-term memory network model.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion as described in any one of the above is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the photovoltaic power generation prediction methods based on dynamic decomposition and digital-analog fusion as described above.

[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the photovoltaic power generation prediction methods based on dynamic decomposition and digital-analog fusion as described above.

[0017] The photovoltaic power generation prediction method and device based on dynamic decomposition and digital-analog fusion provided by the present invention decomposes photovoltaic power generation data into more stable and higher-quality photovoltaic components through feature selection and dynamic decomposition, and uses a prediction model trained by a digital-analog fusion model to predict the dynamically decomposed photovoltaic components. Finally, the photovoltaic component prediction results are fused to obtain the final photovoltaic power generation prediction result. Photovoltaic power generation prediction is performed by improving data quality and multi-model joint prediction to better capture the trend characteristics of photovoltaic power generation, effectively reduce the impact of the randomness of natural factors, and significantly improve the accuracy and robustness of photovoltaic power generation prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is one of the flow charts of the photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion provided by the present invention.

[0020] Figure 2 This is the second flow chart of the power generation prediction method based on time series prediction and hybrid neural network provided by the present invention.

[0021] Figure 3 It is a heat map of parameter similarity distribution provided by the present invention.

[0022] Figure 4 It is a structural schematic diagram of the unscented Kalman filter model provided by the present invention.

[0023] Figure 5 It is a schematic diagram of the structure of the extended long short-term memory network provided by the present invention.

[0024] Figure 6 It is a simulation comparison diagram of the photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion provided by the present invention.

[0025] Figure 7 It is a structural schematic diagram of a photovoltaic power generation prediction device based on dynamic decomposition and digital-analog fusion provided by the present invention.

[0026] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] As a clean and renewable energy source, photovoltaics are essential to meet the growing demand for clean energy. In the past few decades, the randomness and uncertainty of photovoltaic energy have posed challenges to achieving substantial environmental and economic benefits. The main challenges include intermittent power generation, high installation costs, and sensitivity to weather conditions. Therefore, effective photovoltaic power generation forecasting is essential for large-scale photovoltaic energy access to the main grid. Accurate short-term photovoltaic power generation forecasting depends largely on weather conditions and is crucial for efficient scheduling and the reliability of energy systems. Photovoltaic power generation forecasting brings economic benefits to energy companies and is essential for the effective implementation of tasks such as pricing strategies, safety assessments, reserve management, and maintenance plans. Therefore, how to accurately achieve photovoltaic power generation forecasting is an important issue that the industry is currently working to solve.

[0029] In the prediction of photovoltaic power generation, two types of methods are traditionally used: physical modeling methods and data-driven algorithms. Due to the rapid development of data and artificial intelligence, photovoltaic power generation prediction is increasingly dependent on data-driven methods. This shift limits the use of physical methods because physical modeling methods usually require complex calculations. Data-driven methods, such as machine learning, use large amounts of historical data to train prediction models. The performance of these methods depends largely on the quality and time span of the input data. The photovoltaic power generation process itself is affected by a variety of natural factors, resulting in a certain degree of randomness in the relevant parameters of photovoltaic power generation, making it difficult to guarantee the quality of the input data, and thus the accuracy of photovoltaic power generation predicted by machine learning methods is also difficult to guarantee.

[0030] In addition, due to the overfitting and low convergence of a single machine learning model, hybrid models are widely used.

[0031] In this regard, in order to solve the problem of difficulty in ensuring the accuracy of photovoltaic power generation prediction caused by the limitations of the above-mentioned machine learning model, an embodiment of the present application provides a photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion. The method decomposes photovoltaic power generation data into more stable and higher-quality photovoltaic components through feature selection and dynamic decomposition, and uses a prediction model trained by a digital-analog fusion model to predict the dynamically decomposed photovoltaic components. Finally, the photovoltaic component prediction results are fused to obtain the final photovoltaic power generation prediction results, thereby realizing photovoltaic power generation prediction by improving data quality and multi-model joint prediction, so as to better capture the trend characteristics of photovoltaic power generation, effectively reduce the impact of the randomness of natural factors, and significantly improve the accuracy and robustness of photovoltaic power generation prediction.

[0032] It should be noted that the execution subject of the method may be an electronic device, a component in an electronic device, an integrated circuit, or a chip. The electronic device may be a mobile electronic device or a non-mobile electronic device. For example, the mobile electronic device may be a mobile phone, a tablet computer, a laptop computer, a PDA, an in-vehicle electronic device, a wearable device, etc., and the non-mobile electronic device may be a server, a network attached storage, a personal computer, a teller machine, or a self-service machine, etc., which is not specifically limited in the present invention.

[0033] Figure 1 This is one of the flow charts of the photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion provided by the present invention; Figure 1 As shown, the method includes step 110 , step 120 , step 130 and step 140 .

[0034] Step 110 , performing feature selection on the historical photovoltaic parameter sequence of the photovoltaic system to be predicted in the first historical period to obtain a target photovoltaic parameter sequence.

[0035] Here, the photovoltaic system to be predicted is the photovoltaic system for which power generation prediction is required; the first historical period is one or more historical periods before the period to be predicted; the period to be predicted is the period for which power generation prediction is required, such as the current period or one or more future periods after the current period.

[0036] Figure 2 The second flow chart of the power generation prediction method based on time series prediction and hybrid neural network provided by the present invention; Figure 2 As shown, when it is necessary to predict the power generation of the photovoltaic system to be predicted in the period to be predicted, the photovoltaic power generation influencing factors of the photovoltaic system to be predicted at each sampling point in the first historical period can be collected to construct a historical photovoltaic parameter sequence. The photovoltaic power generation influencing factors include but are not limited to meteorological data and power generation related data, such as operating parameters.

[0037] Exemplarily, in some embodiments, the historical photovoltaic parameter sequence is constructed based on the historical irradiation information, historical environmental information, and historical operating parameters of the photovoltaic system to be predicted in the first historical period.

[0038] Among them, historical irradiation information includes but is not limited to front irradiance, back irradiance, front cumulative irradiance, back cumulative irradiance, horizontal total irradiance, horizontal scattered irradiance, and normal irradiance; historical environmental information includes but is not limited to multiple combinations of ambient temperature, reflectivity, wind speed, wind direction, humidity, precipitation, air pressure, cloud cover and dust; historical operating parameters include but are not limited to the DC power of the photovoltaic system to be predicted.

[0039] Among them, the front irradiance and the back irradiance: the radiation intensity of the sun directly hitting the photovoltaic panel, which directly affects the power generation efficiency of the photovoltaic panel. Ambient temperature: the increase in temperature will cause the efficiency of the photovoltaic panel to decrease, so the impact of ambient temperature on power generation needs to be considered. The front cumulative irradiance and the back cumulative irradiance: the cumulative irradiance is also an important factor affecting the power generation of the photovoltaic panel. Reflectivity: the light reflected from the ground or other objects onto the photovoltaic panel will also affect the power generation efficiency of the photovoltaic panel. Wind speed and wind direction: changes in wind speed and wind direction will affect the heat dissipation effect and cleanliness of the photovoltaic panel, and then affect the power generation efficiency. Humidity, horizontal total irradiance, horizontal scattered irradiance, normal irradiance, precipitation, air pressure, cloud cover and dust: these factors will affect the lighting conditions and environmental conditions, and indirectly affect the power generation efficiency of the photovoltaic panel. Therefore, taking the above factors into consideration, it is possible to more accurately predict the photovoltaic power generation, optimize the operation and management of the power generation system, improve the power generation efficiency, save energy resources, and reduce costs, which has a positive significance for the sustainable development of the photovoltaic power generation system.

[0040] The historical photovoltaic parameter sequence here can be user input entered by the user in the front-end interface, or can be information generated by triggering the monitor under certain conditions, such as timed triggering. The so-called user input can be information input through a command line interface, a graphical interface, a touch input, a drop-down selection input, a voice input, a gesture input, a visual input, a brain-computer input, etc., and this embodiment does not specifically limit this.

[0041] Optionally, since the original historical photovoltaic parameter sequence has certain randomness, volatility, and redundancy, directly using these parameters for photovoltaic power generation prediction may affect the prediction results. Therefore, in order to improve data quality and prediction performance, the historical photovoltaic parameter sequence can be analyzed and processed, and feature selection can be performed to select parameters of one or more dimensions that are relatively important for power generation prediction as the target photovoltaic parameter sequence, thereby improving data quality.

[0042] Here, feature selection may be to calculate the similarity between parameters of each dimension in the historical photovoltaic parameter sequence and / or the correlation between parameters of each dimension and the power generation label, and to score the importance of parameters of each dimension in the historical photovoltaic parameter sequence, so as to select one or more parameters of dimensions with relatively high importance for power generation prediction from the multi-dimensional parameters of the historical photovoltaic parameter sequence according to the importance score as the target photovoltaic parameter sequence; or, to input the parameters of each dimension in the historical photovoltaic parameter sequence into the machine learning model, so as to further evaluate the importance of each feature by the feature importance evaluation function of the machine learning model (such as feature importance scoring in algorithms such as random forest and gradient boosting tree), so as to select one or more parameters of dimensions with relatively high importance for power generation prediction from the multi-dimensional parameters of the historical photovoltaic parameter sequence according to the importance score as the target photovoltaic parameter sequence, etc., and this embodiment does not make any specific limitation on this.

[0043] Step 120 : dynamically decompose the target photovoltaic parameter sequence to obtain a historical moving average kernel component in the first historical period and a historical seasonal component in the first historical period.

[0044] Optionally, in the photovoltaic power generation forecast, the photovoltaic power generation system is affected by many factors, such as solar radiation intensity, temperature, cloud cover, etc., which causes the power generation data to show non-stationary and nonlinear characteristics. Therefore, in order to cope with the long-term trend and seasonal fluctuations of photovoltaic data and improve the robustness of the model to photovoltaic data, the idea of ​​decomposition is adopted to learn the complex time pattern in the long-term prediction scenario. The historical photovoltaic parameter sequence can be dynamically decomposed to decompose the historical photovoltaic parameter sequence into trend cycle components and seasonal components, thereby obtaining photovoltaic components and seasonal photovoltaic components that reflect the long-term nature of the sequence data, respectively, and thus more accurately predict photovoltaic power generation.

[0045] The dynamic decomposition here may be implemented based on a weighted moving average decomposition algorithm improved based on exponential decay weights, etc., which is not specifically limited in this embodiment.

[0046] Step 130, input the historical moving average kernel component and the historical seasonal component into a prediction module of a prediction model, and predict the moving average kernel component in the period to be predicted and the seasonal component in the period to be predicted.

[0047] Step 140, input the moving average kernel component under the period to be predicted and the seasonal component under the period to be predicted into the fusion module of the prediction model, and predict the power generation forecast value of the photovoltaic system to be predicted in the period to be predicted; wherein, the prediction model is obtained by joint training based on the sample photovoltaic parameter sequence of the sample photovoltaic system in the second historical period, and the moving average kernel component label, seasonal component label and power generation label in the third historical period on the basis of the digital-analog fusion model; the digital-analog fusion model is constructed based on the fusion of the unscented Kalman filter model and the extended long short-term memory network model.

[0048] Optionally, after the historical moving average kernel component and the historical seasonal component are decomposed, they can be input into the prediction module of the prediction model, and the prediction module predicts the moving average kernel component of the period to be predicted and the seasonal component of the period to be predicted. The prediction module here can be two branch networks constructed by respectively including an unscented Kalman filter model and a trained extended long short-term memory network model, so that one of the branch networks predicts the moving average kernel component of the period to be predicted, and the other branch network predicts the seasonal component of the period to be predicted.

[0049] Optionally, after predicting the moving average kernel component under the period to be predicted and the seasonal component under the period to be predicted, the predicted moving average kernel component and seasonal component can be input into the fusion module of the prediction model, so that the fusion module can sum the predicted moving average kernel component and seasonal component to predict the predicted value of power generation of the photovoltaic system to be predicted in the period to be predicted.

[0050] The prediction model here can be obtained by training the digital-analog fusion model through the sample photovoltaic parameter sequence of the sample photovoltaic system in the second historical period, and the moving average kernel component label, seasonal component label and power generation label in the third historical period. The specific training steps are as follows: First, in order to improve the performance of power generation prediction, it is necessary to collect sample photovoltaic parameter sequences of sample photovoltaic systems of different models in different scenarios and environments in each second historical period.

[0051] Subsequently, the sample photovoltaic parameter sequences of each second historical period are preprocessed, such as outlier detection and outlier filling processing, and feature selection processing. The feature selection processing here can refer to step 110, which will not be described in detail here.

[0052] Subsequently, the preprocessed sample photovoltaic parameter sequence is dynamically decomposed to obtain the sample moving average kernel component under the second historical period and the sample seasonal component under the second historical period. The specific steps can be found in step 120 and will not be described in detail here.

[0053] In addition, the moving average kernel component label and seasonal component label of the sample photovoltaic parameter sequence of the second historical period are marked in the third historical period, and the power generation label of the third historical period is marked; based on the sample moving average kernel component of the second historical period and the sample seasonal component of the second historical period, and the corresponding moving average kernel component label, seasonal component label and corresponding power generation label, a sample data set is constructed so that the constructed sample data set has sufficient depth and breadth, so that the trained prediction model can accurately predict trend characteristics and power generation in different scenarios. The so-called second historical period is one or more historical periods before the third historical period.

[0054] For the sample data set, it can be obtained through historical collection, such as making the sample photovoltaic parameter sequences of sample photovoltaic systems of different models under different scenarios and environments in each second historical period into a sample data set; or it can be obtained by loading and acquiring the sample data set in an open source database, which is not specifically limited in this embodiment.

[0055] Secondly, the sample moving average kernel component of the second historical period and the sample seasonal component of the second historical period in the sample data set are first input into the digital-analog fusion model, and after the digital-analog fusion model predicts and outputs the sample moving average kernel component of the third historical period and the sample seasonal component of the third historical period, the sample moving average kernel component of the third historical period and the sample seasonal component of the third historical period are fused to obtain the power generation forecast value of the third historical period; then, the deviation values ​​between the sample moving average kernel component of the third historical period and the sample seasonal component of the third historical period and the moving average kernel component label of the third historical period and the seasonal component label of the third historical period, as well as the deviation value between the power generation forecast value of the third historical period and the power generation label of the third historical period are fused, and the digital-analog fusion model is iteratively trained to obtain a prediction model based on the trained digital-analog fusion model that can accurately perform trend feature prediction and power generation prediction.

[0056] The method provided in this embodiment decomposes photovoltaic power generation data into more stable and higher-quality photovoltaic components through feature selection and dynamic decomposition, and uses a prediction model trained by a digital-analog fusion model to predict the dynamically decomposed photovoltaic components. Finally, the photovoltaic component prediction results are fused to obtain the final photovoltaic power generation prediction results. Photovoltaic power generation is predicted by improving data quality and multi-model joint prediction to better capture the trend characteristics of photovoltaic power generation, effectively reduce the impact of the randomness of natural factors, and significantly improve the accuracy and robustness of photovoltaic power generation prediction.

[0057] In some embodiments, the dynamic decomposition step in step 120 can be implemented based on the following steps: Based on the exponential decay model, calculating the weight coefficient of each parameter in the target photovoltaic parameter sequence; According to the weight coefficient, performing weighted moving average decomposition on the target photovoltaic parameter sequence to obtain a historical moving average kernel component in the first historical period; According to the target photovoltaic parameter sequence and the historical moving average kernel component in the first historical period, the historical seasonal component in the first historical period is decomposed to obtain.

[0058] Optionally, in order to improve the accuracy and interpretability of time series analysis and enable the model to better capture the characteristics and trends of time series data, the moving average decomposition algorithm can be improved by exponential decay weights to obtain an improved moving average decomposition algorithm, thereby smoothing periodic fluctuations and highlighting long-term trends by adjusting the moving average. The dynamic decomposition steps are as follows: First, in order to better perform dynamic decomposition, the weight coefficient of each parameter in the target photovoltaic parameter sequence can be calculated based on the exponential decay model in the improved moving average decomposition algorithm. The exponential decay model here takes into account the temporal correlation, so that the data points closer to the current time point have a greater weight, while the data points farther from the current time point have a smaller weight. This can better reflect the characteristics of time series data and is more in line with the dynamics of time series data, so as to assist in the dynamic decomposition of time series, so that the model can better capture the characteristics and trends of time series data; the exponential decay model can be constructed based on the temporal position of each parameter in the target photovoltaic parameter sequence and the decay coefficient.

[0059] Exemplarily, the calculation formula of the exponential decay model is as follows: ; in, The timing position in the target photovoltaic parameter sequence is The weight coefficient of the parameter; is the attenuation coefficient; is the number of parameters in the target photovoltaic parameter sequence; and The position number of the parameter.

[0060] Subsequently, based on the weight coefficient, the idea of ​​the moving average decomposition algorithm in the improved moving average decomposition algorithm is used to extract the historical moving average kernel component (or trend component) of the target photovoltaic parameter series: ; in, is the historical moving average kernel component, is the target photovoltaic parameter sequence; Secondly, the residual between the target photovoltaic parameter series and the historical moving average kernel component in the first historical period is taken as the historical seasonal component in the first historical period. The specific calculation formula is as follows: ; in, is the historical seasonal component.

[0061] In summary, the improved moving average decomposition method uses the moving average time series analysis method to deal with the long-term trend changes and seasonal fluctuations of photovoltaic data, and uses exponential decay to perform dynamic and more accurate decomposition, which can improve the accuracy and interpretability of time series analysis, thereby improving the quality of data input to the model to effectively improve the prediction accuracy.

[0062] In some embodiments, the feature selection step in step 110 is implemented based on the following steps: Performing outlier detection and outlier filling processing on the historical photovoltaic parameter sequence to obtain a preprocessed historical photovoltaic parameter sequence; Performing importance scoring on each dimension parameter in the sample photovoltaic parameter sequence to obtain an importance score value corresponding to each dimension parameter in the sample photovoltaic parameter sequence; Determining a target parameter from among the multi-dimensional parameters in the sample photovoltaic parameter sequence according to the importance score value; In the preprocessed historical photovoltaic parameter sequence, parameters matching the target parameters are selected to construct the target photovoltaic parameter sequence.

[0063] Optionally, since there may be a lot of abnormal data in the historical photovoltaic parameter sequence, which interferes with the prediction results, it is first necessary to detect and remove abnormal values ​​from the historical photovoltaic parameter sequence, which can remove noise and unreasonable data points and improve the accuracy of subsequent predictions.

[0064] After that, in order to find out which factors affect the power generation, so as to better build the model and improve the prediction accuracy of the model, the importance of each dimensional parameter in the sample photovoltaic parameter sequence can be scored to calculate the importance score value corresponding to each dimensional parameter in the sample photovoltaic parameter sequence, and one or more dimensional parameters with larger importance score values ​​are determined as target parameters in the multi-dimensional parameters in the sample photovoltaic parameter sequence; then, in the preprocessed historical photovoltaic parameter sequence, the parameters that match the target parameters are selected to construct the target photovoltaic parameter sequence based on the matching parameters.

[0065] Here, the importance score value can be calculated based on a machine learning model, or calculated based on the similarity between the parameters of each dimension in the sample photovoltaic parameter sequence and / or the parameters of each dimension in the sample photovoltaic parameter sequence and the corresponding power generation label, etc. This embodiment does not specifically limit this.

[0066] Exemplarily, in some embodiments, the importance score value may be further implemented based on the following steps: Calculating the correlation between every two dimensional parameters in the sample photovoltaic parameter sequence to obtain a first correlation value corresponding to each dimensional parameter in the sample photovoltaic parameter sequence; Calculating the correlation between each dimensional parameter in the sample photovoltaic parameter sequence and the power generation label to obtain a second correlation value corresponding to each dimensional parameter in the sample photovoltaic parameter sequence; The first correlation value and the second correlation value are combined to obtain the importance score value corresponding to each dimensional parameter in the sample photovoltaic parameter sequence.

[0067] Optionally, the correlation between every two dimensional parameters in the sample photovoltaic parameter sequence is calculated to obtain the intrinsic connection and mutual influence degree between the parameters of different dimensions in the sample photovoltaic parameter sequence, such as calculating the correlation between light intensity and temperature, the correlation between light intensity and wind speed, etc., until all possible combinations of dimension pairs are covered, thereby obtaining the first correlation value corresponding to each dimensional parameter in the sample photovoltaic parameter sequence.

[0068] In addition, the power generation label of each dimensional parameter in the sample photovoltaic parameter sequence can be subjected to correlation analysis to obtain the contribution and importance of each photovoltaic parameter to the power generation prediction, thereby obtaining the second correlation value corresponding to each dimensional parameter in the sample photovoltaic parameter sequence.

[0069] Next, in order to comprehensively evaluate the importance of the photovoltaic parameters of each dimension in the overall prediction model, the first correlation value (correlation between dimensions) and the second correlation value (correlation between dimensions and power generation) can be weighted added or model fused to generate a corresponding importance score value for the photovoltaic parameters of each dimension, which combines its position in the relationship between parameters and its direct contribution to power generation prediction. The weighted addition here can be adaptively determined based on the importance of the first correlation value and the second correlation value to the photovoltaic power generation prediction.

[0070] The similarity calculation here can be achieved by introducing correlation calculation methods such as the Pearson correlation coefficient to obtain a heat map, and the correlation between parameters and parameters and between parameters and labels can be obtained through the heat map. Based on the calculated correlation, key parameters with higher importance scores are selected to predict photovoltaic power generation. Figure 3is a heat map of parameter similarity distribution provided by the present invention; illustratively, Figure 3 As shown in the figure, it can be observed from the heat map that the importance score corresponding to the irradiation intensity is the highest. Therefore, the parameter sequence associated with the irradiation intensity can be selected from the historical photovoltaic parameter sequence as the target photovoltaic parameter sequence to participate in the subsequent photovoltaic power generation prediction.

[0071] Among them, the Pearson correlation coefficient is a statistical method used to measure the strength and direction of the linear relationship between two continuous variables. It is one of the most commonly used correlation measurement methods and is usually used to understand the degree of correlation between two variables. The value range of the Pearson correlation coefficient is -1 to 1. A positive value indicates a positive correlation, which means that the two variables change in the same direction, while a negative value indicates the opposite. The formula is as follows: ; in, For the A dimension parameter in the sample photovoltaic parameter sequence under a cycle; For the The power generation label under a cycle or another dimension parameter in the sample photovoltaic parameter sequence; It is the average value of a certain dimension parameter in the sample photovoltaic parameter sequence under multiple cycles; It is the average value of another dimension parameter or the average value of power generation label in the sample photovoltaic parameter sequence under multiple cycles.

[0072] In summary, the method provided in the embodiment of the present application can accurately select key target parameters by detecting and processing outliers in historical photovoltaic parameter sequences, and comprehensively considering the internal correlation of parameters in each dimension of the sample photovoltaic parameter sequence and the importance score of their impact on power generation prediction, thereby constructing a more accurate and effective target photovoltaic parameter sequence to improve the accuracy and robustness of photovoltaic power generation prediction.

[0073] In some embodiments, the step of inputting the historical moving average kernel component and the historical seasonal component into a prediction module of a prediction model to predict the moving average kernel component under a period to be predicted and the seasonal component under the period to be predicted includes: Inputting the historical moving average kernel component into the first branch network of the prediction module to predict the moving average kernel component in the period to be predicted; Inputting the historical seasonal component into the second branch network of the prediction module to predict the seasonal component under the period to be predicted; Among them, the first branch network is constructed based on the unscented Kalman filter model; the second branch network is constructed based on the trained extended long short-term memory network model.

[0074] Optionally, in order to overcome the limitations of a single model, compared with traditional prediction methods, the embodiment of the present application trains and constructs a prediction model based on the digital-analog fusion model formed by the digital-analog fusion of the unscented Kalman filter model (Unscented Kalman Filter, UKF) and the extended long short-term memory network (Extended Long Short-Term Memory, xLSTM), and performs moving average kernel component prediction through the first branch network constructed based on the unscented Kalman filter model in the prediction model thus constructed, and performs seasonal component prediction through the second branch network constructed by the extended long short-term memory network trained in the prediction model thus constructed. Finally, the predicted moving average kernel component and seasonal component are fused to obtain the final photovoltaic power generation prediction result, thereby realizing the use of the comprehensive advantages of multiple models to improve the accuracy and reliability of modeling and prediction.

[0075] It should be noted that the Kalman filter obtains the optimal state estimate under the assumption of linear Gaussian distribution. In practical applications, the situation is often nonlinear. In order to solve the state estimation problem under nonlinear conditions, it is necessary to explore filtering methods suitable for nonlinear scenarios. Both the Extended Kalman Filter (EKF) and the Unscented Kalman Filter can handle nonlinear problems, but they use different principles. The EKF uses Taylor series to expand the state equation and the observation equation, retaining only the first-order terms to linearize the nonlinear filtering problem. However, this method can introduce significant errors, resulting in reduced or ineffective filtering performance. Generally, the Jacobian matrix required to calculate the EKF may be complex, thereby increasing the computational complexity. The UKF algorithm replaces the traditional linearization method with the probability distribution of the nonlinear function. Within the framework of the Kalman linear filter, the UKF uses the unscented transformation (UT) strategy to achieve nonlinear transformation of the mean and covariance without calculating the Jacobian matrix or considering high-order terms. This approach ensures effective nonlinear filtering performance without the disadvantages associated with linearization. Therefore, for the prediction of trend component (or moving average kernel component), the present application can use the unscented Kalman filter model for prediction implementation.

[0076] Figure 4 Schematic diagram of the structure of the unscented Kalman filter model provided by the present invention; Figure 4 As shown, the steps of traceless transformation include: First, 2n+1 Sigma points, also known as sampling points, are calculated using the following formula, where n represents the dimension of the state: ; in, , represents the covariance matrix List; represents the state vector at time k, Indicates The average weight of the sigama points, Indicates Covariance weights for the sigma points. represents the predicted value, Represents model parameters that need to be initialized with actual data.

[0077] Secondly, the weight corresponding to the sampling point is calculated by the following formula: ; In the formula, k represents the time, and the parameter is a scaling factor that reduces the overall prediction error, Determines the distribution of sampling points. It is a parameter to be selected, and should usually be guaranteed is a semi-positive definite matrix. The parameter can incorporate the moments of higher-order terms and is usually not less than 0.

[0078] In addition, since it is smooth enough, a double exponential model can be applied to build the model. The specific calculation formula is as follows: .

[0079] Figure 5 is a schematic diagram of the structure of the extended long short-term memory network provided by the present invention; Figure 5 As shown in the figure, for seasonal component prediction, the trained extended long short-term memory network can be used for prediction implementation. This network is designed to overcome the limitations of the long short-term memory network (Long Short-Term Memory, LSTM), so two major modifications are introduced: first, a new LSTM variant with scalar memory, scalar update and memory hybrid, namely the Scalar Long Short-Term Memory Network (ScalarLong Short-Term Memory, sLSTM); second, a new LSTM variant with matrix memory and covariance (outer product) update rule, namely the Matrix Long Short-Term Memory Network (Matrix Long Short-Term Memory, mlSTM). By integrating these new LSTM variants into the residual block module, an xLSTM block can be created.

[0080] To further enable LSTM to modify storage decisions, exponential gates are introduced along with normalization and stabilization. Specifically, the input gate and forget gate can have exponential activation functions. For normalization, a normalization state is introduced that adds the product of the input gate to the product of all future forget gates. The forward propagation calculation formula of sLSTM is as follows: ; ; ; ; ; ; ; in, For unit input, is the unit state, represents the input gate, Represented as the forget gate, is the hidden state, represents the unit activation function, log represents the logarithmic function, and the weight vector , , and Corresponding to the input The input weight vector between the unit input, input gate, forget gate, and output gate. Weight , , and Corresponding to the hidden state and the recursive weights between the unit input, input gate, forget gate, and output gate. , , and are the corresponding bias terms for the unit input, input gate, forget gate, and output gate. is the cell input. Use additional state To stabilize the door, the specific calculation formula is as follows: ; ; ; In order to increase the storage capacity of LSTM, matrix memory is used instead of scalar memory units to increase storage capacity and improve information retrieval efficiency. Drawing inspiration from bidirectional associative memory (BAM), the covariance update rule is used to efficiently store and retrieve key-value pairs. The covariance update for storing key-value pairs is shown below: ; The forward propagation of mLSTM is as follows: ; ; ; ; ; ; ; To stabilize the exponential gate of mLSTM, mLSTM can use the same stabilization technique as sLSTM.

[0081] In summary, the method provided in the embodiment of the present application effectively overcomes the defects of the traditional single prediction method by using the method of digital-analog fusion, compensates for the impact of the non-stationarity and nonlinearity of photovoltaic data, and further improves the accuracy and reliability of modeling and prediction.

[0082] The following uses photovoltaic data over a period of time as an example to experimentally verify the method provided in the application example, and compares it with other models and methods to verify the effectiveness of the method provided in the application example.

[0083] according to Figure 2 The steps shown use photovoltaic data and meteorological data over a period of time and design a certain amount of experiments to verify the accuracy of the entire algorithm. Here, only an experimental result chart after parameter adjustment is given to verify the effectiveness of the algorithm.

[0084] Figure 6 is a simulation comparison diagram of the photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion provided by the present invention; Figure 6As shown, the method provided in the embodiment of the present application (or Proposed Method) is to obtain the prediction results of each component by inputting the historical moving average kernel component and the historical seasonal component obtained by decomposition into the first branch network constructed by the unscented Kalman filter model, and the second branch network constructed based on the trained extended long short-term memory network. After the prediction results of each component are obtained, the prediction results of the components are summed to obtain the final photovoltaic power generation prediction result. After multiple verifications, the accuracy of the power generation prediction result finally obtained can reach 97% or more; and the method provided in the embodiment of the present application is compared with several existing methods, such as LSTM, LSTM and UKF fusion model (or LSTM+UKF), exponential moving average (EMA), LSTM and UKF fusion model (or EMA+LSTM+UKF), and Informer time series prediction model, etc. It can be seen that the power generation prediction result predicted by the method provided in the embodiment of the present application is closer to the actual power generation (or TURE), thereby effectively verifying the effectiveness of the method provided in the embodiment of the present application.

[0085] The photovoltaic power generation prediction device based on dynamic decomposition and digital-analog fusion provided by the present invention is described below. The photovoltaic power generation prediction device based on dynamic decomposition and digital-analog fusion described below and the photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion described above can be referenced to each other.

[0086] Figure 7 is a schematic diagram of the structure of a photovoltaic power generation prediction device based on dynamic decomposition and digital-analog fusion provided by the present invention; Figure 7 As shown, the device comprises: The feature selection unit 710 is used to perform feature selection on the historical photovoltaic parameter sequence of the photovoltaic system to be predicted in the first historical period to obtain a target photovoltaic parameter sequence; The decomposition unit 720 is used to dynamically decompose the target photovoltaic parameter sequence to obtain a historical moving average kernel component in the first historical period and a historical seasonal component in the first historical period; The first prediction unit 730 is used to input the historical moving average kernel component and the historical seasonal component into the prediction module of the prediction model, and predict the moving average kernel component under the period to be predicted and the seasonal component under the period to be predicted; The second prediction unit 740 is used to input the moving average kernel component under the period to be predicted and the seasonal component under the period to be predicted into the fusion module of the prediction model, and predict the power generation prediction value of the photovoltaic system to be predicted in the period to be predicted; Among them, the prediction model is obtained by jointly training the sample photovoltaic parameter sequence of the sample photovoltaic system in the second historical period, and the moving average kernel component label, seasonal component label and power generation label in the third historical period on the basis of the digital-analog fusion model; the digital-analog fusion model is constructed based on the fusion of the unscented Kalman filter model and the extended long short-term memory network model.

[0087] The device provided in this embodiment decomposes photovoltaic power generation data into more stable and higher-quality photovoltaic components through feature selection and dynamic decomposition, and uses a prediction model trained by a digital-analog fusion model to predict the dynamically decomposed photovoltaic components. Finally, the photovoltaic component prediction results are fused to obtain the final photovoltaic power generation prediction results. Photovoltaic power generation is predicted by improving data quality and multi-model joint prediction to better capture the trend characteristics of photovoltaic power generation, effectively reduce the impact of the randomness of natural factors, and significantly improve the accuracy and robustness of photovoltaic power generation prediction.

[0088] The device provided by the present invention is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for the specific processes and detailed contents, which will not be repeated here.

[0089] Figure 8 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 8As shown, the electronic device may include: a processor (processor) 810, a communication interface (Communications Interface) 820, a memory (memory) 830 and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call the logic instructions in the memory 830 to execute a photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion, the method comprising: performing feature selection on the historical photovoltaic parameter sequence of the photovoltaic system to be predicted in the first historical period to obtain a target photovoltaic parameter sequence; dynamically decomposing the target photovoltaic parameter sequence to obtain a historical moving average kernel component under the first historical period and a historical seasonal component under the first historical period; inputting the historical moving average kernel component and the historical seasonal component into the prediction module of the prediction model to predict the moving average kernel component under the period to be predicted and the seasonal component under the period to be predicted; inputting the moving average kernel component under the period to be predicted and the seasonal component under the period to be predicted into the fusion module of the prediction model to predict the power generation prediction value of the photovoltaic system to be predicted in the period to be predicted; wherein, the prediction model is obtained by joint training based on the digital-analog fusion model, based on the sample photovoltaic parameter sequence of the sample photovoltaic system in the second historical period, and the moving average kernel component label, seasonal component label and power generation label in the third historical period; the digital-analog fusion model is constructed based on the fusion of the unscented Kalman filter model and the extended long short-term memory network model.

[0090] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0091] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion provided by the above methods, the method comprising: performing feature selection on a historical photovoltaic parameter sequence of a photovoltaic system to be predicted in the first historical period to obtain a target photovoltaic parameter sequence; dynamically decomposing the target photovoltaic parameter sequence to obtain a historical moving average kernel component under the first historical period and a historical seasonal component under the first historical period; inputting the historical moving average kernel component and the historical seasonal component into the prediction The prediction module of the model predicts the moving average kernel component under the period to be predicted and the seasonal component under the period to be predicted; the moving average kernel component under the period to be predicted and the seasonal component under the period to be predicted are input into the fusion module of the prediction model to predict the power generation forecast value of the photovoltaic system to be predicted in the period to be predicted; wherein, the prediction model is obtained by joint training based on the sample photovoltaic parameter sequence of the sample photovoltaic system in the second historical period, and the moving average kernel component label, seasonal component label and power generation label in the third historical period on the basis of the digital-analog fusion model; the digital-analog fusion model is constructed based on the fusion of the unscented Kalman filter model and the extended long short-term memory network model.

[0092] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion provided by the above-mentioned methods, the method comprising: performing feature selection on a historical photovoltaic parameter sequence of a photovoltaic system to be predicted in a first historical period to obtain a target photovoltaic parameter sequence; dynamically decomposing the target photovoltaic parameter sequence to obtain a historical moving average kernel component under the first historical period and a historical seasonal component under the first historical period; inputting the historical moving average kernel component and the historical seasonal component into a prediction module of a prediction model to predict the photovoltaic system to be predicted. The moving average kernel component of the period to be predicted and the seasonal component of the period to be predicted are input into the fusion module of the prediction model to predict the power generation forecast value of the photovoltaic system to be predicted in the period to be predicted; wherein, the prediction model is obtained by jointly training the sample photovoltaic parameter sequence of the sample photovoltaic system in the second historical period and the moving average kernel component label, seasonal component label and power generation label in the third historical period on the basis of the digital-analog fusion model; the digital-analog fusion model is constructed based on the fusion of the unscented Kalman filter model and the extended long short-term memory network model.

[0093] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0094] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion, characterized in that: include: Perform feature selection on the historical photovoltaic parameter sequence of the photovoltaic system to be predicted in the first historical period to obtain the target photovoltaic parameter sequence; Dynamically decomposing the target photovoltaic parameter sequence to obtain a historical moving average kernel component in the first historical period and a historical seasonal component in the first historical period; Inputting the historical moving average kernel component and the historical seasonal component into a prediction module of a prediction model, and predicting the moving average kernel component under a period to be predicted and the seasonal component under the period to be predicted; Inputting the moving average kernel component under the period to be predicted and the seasonal component under the period to be predicted into the fusion module of the prediction model, and predicting the power generation forecast value of the photovoltaic system to be predicted in the period to be predicted; Among them, the prediction model is obtained by jointly training the sample photovoltaic parameter sequence of the sample photovoltaic system in the second historical period, and the moving average kernel component label, seasonal component label and power generation label in the third historical period on the basis of the digital-analog fusion model; the digital-analog fusion model is constructed based on the fusion of the unscented Kalman filter model and the extended long short-term memory network model.

2. The photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion according to claim 1 is characterized in that: The dynamically decomposing the target photovoltaic parameter sequence to obtain the historical moving average kernel component in the first historical period and the historical seasonal component in the first historical period includes: Based on the exponential decay model, calculating the weight coefficient of each parameter in the target photovoltaic parameter sequence; According to the weight coefficient, performing weighted moving average decomposition on the target photovoltaic parameter sequence to obtain a historical moving average kernel component in the first historical period; According to the target photovoltaic parameter sequence and the historical moving average kernel component in the first historical period, the historical seasonal component in the first historical period is decomposed to obtain.

3. The photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion according to claim 2 is characterized in that: The calculation formula of the exponential decay model is as follows: ； in, The timing position in the target photovoltaic parameter sequence is The weight coefficient of the parameter; is the attenuation coefficient; is the number of parameters in the target photovoltaic parameter sequence; and The position number of the parameter.

4. The photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion according to any one of claims 1 to 3, characterized in that: The feature selection is performed on the historical photovoltaic parameter sequence of the photovoltaic system to be predicted in the first historical period to obtain the target photovoltaic parameter sequence, including: Performing outlier detection and outlier filling processing on the historical photovoltaic parameter sequence to obtain a preprocessed historical photovoltaic parameter sequence; Performing importance scoring on each dimension parameter in the sample photovoltaic parameter sequence to obtain an importance score value corresponding to each dimension parameter in the sample photovoltaic parameter sequence; Determining a target parameter from among the multi-dimensional parameters in the sample photovoltaic parameter sequence according to the importance score value; In the preprocessed historical photovoltaic parameter sequence, parameters matching the target parameters are selected to construct the target photovoltaic parameter sequence.

5. The photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion according to any one of claims 1 to 3, characterized in that: The step of performing importance scoring on each dimension parameter in the sample photovoltaic parameter sequence to obtain the importance score value corresponding to each dimension parameter in the sample photovoltaic parameter sequence includes: Calculating the correlation between every two dimensional parameters in the sample photovoltaic parameter sequence to obtain a first correlation value corresponding to each dimensional parameter in the sample photovoltaic parameter sequence; Calculating the correlation between each dimensional parameter in the sample photovoltaic parameter sequence and the power generation label to obtain a second correlation value corresponding to each dimensional parameter in the sample photovoltaic parameter sequence; The first correlation value and the second correlation value are combined to obtain the importance score value corresponding to each dimensional parameter in the sample photovoltaic parameter sequence.

6. The photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion according to any one of claims 1 to 3, characterized in that: The step of inputting the historical moving average kernel component and the historical seasonal component into a prediction module of a prediction model to predict the moving average kernel component under a period to be predicted and the seasonal component under the period to be predicted includes: Inputting the historical moving average kernel component into the first branch network of the prediction module to predict the moving average kernel component in the period to be predicted; Inputting the historical seasonal component into the second branch network of the prediction module to predict the seasonal component under the period to be predicted; Among them, the first branch network is constructed based on the unscented Kalman filter model; the second branch network is constructed based on the trained extended long short-term memory network model.

7. The photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion according to any one of claims 1 to 3, characterized in that: The historical photovoltaic parameter sequence is constructed based on the historical irradiation information, historical environmental information and historical operating parameters of the photovoltaic system to be predicted in the first historical period.

8. A photovoltaic power generation prediction device based on dynamic decomposition and digital-analog fusion, characterized in that: include: A feature selection unit is used to perform feature selection on a historical photovoltaic parameter sequence of a photovoltaic system to be predicted in a first historical period to obtain a target photovoltaic parameter sequence; A decomposition unit, used for dynamically decomposing the target photovoltaic parameter sequence to obtain a historical moving average kernel component in the first historical period and a historical seasonal component in the first historical period; A first prediction unit, used for inputting the historical moving average kernel component and the historical seasonal component into a prediction module of a prediction model, and predicting the moving average kernel component under a period to be predicted and the seasonal component under the period to be predicted; A second prediction unit is used to input the moving average kernel component of the period to be predicted and the seasonal component of the period to be predicted into the fusion module of the prediction model to predict the power generation forecast value of the photovoltaic system to be predicted in the period to be predicted; Among them, the prediction model is obtained by jointly training the sample photovoltaic parameter sequence of the sample photovoltaic system in the second historical period, and the moving average kernel component label, seasonal component label and power generation label in the third historical period on the basis of the digital-analog fusion model; the digital-analog fusion model is constructed based on the fusion of the unscented Kalman filter model and the extended long short-term memory network model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion as described in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the photovoltaic power generation prediction method based on dynamic decomposition and digital-analog fusion as described in any one of claims 1 to 7 is implemented.

Images