Physical constraint photovoltaic power probability prediction method, device, equipment and medium

Abstract

The invention relates to a physically constrained photovoltaic power probability prediction method, device and equipment and a medium, and belongs to the technical field of new energy power generation prediction.The method comprises the steps that preprocessed regional operation data is acquired; on the basis of the preprocessed regional operation data, regional constraint characteristics at a preset future moment are calculated, an input characteristic set is generated according to the regional operation data and the preprocessed regional constraint characteristics, and the regional constraint characteristics comprise theoretical maximum power and irradiance uniformity indexes; inputting the input feature set into a preset quantile prediction model to obtain initial prediction results under different confidence degrees; constraining the power range boundary of the initial prediction result through a preset physical constraint and a theoretical maximum power to obtain a prediction optimization result; and performing uncertainty correction on the prediction optimization result based on the irradiance uniformity index, and dynamically adjusting the interval width of the prediction optimization result to obtain a final prediction result. The method has the effect of improving the power generation prediction accuracy.

Description

Technical Field

[0001] This invention belongs to the technical field of new energy power generation prediction, and in particular relates to a method, device, equipment and medium for predicting photovoltaic power probability with physical constraints. Background Technology

[0002] Currently, photovoltaic power generation has become an important part of new energy consumption. With the continuous grid connection of centralized and distributed power stations, multiple power stations have formed large-scale photovoltaic clusters in the same area. Power dispatch centers need to make operational decisions based on the overall output characteristics of the clusters, which puts forward higher requirements for the power prediction of photovoltaic clusters in the future.

[0003] Existing methods for predicting photovoltaic (PV) cluster power are mostly based on modeling using historical power data and numerical weather prediction data from individual power plants, outputting single-point prediction values ​​through time series methods or machine learning models. However, in cluster scenarios, different power plants exhibit differences in installed capacity, geographical distribution, and spatial unevenness of irradiance. Prediction methods at the level of a single power plant cannot accurately characterize the random fluctuations in the overall power of the cluster, resulting in significant deviations between prediction results and actual operation. It is difficult to quantify the uncertainty of the prediction. Furthermore, some methods attempt to introduce statistical interval estimation or physical correction, but they often lack constraints closely integrated with the operating characteristics of the power plant cluster, easily leading to prediction intervals that are too wide or too narrow, affecting the reference value of the results.

[0004] The existing technical solutions mentioned above have the following drawbacks: the existing photovoltaic cluster power prediction methods lack effective characterization of cluster spatial differences and physical constraints, resulting in insufficient physical rationality and unstable interval coverage in the prediction results, thus there is room for improvement. Summary of the Invention

[0005] The purpose of this invention is to provide a method, apparatus, device, and medium for predicting photovoltaic power probability under physical constraints, so as to solve the technical problem of inaccurate prediction results in existing photovoltaic cluster power prediction methods.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a physically constrained photovoltaic power probabilistic prediction method, the method comprising: Obtain preprocessed regional operation data; Based on the preprocessed regional operation data, the regional constraint characteristics at a preset future time are calculated, and an input feature set is generated according to the regional operation data and the preprocessed regional constraint characteristics. The regional constraint characteristics include theoretical maximum power and irradiance uniformity index. The input feature set is input into a preset quantile prediction model to obtain initial prediction results at different confidence levels; By constraining the power range boundary of the initial prediction result with preset physical constraints and the theoretical maximum power, the prediction optimization result is obtained; The uncertainty of the prediction optimization result is corrected based on the irradiance uniformity index, and the interval width of the prediction optimization result is dynamically adjusted to obtain the final prediction result.

[0007] By adopting the above technical solutions, and by acquiring preprocessed regional operational data, it is possible to ensure that the input data sources are multi-dimensional and consistent in terms of spatiotemporal dimensions, thereby improving the reliability and stability of subsequent predictions. By calculating regional constraint features and generating input feature sets, physical constraints can be introduced to avoid prediction results deviating from the physically reasonable range, thereby enhancing the interpretability and credibility of the prediction model. By inputting the input feature set into the quantile prediction model, prediction results at different confidence levels can be output, thereby providing interval-based information support for scheduling and operation. By applying physical constraints and uncertainty correction to the initial prediction results, the prediction interval width can be dynamically optimized, thereby improving interval coverage and prediction accuracy while ensuring physical reasonableness.

[0008] In one example, the present invention can be further configured such that: obtaining the preprocessed regional running data includes: The target area and each grid's NWP data, total photovoltaic installed capacity, historical cluster total power data, and the latitude and longitude information of the center point of the power plant cluster are obtained as the regional operation data; The outlier data of the region's operation data is removed by statistical threshold principle and preset physical range verification, and the missing data is filled by linear interpolation of data from adjacent time points. The short-term fluctuations in the regional operation data are reduced by using a moving average method, and the time resolution of the regional operation data is uniformly processed according to a preset time interval. The mean irradiance G of the NWP data in each grid was calculated. i and average temperature T i This creates a spatial scale that matches the operating range of the photovoltaic cluster; Based on the results of outlier handling, sliding processing, temporal resolution unification, and spatial scale matching, preprocessed regional operational data are obtained.

[0009] By adopting the above technical solution, a complete input data foundation can be formed by acquiring NWP grid data of each grid in the target area, the total installed capacity of photovoltaics in the area, the historical total power data of the cluster, and the latitude and longitude of the center point of the power station cluster. This provides comprehensive data support for subsequent feature extraction and model training. By performing outlier removal, interpolation completion, sliding smoothing, unified temporal resolution, and spatial scale matching on the data, the continuity and consistency of the data can be guaranteed, thereby eliminating the impact of noise interference and sampling differences on the prediction results.

[0010] In one example, the present invention can be further configured as follows: calculating the regional constraint characteristics at a preset future time based on the preprocessed regional operation data includes: Calculate the weighted average solar irradiance G of all grids in the region at prediction time t. avg (t), and according to the preset theoretical maximum power formula P max (t)=η avg ×P total ×G avg (t), calculate the theoretical maximum power P at the predicted time t. max (t), where η avg P is the preset regional average efficiency coefficient. total The total installed photovoltaic capacity in the region; By using a preset formula for irradiance uniformity index Calculate the irradiance uniformity index U i G i (t) represents the mean irradiance of each grid, std() is the standard deviation function, and mean() is the mean value function; By using the preset temperature difference characteristic formula T delta (t)=mean(T i (t)-T ref ), calculate the characteristic temperature difference T at the predicted time t. delta (t), where T i (t) represents the average temperature of each grid at time t, where T ref This is the preset standard temperature for the components.

[0011] By adopting the above technical solutions, the theoretical maximum power of the region can be calculated at the prediction time, and the upper limit constraint on the prediction results can be imposed according to the physical boundary, thereby avoiding the predicted value from exceeding the possible range of the installed capacity. By calculating the regional irradiance uniformity index, the irradiance difference between different grids can be quantitatively characterized, thus providing a basis for uncertainty modeling. By calculating the temperature difference characteristics, the deviation between the regional temperature level and the standard operating conditions can be reflected, thereby compensating for the power prediction error caused by temperature changes.

[0012] In one example, the present invention can be further configured as follows: the weighted average solar irradiance G of all grids in the region at prediction time t is calculated. avg (t), including: pass Calculate the weighted average value G of the solar irradiance. avg (t), where, This represents the percentage of the total installed photovoltaic capacity of each grid within the region.

[0013] By adopting the above technical solution, the weighted average of solar irradiance of each grid in the region at the prediction time can comprehensively reflect the contribution of different grids to the overall illumination level, thereby avoiding the lack of representativeness caused by single grid data. By using the installed capacity ratio of each grid as a weighting factor, the irradiance characteristics can be made closer to the actual power generation capacity distribution, thereby improving the physical rationality of the prediction characteristics and the accuracy of the prediction results.

[0014] In one example, the present invention can be further configured such that the construction of the quantile prediction model includes: Based on several preset confidence levels, corresponding quantile regression models are established respectively, wherein the confidence levels include lower quantile, median quantile and upper quantile; Historical power features and corresponding meteorological features, regional constraint features, and temporal features at the same timestamp are obtained as training datasets, and each quantile regression model is trained independently based on the training datasets. The difference between the predicted value and the true value is weighted and measured based on the quantile loss function to iteratively optimize the parameters of the quantile regression model until the predicted value output by each quantile regression model reaches a preset accuracy, and then the quantile prediction model is generated based on each quantile regression model.

[0015] By adopting the above technical solution, the uncertainty distribution of photovoltaic power can be characterized by inputting the input feature set into the quantile prediction model and outputting the prediction results at different confidence levels, thus providing more comprehensive power information than single-point prediction. By establishing independent quantile regression models and training them separately, the model's fitting accuracy to the nonlinear relationship between input and output can be improved, thereby enhancing the credibility of the prediction results at different quantiles.

[0016] In one example, the present invention can be further configured as follows: constraining the power range boundary of the initial prediction result by means of preset physical constraints and the theoretical maximum power includes: Obtain the UTC time data of the initial prediction result, calculate the corresponding solar altitude angle based on the latitude and longitude information of the center point of the power plant cluster and the UTC time data, and introduce a nighttime zero value constraint on the initial prediction result if the solar altitude angle does not exceed the preset altitude angle threshold, and uniformly modify the initial prediction result to 0. If the solar altitude angle is greater than a preset altitude angle threshold, the initial prediction result is subject to an upper limit constraint based on the theoretical maximum power, and the initial prediction result exceeding the theoretical maximum power is cropped to the theoretical maximum power.

[0017] By adopting the above technical solutions, and by applying a zero-value constraint at night to the prediction results when the solar altitude angle is less than the threshold, non-zero predictions under no-sunlight conditions can be avoided, thus ensuring the physical rationality of the results. By using the theoretical maximum power to prune the prediction values ​​when the solar altitude angle is greater than the threshold, the prediction values ​​can be ensured not to exceed the regional installed capacity limit, thereby enhancing the actual usability of the prediction results.

[0018] In one example, the present invention can be further configured as follows: performing uncertainty correction on the prediction optimization result based on the irradiance uniformity index, dynamically adjusting the interval width of the prediction optimization result, and obtaining the final prediction result includes: Obtain the lower quantile optimization prediction value from the prediction optimization results. Median optimized prediction value and upper quantile optimized prediction value ,pass , Calculate the radius Up of the constrained upper interval respectively. raw (t) and lower interval radius Low raw (t); According to the irradiance uniformity index U i (t), through S(t) = 1 + δ × [U i (t)-U m Determine the scaling factor S(t), where δ is the adjustment coefficient, U m The median of the historical irradiance uniformity index; Based on the upper interval radius Up raw (t), the lower interval radius Low raw (t) and the scaling factor, through , Calculate the final upper confidence boundary separately and the final lower confidence boundary This is done to generate the final prediction result.

[0019] By adopting the above technical solutions, the difference between the median predicted value and the upper and lower quantile boundaries can be obtained by calculating the upper and lower radii of the initial prediction results, thus providing a quantitative basis for adjusting the interval width. By constructing a scaling factor based on the irradiance uniformity index and dynamically adjusting the interval radius, the interval width can be made to change with the size of spatial differences, thereby avoiding the prediction interval being too wide or too narrow. By applying zero and maximum power constraints and correcting the final interval boundary, it can be ensured that the prediction interval satisfies both physical rationality and dynamically reflects regional uncertainty, thereby improving the practicality and stability of the final prediction results.

[0020] In a second aspect, the present invention provides a physically constrained photovoltaic power probability prediction device, the device comprising: The data acquisition module is used to acquire preprocessed regional operational data; The feature generation module is used to calculate the regional constraint features at a preset future time based on the preprocessed regional operation data, and generate an input feature set according to the regional operation data and the preprocessed regional constraint features, wherein the regional constraint features include theoretical maximum power and irradiance uniformity index. The model input module is used to input the input feature set into a preset quantile prediction model to obtain initial prediction results at different confidence levels. The physical constraint module is used to constrain the power range boundary of the initial prediction result by means of preset physical constraints and the theoretical maximum power, so as to obtain the prediction optimization result; An uncertainty correction module is used to perform uncertainty correction on the prediction optimization results based on the irradiance uniformity index, dynamically adjust the interval width of the prediction optimization results, and obtain the final prediction results.

[0021] By adopting the above technical solutions, and by acquiring preprocessed regional operational data, it is possible to ensure that the input data sources are multi-dimensional and consistent in terms of spatiotemporal dimensions, thereby improving the reliability and stability of subsequent predictions. By calculating regional constraint features and generating input feature sets, physical constraints can be introduced to avoid prediction results deviating from the physically reasonable range, thereby enhancing the interpretability and credibility of the prediction model. By inputting the input feature set into the quantile prediction model, prediction results at different confidence levels can be output, thereby providing interval-based information support for scheduling and operation. By applying physical constraints and uncertainty correction to the initial prediction results, the prediction interval width can be dynamically optimized, thereby improving interval coverage and prediction accuracy while ensuring physical reasonableness.

[0022] In a third aspect, the present invention provides an electronic device including a memory and a processor, the memory storing a computer program, characterized in that the processor executes the computer program to implement the steps of the physically constrained photovoltaic power probability prediction method.

[0023] In a fourth aspect, the present invention provides a storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the physically constrained photovoltaic power probability prediction method described above.

[0024] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By calculating the upper and lower radii of the initial prediction results, the difference between the median prediction value and the upper and lower quantile boundaries can be obtained, thus providing a quantitative basis for adjusting the interval width. By constructing a scaling factor based on the irradiance uniformity index and dynamically adjusting the interval radius, the interval width can be made to vary with the size of spatial differences, thereby avoiding the prediction interval being too wide or too narrow. By applying zero and maximum power constraints and correcting the final interval boundary, it can be ensured that the prediction interval satisfies both physical rationality and dynamically reflects regional uncertainty, thereby improving the practicality and stability of the final prediction results. 2. By applying a zero-value constraint to the prediction results when the solar altitude angle is less than the threshold, non-zero predictions under no-sunlight conditions can be avoided, thus ensuring the physical rationality of the results; by using the theoretical maximum power to prune the prediction values ​​when the solar altitude angle is greater than the threshold, the prediction values ​​can be ensured not to exceed the regional installed capacity limit, thus enhancing the actual usability of the prediction results. 3. By calculating the theoretical maximum power of the region at the prediction time, the upper limit constraint of the prediction results can be imposed according to the physical boundary, thereby avoiding the predicted value from exceeding the possible range of the installed capacity; by calculating the regional irradiance uniformity index, the irradiance difference between different grids can be quantitatively characterized, thereby providing a basis for uncertainty modeling; by calculating the temperature difference characteristics, the deviation between the regional temperature level and the standard operating conditions can be reflected, thereby compensating for the power prediction error caused by temperature changes. Attached Figure Description

[0025] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of a physically constrained photovoltaic power probability prediction method in an embodiment of the present invention; Figure 2 This is a structural block diagram of a photovoltaic power probability prediction device with physical constraints according to an embodiment of the present invention; Figure 3This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0026] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0027] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0028] Example 1 like Figure 1 As shown, this invention discloses a physically constrained photovoltaic power probabilistic prediction method, which specifically includes the following steps: S10: Obtain preprocessed regional operation data.

[0029] Specifically, the preprocessed regional operational data is acquired, including irradiance and temperature information collected from numerical weather prediction (NWP) grid data at each grid point within the target area, as well as the total installed capacity of photovoltaic power in the region, historical total power data of the cluster, and the latitude and longitude information of the center point of the power plant cluster.

[0030] S20: Based on the preprocessed regional operation data, calculate the regional constraint characteristics at a preset future time, and generate an input feature set according to the regional operation data and the preprocessed regional constraint characteristics. The regional constraint characteristics include the theoretical maximum power and the irradiance uniformity index.

[0031] Specifically, based on the preprocessed regional operational data, regional constraint features are calculated at the prediction time, including the regional theoretical maximum power and the regional irradiance uniformity index. The calculated regional constraint features are then combined with the preprocessed regional operational data to generate an input feature set. This input feature set can be used as the input to the subsequent quantile prediction model, providing physical boundary constraints and spatial distribution information for the model to output prediction results at different confidence levels.

[0032] S30: Input the input feature set into the preset quantile prediction model to obtain the initial prediction results under different confidence levels.

[0033] Specifically, the input feature set is input into a preset quantile prediction model. The model calculates the following quantile confidence, middle quantile confidence, and upper quantile confidence for the input feature set at different confidence levels, thereby obtaining the initial prediction results for the corresponding multiple confidence levels.

[0034] S40: By setting physical constraints and theoretical maximum power, the power range boundary of the initial prediction result is constrained to obtain the prediction optimization result.

[0035] Specifically, the power range boundary of the initial prediction result is constrained by preset physical constraints to ensure that the prediction result is physically reasonable. When the solar altitude angle does not exceed the preset threshold, a zero-value constraint is introduced for the initial prediction result at night, and the prediction value is uniformly corrected to 0. When the solar altitude angle is greater than the preset threshold, the prediction value is limited to within the theoretical maximum power Pmax(t) of the region.

[0036] S50: Based on the irradiance uniformity index, the uncertainty of the prediction optimization results is corrected, and the interval width of the prediction optimization results is dynamically adjusted to obtain the final prediction result.

[0037] Specifically, uncertainty correction is performed on the prediction optimization results based on the regional irradiance uniformity index. The upper and lower boundary widths of the prediction interval are dynamically adjusted according to the size of the spatial difference in irradiance. When the difference is large, the prediction interval is appropriately widened, and when the difference is small, the prediction interval is narrowed, so as to obtain the final prediction result that takes into account both physical rationality and interval coverage.

[0038] In one embodiment, step S10, namely acquiring the preprocessed region operation data, includes: S11: Obtain NWP data, total photovoltaic installed capacity, historical cluster total power data, and latitude and longitude information of the center point of the power plant cluster for the target area and each grid as regional operation data.

[0039] Specifically, the system acquires NWP data such as irradiance and temperature for each numerical weather forecast grid point within the target area, obtains the total installed photovoltaic capacity of the area, acquires historical cluster total power data, and acquires the latitude and longitude information of the power plant cluster center point. The above data are used as regional operation data for subsequent processing and modeling.

[0040] S12: Outlier removal is performed on regional operation data using statistical threshold principles and preset physical range verification, and missing data is filled in by linear interpolation of data from adjacent time points.

[0041] Specifically, outlier removal and missing value completion are performed on the regional operation data. This includes filtering NWP grid point data according to statistical threshold principles and preset physical ranges, removing records with wind speeds greater than 40m / s or irradiance less than 0W / ㎡, removing outliers that exceed the range of [0, total installed photovoltaic capacity of the region] for historical cluster total power data, and using linear interpolation of power at adjacent times to fill in missing values ​​to ensure the physical rationality and time series continuity of the data.

[0042] S13: Reduce short-term fluctuations in regional operation data by using a moving average method, and perform time resolution uniform processing on regional operation data according to a preset time interval.

[0043] Specifically, the regional operational data is smoothed and its time resolution is unified. This includes smoothing short-term fluctuations in NWP grid point data using a moving average method to suppress high-frequency noise interference, and unifying historical power data and meteorological data to a preset 15-minute time interval through linear interpolation or sampling, so that data from different sources are fully aligned in timestamps for subsequent joint use.

[0044] S14: Calculate the mean irradiance G of NWP data for each grid. i and average temperature T i This creates a spatial scale that matches the operating range of the photovoltaic cluster.

[0045] Specifically, spatial scale matching processing is performed, including collecting all NWP grid point data within the geographical coverage area of ​​the photovoltaic cluster and calculating the average irradiance G within the coverage area. i With the average temperature T i This will enable the acquisition of macro-meteorological quantities that can represent the cluster scale, achieve spatial scale consistency mapping between regional NWP data and the operating range of photovoltaic clusters, and provide a spatially matched data baseline for subsequent feature extraction and modeling.

[0046] S15: Based on the outlier processing results, sliding processing results, temporal resolution unification results, and spatial scale matching results, the preprocessed regional operation data is obtained.

[0047] Specifically, data fusion is performed based on outlier removal results, smoothing results, temporal resolution unification results, and spatial scale matching results to form a multi-source preprocessed dataset that is consistent and continuous in both temporal and spatial dimensions. This dataset serves as the basic input for subsequent calculation of regional constraint features and construction of input feature sets, thereby obtaining preprocessed regional operation data.

[0048] In one embodiment, step S20, namely, calculating the regional constraint characteristics at a preset future time based on the preprocessed regional operation data, includes: S21: Calculate the weighted average solar irradiance G of all grids within the region at prediction time t. avg (t), and according to the preset theoretical maximum power formula P max (t)=η avg ×P total ×G avg (t), calculate the theoretical maximum power P at the predicted time t. max (t), where η avg P is the preset regional average efficiency coefficient.total This represents the total installed capacity of photovoltaic power in the region.

[0049] Specifically, at the predicted time t, the weighted average solar irradiance G of all grid points in the region is calculated. avg (t), then based on the preset regional average efficiency coefficient η avg With the total installed capacity of photovoltaic power in the region P total , will G avg (t) Substitute into the theoretical maximum power calculation formula P max (t)=η avg ×P total ×G avg (t) Obtain the theoretical maximum power P in the region at the predicted time t. max (t), the calculation result provides physical boundary constraints for subsequent probability prediction.

[0050] S22: By using a preset formula for irradiance uniformity index Calculate the irradiance uniformity index U i G i (t) represents the mean irradiance of each grid, std() is the standard deviation function, and mean() is the mean value function.

[0051] Specifically, to reflect the consistency of irradiance distribution at different grid points, the regional irradiance uniformity index Ui is calculated, and the mean irradiance sequence {G1(t),G2(t),…,G...} of all grid points is obtained at the prediction time t. i Using the standard deviation function std() and the mean function mean(), the irradiance uniformity index U is calculated using the above formula. i To measure the dispersion of irradiance between different grid points, when U i A larger value indicates uneven irradiance distribution and relatively higher prediction uncertainty. i A smaller value indicates a more uniform irradiance distribution, which helps improve the stability of the prediction.

[0052] S23: By using the preset temperature difference characteristic formula T delta (t)=mean(T i (t)-T ref ), calculate the characteristic temperature difference T at the predicted time t. delta (t), where T i (t) represents the average temperature of each grid at time t, where T is the mean temperature of the grid. ref This is the preset standard temperature for the components.

[0053] Specifically, in order to characterize the deviation between the temperature at different grid points and the standard operating conditions, the characteristic temperature difference T of the region is calculated. delta(t), where the average temperature T of each grid point is obtained at the prediction time t. i (t), and compared with the preset component standard temperature T. ref By comparing the values, the characteristic value T of the average temperature difference in the region can be calculated. delta (t) is used to reflect the degree of difference between the actual temperature and the standard temperature in the region, thereby providing an environmental correction reference for subsequent model inputs.

[0054] In one embodiment, in step S21, the weighted average solar irradiance G of all grids in the region at the predicted time t is calculated. avg (t), including: S211: Through Calculate the weighted average value of solar irradiance G avg (t), where, This represents the percentage of the total installed photovoltaic capacity of each grid within the region.

[0055] Specifically, at the predicted time t, the weighted average solar irradiance G of all grid points in the region is calculated. avg (t), the irradiance G at each grid point i (t) is the percentage of installed capacity corresponding to this grid. The weighted average irradiance G at time t is then calculated. avg (t), where This represents the ratio of the total installed photovoltaic capacity of the i-th grid region to the total installed capacity of the region.

[0056] In one embodiment, step S30, namely the construction of the quantile prediction model, includes: S31: Based on several preset confidence levels, establish corresponding quantile regression models, where the confidence levels include lower quantile, median quantile, and upper quantile.

[0057] Specifically, corresponding quantile regression models are established according to multiple preset confidence levels. After the input feature set is sent in, it is mapped to model channels at different confidence levels, including low quantile, middle quantile and high quantile channels. Each channel independently generates a predicted value corresponding to the confidence level, forming a model structure that can output multiple quantile prediction results at the same time. The quantile τ takes values ​​between (0,1) and is used to define different confidence levels, thereby ensuring that the model can cover different needs of point prediction and interval prediction.

[0058] S32: Obtain historical power features and corresponding meteorological features, regional constraint features, and temporal features at the same timestamp as training datasets, and train each quantile regression model independently based on the training datasets.

[0059] Specifically, historical power features are aligned with meteorological features, regional constraint features, and time features under the corresponding timestamps according to the sample sequence to form sample pairs containing input and target output as training datasets. The training datasets are then input into each quantile regression model for independent training. During the training process, the relationship between input features and output power prediction values ​​is gradually fitted, enabling the model to learn the distribution features under different quantiles and to generate prediction values ​​with different confidence levels based on the input feature set.

[0060] S33: The difference between the predicted value and the true value is weighted and measured based on the quantile loss function to iteratively optimize the parameters of the quantile regression model until the predicted value output by each quantile regression model reaches the preset accuracy, and then a quantile prediction model is generated based on each quantile regression model.

[0061] Specifically, the quantile loss function is used as the optimization objective to weight the error between the predicted quantile value and the actual power value. The quantile loss function is defined as follows: , where τ is the quantile level and y is the true value. The parameters of each quantile regression model are continuously adjusted through iterative optimization to make them converge to the optimal solution on the training samples, until the prediction results at different confidence levels reach the preset accuracy requirements, thereby completing the training of the quantile prediction model.

[0062] In one embodiment, step S40, which involves constraining the power range boundary of the initial prediction result by pre-set physical constraints and theoretical maximum power, includes: S41: Obtain the UTC time data of the initial prediction result, calculate the corresponding solar altitude angle based on the latitude and longitude information of the center point of the power plant cluster and the UTC time data, and introduce a nighttime zero value constraint to the initial prediction result if the solar altitude angle does not exceed the preset altitude angle threshold, and uniformly modify the initial prediction result to 0.

[0063] Specifically, at the prediction time, the UTC time data corresponding to the initial prediction result is obtained, and the solar altitude angle at that time is calculated by combining the latitude and longitude information of the center point of the power plant cluster. When the solar altitude angle α(t) is less than or equal to the preset altitude angle threshold, such as when the solar altitude angle is less than 0°, it is determined that the current condition is nighttime with no sunlight. The initial prediction results under all confidence levels are uniformly corrected to 0 to ensure that the prediction value will not have physically unreasonable non-zero power output under the condition of no sunlight.

[0064] S42: When the solar altitude angle is greater than the preset altitude angle threshold, the initial prediction results are subject to an upper limit constraint based on the theoretical maximum power, and the initial prediction results that exceed the theoretical maximum power are cropped to the theoretical maximum power.

[0065] Specifically, when the solar altitude angle α(t) is greater than a preset altitude angle threshold such as 0°, the maximum power P of the region theory is used. max (t) Apply a power upper limit constraint to the initial prediction results, and prune prediction values ​​that exceed this upper limit to correct them to the corresponding P value at that time. max (t), thereby ensuring that the prediction results at different confidence levels do not exceed the maximum power range that the regional physical conditions can support, and avoiding the prediction values ​​from exceeding the physical reasonable boundaries.

[0066] In one embodiment, step S50, which involves performing uncertainty correction on the prediction optimization results based on the irradiance uniformity index and dynamically adjusting the interval width of the prediction optimization results to obtain the final prediction result, includes: S51: Obtain the lower quantile optimization prediction value from the prediction optimization results. Median optimized prediction value and upper quantile optimized prediction value ,pass , Calculate the radius Up of the constrained upper interval respectively. raw (t) and lower interval radius Low raw (t).

[0067] Specifically, predicted values ​​at different confidence levels are extracted from the prediction optimization results, and the interval radius is calculated for the prediction optimization results. The upper interval radius is calculated using the upper quantile optimized predicted value and the median optimized predicted value, and the lower interval radius is calculated using the median optimized predicted value and the lower quantile optimized predicted value, so as to reflect the difference between the median value and the upper and lower confidence boundaries and provide a benchmark for subsequent adjustment of the interval width.

[0068] S52: Based on the irradiance uniformity index U i (t), through S(t) = 1 + δ × [U i (t)-U m Determine the scaling factor S(t), where δ is the adjustment coefficient, U m This represents the median of the historical irradiance uniformity index.

[0069] Specifically, based on the regional irradiance uniformity index U i (t) Construct a scaling factor to dynamically adjust the original interval radius, when U i (t) is greater than U m When the irradiance difference within the region is large, the scaling factor S(t) is greater than 1, and the interval radius will be magnified. i (t) is less than U mWhen the irradiance distribution is relatively uniform, the scaling factor S(t) is less than 1, and the interval radius will be shrunk, so that the prediction interval can be dynamically expanded or contracted according to spatial differences.

[0070] S53: Based on the upper interval radius Up raw (t), lower interval radius Low raw (t) and scaling factor, through , Calculate the final upper confidence boundary separately and the final lower confidence boundary This is used to generate the final prediction results.

[0071] Specifically, the final prediction interval boundary is corrected using the scaled interval radius, and the final upper confidence boundary and lower confidence boundary are calculated. The final prediction interval is obtained by combining the upper and lower boundaries as the final prediction result.

[0072] Example 2 like Figure 2 As shown, based on the same inventive concept as the above embodiments, the present invention also provides a physically constrained photovoltaic power probability prediction device, comprising: The data acquisition module is used to acquire preprocessed regional operational data; The feature generation module is used to calculate the regional constraint features at a preset future time based on the preprocessed regional operation data, and generate an input feature set based on the regional operation data and the preprocessed regional constraint features. The regional constraint features include the theoretical maximum power and the irradiance uniformity index. The model input module is used to input the input feature set into the preset quantile prediction model to obtain the initial prediction results at different confidence levels. The physical constraint module is used to constrain the power range boundary of the initial prediction result by pre-set physical constraints and theoretical maximum power, so as to obtain the prediction optimization result; The uncertainty correction module is used to correct the uncertainty of the prediction optimization results based on the irradiance uniformity index, dynamically adjust the interval width of the prediction optimization results, and obtain the final prediction result.

[0073] Optionally, the data acquisition module includes: The data acquisition submodule is used to acquire NWP data, total photovoltaic installed capacity, historical cluster total power data, and latitude and longitude information of the center point of the power plant cluster for the target area and each grid as regional operation data; The anomaly handling submodule is used to remove outliers from regional operation data by using statistical threshold principles and preset physical range verification, and to fill in the corresponding missing data by linear interpolation of data from adjacent time points. The smoothing and unification submodule is used to reduce short-term fluctuations in regional operation data through a moving average method, and to perform time resolution unification processing on regional operation data according to a preset time interval. The spatial matching submodule is used to calculate the mean irradiance G of the NWP data for each grid. i and average temperature T i This creates a spatial scale that matches the operating range of the photovoltaic cluster; The data fusion submodule is used to obtain preprocessed regional operational data based on outlier processing results, sliding processing results, temporal resolution unification results, and spatial scale matching results.

[0074] Optionally, the feature generation module includes: The power calculation submodule is used to calculate the weighted average solar irradiance G of all grids in the region at prediction time t. avg (t), and according to the preset theoretical maximum power formula P max (t)=η avg ×P total ×G avg (t), calculate the theoretical maximum power P at the predicted time t. max (t), where η avg P is the preset regional average efficiency coefficient. total The total installed capacity of photovoltaic power in the region; The uniformity calculation submodule is used to calculate the uniformity index using a preset formula. Calculate the irradiance uniformity index U i G i (t) represents the mean irradiance of each grid, std() is the standard deviation function, and mean() is the mean function; The temperature difference calculation submodule is used to calculate the temperature difference using a preset characteristic formula T. delta (t)=mean(T i (t)-T ref ), calculate the characteristic temperature difference T at the predicted time t. delta (t), where T i (t) represents the average temperature of each grid at time t, where T is the mean temperature of the grid. ref This is the preset standard temperature for the components.

[0075] Optionally, the power calculation submodule includes: Weighted average calculation unit, used to calculate the weighted average value by... Calculate the weighted average value of solar irradiance G avg (t), where, This represents the percentage of the total installed photovoltaic capacity of each grid within the region.

[0076] Optional, the construction of the quantile prediction model includes: The multiquantile submodule is used to establish corresponding quantile regression models based on several preset confidence levels, where the confidence levels include lower quantile, median quantile, and upper quantile. The sample construction submodule is used to obtain historical power features and meteorological features, regional constraint features and time features corresponding to the same timestamp as training datasets, and to train each quantile regression model independently based on the training datasets. The parameter optimization submodule is used to perform a weighted measurement of the difference between the predicted value and the true value based on the quantile loss function, so as to iteratively optimize the parameters of the quantile regression model until the predicted value output by each quantile regression model reaches the preset accuracy, and then generate a quantile prediction model based on each quantile regression model.

[0077] Optional, the physical constraint module includes: The zero-value constraint submodule is used to obtain the UTC time data of the initial prediction results, calculate the corresponding solar altitude angle based on the latitude and longitude information of the center point of the power plant cluster and the UTC time data, and introduce a nighttime zero-value constraint on the initial prediction results if the solar altitude angle does not exceed the preset altitude angle threshold, and uniformly modify the initial prediction results to 0. The upper limit constraint submodule is used to impose an upper limit constraint on the initial prediction results based on the theoretical maximum power when the solar altitude angle is greater than the preset altitude angle threshold. Initial prediction results that exceed the theoretical maximum power are then cropped to the theoretical maximum power.

[0078] Optional, uncertainty modules include: The radius calculation submodule is used to obtain the lower quantile optimization prediction value from the prediction optimization results. Median optimized prediction value and upper quantile optimized prediction value ,pass , Calculate the radius Up of the constrained upper interval respectively. raw (t) and lower interval radius Low raw (t); The scaling factor calculation submodule is used to calculate the scaling factor based on the irradiance uniformity index U. i (t), through S(t) = 1 + δ × [U i (t)-U m Determine the scaling factor S(t), where δ is the adjustment coefficient, U m The median of the historical irradiance uniformity index; The boundary correction submodule is used to adjust the boundary based on the upper interval radius Up. raw (t), lower interval radius Low raw (t) and scaling factor, through , Calculate the final upper confidence boundary separately and the final lower confidence boundary This is used to generate the final prediction results.

[0079] Example 3 like Figure 3 As shown, the present invention also provides an electronic device 100 for implementing a photovoltaic power probability prediction method with physical constraints; The electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on at least one processor 102, and at least one communication bus 104.

[0080] The memory 101 can be used to store the computer program 103. The processor 102 implements the steps of the physically constrained photovoltaic power probability prediction method of Embodiment 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101.

[0081] The memory 101 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0082] At least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 102 may be a microprocessor or any conventional processor. Processor 102 is the control center of electronic device 100, connecting various parts of electronic device 100 via various interfaces and lines.

[0083] The memory 101 in the electronic device 100 stores multiple instructions to implement a physically constrained photovoltaic power probabilistic prediction method, and the processor 102 can execute multiple instructions to achieve the following: Obtain preprocessed regional operation data; Based on the preprocessed regional operation data, the regional constraint characteristics at a preset future time are calculated, and an input feature set is generated according to the regional operation data and the preprocessed regional constraint characteristics. The regional constraint characteristics include theoretical maximum power and irradiance uniformity index. The input feature set is input into a preset quantile prediction model to obtain initial prediction results at different confidence levels; By constraining the power range boundary of the initial prediction result with preset physical constraints and the theoretical maximum power, the prediction optimization result is obtained; The uncertainty of the prediction optimization result is corrected based on the irradiance uniformity index, and the interval width of the prediction optimization result is dynamically adjusted to obtain the final prediction result.

[0084] Example 4 If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0085] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0086] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0087] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0088] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0089] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A physically constrained photovoltaic power probabilistic prediction method, characterized in that, The method includes: Obtain preprocessed regional operation data; Based on the preprocessed regional operation data, the regional constraint characteristics at a preset future time are calculated, and an input feature set is generated according to the regional operation data and the preprocessed regional constraint characteristics. The regional constraint characteristics include theoretical maximum power and irradiance uniformity index. The input feature set is input into a preset quantile prediction model to obtain initial prediction results at different confidence levels; By constraining the power range boundary of the initial prediction result with preset physical constraints and the theoretical maximum power, the prediction optimization result is obtained; The uncertainty of the prediction optimization result is corrected based on the irradiance uniformity index, and the interval width of the prediction optimization result is dynamically adjusted to obtain the final prediction result.

2. The physically constrained photovoltaic power probabilistic prediction method according to claim 1, characterized in that, The acquisition of preprocessed regional operation data includes: The target area and each grid's NWP data, total photovoltaic installed capacity, historical cluster total power data, and the latitude and longitude information of the center point of the power plant cluster are obtained as the regional operation data; The outlier data of the region's operation data is removed by statistical threshold principle and preset physical range verification, and the missing data is filled by linear interpolation of data from adjacent time points. The short-term fluctuations in the regional operation data are reduced by using a moving average method, and the time resolution of the regional operation data is uniformly processed according to a preset time interval. The mean irradiance G of the NWP data in each grid was calculated. i and average temperature T i This creates a spatial scale that matches the operating range of the photovoltaic cluster; Based on the results of outlier handling, sliding processing, temporal resolution unification, and spatial scale matching, preprocessed regional operational data are obtained.

3. The physically constrained photovoltaic power probabilistic prediction method according to claim 2, characterized in that, The calculation of regional constraint characteristics at a preset future time based on the preprocessed regional operational data includes: Calculate the weighted average solar irradiance G of all grids in the region at prediction time t. avg (t), and according to the preset theoretical maximum power formula P max (t)=η avg ×P total ×G avg (t), calculate the theoretical maximum power P at the predicted time t. max (t), where η avg P is the preset regional average efficiency coefficient. total The total installed photovoltaic capacity in the region; By using a preset formula for irradiance uniformity index Calculate the irradiance uniformity index U i G i (t) represents the mean irradiance of each grid, std() is the standard deviation function, and mean() is the mean value function; By using the preset temperature difference characteristic formula T delta (t)=mean(T i (t)-T ref ), calculate the characteristic temperature difference T at the predicted time t. delta (t), where T i (t) represents the average temperature of each grid at time t, where T ref This is the preset standard temperature for the components.

4. The physically constrained photovoltaic power probabilistic prediction method according to claim 3, characterized in that, The weighted average solar irradiance G of all grids in the region at the predicted time t is calculated. avg (t), including: pass Calculate the weighted average value G of the solar irradiance. avg (t), where, This represents the percentage of the total installed photovoltaic capacity of each grid within the region.

5. The physically constrained photovoltaic power probabilistic prediction method according to claim 1, characterized in that, The construction of the quantile prediction model includes: Based on several preset confidence levels, corresponding quantile regression models are established respectively, wherein the confidence levels include lower quantile, median quantile and upper quantile; Historical power features and corresponding meteorological features, regional constraint features, and temporal features at the same timestamp are obtained as training datasets, and each quantile regression model is trained independently based on the training datasets. The difference between the predicted value and the true value is weighted and measured based on the quantile loss function to iteratively optimize the parameters of the quantile regression model until the predicted value output by each quantile regression model reaches a preset accuracy, and then the quantile prediction model is generated based on each quantile regression model.

6. The physically constrained photovoltaic power probabilistic prediction method according to claim 3, characterized in that, The constraint on the power range boundary of the initial prediction result by using preset physical constraints and the theoretical maximum power includes: Obtain the UTC time data of the initial prediction result, calculate the corresponding solar altitude angle based on the latitude and longitude information of the center point of the power plant cluster and the UTC time data, and introduce a nighttime zero value constraint on the initial prediction result if the solar altitude angle does not exceed the preset altitude angle threshold, and uniformly modify the initial prediction result to 0. If the solar altitude angle is greater than a preset altitude angle threshold, the initial prediction result is subject to an upper limit constraint based on the theoretical maximum power, and the initial prediction result exceeding the theoretical maximum power is cropped to the theoretical maximum power.

7. The physically constrained photovoltaic power probabilistic prediction method according to claim 1, characterized in that, The process of performing uncertainty correction on the prediction optimization results based on the irradiance uniformity index, dynamically adjusting the interval width of the prediction optimization results, and obtaining the final prediction result includes: Obtain the lower quantile optimization prediction value from the prediction optimization results. Median optimized prediction value and upper quantile optimized prediction value ,pass , Calculate the radius Up of the constrained upper interval respectively. raw (t) and lower interval radius Low raw (t); According to the irradiance uniformity index U i (t), through S(t) = 1 + δ × [U i (t)-U m Determine the scaling factor S(t), where δ is the adjustment coefficient, U m The median of the historical irradiance uniformity index; Based on the upper interval radius Up raw (t), the lower interval radius Low raw (t) and the scaling factor, through , Calculate the final upper confidence boundary separately and the final lower confidence boundary This is done to generate the final prediction result.

8. A physically constrained photovoltaic power probability prediction device, characterized in that, The device includes: The data acquisition module is used to acquire preprocessed regional operational data; The feature generation module is used to calculate the regional constraint features at a preset future time based on the preprocessed regional operation data, and generate an input feature set according to the regional operation data and the preprocessed regional constraint features, wherein the regional constraint features include theoretical maximum power and irradiance uniformity index. The model input module is used to input the input feature set into a preset quantile prediction model to obtain initial prediction results at different confidence levels. The physical constraint module is used to constrain the power range boundary of the initial prediction result by means of preset physical constraints and the theoretical maximum power, so as to obtain the prediction optimization result; An uncertainty correction module is used to perform uncertainty correction on the prediction optimization results based on the irradiance uniformity index, dynamically adjust the interval width of the prediction optimization results, and obtain the final prediction results.

9. An electronic device, characterized in that, It includes a processor and a memory, the processor being configured to execute a computer program stored in the memory to implement the steps of the photovoltaic power probability prediction method with physical constraints as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements the steps of the photovoltaic power probability prediction method with physical constraints as described in any one of claims 1 to 7.

Images