A photovoltaic power prediction method and device

Abstract

The application provides a photovoltaic power prediction method and device, the method comprising: obtaining power increment data corresponding to the current time by preprocessing real-time photovoltaic power data, meteorological observation data and weather forecast data at the current time; generating a real-time feature set based on the power increment data; determining the predicted power increment corresponding to each prediction step according to the real-time feature set, and superimposing it with the real-time photovoltaic power data at the current time to generate the initial predicted power result corresponding to each prediction step; and outputting the photovoltaic power prediction result by performing power constraint and smoothing processing on the initial predicted power result. The application superimposes the power increment under multiple prediction steps, simultaneously applies physical constraints and time sequence verification to the prediction result, can maintain the continuity and physical rationality of the power change in the multi-time scale prediction process, and enhances the accuracy of the prediction result in the actual scheduling scenario.

Description

Technical Field

[0001] This application relates to the field of photovoltaic power technology, and in particular to a photovoltaic power prediction method and apparatus. Background Technology

[0002] Existing photovoltaic power prediction methods mostly use historical power data and meteorological data as prediction inputs, focusing on modeling and extrapolating the absolute value of photovoltaic power. Under conditions of rapid changes in irradiance or approaching the upper limit of installed capacity, the prediction results are easily affected by physical saturation constraints, making it difficult to simultaneously characterize the relationship between power change trends and physical boundaries. Especially when making continuous predictions with multiple prediction steps, the power prediction results are prone to discontinuity or exceed the physical feasible range between different time steps, thus limiting the accuracy of the prediction results in actual scheduling scenarios. Summary of the Invention

[0003] This application provides a photovoltaic power prediction method and apparatus to solve the problem that existing prediction methods in related technologies limit the accuracy of prediction results in actual scheduling scenarios.

[0004] The first aspect of this application provides a photovoltaic power prediction method, the photovoltaic power prediction method comprising: By preprocessing the real-time photovoltaic power data, meteorological observation data, and weather forecast data of the photovoltaic power station at the current moment, the power increment data corresponding to the current moment is obtained. Based on the power increment data, multi-dimensional features are constructed to generate a real-time feature set; The predicted power increment corresponding to each prediction step is determined based on the real-time feature set, and then superimposed with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step. By performing power constraint and smoothing processing on the initial predicted power results, the photovoltaic power prediction results are output.

[0005] Optionally, in the first implementation of the first aspect of this application, the step of preprocessing the real-time photovoltaic power data, meteorological observation data, and weather forecast data of the photovoltaic power station at the current moment to obtain the power increment data corresponding to the current moment includes: Acquire real-time photovoltaic power data corresponding to the current moment of the photovoltaic power station, and simultaneously acquire meteorological observation data and weather forecast data corresponding to the current moment; By performing missing data interpolation and outlier data filtering on the real-time photovoltaic power data, meteorological observation data, and weather forecast data, a valid data set is determined. The theoretical saturation power value corresponding to the irradiance data of the weather forecast data in the effective data set at the current moment is determined according to the preset saturation power calculation rules. The power increment data corresponding to the current moment is determined by performing a difference calculation between the real-time photovoltaic power data of the effective data set and the theoretical saturation power value.

[0006] Optionally, in the second implementation of the first aspect of this application, the step of constructing multi-dimensional features based on the power increment data to generate a real-time feature set includes: Based on the current power increment data and the corresponding historical power increment data, by performing time backtracking on the historical power increment data within multiple preset time windows, the lagged power increment data corresponding to different time scales are obtained respectively, and multi-time scale power increment features are generated. By statistically analyzing the power increment characteristics at the multiple time scales, the statistical characteristics of the power increment corresponding to each time scale are determined. By performing correlation processing on the theoretical saturation power value, the real-time photovoltaic power data, and the corresponding time information, physical correlation features and time correlation features are generated. The power increment features, power increment statistical features, physical correlation features, and time correlation features are integrated to obtain the real-time feature set.

[0007] Optionally, in the third implementation of the first aspect of this application, the step of determining the predicted power increment corresponding to each prediction step based on the real-time feature set, and superimposing it with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step includes: Based on the time span information corresponding to each preset prediction step, the real-time feature set is reorganized to generate prediction input feature sets corresponding to each prediction step. By calculating the power change of the prediction input feature set corresponding to each prediction step, the prediction power increment corresponding to each prediction step is determined respectively. The predicted power increment corresponding to each prediction step is superimposed with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step.

[0008] Optionally, in the fourth implementation of the first aspect of this application, the step of outputting the photovoltaic power prediction result by performing power constraint and smoothing processing on the initial predicted power result includes: Based on the preset photovoltaic power plant installed capacity parameters and physical operation constraints, the predicted power values ​​that exceed the preset power range in the initial predicted power results are limited to the corresponding power boundary values ​​to generate constrained predicted power results. The predicted power result after the constraint is calculated by a sliding window between adjacent prediction steps, and the predicted power sequence is smoothed to generate a smoothed predicted power result. The predicted power change amplitude between adjacent prediction steps is determined based on the smoothed predicted power result, and the predicted power change amplitude is compared with a preset change threshold to determine the predicted power value that meets the change constraint condition in the predicted power result, thereby generating the photovoltaic power prediction result.

[0009] Optionally, in the fifth implementation of the first aspect of this application, before the step of determining the predicted power increment corresponding to each prediction step size based on the real-time feature set and superimposing it with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step size, the method further includes: Acquire historical photovoltaic power data corresponding to the historical operating cycle of the photovoltaic power station and historical meteorological observation data corresponding to the timestamp of the historical photovoltaic power data; Based on the correspondence between historical meteorological observation data and historical photovoltaic power data, the photovoltaic power data in different irradiance intervals are statistically segmented to determine the power change characteristics corresponding to each irradiance interval. Based on the power change characteristics, the nonlinear mapping relationship between irradiance data and photovoltaic power data is fitted with parameters to determine the saturation power calculation rules.

[0010] Optionally, in a sixth implementation of the first aspect of this application, the method further includes: The theoretical saturated power values ​​of the photovoltaic power station at each moment in the historical operating cycle are obtained according to the saturated power calculation rules. The historical photovoltaic power data is compared with the theoretical saturation power value at the corresponding time to obtain the historical power increment dataset. By performing time backtracking on the historical power increment dataset within multiple preset time windows, the power increment change sequence corresponding to different time scales is determined, and multi-time scale power increment feature samples are generated. By performing statistical calculations on the power increment feature samples at multiple time scales, statistical feature rules corresponding to the power increment change characteristics are determined, and these statistical feature rules are stored as power increment feature construction rules.

[0011] A second aspect of this application provides a photovoltaic power prediction device, which is used to implement a photovoltaic power prediction method. The photovoltaic power prediction device includes: The processing module is used to preprocess the real-time photovoltaic power data, meteorological observation data and weather forecast data of the photovoltaic power station at the current moment to obtain the power increment data corresponding to the current moment; The construction module is used to construct multi-dimensional features based on the power increment data and generate a real-time feature set; The generation module is used to determine the predicted power increment corresponding to each prediction step based on the real-time feature set, and superimpose it with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step. The output module is used to output the photovoltaic power prediction result by performing power constraint and smoothing processing on the initial predicted power result.

[0012] A third aspect of this application provides an electronic device, including a memory and a processor, wherein the processor is configured to execute a computer program stored in the memory, and when the processor executes the computer program, it implements the steps in the photovoltaic power prediction method provided in the first aspect of this application.

[0013] The fourth aspect of this application provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the steps in the photovoltaic power prediction method provided in the first aspect of this application.

[0014] In summary, the photovoltaic power prediction method and apparatus provided in this application preprocesses real-time photovoltaic power data, meteorological observation data, and weather forecast data of the photovoltaic power station at the current moment to obtain the power increment data corresponding to the current moment; based on the power increment data, multi-dimensional features are constructed to generate a real-time feature set; the predicted power increment corresponding to each prediction step is determined according to the real-time feature set, and superimposed with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step; by performing power constraint and smoothing processing on the initial predicted power result, the photovoltaic power prediction result is output. This application calculates and superimposes the power increment under multiple prediction steps, and applies physical constraints and time-series verification to the prediction result, which can maintain the continuity and physical rationality of power changes in the multi-timescale prediction process, and enhance the accuracy of the prediction result in actual scheduling scenarios. Attached Figure Description

[0015] Figure 1 A schematic flowchart illustrating the photovoltaic power prediction method provided in the embodiments of this application; Figure 2This is a schematic diagram of the program modules of the photovoltaic power prediction device provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0016] To make the inventive objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] To address the limitation of existing prediction methods in practical scheduling scenarios, this application provides a photovoltaic power prediction method, such as... Figure 1 This is a flowchart illustrating the photovoltaic power prediction method provided in this embodiment. The photovoltaic power prediction method includes the following steps: Step 110: By preprocessing the real-time photovoltaic power data, meteorological observation data and weather forecast data of the photovoltaic power station at the current moment, the power increment data corresponding to the current moment is obtained.

[0018] Specifically, photovoltaic (PV) power prediction requires unified processing of real-time PV power data, meteorological observation data, and weather forecast data from the PV power plant at the current moment. This processing is based on time alignment, mapping data from different sources and with different sampling frequencies to the same time base. Simultaneously, to address unavoidable data gaps and abnormal fluctuations during operation, interpolation and filtering methods are used to form a continuous and effective data set. Building upon this, a saturation power calculation rule based on irradiance information is introduced to calculate the theoretical saturation power corresponding to the current moment. By performing a difference calculation with the real-time PV power, the power increment data characterizing the PV power change state is obtained.

[0019] Step 120: Construct multi-dimensional features based on power increment data to generate a real-time feature set.

[0020] Specifically, after obtaining the power increment data, the photovoltaic power change information is described through multi-dimensional feature construction. This process focuses on the temporal evolution characteristics of the power increment, retrospectively analyzing historical power increments within multiple preset time windows to form lagged power increment expressions at different time scales, thereby characterizing the changes in photovoltaic power in the short and medium term. Simultaneously, statistical calculations are performed on the power increments at each time scale to extract statistics reflecting the level of change and fluctuation characteristics. These statistics are then correlated with theoretical saturation power, real-time photovoltaic power, and time attribute information, ensuring that the resulting features encompass both power change trends and reflect physical operating boundaries and time periodic factors. Ultimately, this constitutes a real-time feature set for prediction calculations.

[0021] Step 130: Determine the predicted power increment corresponding to each prediction step based on the real-time feature set, and overlay it with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step. Specifically, in the prediction calculation stage, based on the real-time feature set and combined with the time span information corresponding to different prediction step sizes, the feature content is organized in a targeted manner, so that each prediction time point corresponds to the corresponding input feature form. On this basis, by calculating the information reflecting power changes in the features, the predicted power increment under different prediction step sizes is determined, and the obtained predicted power increment is superimposed with the real-time photovoltaic power at the current moment, thereby naturally mapping the power change to the power prediction value at the future moment, so that the prediction results under multiple prediction step size conditions maintain a consistent calculation logic and time correlation.

[0022] Step 140: By performing power constraint and smoothing processing on the initial predicted power results, output the photovoltaic power prediction results.

[0023] Specifically, to ensure that the predicted results meet the actual operating conditions of the photovoltaic power plant, physical constraints and time-series verification are introduced into the initial predicted power results. This process sets upper and lower limits on the predicted power based on the installed capacity and operating boundaries of the photovoltaic power plant to prevent the predicted values ​​from exceeding the physically feasible range. Furthermore, a sliding window calculation between adjacent prediction time points is used to smooth the power sequence, reducing unreasonable abrupt changes. Simultaneously, the consistency of the prediction sequence is verified by combining the power variation amplitude constraints between adjacent prediction times, ensuring that the final output photovoltaic power prediction results meet the practical application requirements in terms of both numerical continuity and physical rationality.

[0024] In one optional implementation of this embodiment, the step of preprocessing the real-time photovoltaic power data, meteorological observation data, and weather forecast data of the photovoltaic power station at the current moment to obtain the power increment data corresponding to the current moment includes: acquiring the real-time photovoltaic power data corresponding to the photovoltaic power station at the current moment, and simultaneously acquiring the meteorological observation data and weather forecast data corresponding to the current moment; determining the effective data set by performing missing data interpolation processing and abnormal data screening processing on the real-time photovoltaic power data, meteorological observation data, and weather forecast data; determining the theoretical saturated power value corresponding to the irradiance data of the weather forecast data in the effective data set at the current moment according to the preset saturated power calculation rules; and determining the power increment data corresponding to the current moment by performing difference calculation between the real-time photovoltaic power data and the theoretical saturated power value in the effective data set.

[0025] In this embodiment, during the operation of the photovoltaic power station, real-time photovoltaic power data refers to the actual active power value collected and output by the power station's inverter or energy management system at the current time. This data directly reflects the power generation level of the photovoltaic modules under the current irradiance and operating conditions. Meteorological observation data acquired synchronously at this time refers to environmental information collected by the power station site or nearby meteorological stations, including parameters reflecting power generation conditions such as solar irradiance intensity, ambient temperature, and module temperature. Weather forecast data refers to meteorological prediction information given for a future time range, where irradiance data is used to describe the theoretical level of solar energy input at a given time. Synchronous acquisition means that all types of data have a consistent time identifier in the time dimension, thereby avoiding calculation deviations caused by time misalignment. After data acquisition is completed, the real-time photovoltaic power data, meteorological observation data, and weather forecast data need to undergo quality processing. Missing data refers to numerical gaps caused by communication interruptions or equipment malfunctions during the acquisition process, while abnormal data refers to values ​​that significantly deviate from the normal physical range or operating patterns. Missing data interpolation calculates reasonable replacement values ​​from valid data at adjacent time points, thus maintaining data continuity. Outlier data removal involves judging the magnitude of data changes or physical boundaries to eliminate values ​​that do not conform to operational characteristics from the calculations. The resulting valid data set accurately reflects the current operating status of the photovoltaic power station. After obtaining the valid data set, the theoretical saturation power value is calculated based on irradiance information from weather forecast data. The saturation power calculation rule states that the relationship between irradiance and photovoltaic power is not linear. When irradiance reaches a certain level, the output power of the photovoltaic modules gradually approaches the upper limit determined by the installed capacity and module characteristics; this upper limit is the theoretical saturation power. This calculation relationship is determined through historical operating data, and its function is to map the environmental input of irradiance to the maximum power level that the photovoltaic power station can achieve under ideal conditions. For example, under high irradiance conditions at noon, even if irradiance continues to increase, the output power of the photovoltaic power station will be limited by the rated capacity of the modules and tend to stabilize; this stable value is the theoretical saturation power at the corresponding moment. After determining the theoretical saturation power, the difference between the real-time photovoltaic power in the effective dataset and the theoretical saturation power is calculated to obtain the power increment data at the current moment. The power increment data is used to characterize the degree of deviation of the actual power generation from the theoretical saturation power. Its value reflects both the power fluctuations caused by weather changes and the relationship between the photovoltaic system's operating state and its physical boundaries.

[0026] In an optional implementation of this embodiment, before the step of determining the predicted power increment corresponding to each prediction step based on the real-time feature set and superimposing it with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step, the method further includes: acquiring historical photovoltaic power data corresponding to the photovoltaic power station within its historical operating cycle and historical meteorological observation data corresponding to the timestamps of the historical photovoltaic power data; performing segmented statistics on photovoltaic power data in different irradiance intervals based on the correspondence between the irradiance data of the historical meteorological observation data and the historical photovoltaic power data to determine the power change characteristics corresponding to each irradiance interval; and performing parameter fitting on the nonlinear mapping relationship between the irradiance data and the photovoltaic power data based on the power change characteristics to determine the saturation power calculation rules.

[0027] In this embodiment, before determining the saturation power calculation rules, it is necessary to analyze the data foundation formed by the photovoltaic power station during its historical operating cycle. Historical photovoltaic power data refers to the sequence of actual active power values ​​continuously recorded and stored by the power generation equipment within the past operating time range of the photovoltaic power station. This data reflects the power generation level of the photovoltaic power station under different environmental conditions and operating states. Historical meteorological observation data corresponding to the timestamps of historical photovoltaic power data refers to environmental information recorded by meteorological acquisition equipment at the same time point. Among them, irradiance data is used to describe the input intensity of solar radiation energy per unit area and is a key external factor determining the photovoltaic power generation capacity. The establishment of the timestamp correspondence relationship allows each historical photovoltaic power value to be matched one-to-one with the irradiance conditions at the time of its generation, thereby forming a data pair with physical correlation. After obtaining the historical data with time correspondence, the historical data is divided into intervals by analyzing the correspondence between irradiance data and historical photovoltaic power data. Irradiance intervals refer to grouping continuous irradiance data according to the range of irradiance values. For example, low irradiance, medium irradiance, and high irradiance states are divided into different numerical intervals, and each interval represents the operating state of the photovoltaic power station under similar irradiance conditions. Within each irradiance interval, the corresponding photovoltaic power data is segmented and statistically processed. This processing calculates the distribution, variation range, and central tendency of the power data to extract statistical results that reflect the power variation characteristics with irradiance, thus obtaining the power variation characteristics corresponding to each irradiance interval. The power variation characteristics describe the overall behavioral characteristics of photovoltaic power under given irradiance conditions. Taking a centralized photovoltaic power station with an installed capacity of 50 MW as an example, in historical operating data, when the irradiance is in the range of 300 to 500 watts per square meter, the photovoltaic power shows an approximately linear increase with increasing irradiance; while when the irradiance is above 800 watts per square meter, the power growth rate gradually decreases and tends to stabilize. By statistically analyzing the power data within different irradiance intervals, the variation characteristics of power transitioning from growth to saturation can be clearly identified. This characteristic reflects the output boundary of photovoltaic modules limited by installed capacity and physical characteristics. After obtaining the power variation characteristics corresponding to each irradiance interval, parameter fitting processing is further performed on the nonlinear mapping relationship between irradiance data and photovoltaic power data. A nonlinear mapping relationship refers to a non-linear correspondence between changes in irradiance and changes in photovoltaic power. This relationship manifests as rapid power growth in low irradiance ranges and a gradual approach to the upper limit in high irradiance ranges. Parameter fitting involves adjusting key parameters in the mapping relationship to ensure that the calculated power values ​​remain consistent with historically obtained power change characteristics across various irradiance ranges, thus forming a power change curve that reflects the actual operating characteristics of the photovoltaic power plant. The saturated power calculation rules determined through this process solidify the physical constraints between irradiance input and photovoltaic power output into a calculable relationship.

[0028] In one optional implementation of this embodiment, the step of constructing multi-dimensional features based on power increment data to generate a real-time feature set includes: based on the power increment data at the current moment and the corresponding historical power increment data, by performing time backtracking on the historical power increment data within multiple preset time windows, obtaining the lagged power increment data corresponding to different time scales, and generating power increment features at multiple time scales; by statistically analyzing the power increment features at multiple time scales, determining the power increment statistical features corresponding to each time scale; by performing correlation processing on the theoretical saturation power value, real-time photovoltaic power data, and corresponding time information, generating physical correlation features and time correlation features; and integrating the power increment features, power increment statistical features, physical correlation features, and time correlation features to obtain a real-time feature set.

[0029] In this embodiment, after obtaining the power increment data corresponding to the current moment, it is necessary to combine it with the power increment data formed by the photovoltaic power station during its historical operating cycle to characterize the temporal evolution characteristics of power changes. Time backtracking refers to selecting a continuous historical data interval along the time axis in the past direction, using the current moment as a reference point, to obtain the power increment changes within different time spans. Multiple preset time windows correspond to different time scales; for example, a short time scale is used to reflect instantaneous fluctuation characteristics, while a longer time scale is used to reflect the continuous trend of power changes. By extracting historical power increment data within each time window, it is possible to form lagged power increment data corresponding to different time scales, thereby generating multi-time-scale power increment characteristics. Taking a centralized photovoltaic power station as an example, if the current moment is 12:00, it is possible to backtrack to different moments such as 11:55, 11:45, and 11:30 to obtain the power increment values ​​at the corresponding time points. The short-term data from 11:55 to 12:00 reflects power fluctuations caused by rapid changes in cloud cover, while the longer-term data from 11:30 to 12:00 reflects the overall trend of gradually increasing or decreasing sunlight conditions. This multi-timescale power increment feature construction preserves the complete information about power changes over time. After forming the multi-timescale power increment features, statistical processing is performed on the data corresponding to each timescale to extract statistics that characterize the power change state. The power increment statistical feature is a value obtained by performing centralized and discrete calculations on the power increment data within the same timescale, used to reflect the average level, fluctuation range, and direction of power change. This statistical result compresses the originally scattered time series information into a representative numerical expression, thereby enhancing the stability of the feature in subsequent calculations. For example, if the power increment is continuously negative and gradually decreases in the 30 minutes before 12:00, the statistical result clearly reflects the state of power gradually approaching the theoretical saturation level. Based on the power increment and its statistical features, theoretical saturation power values, real-time photovoltaic power data, and time information are introduced for correlation processing. Physical correlation features are calculated by combining real-time photovoltaic power with theoretical saturation power, reflecting the relative relationship between the current operating state and the physical upper limit. Temporal correlation features are numerical expressions obtained by encoding the time attribute of the current moment, reflecting the impact of intraday solar cycle and seasonal variations on power behavior. For example, in the same photovoltaic power plant, there are significant differences in irradiance conditions and power variation patterns between noon and early morning; introducing time information allows the feature expression to include these periodic differences.After completing the above correlation processing, the power increment features, power increment statistical features, physical correlation features and time correlation features of multiple time scales are integrated into a unified real-time feature set with consistent structure and complementary information. This feature set includes power change amplitude, change trend, physical boundary relationship and time attribute information at the numerical level.

[0030] In one optional implementation of this embodiment, the step of determining the predicted power increment corresponding to each prediction step based on the real-time feature set and superimposing it with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step includes: performing feature recombination processing on the real-time feature set according to the preset time span information corresponding to each prediction step to generate prediction input feature sets corresponding to each prediction step; calculating the power change of the prediction input feature sets corresponding to each prediction step to determine the predicted power increment corresponding to each prediction step; and superimposing the predicted power increment corresponding to each prediction step with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step.

[0031] In this embodiment, after forming the real-time feature set, the feature content needs to be organized in a targeted manner according to the specific requirements of the prediction task for the future time range. The prediction step size refers to the time span extending into the future from the current moment, used to distinguish different prediction time points, such as the photovoltaic power status a few minutes later, a dozen minutes later, or even longer later. The time span information corresponding to the preset prediction step size clarifies the time distance between each prediction time point and the current moment, thus providing constraints on the way features are used. Based on this, feature reorganization processing of the real-time feature set refers to rearranging and selecting time-related content in the feature set according to the time span of the prediction step size, so that it is consistent with the corresponding prediction time point. For example, for shorter prediction time spans, more emphasis is placed on power increment features of short time scales, while for longer prediction time spans, power increment features and their statistical features of longer time scales are introduced. Through this reorganization processing, prediction input feature sets corresponding one-to-one with different prediction step sizes are formed, so that each prediction time point has an input expression that matches its time span characteristics. After obtaining the prediction input feature set corresponding to each prediction step size, the predicted power increment is determined by calculating the power change amount based on the information reflecting the power change state in the features. The power change amount refers to the possible increase or decrease in photovoltaic power relative to the current moment, and its value comes from the comprehensive calculation results of power increment features, statistical features, and physical correlation features included in the prediction input feature set. This calculation process is based on the current power increment state and combines the continuity and volatility of power changes at different time scales to quantitatively describe the degree to which the power deviates from the theoretical saturation power in the future. Taking the same centralized photovoltaic power station as an example, when the power increment shows a continuous decreasing trend in a short time scale, the predicted power increment corresponding to a shorter prediction step size will maintain a similar direction of change, while the predicted power increment corresponding to a longer prediction step size will comprehensively consider the change trend over a longer time scale, thus reflecting the differences between different prediction time points. After determining the predicted power increment corresponding to each prediction step size, this predicted power increment is superimposed with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to the prediction time point. The meaning of superposition is to map the relative change of power increment back to the actual power value space, so that the prediction result maintains the same dimensions and physical meaning as the current operating power of the photovoltaic power station. Taking the current output power of the photovoltaic power station as 40 MW as an example, if the predicted power increment corresponding to a certain prediction step is positive 3 MW, then the initial predicted power result at that prediction time point is 43 MW; if the predicted power increment corresponding to another prediction step is negative 2 MW, then the corresponding initial predicted power result is 38 MW. In this way, the prediction results under different prediction step sizes are all based on the actual operating state at the current moment, ensuring the continuity and consistency between prediction results at multiple time points.Through the above process, a clear correspondence is formed between the predicted input feature set, the predicted power increment, and the initial predicted power result under different prediction time spans, enabling photovoltaic power prediction to be carried out on multiple time scales while maintaining the consistency of power change calculation logic and physical meaning, and providing a stable foundation for the constraint and verification of power prediction results.

[0032] In one optional implementation of this embodiment, the step of outputting photovoltaic power prediction results by performing power constraint and smoothing processing on the initial predicted power results includes: limiting the predicted power values ​​in the initial predicted power results that exceed the preset power range to the corresponding power boundary values ​​according to preset photovoltaic power plant installed capacity parameters and physical operation constraints, thereby generating constrained predicted power results; performing sliding window calculation on the predicted power sequence using the constrained predicted power results between adjacent prediction steps to generate smoothed predicted power results; determining the predicted power change amplitude between adjacent prediction steps based on the smoothed predicted power results, comparing the predicted power change amplitude with a preset change threshold, determining the predicted power values ​​in the predicted power results that meet the change constraint conditions, and generating photovoltaic power prediction results.

[0033] In this embodiment, after obtaining the initial predicted power results corresponding to each prediction step, it is necessary to constrain the prediction results in conjunction with the physical operating boundaries of the photovoltaic power station. The installed capacity parameter of the photovoltaic power station refers to the upper limit of the maximum power generation capacity determined during the design and construction phases of the power station. This parameter is jointly determined by the number of modules, the rated power of a single module, and grid connection conditions, constituting the maximum insurmountable boundary of photovoltaic power at the physical level. The physical operating constraints describe the basic limitations of the photovoltaic power station during operation, ensuring that the power output cannot be lower than zero and cannot exceed the installed capacity. Based on these parameters, the initial predicted power results are range-checked. When the predicted power value exceeds the upper limit of the installed capacity, the value is limited to the power boundary value corresponding to the installed capacity; when the predicted power value is lower than zero, it is limited to zero power, thereby generating a constrained predicted power result that satisfies the physical boundary conditions. Taking a centralized photovoltaic power station with an installed capacity of 50 MW as an example, if the initial predicted power result at a certain prediction time point is 52 MW, then after boundary constraints, the predicted power is corrected to 50 MW. After defining the physical boundaries, the constrained predicted power results need to be smoothed over time to mitigate potential unreasonable fluctuations in the prediction sequence. Sliding window calculation involves selecting a continuous window of multiple adjacent prediction time points within the predicted power sequence, jointly calculating the predicted power values ​​within the window, and generating a new power value to replace the original prediction. The purpose of the sliding window is to introduce information from adjacent time points, making the predicted power show a more continuous trend over time. For example, in the same photovoltaic power plant, if the power at a certain prediction time point is significantly higher than at previous and subsequent time points, the sliding window calculation can adjust this abrupt change to a level more consistent with adjacent prediction results. The smoothed predicted power results obtained after this process exhibit stronger continuity and stability over time. After obtaining the smoothed predicted power results, the power changes between adjacent prediction step sizes are further verified. The magnitude of the predicted power change refers to the difference in power values ​​between two adjacent prediction time points, reflecting the rate of change of the predicted power over a short period. A preset change threshold is used to limit the maximum allowable power change range between adjacent prediction time points. This threshold is determined based on the actual level of power change during the historical operation of the photovoltaic power plant, and is used to constrain drastic changes that do not conform to the physical operating characteristics. When the change in the smoothed predicted power result between adjacent prediction time points exceeds the threshold, the corresponding predicted power value is corrected to bring the change back within the threshold range. Taking a photovoltaic power plant with the same installed capacity of 50 MW as an example, if the predicted power jumps from 30 MW to 45 MW between two adjacent prediction time points, and this change exceeds the preset threshold, the predicted power at the later time point is adjusted so that the change between it and the previous time point meets the constraint requirements.By sequentially applying physical boundary constraints, time series smoothing, and change magnitude verification to the predicted power results, a photovoltaic power prediction result that meets the actual operating characteristics of a photovoltaic power plant in terms of numerical range, time continuity, and rate of change is finally obtained, enabling the prediction result to directly correspond to the power change state under the actual operating scenario.

[0034] In one optional implementation of this embodiment, the theoretical saturated power value corresponding to each moment in the historical operating cycle of the photovoltaic power station is obtained according to the saturated power calculation rules; the difference between the historical photovoltaic power data and the theoretical saturated power value at the corresponding moment is calculated to obtain the historical power increment dataset; by performing time backtracking on the historical power increment dataset within multiple preset time windows, the power increment change sequence corresponding to different time scales is determined, and multi-time scale power increment feature samples are generated; by performing statistical calculations on the multi-time scale power increment feature samples, the statistical feature rules corresponding to the power increment change characteristics are determined, and the statistical feature rules are stored as power increment feature construction rules.

[0035] In this embodiment, within the historical operating cycle of the photovoltaic power station, the meteorological irradiance conditions corresponding to each historical moment are first calculated according to the established saturation power calculation rules, thereby obtaining the theoretical saturation power value corresponding to each historical timestamp. The theoretical saturation power value refers to the upper limit of power that photovoltaic modules can achieve under given irradiance levels and inherent installed capacity conditions of the power station, without being affected by random disturbances, and is used to characterize the physical boundary characteristics of photovoltaic power generation. This calculation process uses timestamps as indexes, so that historical photovoltaic power data and theoretical saturation power values ​​form a strict correspondence in the time dimension. After obtaining the theoretical saturation power value, the difference between the historical photovoltaic power data and the theoretical saturation power value at the corresponding moment is calculated, and the result is the historical power increment data. The historical power increment data corresponding to all moments are integrated to obtain the historical power increment dataset. Based on the formed historical power increment dataset, the data is processed by time backtracking within multiple preset time windows. A time window is a fixed-length time interval extending in the historical direction, used to extract the power change trajectory at different time scales. By setting short, medium, and long time windows, power increment sequences are extracted within their respective time ranges, resulting in multiple sets of power increment change sequences reflecting the characteristics of changes at different time scales. Short-time-scale sequences emphasize rapid power fluctuations, while longer-time-scale sequences reflect the overall trend of power changes. After obtaining the power increment change sequences at multiple time scales, statistical calculations are performed on each sequence to extract statistical feature rules that stably characterize power change features. Statistical features are feature quantities obtained through numerical calculations on the power increment sequences, used to describe attributes such as the magnitude of change, the degree of concentration of change, and the symmetry of change. By performing statistical calculations on the power increment sequences at different time scales, a set of feature description rules corresponding to each time scale can be obtained, enabling the quantitative expression of the behavioral characteristics of power increments across different time ranges. Finally, these statistical feature rules are organized and stored to form directly callable power increment feature construction rules, ensuring that feature generation during the online phase remains consistent with historical operational characteristics.

[0036] It should be noted that, based on the completion of historical power increment data, real-time feature sets, and prediction step size division, the construction and optimization of the multi-step increment calculation structure revolves around the evolution law of power increment under different prediction time spans. Multi-step refers to establishing corresponding power increment calculation relationships for multiple future prediction time points, enabling the power change characteristics at short and long time scales to be characterized separately. The training process uses power increment data formed within the historical operating cycle as input. For different prediction step sizes, the historical power increment data is time-aligned according to the corresponding time span, so that each prediction step size corresponds to a complete set of historical power increment sequences, thus forming multiple sets of increment samples corresponding to the prediction time span. Based on this, the multi-time-scale power increment features, power increment statistical features, and physical correlation features constructed earlier are introduced and combined according to the prediction step size to form the increment calculation input structure corresponding to each prediction step size. This input structure jointly characterizes the magnitude and direction of power increment changes at different time scales, as well as their relationship with theoretical saturation power, enabling the calculation results to reflect the changing trend of photovoltaic power in different future time spans. Taking a centralized photovoltaic power plant as an example, for a 5-minute prediction step, the input structure focuses more on the correspondence between short-term irradiance changes and rapid power fluctuations; for a 30-minute prediction step, it incorporates more statistical characteristics of power increments within a longer time window to describe the overall trend of power changes. After the incremental calculation structure is formed, its parameters are adjusted to gradually converge the deviation between the calculated results and historical actual power increments. This process corrects the difference between historical prediction results and actual power increments by providing feedback, allowing the parameter configuration to gradually align with the power change characteristics under different prediction step lengths. The deviation can be continuously adjusted to keep the difference within a stable range for each prediction step length. To avoid mutual interference between different prediction step lengths, the parameter configurations for each prediction step length remain independent, thus ensuring consistency and stability in the incremental change characterization between short-term and medium-to-long-term predictions. During the optimization process, upper and lower power limits and variation amplitude constraints are also introduced to verify the incremental calculation results, ensuring that the calculation structure conforms to the physical operating characteristics of the photovoltaic power plant during the training phase. For example, in the historical data of the same photovoltaic power plant, when the power increment corresponding to a certain prediction step causes the predicted power to exceed the installed capacity, the increment result is rolled back to the allowable range and used as the basis for correction in subsequent parameter adjustments. By repeatedly rolling back historical data and correcting parameters, the multi-step power increment calculation structure gradually forms a stable mapping relationship, enabling it to output power increment results that conform to physical constraints and are consistent for different prediction time spans in the actual prediction process.

[0037] According to the photovoltaic power prediction method provided in this application, the real-time photovoltaic power data, meteorological observation data, and weather forecast data of the photovoltaic power station at the current moment are preprocessed to obtain the power increment data corresponding to the current moment; multi-dimensional features are constructed based on the power increment data to generate a real-time feature set; the predicted power increment corresponding to each prediction step is determined according to the real-time feature set, and is superimposed with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step; by performing power constraint and smoothing processing on the initial predicted power result, the photovoltaic power prediction result is output. This application calculates and superimposes the power increment under multiple prediction steps, and applies physical constraints and time-series verification to the prediction result, which can maintain the continuity and physical rationality of power changes in the multi-timescale prediction process, and enhance the accuracy of the prediction result in actual scheduling scenarios.

[0038] Figure 2 A photovoltaic power prediction device is provided as an embodiment of this application. This photovoltaic power prediction device can be used to implement the photovoltaic power prediction method in the foregoing embodiments. Figure 2 As shown, the photovoltaic power prediction device mainly includes: Processing module 10 is used to preprocess the real-time photovoltaic power data, meteorological observation data and weather forecast data of the photovoltaic power station at the current moment to obtain the power increment data corresponding to the current moment; Module 20 is used to construct multi-dimensional features based on power increment data and generate a real-time feature set; The generation module 30 is used to determine the predicted power increment corresponding to each prediction step based on the real-time feature set, and superimpose it with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step. The output module 40 is used to output the photovoltaic power prediction result by performing power constraint and smoothing processing on the initial predicted power result.

[0039] In one optional implementation of this embodiment, the processing module is specifically used to: acquire real-time photovoltaic power data corresponding to the current moment of the photovoltaic power station, and simultaneously acquire meteorological observation data and weather forecast data corresponding to the current moment; determine the effective data set by performing missing data interpolation processing and abnormal data screening processing on the real-time photovoltaic power data, meteorological observation data, and weather forecast data; determine the theoretical saturated power value corresponding to the irradiance data of the weather forecast data in the effective data set at the current moment according to the preset saturated power calculation rules; and determine the power increment data corresponding to the current moment by performing difference calculation between the real-time photovoltaic power data and the theoretical saturated power value in the effective data set.

[0040] In one optional implementation of this embodiment, the construction module is specifically used to: based on the current power increment data and the corresponding historical power increment data, by performing time backtracking on the historical power increment data within multiple preset time windows, obtain the lagged power increment data corresponding to different time scales, and generate power increment features at multiple time scales; by statistically analyzing the power increment features at multiple time scales, determine the power increment statistical features corresponding to each time scale; by performing correlation processing on the theoretical saturation power value, real-time photovoltaic power data, and corresponding time information, generate physical correlation features and time correlation features; and integrate the power increment features, power increment statistical features, physical correlation features, and time correlation features to obtain a real-time feature set.

[0041] In one optional implementation of this embodiment, the generation module is specifically used to: perform feature recombination processing on the real-time feature set according to the preset time span information corresponding to each prediction step, and generate prediction input feature sets corresponding to each prediction step; calculate the power change amount of the prediction input feature sets corresponding to each prediction step, and determine the prediction power increment corresponding to each prediction step; and perform superposition calculation between the prediction power increment corresponding to each prediction step and the real-time photovoltaic power data at the current moment to generate the initial prediction power result corresponding to each prediction step.

[0042] In one optional implementation of this embodiment, the output module is specifically used to: limit the predicted power values ​​that exceed the preset power range in the initial predicted power results to the corresponding power boundary values ​​according to the preset photovoltaic power station installed capacity parameters and physical operation constraints, and generate constrained predicted power results; perform sliding window calculation on the predicted power sequence through the constrained predicted power results between adjacent prediction steps to generate smoothed predicted power results; determine the predicted power change amplitude between adjacent prediction steps based on the smoothed predicted power results, compare the predicted power change amplitude with the preset change threshold, determine the predicted power values ​​in the predicted power results that meet the change constraint conditions, and generate photovoltaic power prediction results.

[0043] In an optional implementation of this embodiment, the generation module is further configured to: acquire historical photovoltaic power data corresponding to the photovoltaic power station within its historical operating cycle and historical meteorological observation data corresponding to the timestamps of the historical photovoltaic power data; based on the correspondence between the irradiance data of the historical meteorological observation data and the historical photovoltaic power data, segment and statistically analyze the photovoltaic power data in different irradiance intervals to determine the power change characteristics corresponding to each irradiance interval; and perform parameter fitting on the nonlinear mapping relationship between the irradiance data and the photovoltaic power data based on the power change characteristics to determine the saturation power calculation rules.

[0044] In an optional embodiment of this example, the generation module is further configured to: obtain the theoretical saturated power value corresponding to each moment in the historical operating cycle of the photovoltaic power station according to the saturated power calculation rules; perform difference calculation between the historical photovoltaic power data and the theoretical saturated power value at the corresponding moment to obtain a historical power increment dataset; determine the power increment change sequence corresponding to different time scales by performing time backtracking on the historical power increment dataset within multiple preset time windows, and generate power increment feature samples at multiple time scales; determine the statistical feature rules of the corresponding power increment change characteristics by performing statistical calculation on the power increment feature samples at multiple time scales, and store the statistical feature rules as power increment feature construction rules.

[0045] According to the photovoltaic power prediction device provided in this application, the real-time photovoltaic power data, meteorological observation data, and weather forecast data of the photovoltaic power station at the current moment are preprocessed to obtain the power increment data corresponding to the current moment; multi-dimensional features are constructed based on the power increment data to generate a real-time feature set; the predicted power increment corresponding to each prediction step is determined according to the real-time feature set, and superimposed with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step; by performing power constraint and smoothing processing on the initial predicted power result, the photovoltaic power prediction result is output. This application calculates and superimposes the power increment under multiple prediction steps, and applies physical constraints and time-series verification to the prediction result, which can maintain the continuity and physical rationality of power changes in the multi-timescale prediction process, and enhance the accuracy of the prediction result in actual scheduling scenarios.

[0046] According to the scheme provided in this application Figure 3 An electronic device is provided as an embodiment of this application. This electronic device can be used to implement the photovoltaic power prediction method in the foregoing embodiments, and mainly includes: The system includes a memory 301, a processor 302, and a computer program 303 stored in the memory 301 and executable on the processor 302. The memory 301 and the processor 302 are connected via communication. When the processor 302 executes the computer program 303, it implements the photovoltaic power prediction method described in the foregoing embodiments. The number of processors can be one or more.

[0047] The memory 301 can be a high-speed random access memory (RAM) or a non-volatile memory, such as a disk storage device. The memory 301 is used to store executable program code, and the processor 302 is coupled to the memory 301.

[0048] Furthermore, embodiments of this application also provide a computer-readable storage medium, which may be disposed in the electronic device described in the above embodiments, and the computer-readable storage medium may be as described above. Figure 3 The memory in the illustrated embodiment.

[0049] The computer-readable storage medium stores a computer program that, when executed by a processor, implements the photovoltaic power prediction method described in the foregoing embodiments. Furthermore, the computer-readable storage medium can also be a USB flash drive, external hard drive, read-only memory (ROM), RAM, magnetic disk, or optical disk, or any other medium capable of storing program code.

[0050] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0051] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0052] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A photovoltaic power prediction method, characterized in that, include: By preprocessing the real-time photovoltaic power data, meteorological observation data, and weather forecast data of the photovoltaic power station at the current moment, the power increment data corresponding to the current moment is obtained. Specifically, this includes: acquiring real-time photovoltaic power data corresponding to the current moment of the photovoltaic power station, and simultaneously acquiring meteorological observation data and weather forecast data corresponding to the current moment; determining the effective data set by performing missing data interpolation and abnormal data filtering on the real-time photovoltaic power data, meteorological observation data, and weather forecast data; and determining the theoretical saturated power value corresponding to the irradiance data of the weather forecast data in the effective data set at the current moment according to a preset saturated power calculation rule. The saturated power calculation rule refers to the non-linear growth relationship between irradiance and photovoltaic power. When the irradiance reaches a certain level, the output power of the photovoltaic module will gradually approach the upper limit determined by the installed capacity and module characteristics. This upper limit is the theoretical saturated power value. The process of constructing the saturated power calculation rule is as follows: Historical photovoltaic power data corresponding to the photovoltaic power station within its historical operating cycle and historical meteorological observation data corresponding to the timestamps of the historical photovoltaic power data are obtained; based on the correspondence between the irradiance data and the historical photovoltaic power data in the historical meteorological observation data, the photovoltaic power data in different irradiance intervals are segmented and statistically analyzed to determine the power change characteristics corresponding to each irradiance interval; based on the power change characteristics, the nonlinear mapping relationship between the irradiance data and the photovoltaic power data is fitted with parameters to determine the saturated power calculation rule; by performing a difference calculation between the real-time photovoltaic power data of the effective data set and the theoretical saturated power value, the power increment data corresponding to the current moment is determined. Based on the power increment data, multi-dimensional features are constructed to generate a real-time feature set. Specifically, this includes: based on the current power increment data and the corresponding historical power increment data, by performing time backtracking on the historical power increment data within multiple preset time windows, obtaining the lagged power increment data corresponding to different time scales, and generating multi-time-scale power increment features; by statistically analyzing the multi-time-scale power increment features, determining the power increment statistical features corresponding to each time scale; by performing correlation processing on the theoretical saturation power value, the real-time photovoltaic power data, and the corresponding time information, generating physical correlation features and time correlation features; and integrating the power increment features, power increment statistical features, physical correlation features, and time correlation features to obtain the real-time feature set. The predicted power increment corresponding to each prediction step is determined based on the real-time feature set, and then superimposed with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step. Specifically, this includes: performing feature recombination processing on the real-time feature set according to the preset time span information corresponding to each prediction step, generating prediction input feature sets corresponding to each prediction step; calculating the power change of the prediction input feature sets corresponding to each prediction step to determine the predicted power increment corresponding to each prediction step; and superimposing the predicted power increment corresponding to each prediction step with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step. By performing power constraint and smoothing processing on the initial predicted power results, the photovoltaic power prediction results are output.

2. The photovoltaic power prediction method according to claim 1, characterized in that, The step of outputting photovoltaic power prediction results by performing power constraint and smoothing processing on the initial predicted power results includes: Based on the preset photovoltaic power plant installed capacity parameters and physical operation constraints, the predicted power values ​​that exceed the preset power range in the initial predicted power results are limited to the corresponding power boundary values ​​to generate constrained predicted power results. The predicted power result after the constraint is calculated by a sliding window between adjacent prediction steps, and the predicted power sequence is smoothed to generate a smoothed predicted power result. The predicted power change amplitude between adjacent prediction steps is determined based on the smoothed predicted power result, and the predicted power change amplitude is compared with a preset change threshold to determine the predicted power value that meets the change constraint condition in the predicted power result, thereby generating the photovoltaic power prediction result.

3. The photovoltaic power prediction method according to claim 1, characterized in that, The method further includes: The theoretical saturated power values ​​of the photovoltaic power station at each moment in the historical operating cycle are obtained according to the saturated power calculation rules. The historical photovoltaic power data is compared with the theoretical saturation power value at the corresponding time to obtain the historical power increment dataset. By performing time backtracking on the historical power increment dataset within multiple preset time windows, the power increment change sequence corresponding to different time scales is determined, and multi-time scale power increment feature samples are generated. By performing statistical calculations on the power increment feature samples at multiple time scales, statistical feature rules corresponding to the power increment change characteristics are determined, and these statistical feature rules are stored as power increment feature construction rules.

4. A photovoltaic power prediction device, characterized in that, The photovoltaic power prediction device is used to implement the photovoltaic power prediction method according to claim 1, and the photovoltaic power prediction device includes: The processing module is used to preprocess the real-time photovoltaic power data, meteorological observation data and weather forecast data of the photovoltaic power station at the current moment to obtain the power increment data corresponding to the current moment; The construction module is used to construct multi-dimensional features based on the power increment data and generate a real-time feature set; The generation module is used to determine the predicted power increment corresponding to each prediction step based on the real-time feature set, and superimpose it with the real-time photovoltaic power data at the current moment to generate the initial predicted power result corresponding to each prediction step. The output module is used to output the photovoltaic power prediction result by performing power constraint and smoothing processing on the initial predicted power result.

5. An electronic device, characterized in that, Includes memory and processor, of which: The processor is used to execute computer programs stored in the memory; When the processor executes the computer program, it implements the steps in the photovoltaic power prediction method according to any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps in the photovoltaic power prediction method according to any one of claims 1 to 3.

Images