Power prediction methods, systems, devices, media, and products

Abstract

The application discloses a power prediction method, system, device, medium and product, comprising: comparing the current meteorological characteristics of the environment where the target object is located with a plurality of historical daily meteorological characteristics, and screening out target meteorological characteristics with similarity not less than a preset threshold; determining a historical similar day corresponding to the target meteorological characteristics, and constructing a similar day boundary power curve based on the power generation of the historical similar day; constructing a clear sky boundary power curve based on the historical power generation of the target object; generating a dynamic calibration coefficient according to the deviation of the actual power generation of the target object in a first period before the current time from the power value of the corresponding time in the similar day boundary power curve; and determining the predicted power of the target object in a second period after the current time based on the dynamic calibration coefficient, the similar day boundary power curve and the clear sky boundary power curve. The application improves the prediction accuracy through similar day matching and dynamic calibration.

Description

Technical Field

[0001] This application belongs to the field of photovoltaic power generation technology, and in particular relates to a power prediction method, system, equipment, medium and product. Background Technology

[0002] In ultra-short-term photovoltaic power forecasting, physical parameterization schemes or statistical machine learning methods are typically employed to predict future power output on a minute-by-minute basis by establishing the time-series dependence of historical power on meteorological variables. Within this framework, the prediction accuracy is directly limited by the irradiance prediction bias and the model's generalization ability.

[0003] In related technologies, ultra-short-term power prediction is mostly based on meteorological elements provided by numerical weather prediction combined with historical power data, and future power output is extrapolated through time series models or neural networks. However, atmospheric systems have chaotic characteristics, and cloud formation and dissipation are difficult to capture accurately on a minute-scale scale. The simplification assumptions of parameterization schemes lead to systematic biases in irradiance prediction. At the same time, machine learning models are prone to overfitting under conditions of few samples, making it difficult to adapt to the needs of accurate ultra-short-term prediction under the constraint of sparse data. Summary of the Invention

[0004] In view of the above, this application provides a power prediction method, system, device, medium, and product to address some of the deficiencies mentioned in the background art.

[0005] In a first aspect, embodiments of this application provide a power prediction method, including: The current meteorological characteristics of the environment where the target object is located are compared with the meteorological characteristics of multiple historical days, and the target meteorological characteristics with a similarity of not less than a preset threshold are selected. Determine the historical similar days corresponding to the target meteorological characteristics, and construct the boundary power curve of the similar days based on the power generation of the historical similar days; Construct a clear-sky boundary power curve based on the historical power generation of the target object; Based on the deviation between the actual power generation of the target object in the first time period before the current time and the power value at the corresponding time in the similar daily boundary power curve, a dynamic calibration coefficient is generated. Based on the dynamic calibration coefficients, the similar day boundary power curves, and the clear sky boundary power curves, the predicted power of the target object in the second time period after the current moment is determined.

[0006] In a second aspect, embodiments of this application provide a power prediction system, comprising: The feature filtering module is configured to: compare the current meteorological features of the environment where the target object is located with the meteorological features of multiple historical days, and filter out the target meteorological features with a similarity of not less than a preset threshold. The similar day boundary power curve determination module is configured to: determine the historical similar days corresponding to the target meteorological feature, and construct the similar day boundary power curve based on the power generation of the historical similar days; The clear-sky boundary power curve determination module is configured to generate dynamic calibration coefficients based on the deviation between the actual power generation of the target object in the first time period before the current time and the power value at the corresponding time in the similar daily boundary power curve. The calibration coefficient determination module is configured to generate dynamic calibration coefficients based on the deviation between the actual power generation of the target object in the first time period before the current time and the power value at the corresponding time in the similar daily boundary power curve. The power prediction module is configured to determine the predicted power of the target object in the second time period after the current time, based on the dynamic calibration coefficient, the similar day boundary power curve, and the clear sky boundary power curve.

[0007] In a third aspect, this application provides a computer device including one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and executed by the one or more processors, and the one or more programs include instructions for performing the method described in the first aspect.

[0008] In a fourth aspect, this application provides a non-volatile computer-readable storage medium containing a computer program that, when executed by one or more processors, causes the one or more processors to perform the method described in the first aspect.

[0009] In a fifth aspect, this application provides a computer program product comprising one or more computer programs that, when executed by one or more processors, implement the method described in the first aspect.

[0010] The power prediction method, system, device, medium, and product provided in this disclosure compare the current meteorological characteristics of the target object's environment with the meteorological characteristics of multiple historical days to screen out similar days and construct similar day boundary power curves. At the same time, a clear sky boundary power curve is constructed based on historical power generation. Then, a dynamic calibration coefficient is generated by using the deviation between the actual power generation in the first time period before the current time and the similar day boundary power curve. Finally, the predicted power for the second time period is determined based on the calibration coefficient, the similar day boundary power curve, and the clear sky boundary power curve. This can effectively suppress the systematic bias of irradiance prediction, avoid the overfitting problem of machine learning models under few sample conditions, and improve the accuracy and numerical stability of ultra-short-term power prediction. Attached Figure Description

[0011] Figure 1A flowchart illustrating a power prediction method provided in an embodiment of this application is shown; Figure 2 A schematic diagram illustrating a process for determining similar dates provided in an embodiment of this application is shown; Figure 3 A schematic diagram of the structure of a power prediction system provided in an embodiment of this application is shown; Figure 4 A more specific device hardware structure diagram provided in an embodiment of this application is shown. Detailed Implementation

[0012] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0013] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0015] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0016] The term "in response to" as used herein refers to a state in which a corresponding event occurs or a condition is satisfied. It will be understood that the timing of subsequent actions performed in response to such event or condition is not necessarily strongly correlated with the time when the event occurs or the condition is met. For example, in some cases, subsequent actions may be performed immediately upon the occurrence of the event or the fulfillment of the condition; while in others, they may be performed some time after the occurrence of the event or the fulfillment of the condition.

[0017] For clarity, before describing the specific technical solutions of the embodiments of this application, the technical terms involved in the embodiments of this application will be explained first.

[0018] Similar Day: Refers to a historical day that has similar meteorological characteristics to the predicted day. In this embodiment, the similar day is one or more historical days whose similarity meets preset conditions by comparing the current meteorological characteristics of the target object's environment with the meteorological characteristics of multiple historical days. The corresponding historical power generation is used to construct the boundary power curve of the similar day.

[0019] Boundary Power Curve: This refers to a power curve obtained by sorting the power generation at various sampling times throughout the day on multiple historical days, taking the non-first value after sorting by each time moment, and then smoothing it. In the embodiments of this application, the boundary power curve includes a clear-sky boundary power curve and a similar-day boundary power curve. The former is used to characterize the power envelope under clear-sky conditions, and the latter is used to characterize the power envelope under similar-day conditions.

[0020] Dynamic Calibration Coefficient: This refers to a correction coefficient generated based on the deviation between the actual power generation during the first time period prior to the current time and the corresponding power value in the similar daily boundary power curve. In the embodiments of this application, the dynamic calibration coefficient is used to correct the power value of the similar daily boundary power curve in the future prediction period.

[0021] Meteorological Feature Vector: A vector composed of multiple meteorological factors. In this embodiment, the meteorological feature vector includes factors in four dimensions: maximum irradiance, maximum temperature, minimum temperature, and average temperature, used to characterize the meteorological state of the environment in which the target object is located.

[0022] Euclidean distance: refers to the straight-line distance between two points in Euclidean space. In this embodiment, the Euclidean distance is used to measure the similarity between the current meteorological feature vector and the historical daily meteorological feature vector. The smaller the Euclidean distance, the more similar the two meteorological features are.

[0023] Limiting / Clamping: refers to a processing method that restricts a value to a specified range. In the embodiments of this application, limiting refers to restricting the corrected predicted power value to between 0 and the power value at the corresponding moment on the clear sky boundary power curve.

[0024] As described in the background section, the relevant technologies have many problems in the photovoltaic ultra-short-term power prediction scenario.

[0025] First, the prediction methods based on physical parameterization schemes in related technologies rely on parameterized models such as cloud microphysical processes and aerosol radiation effects. However, the atmospheric system has chaotic characteristics, and the formation, dissipation, movement speed, and thickness changes of clouds are difficult to capture accurately on a minute-scale. The simplification assumptions of parameterization schemes lead to systematic biases in irradiance prediction. The prediction error of solar radiation increases significantly under cloudy weather, which directly affects the baseline accuracy of ultra-short-term power prediction.

[0026] Secondly, the prediction methods based on statistical machine learning in related technologies rely on a large amount of high-quality historical power data to establish time-series dependencies. However, in actual business, there is a general problem of sample scarcity. Under the condition of few samples, the model's ability to adapt to local micro-meteorological conditions is greatly reduced. It is prone to overfitting or under-smoothing at rapidly fluctuating "mutation points", making it difficult to support robust modeling of cloudy or alternating cloudy and sunny scenarios, and its generalization ability is severely limited.

[0027] Furthermore, the relevant technologies lack an effective pre-classification mechanism for weather types. The statistical distribution of prediction errors varies significantly under different weather types. Conventional statistical methods and lightweight neural networks both adopt a unified modeling strategy to deal with all weather scenarios, which cannot adaptively adjust the prediction logic according to the differences in weather types. This results in inconsistent prediction accuracy under scenarios such as clear skies, cloudy skies, and rainy skies, and the overall error is difficult to control within an acceptable range.

[0028] Furthermore, while some related technologies employ prediction approaches based on screening historical similar days, they often use a single similar day or a simple multi-day average method, directly outputting the power of similar days as the predicted value. This lacks a dynamic calibration process to assess future power output fluctuations and makes it difficult to respond to the need for power output correction based on real-time weather changes on the predicted day. At the same time, the prediction results lack an effective physical constraint quality control mechanism, and the predicted power is prone to extreme fluctuations under extreme weather conditions, which is inconsistent with the actual physical laws of photovoltaic power output and further reduces the engineering reliability of the prediction results.

[0029] Therefore, how to overcome the systematic bias of physical parameterization schemes and the generalization bottleneck of statistical machine learning methods under the constraint of data sparsity, achieve adaptive forecasting under different weather types, and introduce dynamic calibration and physical constraint mechanisms to improve the consistency and reliability of forecast results has become a technical problem that urgently needs to be solved in this field.

[0030] Figure 1 A schematic flowchart of a power prediction method provided in an embodiment of this application is shown. Figure 1 As shown, the power prediction method in this disclosure further includes the following steps: S102: Compare the current meteorological characteristics of the environment where the target object is located with the meteorological characteristics of multiple historical days, and select the target meteorological characteristics with a similarity of not less than a preset threshold.

[0031] In some optional embodiments, the target object is a photovoltaic power plant. Current meteorological characteristics include the maximum irradiance, maximum temperature, minimum temperature, and average temperature for the current day; historical daily meteorological characteristics include the maximum irradiance, maximum temperature, minimum temperature, and average temperature for each day in the historical timeframe. The reason for selecting these four meteorological factors is that the maximum irradiance determines the upper limit potential of photovoltaic output; the maximum temperature reflects the most unfavorable photoelectric conversion efficiency conditions for the day; the minimum temperature reflects the baseline ambient temperature during the early morning hours of the day; and the average temperature comprehensively characterizes the average impact of the daily thermal environment on module efficiency. These four factors together constitute a representative set of factors describing the meteorological conditions of a day.

[0032] Specifically, a current meteorological feature vector is generated based on current meteorological characteristics, and multiple historical meteorological feature vectors are generated based on meteorological characteristics from multiple historical days. The current meteorological feature vector can be represented as:

[0033] Where Rmax represents the maximum irradiance, Tmax represents the maximum temperature, Tmin represents the minimum temperature, and Tmean represents the average temperature.

[0034] Multiple historical daily meteorological feature vectors can be represented as:

[0035] Calculate the Euclidean distance between the current meteorological feature vector and the meteorological feature vectors of each historical day. The Euclidean distance is calculated as the square root of the sum of the squares of the differences between the corresponding components; mathematically, it represents the true distance between two vectors in four-dimensional space. A smaller Euclidean distance indicates that the two meteorological feature vectors are closer in four-dimensional space, meaning the weather type of the historical day is more similar to the predicted day. Since Euclidean distance and similarity are negatively correlated (smaller distance indicates greater similarity, and larger distance indicates less similarity), the similarity between the current meteorological feature and the meteorological features of each historical day can be determined based on the Euclidean distance. Historical day meteorological features with a similarity greater than or equal to a preset threshold are selected as target meteorological features—in other words, historical days with sufficiently small Euclidean distances are selected as samples similar to the weather type of the predicted day.

[0036] Figure 2 This illustration shows a schematic diagram of a process for determining similar dates according to an embodiment of this application. Figure 2 As shown, the Euclidean distance is calculated between the current meteorological feature vector and the meteorological feature vector of each day in the past 15 days. After sorting the distances in ascending order, the historical days corresponding to the top 5 distances are selected as similar days. The consideration for choosing 15 days as the historical search range is that, in terms of time span, 15 days can cover samples of various weather types, while avoiding the inclusion of samples from too distant dates that differ too much from the current seasonal characteristics. Of course, those skilled in the art can choose other numbers of historical days as the search range according to the needs of actual application scenarios, and this application does not limit this.

[0037] The historical days corresponding to the selected target meteorological characteristics are the samples that are closest to the predicted day in terms of meteorological characteristics. The power generation data of these historical days reflects the actual output level of photovoltaic power plants under this weather type and can serve as an important reference benchmark for predicting future power on the predicted day.

[0038] S104: Determine the historical similar days corresponding to the target meteorological characteristics, and construct the boundary power curve of the similar days based on the power generation of the historical similar days.

[0039] The target meteorological features selected through the above steps correspond to certain historical days, which are "similar days" with weather types similar to the predicted day. This step first identifies the historical days corresponding to these target meteorological features and designates them as historical similar days. There are usually multiple historical similar days (e.g., the 5 similar days selected in the previous step).

[0040] Because photovoltaic power output varies throughout the day—increasing at sunrise, peaking at noon, and decreasing at sunset—a single power value only reflects the instantaneous output at that moment and cannot represent the overall output pattern for the entire day. For example, a similar day might have higher power at 10 AM but lower power at 3 PM, while the predicted day might still maintain high output at 3 PM. Relying solely on the similarity of a single moment for prediction would inevitably lead to significant deviations. Therefore, a power curve covering all sampling times throughout the day is needed to comprehensively characterize the overall output pattern of similar days. Furthermore, power values ​​may differ between multiple similar days at the same time—some similar days might have higher output at that moment, while others might have lower output. Simply including all similar days' power is insufficient; a selection rule representing the "typical" output level under that weather type is required. Many related technologies use direct averaging, but averages are easily affected by extreme values ​​(abnormally high or low values ​​on individual similar days due to localized meteorological changes), causing the benchmark value to deviate from the typical level.

[0041] To address the aforementioned issues, this application employs a sorted boundary value method to construct the boundary power curve. Specifically, it acquires the power generation at various sampling times throughout the day on multiple historically similar days. The sampling times throughout the day can be 96 points, i.e., one sampling point every 15 minutes. The reason for using a 15-minute sampling interval is that the time granularity of ultra-short-term forecasts is typically 15 minutes, which matches the dispatch cycle of the power grid dispatch system, while also capturing the impact of minute-level meteorological changes on power output.

[0042] For each sampling time, the power generation of multiple historical similar days at that sampling time is sorted, and the boundary power value for that sampling time is selected based on the sorting result. In some optional embodiments, multiple power values ​​can be sorted from largest to smallest, and the second value is taken as the boundary power value for that time. The reason for taking the second value instead of the maximum value after sorting is that the maximum value may come from an extreme power value generated by local meteorological conditions (such as the cloud gap effect causing a sudden increase in irradiance) on a similar day. This value does not represent the typical output level of this type of weather. If the maximum value is taken as the boundary, it will lead to an overly high baseline and generally overestimated subsequent forecast values; taking the minimum value will lead to an overly low baseline and generally underestimated forecast values. Taking the second value (i.e., the second largest value) after sorting can eliminate the interference of the maximum extreme value and retain the relatively high output level under this type of weather, resulting in a more representative boundary power value.

[0043] The initial similar daily boundary power sequence generated from the boundary power values ​​at each sampling time can be expressed as:

[0044] Because the boundary power values ​​are selected independently at each sampling time, unnatural jumps may occur between adjacent times (for example, the boundary value taken at a certain time comes from similar day A, while the boundary value taken at an adjacent time comes from similar day B, and there may be a lack of temporal continuity between the two). Such jumps do not conform to the physical law that photovoltaic power output changes continuously over time. Therefore, the initial similar day boundary power sequence is smoothed to obtain a smooth and continuous similar day boundary power curve.

[0045] In some optional embodiments, the smoothing process employs a local polynomial fitting method: for each point in the sequence, a window is formed with 7 points to its left and 7 points to its right (a total of 15 points), and a second-order polynomial is used for local fitting, replacing the original value of that point with the fitted value. The selection of a window length of 15 and an order of 2 is based on the following: with a 15-minute sampling interval, 15 points cover a time span of 3.75 hours, which is sufficient to capture the macroscopic trend of the photovoltaic power output curve while effectively filtering out random fluctuations on short time scales; the second-order polynomial can fit the parabolic change trend of the power output curve within this window, which conforms to the physical characteristics of smooth changes in photovoltaic power output within a local time period. The polynomial coefficients are determined by least-squares fitting and are the same for all windows. The smoothed similar daily boundary power curve can be expressed as:

[0046] This curve represents the typical power output pattern of photovoltaic power stations under this type of weather, and serves as a baseline reference line for subsequent forecasts.

[0047] S106: Construct a clear-sky boundary power curve based on the historical power generation of the target object.

[0048] The similar-day boundary power curve reflects the typical power output pattern of photovoltaic (PV) power plants under similar weather conditions as the forecast day. However, this curve is always generated based on actual power data from historical similar days, and its value is limited by the actual weather conditions on that day. When the weather improves on the forecast day (e.g., less cloud cover than on similar days), the actual power output on the forecast day may exceed the level indicated by the similar-day boundary power curve; conversely, when the weather worsens on the forecast day, the actual power output may be lower than the curve. In other words, the similar-day boundary power curve itself cannot characterize the maximum possible power output of a PV power plant under the ideal condition of "clear skies and no clouds," while the maximum power output under clear skies is an indispensable physical upper limit reference value in PV power output prediction.

[0049] In related technologies, prediction models directly output power values, lacking effective physical constraint mechanisms. Under extreme weather conditions, the predicted power is prone to extreme fluctuations—for example, during the transition from cloudy to sunny weather, the predicted power may jump from a low value to a high value in a short period, even exceeding the rated installed capacity of the photovoltaic modules, which is impossible in actual physical processes. To address this issue, this application constructs an independent clear-sky boundary power curve to characterize the maximum output envelope of the photovoltaic power station under clear-sky conditions, providing a physical upper limit constraint for the prediction results.

[0050] The core idea of ​​the clear-sky boundary power curve is that, over a historical period (e.g., the last 15 days), a photovoltaic power plant must have experienced several clear-sky days. The power generation on each of these clear-sky days represents the maximum output level of the plant under different seasons and aerosol conditions. By taking boundary values ​​from the power values ​​at the same moment on multiple clear-sky days and smoothing them, an envelope curve representing the maximum output potential of the plant under clear-sky conditions in the current season can be obtained. This curve is independent of the similar day boundary power curve, does not depend on the weather type on the predicted day, and always reflects the upper limit of the plant's physical installed capacity and the upper limit of its photovoltaic conversion capability.

[0051] Specifically, the power generation of the target object is obtained at various sampling times throughout the day on multiple historical days within a preset historical period. The preset historical period can be the most recent 15 days, which reflects the current season's sunlight conditions and temperature levels while avoiding the introduction of seasonal biases caused by changes in solar altitude angle from distant dates. For each sampling time, the power generation of multiple historical days at that sampling time is sorted, and the boundary power value for that sampling time is selected based on the sorting result. In some optional embodiments, multiple power values ​​can be sorted from largest to smallest, and the second value is taken as the boundary power value for that time. An initial clear-sky boundary power sequence is generated from the boundary power values ​​at each sampling time, which can be expressed as:

[0052] Then, the initial clear-sky boundary power sequence is smoothed to obtain the clear-sky boundary power curve. The smoothing method is the same as described above, using a local polynomial fitting method (15 points in the window, order 2). The smoothed clear-sky boundary power curve can be represented as:

[0053] This curve represents the maximum possible output envelope of a photovoltaic (PV) power plant under clear-sky conditions in the current season. The actual output under any weather conditions cannot exceed this envelope; this is the most fundamental physical constraint on PV output—that is, the actual power cannot exceed the theoretical maximum value under clear-sky conditions. This physical constraint will be used as a limiting upper limit in subsequent prediction steps.

[0054] Thus far, this application has constructed two boundary power curves: the similar-day boundary power curve serves as a typical power output benchmark under the same weather type, and the clear-sky boundary power curve serves as a physical upper limit constraint. However, the similar-day boundary power curve is a static benchmark generated based on historical similar-day data. It reflects the typical power output pattern throughout the similar day, but there will inevitably be a certain deviation between the actual weather conditions on the prediction date and historical similar days—it could be that more cloud cover leads to lower actual power output than similar days, or it could be that improved weather leads to higher actual power output than similar days. If the similar-day boundary power curve is directly used as the basis for future predictions without any correction, this deviation will be transmitted to the prediction results, reducing prediction accuracy.

[0055] S108: Generate dynamic calibration coefficients based on the deviation between the actual power generation of the target object in the first time period before the current time and the power value at the corresponding time in the similar daily boundary power curve.

[0056] In related technologies, the lack of a dynamic calibration step for future power output fluctuations—that is, comparing the static forecast baseline with the real-time monitored actual power and dynamically adjusting the future forecast value based on the deviation—makes it impossible to respond to the need for power output correction due to real-time weather changes on the forecast day. For example, if cloud cover is higher than normal at 9:00 AM on the forecast day, resulting in actual power output lower than the same period on similar days, without any correction, the forecast value at 10:00 AM will still be based on the uncorrected boundary power curve of similar days, remaining consistently higher, until the deviation reverses when the actual weather improves at some point in the future, but at this time the forecast value may be lower. This deviation fluctuates repeatedly throughout the forecast day, seriously affecting forecast accuracy. To address this problem, this application introduces a dynamic calibration coefficient mechanism.

[0057] The core idea of ​​the dynamic calibration coefficient mechanism is to compare the actual power generation of the photovoltaic power station with the power value at the corresponding moment in the similar daily boundary power curve within the most recent period before the current moment, and calculate the degree of deviation between the two. Since weather changes are continuous and gradually variable (cloud movement and irradiance changes are gradual rather than abrupt on a minute-scale scale), it is reasonable to assume that this deviation will remain relatively stable in the short term (e.g., the next hour). Therefore, this deviation can be used as the calibration basis for future predictions—using the deviation at the current moment to correct the predicted value at future moments.

[0058] Specifically, the actual power generation of the target object is obtained at multiple sampling times within the first time period to form an actual power sequence. The first time period refers to a time window preceding the current time. Its selection should consider two factors: the time window should not be too short (otherwise the sample size is insufficient and random fluctuations will dominate the deviation calculation), nor should it be too long (otherwise outdated meteorological information will be introduced, failing to reflect the current real-time weather conditions). In some optional embodiments, the first time period is selected as four time points: 15 minutes, 30 minutes, 45 minutes, and 60 minutes before the current time. This time window covers the power output change trend within the most recent hour, containing a sufficient sample size while also responding promptly to minute-level meteorological changes.

[0059] The actual power sequence can be represented as:

[0060] The power values ​​corresponding to each sampling time within the first time period (i.e., the power values ​​at the same time) are obtained from the power curves of similar daily boundaries to form a reference power sequence:

[0061] Based on the comparison between the actual power sequence and the reference power sequence, dynamic calibration coefficients are generated. This application uses the mean ratio as a deviation measure—calculating the ratio of the mean of the actual power sequence to the mean of the reference power sequence:

[0062] The reason for choosing the mean ratio instead of the point-by-point ratio is that the point-by-point ratio introduces random fluctuation noise at each moment, while the mean ratio, by averaging the values ​​at multiple moments, effectively smooths out random fluctuations on a short time scale, resulting in a more robust and reliable deviation measure. When the ratio is greater than 1, it indicates that the current actual output is higher than the typical output level of the same period on similar days, i.e., the weather is favorable for the day; when the ratio is less than 1, it indicates that the current actual output is lower than the typical output level of the same period on similar days, i.e., the weather is unfavorable for the day.

[0063] In some optional embodiments, the ratio is limited to a preset numerical range. The selection of the preset range should follow the physical laws of photovoltaic power output: the actual power cannot be negative, so the lower limit of the coefficient is 0; the actual power also cannot exceed the maximum output under clear sky conditions, so the upper limit of the coefficient is 1 (when the ratio is greater than 1, it means that the current actual output has exceeded the level of the same period on a similar day, but the excess needs to be constrained by the clear sky boundary, rather than reflected by amplifying the calibration coefficient). Forcibly limiting the ratio to between 0 and 1 as a dynamic calibration coefficient can avoid coefficient distortion caused by abnormal data.

[0064] By calculating the aforementioned dynamic calibration coefficient, this application achieves a quantitative perception of the deviation between real-time weather changes on the forecast date and similar days. This coefficient is continuously updated as the current time progresses—each time a new current time is reached, the calibration coefficient can be recalculated using the latest actual power data prior to that time—achieving real-time tracking of weather changes.

[0065] S110: Based on the dynamic calibration coefficient, the similar day boundary power curve, and the clear sky boundary power curve, determine the predicted power of the target object in the second time period after the current moment.

[0066] Thus far, this application has obtained three key elements: the power curve of the similar day boundary (characterizing the typical power output benchmark under similar weather conditions), the power curve of the clear sky boundary (characterizing the physical upper limit constraint), and the dynamic calibration coefficient (characterizing the real-time deviation of the current weather relative to similar days). By synergistically integrating these three elements, the predicted power for future periods can be determined.

[0067] It should be noted that these three elements each carry different information dimensions: the similar daily boundary power curve provides baseline information on "what is the typical power output under this type of weather"; the dynamic calibration coefficient provides correction information on "whether today's power output is stronger or weaker than this type of weather"; and the clear-sky boundary power curve provides constraint information on "how much power output cannot be exceeded under any circumstances." All three are indispensable—without the similar daily boundary power curve, prediction lacks a baseline; without the dynamic calibration coefficient, prediction cannot respond to real-time weather changes; and without the clear-sky boundary power curve, prediction may exceed physical limits. The synergistic integration of these three elements constitutes the core technical solution of this application.

[0068] Specifically, the power values ​​of the similar diurnal boundary power curves in the second time period are corrected using dynamic calibration coefficients to obtain preliminary power predictions. The second time period refers to the future prediction period after the current moment. Its time span should be consistent with the first time period (e.g., four time points: 15 minutes, 30 minutes, 45 minutes, and 60 minutes in the future) to ensure the symmetry of the assumption that "the deviation remains stable in the future" on the time scale—since the deviation of the most recent hour is used to calibrate the future, predicting the next hour is reasonable; beyond one hour, the deviation may change significantly due to weather changes.

[0069] In some optional embodiments, for each predicted time t1, t2, t3, t4 within the future second time period, the preliminary predicted power is the product of the dynamic calibration coefficient and the corresponding time value of the power curve of the similar day boundary: Pt1′=coffQY mpt1 Pt2′=coffQY mpt2 Pt3′=coffQY mpt3 Pt4′=coffQY mpt4 Since the dynamic calibration coefficient is between 0 and 1, the preliminary predicted power is reduced proportionally (compared to the original similar day boundary power curve), but will not exceed the original value. This means that when the actual power output is higher than the level of the same period on similar days (ratio greater than 1), the coefficient is set to 1, and the preliminary predicted power equals the value of the similar day boundary power curve; when the actual power output is lower than the level of the same period on similar days (ratio less than 1), the coefficient is set to a value less than 1, and the preliminary predicted power is reduced proportionally based on the value of the similar day boundary power curve. This unidirectional reduction calibration logic conforms to engineering practice—the similar day boundary power curve itself represents a relatively high power output level under this type of weather (taking the second largest value rather than the average value), and in most cases, the actual power output on the predicted day is unlikely to exceed this benchmark; if the weather does improve and the actual power output exceeds the level of the same period on similar days, then the subsequent clear sky boundary limit is used to ensure that the predicted value does not exceed the physical upper limit, rather than reflecting this by amplifying the calibration coefficient.

[0070] Then, the preliminary predicted power is limited using the clear-sky boundary power curve, and the limited power value is used as the predicted power for the second time period. The limiting process is based on the fundamental physical laws of photovoltaic power generation: under any weather conditions, the actual power generation of a photovoltaic power station can never be lower than 0 (in shutdown mode), nor can it exceed the theoretical maximum output under clear-sky conditions (limited by the rated capacity of the photovoltaic modules and the physical upper limit of incident irradiance). For each prediction time, the final predicted power is: ; ; ;

[0071] The final predicted power is limited to between 0 and the corresponding clear-sky boundary power value at that time. When the initial predicted power is less than 0, it is taken as 0; when the initial predicted power is greater than the clear-sky boundary value, the clear-sky boundary value is taken; and when it is in between, the initial predicted power itself is taken. This process ensures that the prediction results always conform to the physical laws of photovoltaic power output and avoids the problem of predicted values ​​jumping out of bounds under extreme weather conditions.

[0072] Compared with related technologies, this application has the following advantages: First, this application achieves automatic pre-classification of weather types by comparing the current meteorological characteristics of the target object's environment with the meteorological characteristics of multiple historical days, thus filtering out similar days. This effectively solves the problem of inconsistent prediction accuracy caused by the lack of a weather type classification mechanism and the use of a uniform model to deal with all weather scenarios in related technologies. Through the similarity measurement of multi-dimensional meteorological feature vectors and Euclidean distance, the predicted day is accurately classified into the weather type of the historical day with the most similar meteorological characteristics, laying the foundation for subsequent differentiated predictions for different weather types.

[0073] Secondly, this application effectively solves the problems of lack of physical constraints in prediction results and the tendency for predicted power to jump excessively under extreme weather conditions in related technologies by constructing similar daily boundary power curves and clear-sky boundary power curves. The similar daily boundary power curve extracts typical power output patterns under similar weather conditions through sorting, boundary value extraction, and smoothing, providing a benchmark reference for prediction under similar weather conditions. The clear-sky boundary power curve characterizes the maximum power output envelope of photovoltaic power plants under clear-sky conditions, providing a physical upper limit constraint for prediction. The two curves work together to ensure that the prediction results always conform to the physical laws of photovoltaic power output, eliminating the problem of predicted values ​​exceeding the physical upper limit and rendering the predictions unusable for engineering applications.

[0074] Third, this application employs a dynamic calibration coefficient mechanism. It generates calibration coefficients by utilizing the deviation between the actual power generation during the first time period prior to the current moment and the corresponding power value in the boundary power curve of a similar day. This effectively solves the problem in related technologies where static prediction benchmarks cannot respond to real-time weather changes on the prediction day. This mechanism quantifies the degree of deviation of the current day's weather relative to similar days by comparing the actual power of the most recent period with the benchmark of a similar day in real time. This deviation is then used as a correction factor to dynamically adjust future prediction values, enabling the prediction logic to adaptively correct itself according to real-time weather changes on that day. This avoids the decrease in accuracy caused by the deviation between the static benchmark and the actual power on that day being transmitted to the prediction results.

[0075] Fourth, this application integrates similar day screening, dynamic calibration coefficient correction, and clear sky boundary limiting to form a step-by-step prediction framework of "weather classification first, dynamic calibration second, and physical constraints third". Compared with the scheme of direct prediction by a single model in related technologies, this application plays a role in three dimensions: weather type pre-classification, real-time deviation correction, and physical constraint guarantee. It effectively suppresses the systematic deviation of irradiance prediction, avoids the overfitting problem of machine learning models under few sample conditions, and can still achieve accurate prediction under data sparsity constraints.

[0076] Fifth, this application constructs the boundary power curve by smoothing after sorting and taking the second largest value, and calculates the dynamic calibration coefficient by means ratio. Each step is designed with an outlier suppression mechanism (sorting and taking the second largest value to exclude extreme power interference, means ratio smoothing random fluctuations, and coefficient forced amplitude limiting to avoid distortion). Comprehensive quality control is carried out at the source data level to ensure that no outliers enter the core calculation stage of prediction, further improving the robustness of the prediction program under all weather conditions.

[0077] In summary, this application provides a photovoltaic ultra-short-term power prediction method suitable for small sample conditions. Through the synergistic mechanism of similar day screening, dynamic calibration and physical constraints, it effectively improves the accuracy, stability and engineering usability of ultra-short-term power prediction, and has significant engineering application value.

[0078] Based on the same inventive concept, and corresponding to the methods of any of the above embodiments, this application also provides a power prediction system.

[0079] Figure 3 A schematic diagram of the structure of a power prediction system provided in an embodiment of this application is shown.

[0080] refer to Figure 3 As shown, the power prediction system includes: The feature filtering module 202 is configured to compare the current meteorological features of the environment where the target object is located with the meteorological features of multiple historical days, and filter out the target meteorological features with a similarity of not less than a preset threshold.

[0081] In some optional embodiments, the feature filtering module 202 further performs the following steps: A current meteorological feature vector is generated based on the current meteorological features, and multiple historical day meteorological feature vectors are generated based on multiple historical day meteorological features. Calculate the Euclidean distances between the current meteorological feature vector and the multiple historical daily meteorological feature vectors, respectively. The similarity between the current meteorological feature and multiple historical daily meteorological features is determined based on the Euclidean distance; wherein the Euclidean distance is negatively correlated with the similarity. Historical daily meteorological features with a similarity greater than or equal to a preset threshold are used as target meteorological features.

[0082] The similar day boundary power curve determination module 204 is configured to: determine the historical similar days corresponding to the target meteorological characteristics, and construct the similar day boundary power curve based on the power generation of the historical similar days.

[0083] The construction of the similar day boundary power curve based on the power generation of the historical similar days includes: Obtain the power generation of multiple historically similar days at various sampling times throughout the day; For each sampling time, the power generation of multiple historical similar days at that sampling time is sorted, and the boundary power value of that sampling time is selected according to the sorting result. A similar daily boundary power sequence is generated from the boundary power values ​​at each sampling time, and the similar daily boundary power sequence is smoothed to obtain the similar daily boundary power curve.

[0084] The clear-sky boundary power curve determination module 206 is configured to generate dynamic calibration coefficients based on the deviation between the actual power generation of the target object in the first time period before the current time and the power value at the corresponding time in the similar day boundary power curve.

[0085] In some optional embodiments, the clear-sky boundary power curve determination module 206 further performs the following steps: Obtain the power generation of the target object at various sampling times throughout the day on multiple historical days within a preset historical period; For each sampling time, the power generation of multiple historical days at that sampling time is sorted, and the boundary power value of that sampling time is selected according to the sorting result. A clear-sky boundary power sequence is generated from the boundary power values ​​at each sampling time, and the clear-sky boundary power sequence is smoothed to obtain the clear-sky boundary power curve.

[0086] The calibration coefficient determination module 208 is configured to generate dynamic calibration coefficients based on the deviation between the actual power generation of the target object in the first time period before the current time and the power value at the corresponding time in the similar daily boundary power curve.

[0087] In some optional embodiments, the calibration coefficient determination module 208 further performs the following steps: The actual power generation of the target object at multiple sampling times within the first time period is obtained to form an actual power sequence; The power values ​​corresponding to each sampling time within the first time period are obtained from the similar daily boundary power curves to form a reference power sequence; The dynamic calibration coefficients are generated based on the comparison between the actual power sequence and the reference power sequence.

[0088] The power prediction module 210 is configured to determine the predicted power of the target object in the second time period after the current time, based on the dynamic calibration coefficient, the similar day boundary power curve and the clear sky boundary power curve.

[0089] In some optional embodiments, the power prediction module 210 further performs the following steps: The power values ​​of the similar day boundary power curves in the second time period are corrected using the dynamic calibration coefficients to obtain preliminary predicted power; The preliminary predicted power is limited using the clear-sky boundary power curve, and the limited power value is used as the predicted power for the second time period.

[0090] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.

[0091] The apparatus of the above embodiments can be used to implement the corresponding power prediction method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0092] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the power prediction method described in any of the above embodiments.

[0093] Figure 4 This illustration shows a more specific hardware structure diagram of a computer device provided in this embodiment. The device may include: a processor 810, a memory 820, an input / output interface 830, a communication interface 840, and a bus 850. The processor 810, memory 820, input / output interface 830, and communication interface 840 are interconnected internally via the bus 850.

[0094] The processor 810 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the power prediction method provided in any of the foregoing embodiments of this specification.

[0095] The memory 820 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 820 can store operating devices and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 820 and is called and executed by the processor 810.

[0096] The input / output interface 830 is used to connect input / output modules to enable information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0097] The communication interface 840 is used to connect the communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0098] Bus 850 includes a pathway for transmitting information between various components of the device, such as processor 810, memory 820, input / output interface 830, and communication interface 840.

[0099] It should be noted that although the above-described device only shows the processor 810, memory 820, input / output interface 830, communication interface 840, and bus 850, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0100] The electronic devices described above are used to implement the corresponding power prediction methods in any of the foregoing embodiments and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0101] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this disclosure also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the power prediction method as described in any of the above embodiments.

[0102] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0103] The aforementioned non-transitory computer-readable storage media can be any available medium or data storage device that a computer can access, including but not limited to magnetic storage (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MOs), etc.), optical storage (e.g., CDs, DVDs, BDs, HVDs, etc.), and semiconductor storage (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs)).

[0104] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the power prediction method as described in any of the embodiments in the exemplary method section above, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0105] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.

[0106] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0107] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0108] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A power prediction method, characterized in that, include: The current meteorological characteristics of the environment where the target object is located are compared with the meteorological characteristics of multiple historical days, and the target meteorological characteristics with a similarity of not less than a preset threshold are selected. Determine the historical similar days corresponding to the target meteorological characteristics, and construct the boundary power curve of the similar days based on the power generation of the historical similar days; Construct a clear-sky boundary power curve based on the historical power generation of the target object; Based on the deviation between the actual power generation of the target object in the first time period before the current time and the power value at the corresponding time in the similar daily boundary power curve, a dynamic calibration coefficient is generated. Based on the dynamic calibration coefficients, the similar day boundary power curves, and the clear sky boundary power curves, the predicted power of the target object in the second time period after the current moment is determined.

2. The power prediction method according to claim 1, characterized in that, The step of comparing the current meteorological characteristics of the target object's environment with the similarity characteristics of multiple historical days, and selecting target meteorological characteristics with a similarity not less than a preset threshold, includes: A current meteorological feature vector is generated based on the current meteorological features, and multiple historical day meteorological feature vectors are generated based on multiple historical day meteorological features. Calculate the Euclidean distances between the current meteorological feature vector and the multiple historical daily meteorological feature vectors, respectively. The similarity between the current meteorological feature and multiple historical daily meteorological features is determined based on the Euclidean distance; wherein the Euclidean distance is negatively correlated with the similarity. Historical daily meteorological features with a similarity greater than or equal to a preset threshold are used as target meteorological features.

3. The power prediction method according to claim 1, characterized in that, The historical similar days include multiple ones; The construction of the similar day boundary power curve based on the power generation of the historical similar days includes: Obtain the power generation of multiple historically similar days at various sampling times throughout the day; For each sampling time, the power generation of multiple historical similar days at that sampling time is sorted, and the boundary power value of that sampling time is selected according to the sorting result. A similar daily boundary power sequence is generated from the boundary power values ​​at each sampling time, and the similar daily boundary power sequence is smoothed to obtain the similar daily boundary power curve.

4. The power prediction method according to claim 1, characterized in that, The construction of the clear-sky boundary power curve based on the historical power generation of the target object includes: Obtain the power generation of the target object at various sampling times throughout the day on multiple historical days within a preset historical period; For each sampling time, the power generation of multiple historical days at that sampling time is sorted, and the boundary power value of that sampling time is selected according to the sorting result. A clear-sky boundary power sequence is generated from the boundary power values ​​at each sampling time, and the clear-sky boundary power sequence is smoothed to obtain the clear-sky boundary power curve.

5. The power prediction method according to claim 1, characterized in that, The step of generating dynamic calibration coefficients based on the deviation between the actual power generation of the target object in the first time period before the current time and the corresponding power value in the similar daily boundary power curve includes: The actual power generation of the target object at multiple sampling times within the first time period is obtained to form an actual power sequence; The power values ​​corresponding to each sampling time within the first time period are obtained from the similar daily boundary power curves to form a reference power sequence; The dynamic calibration coefficients are generated based on the comparison between the actual power sequence and the reference power sequence.

6. The power prediction method according to claim 1, characterized in that, Determining the predicted power of the target object in the second time period after the current moment includes: The power values ​​of the similar day boundary power curves in the second time period are corrected using the dynamic calibration coefficients to obtain preliminary predicted power; The preliminary predicted power is limited using the clear-sky boundary power curve, and the limited power value is used as the predicted power for the second time period.

7. A power prediction system, characterized in that, include: The feature filtering module is configured to: compare the current meteorological features of the environment where the target object is located with the meteorological features of multiple historical days, and filter out the target meteorological features with a similarity of not less than a preset threshold. The similar day boundary power curve determination module is configured to: determine the historical similar days corresponding to the target meteorological feature, and construct the similar day boundary power curve based on the power generation of the historical similar days; The clear-sky boundary power curve determination module is configured to generate dynamic calibration coefficients based on the deviation between the actual power generation of the target object in the first time period before the current time and the power value at the corresponding time in the similar daily boundary power curve. The calibration coefficient determination module is configured to generate dynamic calibration coefficients based on the deviation between the actual power generation of the target object in the first time period before the current time and the power value at the corresponding time in the similar daily boundary power curve. The power prediction module is configured to determine the predicted power of the target object in the second time period after the current time, based on the dynamic calibration coefficient, the similar day boundary power curve, and the clear sky boundary power curve.

8. A computer device, characterized in that, It includes one or more processors, memory; and one or more programs, wherein the one or more programs are stored in the memory and executed by the one or more processors, and the one or more programs include instructions for performing the method of any one of claims 1 to 6.

9. A non-volatile computer-readable storage medium containing a computer program, characterized in that, When the computer program is executed by one or more processors, the one or more processors perform the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes one or more computer programs that, when executed by one or more processors, implement the method as described in any one of claims 1 to 6.

Images