A method and system for judging the applicability of photovoltaic power prediction models

By calculating the fractal dimension and rate of change of the photovoltaic power generation model, the applicability of the model is determined, which solves the problem that it is difficult to determine the applicability of photovoltaic power generation models under extreme weather conditions in the existing technology, and realizes a fast and accurate model applicability determination.

CN123115677APending Publication Date: 2026-11-17GUANGXI POWER GRID CORP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611344283.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-01
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing photovoltaic power generation prediction models are difficult to determine their applicability under extreme weather conditions, leading to a decline in prediction performance. Furthermore, existing solutions are unable to detect changes in power signals caused by equipment aging or sensor drift in a timely manner.

Method used

By calculating the fractal dimension of historical measured power time series data in the training set, the baseline interval of the fractal dimension is determined, and the fractal dimension of the current window is calculated in real time. Combined with the rate of change and trend of the fractal dimension, the amplitude deviation criterion and trend criterion are triggered to determine the applicability of the model.

Benefits of technology

It enables rapid determination of the applicability of photovoltaic power generation models without waiting for actual power measurements during the prediction period, taking into account the amplitude deviation and trend drift of the power signal, thereby improving the accuracy of applicability determination of the prediction model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115677A_ABST
    Figure CN123115677A_ABST
Patent Text Reader

Abstract

This application discloses a method and system for judging the applicability of a photovoltaic power prediction model. The method involves acquiring historical measured power time-series data of the target photovoltaic power plant in the training set of the power prediction model to be judged, calculating the fractal dimension of each sliding window, and determining the fractal dimension benchmark interval based on the quantiles of the fractal dimension sequence in the training set; real-time acquisition of the current power signal, calculating the fractal dimension of the current window according to a consistent fractal dimension calculation method and window length, calculating the rate of change of the fractal dimension based on the fractal dimension of the current window at multiple consecutive time points, and determining the trend of change; triggering the amplitude deviation criterion and trend criterion respectively, and obtaining the applicability judgment result based on the triggering result, thereby realizing online applicability judgment of the power prediction model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of photovoltaic power generation prediction technology, and more specifically, to a method and system for judging the applicability of photovoltaic power prediction models. Background Technology

[0002] Photovoltaic power generation forecasting typically establishes a mapping relationship between meteorological elements and actual power based on historical data, predicting photovoltaic power generation based on meteorological data such as wind speed, irradiance, temperature, and humidity. The training data for power prediction models generally only covers a limited range of weather types; when actual operating conditions exceed the coverage of the training data, the predictive performance of the power prediction model may decline.

[0003] Existing solutions mainly include traditional machine learning methods, deep learning methods, and physics-based methods. Traditional machine learning methods use models such as extreme gradient boosting and random forests to extract shallow features between weather forecast data and actual power data. Deep learning methods use models such as bidirectional long short-term memory networks (LSTM) and self-attention-based LSTM networks to extract nonlinear relationships between weather forecast data and between weather forecast data and actual power data. Physics-based methods establish mathematical models based on the physical principles of photovoltaic power generation for power prediction. The training data for these solutions typically consists mainly of clear and partly cloudy weather data, with insufficient coverage of extreme weather events such as frontal processes.

[0004] Existing power prediction model solutions typically calculate prediction errors based on actual power or monitor whether current meteorological data deviates from the model's training set distribution. The former method requires actual power data before error statistics can be performed, resulting in a response lag; the latter method only focuses on changes in the distribution of meteorological input, making it difficult to detect changes in the statistical characteristics of power signals caused by factors such as equipment aging, component degradation, or sensor drift. Moreover, changes in meteorological data distribution do not necessarily lead to increased prediction deviations. Existing solutions struggle to promptly determine whether a power prediction model is suitable for the current operating conditions.

[0005] Therefore, this application provides a method and system for judging the applicability of photovoltaic power prediction models to solve one of the above-mentioned technical problems. Summary of the Invention

[0006] The purpose of this application is to provide a method and system for judging the applicability of photovoltaic power prediction models, which can solve at least one of the technical problems mentioned above. The specific solution is as follows: According to a specific embodiment of this application, in a first aspect, this application provides a method for determining the applicability of a photovoltaic power prediction model, including: Historical measured power time-series data of the target photovoltaic power plant in the training set of the power prediction model to be judged are acquired. The fractal dimension of the historical measured power time-series data within each sliding window is calculated to obtain the fractal dimension sequence of the training set. The fractal dimension reference interval is determined according to the quantile of the fractal dimension sequence of the training set. The current power signal of the target photovoltaic power plant is collected in real time to obtain the current window signal. The fractal dimension of the current window is calculated based on the fractal dimension calculation method and window length consistent with the calculation of the fractal dimension sequence of the training set. The fractal dimension change rate is calculated and the change trend is determined based on the fractal dimension of the current window at multiple recent consecutive moments. If the fractal dimension of the current window exceeds the fractal dimension reference interval adjusted by a preset relaxation coefficient, the amplitude deviation criterion is triggered. If the fractal dimension change rate exceeds a preset change rate threshold and the change trend remains in the same direction, the trend criterion is triggered. Based on the triggering results of the amplitude deviation criterion and the trend criterion, the applicability judgment result of the power prediction model to be judged is obtained.

[0007] According to a specific embodiment of this application, in a second aspect, this application provides a photovoltaic power prediction model applicability judgment system, comprising: The baseline interval determination module acquires historical measured power time-series data of the target photovoltaic power station in the training set of the power prediction model to be judged, calculates the fractal dimension of the historical measured power time-series data within each sliding window, obtains the training set fractal dimension sequence, and determines the fractal dimension baseline interval based on the quantiles of the training set fractal dimension sequence; the current window fractal dimension calculation module collects the current power signal of the target photovoltaic power station in real time, obtains the current window signal, and calculates the current window fractal dimension based on the fractal dimension calculation method and window length consistent with those of the baseline interval determination module in calculating the training set fractal dimension sequence; the trend determination module, based on... The fractal dimension of the current window is calculated based on multiple consecutive time points, and the rate of change of the fractal dimension is determined to identify the trend. The applicability judgment module receives the fractal dimension baseline interval, the current window fractal dimension, the rate of change of the fractal dimension, and the trend. If the current window fractal dimension exceeds the fractal dimension baseline interval adjusted by a preset relaxation coefficient, an amplitude deviation criterion is triggered. If the rate of change of the fractal dimension exceeds a preset rate of change threshold and the trend remains in the same direction, a trend criterion is triggered. Based on the triggering results of the amplitude deviation criterion and the trend criterion, the applicability judgment result of the power prediction model to be judged is obtained.

[0008] Compared with the prior art, the above-described solutions of this application have at least the following beneficial effects: This application provides a method for judging the applicability of a photovoltaic power prediction model. By calculating the fractal dimension of historical measured power time-series data in the training set within each sliding window, and determining the fractal dimension benchmark interval based on the quantiles of the fractal dimension sequence in the training set, the method can characterize the power signal fluctuation range covered by the training set of the power prediction model to be judged. The fractal dimension of the current window is calculated according to a consistent fractal dimension calculation method and window length, making the current power signal comparable to the power signal in the training set. By combining the interval deviation of the fractal dimension of the current window with the rate of change and trend of the fractal dimension over multiple consecutive time periods, the method can take into account both the amplitude deviation and trend drift of the power signal. Therefore, the applicability judgment result of the power prediction model to be judged can be obtained without waiting for the measured power of the prediction period to be calculated for error statistics. Attached Figure Description

[0009] Figure 1 A flowchart of a method for judging the applicability of a photovoltaic power prediction model; Figure 2 The flowchart shows the method for acquiring and preprocessing historical measured power time-series data. Figure 3 A flowchart of the method for constructing the fractal dimension sequence of the training set; Figure 4 A flowchart of the window signal normalization method; Figure 5 Flowchart of the method for determining the reference interval of fractal dimension; Figure 6 A flowchart for constructing the current window signal; Figure 7 A flowchart illustrating the method for determining the rate and trend of change of fractal dimension; Figure 8 A flowchart of the fractal dimension change rate threshold determination method; Figure 9 The flowchart shows the comprehensive method for determining the applicability of a model. Figure 10 This is a structural diagram of a photovoltaic power prediction model applicability discrimination system. Detailed Implementation

[0010] 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. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the application. The singular forms “a,” “said,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0012] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0013] It should be understood that although the terms first, second, third, etc., may be used in the embodiments of this application, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, first may also be referred to as second without departing from the scope of the embodiments of this application, and similarly, second may also be referred to as first.

[0014] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”

[0015] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a product or system comprising a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a product or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the product or system that includes said element.

[0016] It should be noted that any symbols and / or numbers present in the specification that are not marked in the accompanying drawings are not reference numerals.

[0017] The optional embodiments of this application are described in detail below with reference to the accompanying drawings.

[0018] The embodiments provided in this application are embodiments of a photovoltaic power prediction model applicability determination method.

[0019] The following is combined Figure 1 The embodiments of this application will be described in detail.

[0020] Figure 1 This is a flowchart of a method for judging the applicability of photovoltaic power prediction models, referencing... Figure 1 The method includes the following steps: S101. Obtain the historical measured power time series data of the target photovoltaic power station in the training set of the power prediction model to be judged, calculate the fractal dimension of the historical measured power time series data in each sliding window, obtain the fractal dimension sequence of the training set, and determine the fractal dimension benchmark interval based on the quantile of the fractal dimension sequence of the training set.

[0021] S102. Real-time acquisition of the current power signal of the target photovoltaic power station to obtain the current window signal, and calculation of the current window fractal dimension based on the fractal dimension calculation method and window length consistent with the fractal dimension sequence of the training set.

[0022] S103. Based on the fractal dimension of the current window at multiple recent consecutive moments, calculate the rate of change of the fractal dimension and determine the trend of change.

[0023] S104a. If the fractal dimension of the current window exceeds the fractal dimension reference range adjusted by the preset relaxation coefficient, the amplitude deviation criterion is triggered.

[0024] S104b If the rate of change of the fractal dimension exceeds the preset rate of change threshold and the trend of change remains in the same direction, then the trend criterion is triggered.

[0025] S105. Based on the triggering results of the amplitude deviation criterion and the trend criterion, the applicability determination result of the power prediction model to be determined is obtained.

[0026] The embodiments of this application are used to determine whether the power prediction model to be judged is applicable to the current operating conditions of the target photovoltaic power station.

[0027] For example, the training set of the power prediction model to be determined contains measured power time-series data collected by the target photovoltaic power station within a historical time period. When the weather types covered by the training set are limited, the power signal under the current operating conditions may exhibit fluctuation patterns not fully covered by the training set, causing the power prediction model to be determined to be in an extrapolation state.

[0028] It should be noted that the applicability determination result is used to characterize the risk of change in the prediction reliability of the power prediction model under the current operating conditions, and does not mean that the prediction error will necessarily increase when the fractal dimension deviates.

[0029] For example, the photovoltaic power generation of a target photovoltaic power station can be represented as a power signal curve defined on the time and amplitude planes: ; in, Represents the power signal curve. Indicates the power sampling time. Indicates the power sampling time The corresponding photovoltaic power generation capacity, This indicates the number of power sampling points.

[0030] Example, fractal dimension It is used to characterize the degree to which the power signal curve fills the time and amplitude planes, reflecting the fluctuation complexity, self-similarity, and irregularity of the power signal.

[0031] when When the value approaches 1, the power signal curve is close to a one-dimensional smooth curve, the power signal is relatively smooth and the daily variation pattern is relatively obvious.

[0032] when At this time, the power signal curve expands moderately in two-dimensional space, and the power signal has slight fluctuations superimposed on the smooth baseline, which usually corresponds to the power signal on a sunny day.

[0033] when When the power signal curve expands more in two-dimensional space, the power signal fluctuates more frequently and becomes more irregular, which usually corresponds to cloudy days or strong fluctuating weather.

[0034] when When the value approaches 2, the power signal curve nearly covers the two-dimensional plane, and the power signal approaches random noise.

[0035] Under normal operating conditions, the fractal dimension of the target photovoltaic power station's power signal is mainly affected by the station's inherent dynamic characteristics and local typical weather patterns.

[0036] In one example, when the inherent dynamic characteristics of the station and the typical local weather patterns remain relatively stable over a longer period of time, the fractal dimension of the power signal is maintained within a specific range.

[0037] In another example, when a frontal system passes through or severe convective weather causes a significant change in the power signal fluctuation pattern, the current window fractal dimension may deviate from the range of fractal dimensions covered by the training set of the power prediction model to be judged.

[0038] For example, this application sets a sliding window for historical measured power time series data and calculates the fractal dimension of the window signal within each sliding window, thus forming a training set fractal dimension sequence organized according to the sliding window arrangement. Based on this, according to the quantiles of the low and high value sides of the training set fractal dimension sequence, a fractal dimension reference interval representing the distribution range of the training set power signal fractal dimension can be obtained.

[0039] For example, this application generates a current window signal based on the current power signal, and processes it using the same fractal dimension calculation method and window length as the training phase to obtain the current window fractal dimension, so that the current window fractal dimension can be compared with the fractal dimension benchmark interval.

[0040] For example, in this application, the fractal dimensions of the current window at multiple consecutive time points are used to form the current fractal dimension sequence. Based on the moving average of the current fractal dimension sequence, the difference between the first and last fractal dimensions, and the direction of change of each current window fractal dimension, the rate of change and trend of the fractal dimension can be obtained. The rate of change of the fractal dimension represents the difference between the current window fractal dimension at the last time point and the current window fractal dimension at the first time point in the current fractal dimension sequence.

[0041] For example, the amplitude deviation criterion and the trend criterion are parallel criteria at the same level. The amplitude deviation criterion takes the current window fractal dimension and the fractal dimension baseline interval adjusted by a preset relaxation coefficient as input, while the trend criterion takes the fractal dimension change rate, a preset change rate threshold, and the change trend as input. In other words, there is no input-output dependency between the two criteria, and they can be triggered simultaneously, separately, or neither can be triggered.

[0042] Figure 2 This is a flowchart of the method for acquiring and preprocessing historical measured power time series data.

[0043] Reference Figure 2 The acquisition and preprocessing of historical measured power time series data includes the following steps: S201. Obtain the original measured power time series data of the target photovoltaic power station covering multiple weather types within a historical period.

[0044] S202. Remove the data from the original measured power time series data that are in the shutdown period and power restriction period to obtain the data for the normal power generation period.

[0045] S203. Linear interpolation is used to fill in the missing data in the data during normal power generation periods to obtain historical measured power time series data.

[0046] In this embodiment, the original measured power time series data comes from the training set of the power prediction model to be judged and includes power sampling values ​​generated by the target photovoltaic power station within a historical time period. The original measured power time series data covers various weather types so that the fractal dimension sequence of the subsequently formed training set can reflect different power fluctuation patterns in the training set of the power prediction model to be judged.

[0047] For example, power data during outage and power curtailment periods are not part of the normal power generation period data. After removing the data from the original measured power time series data, linear interpolation is used to fill in the gaps, forming historical measured power time series data used to calculate the fractal dimension sequence of the training set.

[0048] As a specific implementation, the sampling period for historical measured power time-series data is fifteen minutes. The sampling period is used to determine the time interval between adjacent power sample values.

[0049] Figure 3 A flowchart of the method for constructing the fractal dimension sequence of the training set.

[0050] Reference Figure 3 The construction of the fractal dimension sequence of the training set includes the following processes: S301. Obtain multiple window signals sequentially from historical measured power timing data according to the sliding window length and sliding step size.

[0051] S302. Normalize the amplitude dimension of each window signal to obtain the normalized signal curve.

[0052] S303. Cover the normalized signal curve with boxes of multiple side length scales, and count the number of boxes required to cover the normalized signal curve at each side length scale.

[0053] S304. For each window signal, take the logarithm of the reciprocal of each side length scale as the x-axis and the logarithm of the corresponding number of boxes as the y-axis to obtain multiple regression data points.

[0054] S305. Perform linear regression on multiple regression data points and determine the slope of the obtained fitted line as the fractal dimension of the window signal.

[0055] S306. Arrange the fractal dimensions of the signals of each sliding window according to their arrangement order in the historical measured power time series data to obtain the training set fractal dimension sequence.

[0056] Example, sliding window length This indicates the number of power samples contained in each window signal, and the sliding step size. This indicates the number of power samples between the starting positions of adjacent sliding windows. (Based on the sliding window length) Continuous power samples are selected from historical measured power time-series data to form a window signal; according to the sliding step size After moving the sliding window, continue to obtain the next window signal from the historical measured power timing data.

[0057] As a specific embodiment, when the sampling period for historical measured power time series data is fifteen minutes, the sliding window length is... Set to 96, corresponding to a complete daily variation cycle; sliding step size The value is set to four, corresponding to a one-hour moving interval. The window length is used to capture the fluctuation characteristics of the power signal within a complete daily variation cycle, and the sliding step size is used to determine the update interval between windows corresponding to adjacent fractal dimensions in the training set fractal dimension sequence.

[0058] Each window signal is represented as: ; in, Indicates window signal, The first of the window signals One power sample value, This indicates the sequence number of the power sample value within the window signal. Indicates the length of the sliding window.

[0059] For example, since the power amplitude ranges of different window signals may differ, each window signal is normalized in the amplitude dimension before performing box counting analysis to obtain a normalized signal curve. The normalization process is based on the maximum and minimum amplitude values ​​of each window signal, ensuring that each window signal is within a uniform amplitude range for box counting analysis.

[0060] For example, this application uses multiple side length scales. The boxes cover the normalized signal curve, and the number of boxes required to cover the entire normalized signal curve at each side length scale is counted.

[0061] As a specific embodiment, the side length scale Take the numbers 1, 2, 4, 8, and so on until... The unit of the side length scale is the corresponding scale of the sampling interval after amplitude normalization.

[0062] According to fractal theory, the number of boxes With side length scale The following power-law relationship is satisfied: ; in, The side length is represented by the following scale: The number of boxes required to cover the normalized signal curve at the time. This represents the fractal dimension of the corresponding window signal.

[0063] For the same window signal, the number of boxes is obtained at each side length scale. As the x-axis, Multiple regression data points are formed using the ordinate as the vertical axis. Linear regression is performed on these multiple data points, and the slope of the resulting fitted line is determined as the fractal dimension of the corresponding window signal.

[0064] For example, this application repeatedly moves a sliding window across historical measured power time-series data with a sliding step size, and repeatedly performs normalization, box covering, box number statistics, and linear regression on the window signal within each sliding window to obtain the fractal dimensions of multiple window signals. Further, the fractal dimensions are arranged according to the order of each sliding window in the historical measured power time-series data to obtain the training set fractal dimension sequence: ; in, Indicates the first The fractal dimension of the window signal corresponding to each sliding window. Indicates the sequence number of the sliding window. This indicates the number of sliding windows.

[0065] Figure 4 This is a flowchart of the window signal normalization method.

[0066] Reference Figure 4 Window signal normalization includes the following processing steps: S401. Determine the maximum and minimum amplitude of each window signal.

[0067] S402. For each power sample value in each window signal, calculate the difference between the power sample value and the minimum amplitude, and normalize the power sample value based on the difference between the maximum amplitude and the minimum amplitude to obtain the normalized power sample value.

[0068] S403. Use each normalized power sample value as a curve value point to generate a normalized signal curve.

[0069] For example, for any window signal Determine the maximum amplitude value in the window signal respectively. and minimum amplitude The maximum and minimum amplitude values ​​together define the power amplitude range of the window signal.

[0070] For example, for each power sample value in the window signal First calculate the power sample value With minimum amplitude The difference between them, and then based on the maximum amplitude. With minimum amplitude The differences between them are normalized.

[0071] The normalized power sample value is calculated according to the following formula: ; in, Indicates the first A normalized power sample value, Indicates the first One power sample value, Indicates the maximum amplitude of the window signal. This represents the minimum amplitude of the window signal. This represents a very small positive number, used to avoid the normalization denominator being zero.

[0072] For example, this application will normalize each power sample value These are used as curve numerical points, and a normalized signal curve is formed according to the order of the power sample values ​​in the window signal.

[0073] Among them, the normalized signal curve is used as the object of processing for multi-scale box coverage and box number statistics.

[0074] Figure 5 A flowchart for determining the reference interval for fractal dimension.

[0075] Reference Figure 5 The determination of the fractal dimension reference interval includes the following processes: S501, Set the tail portion ratio used to determine the reference interval for fractal dimension.

[0076] S502. Calculate the first quantile and the second quantile from the low-value side and the high-value side of the fractal dimension sequence of the training set, respectively, according to the tail quantile ratio.

[0077] S503. The first quantile is determined as the lower limit of the fractal dimension reference interval, and the second quantile is determined as the upper limit of the fractal dimension reference interval.

[0078] For example, the tail quantile ratio is used to determine the position of the quantiles selected from the low-value and high-value sides of the training set fractal dimension sequence.

[0079] In some embodiments, the tail portion ratio is denoted as Calculate from the low-value side of the fractal dimension sequence of the training set. quantiles Calculate the quantile corresponding to the proportion of a reduced-tail quantile from the high-value side of the fractal dimension sequence in the training set. .

[0080] For example, the first quantile Determined as the lower bound of the fractal dimension reference interval , the second quantile The upper limit of the fractal dimension reference interval is determined. ,get: ; in, This represents the lower bound of the fractal dimension reference interval. This represents the upper limit of the fractal dimension reference interval.

[0081] As one specific embodiment, the tail portion ratio The value is set to 0.08, with the first quantile being the 0.08 quantile of the fractal dimension sequence of the training set and the second quantile being the 0.92 quantile of the fractal dimension sequence of the training set.

[0082] In the above embodiments, the fractal dimension reference interval formed by two quantiles represents the range of power signal fractal dimension distribution in the training set of the power prediction model to be judged.

[0083] Figure 6 A flowchart for constructing the signal for the current window.

[0084] Reference Figure 6 The current window signal construction includes the following processing steps: S601. The current power signal is sampled in real time according to a fixed time resolution to obtain multiple current power sampling points arranged according to the acquisition time.

[0085] S602. For multiple current power sampling points, select the most recently continuously collected current power sampling points whose number is equal to the window length.

[0086] S603. According to the acquisition time sequence, the selected current power sampling points are stored in the sliding buffer to obtain the current window signal.

[0087] For example, the current power signal of the target photovoltaic power station is represented as follows: .in, This indicates the current acquisition time of the power signal.

[0088] In this application, the current power signal is sampled in real time according to a fixed time resolution, which can obtain multiple current power sampling points arranged according to the acquisition time.

[0089] For example, select the most recent consecutive samples from multiple current power sampling points, with the number equal to the window length. The current power sampling point is selected and stored in the sliding buffer according to the acquisition time sequence to form the current window signal: .

[0090] in, Indicates the current window signal. This indicates the current power sampling point at the current moment. This indicates the current power sampling point with the earliest acquisition time in the current window signal. Indicates the window length.

[0091] For example, whenever a new current power sampling point is acquired, the sliding buffer updates the current window signal according to the acquisition time, so that the current window signal consists of the most recently acquired consecutive current power sampling points with a number equal to the window length.

[0092] For example, when calculating the fractal dimension of the current window signal, the amplitude normalization method, box counting method, linear regression method, and window length are consistent with the process of constructing the fractal dimension sequence of the training set. The slope of the fitted line obtained from the linear regression is used to determine the fractal dimension of the current window. .

[0093] Figure 7 A flowchart for determining the rate of change and trend of fractal dimension.

[0094] Reference Figure 7 The determination of the rate of change and trend of fractal dimension includes the following processing steps: S701. According to the calculation time order, form a current fractal dimension sequence by the preset number of currently calculated consecutively.

[0095] S702. Calculate the moving average of the current fractal dimension sequence, and determine the difference between the fractal dimension of the first current window and the fractal dimension of the second current window as the fractal dimension change rate.

[0096] Wherein, the first current window fractal dimension is the current window fractal dimension at the last moment in the current fractal dimension sequence, and the second current window fractal dimension is the current window fractal dimension at the first moment in the current fractal dimension sequence.

[0097] S703. Determine the trend of change based on the moving average, the rate of change of fractal dimension, and the direction of change of the fractal dimension of each current window in the current fractal dimension sequence.

[0098] For example, the most recently calculated fractal dimension is selected according to the calculation time sequence of the current window. The fractal dimensions of the current window are used to form a sequence of current fractal dimensions: ; in, This indicates the number of fractal dimensions in the current window used to analyze trends of change. This represents the fractal dimension of the current window at the first time step in the current fractal dimension sequence. This represents the current window fractal dimension at the last moment in the current fractal dimension sequence.

[0099] For example, the moving average of the current fractal dimension sequence is calculated as follows: ; in, This represents the moving average of the current fractal dimension sequence. Represents the th element in the current fractal dimension sequence. The fractal dimension of the current window. This indicates the sequence number of the current window's fractal dimension within the current fractal dimension sequence.

[0100] For example, the difference between the current window fractal dimension at the last time step and the current window fractal dimension at the first time step in the current fractal dimension sequence is determined as the rate of change of fractal dimension: ; in, This represents the rate of change of the fractal dimension. The rate of change of the fractal dimension represents the difference between the first and last values ​​of the current fractal dimension sequence, and does not represent the derivative per unit time or the percentage rate of change.

[0101] In this application, based on the moving average Fractal dimension change rate And the direction of change of the fractal dimension of each current window in the current fractal dimension sequence, to determine whether the fractal dimension has been increasing, decreasing or not in the same direction of change in the most recent consecutive moments, thus obtaining the trend of change.

[0102] Figure 8 This is a flowchart of a method for determining the threshold of the rate of change of fractal dimension.

[0103] Reference Figure 8 The fractal dimension change rate threshold determination includes the following processing steps: S801, the rising rate of change threshold and the falling rate of change threshold are determined based on the standard deviation of the rate of change of the fractal dimension sequence of the training set.

[0104] S802a, the rate of change of fractal dimension is greater than the threshold of the rate of change and continues to rise.

[0105] S802b, the rate of change of fractal dimension is less than the threshold of the rate of change of decrease and continues to decrease.

[0106] The rate of change of fractal dimension during the training phase is calculated based on the fractal dimension sequence of the training set, and the standard deviation of the rate of change is determined based on the obtained rate of change during the training phase. The threshold for the rate of change increasing is positive, and the threshold for the rate of change decreasing is negative.

[0107] As a specific implementation, the rising rate of change threshold and the falling rate of change threshold are determined based on twice the standard deviation of the rate of change of the fractal dimension sequence of the training set. Rising rate of change threshold threshold for decreasing rate of change .

[0108] when Furthermore, when the fractal dimension of the current window in the current fractal dimension sequence continues to increase, the threshold condition of the rate of change in the upward direction is satisfied.

[0109] when Furthermore, when the fractal dimension of the current window in the current fractal dimension sequence continues to decrease, the threshold condition of the rate of change in the direction of decrease is satisfied.

[0110] Among them, the upward direction and the downward direction represent opposite states of change, and one of them is chosen to be satisfied in the same judgment.

[0111] For example, a trend criterion is triggered when the rate of change in the upward or downward direction is met, and the trend has maintained the corresponding direction for multiple consecutive time points. Conversely, a trend criterion is not triggered if the rate of change in the fractal dimension does not exceed the rate of change threshold in the corresponding direction, or if the trend does not maintain the same direction.

[0112] In this application, the amplitude deviation criterion is based on the fractal dimension of the current window. and fractal dimension benchmark interval For input, a preset relaxation coefficient is set based on the width of the fractal dimension reference interval, and the fractal dimension reference interval is adjusted using the preset relaxation coefficient. Furthermore, when the fractal dimension of the current window exceeds the adjusted fractal dimension reference interval, an amplitude deviation criterion is triggered.

[0113] As a specific embodiment, the preset relaxation coefficient corresponds to 15% of the width of the fractal dimension reference interval. Setting the preset relaxation coefficient creates a relaxation range outside the boundary of the fractal dimension reference interval, preventing frequent triggering of the amplitude deviation criterion when the current window's fractal dimension fluctuates near the interval boundary.

[0114] Figure 9 This is a flowchart of the comprehensive method for determining the applicability of a model.

[0115] Reference Figure 9 The comprehensive assessment of model applicability includes the following branch processing: S901, Amplitude Deviation Criterion and Trend Criterion.

[0116] S902a. When both the magnitude deviation criterion and the trend criterion are triggered, the applicability determination result is determined to be model misfit, and the confidence level of the applicability determination result is high confidence.

[0117] S902b: When the amplitude deviation criterion is triggered but the trend criterion is not triggered, the applicability determination result is determined to be that the model may not fit, and the confidence level of the applicability determination result is moderate confidence.

[0118] S902c: When the trend criterion is triggered but the magnitude deviation criterion is not triggered, the applicability determination result is determined to be that the model is in a trend drift, and the confidence level of the applicability determination result is low confidence.

[0119] S902d: When neither the amplitude deviation criterion nor the trend criterion is triggered, the applicability determination result is determined to be model fit.

[0120] In the embodiments of this application, the amplitude deviation criterion and the trend criterion respectively form triggered or untriggered criterion states. Specifically, the two criterion states together form four mutually exclusive trigger combinations, each trigger combination corresponding to an applicability determination result.

[0121] For example, when both the magnitude deviation criterion and the trend criterion are triggered, it indicates that the current window fractal dimension has deviated from the range of fractal dimensions covered by the training set, and the current window fractal dimension has maintained a significant trend of change over several recent consecutive time points. In this case, the power prediction model to be discriminated is determined to be ill-fitting, and the decision confidence level is set to high confidence. When only the magnitude deviation criterion is triggered, it indicates that the current window fractal dimension has exceeded the adjusted fractal dimension baseline range, but the current window fractal dimension has not met the trend criterion over several recent consecutive time points. In this case, the power prediction model to be discriminated is determined to be potentially ill-fitting, and the decision confidence level is set to medium confidence. When only the trend criterion is triggered, it indicates that the current window fractal dimension has not yet triggered the magnitude deviation criterion, but the rate of change of the fractal dimension has exceeded the preset rate of change threshold in the corresponding direction, and the trend of change remains in the same direction. In this case, the power prediction model to be discriminated is determined to be in a trend of drift, and the decision confidence level is set to low confidence. If neither the amplitude deviation criterion nor the trend criterion is triggered, it indicates that the current window fractal dimension does not exceed the adjusted fractal dimension baseline range, and the rate of change and trend of the fractal dimension do not meet the trend criterion. In this case, the power prediction model to be determined is identified as the model fit.

[0122] In the above embodiments, model misfit indicates a high risk of reduced prediction reliability of the power prediction model to be judged, but does not necessarily mean that the prediction error will increase.

[0123] In addition, the applicability assessment results can serve as risk warning information, providing a basis for parameter optimization of the power prediction model to be assessed, model replacement, or for dispatchers to differentiate the power prediction results.

[0124] Figure 10 This is a structural diagram of a photovoltaic power prediction model applicability discrimination system.

[0125] Reference Figure 10The photovoltaic power prediction model applicability judgment system includes a benchmark interval determination module 1001, a current window fractal dimension calculation module 1002, a trend determination module 1003, and an applicability judgment module 1004.

[0126] The benchmark interval determination module 1001 acquires the historical measured power time series data of the target photovoltaic power station in the training set of the power prediction model to be judged, calculates the fractal dimension of the historical measured power time series data in each sliding window, obtains the training set fractal dimension sequence, and determines the fractal dimension benchmark interval based on the quantile of the training set fractal dimension sequence. The current window fractal dimension calculation module 1002 collects the current power signal of the target photovoltaic power station in real time, obtains the current window signal, and calculates the current window fractal dimension based on the fractal dimension calculation method and window length that are consistent with the fractal dimension sequence of the training set determined by the reference interval module 1001. The trend determination module 1003 calculates the rate of change of the fractal dimension and determines the trend of change based on the fractal dimension of the current window at multiple recent consecutive moments. The applicability discrimination module 1004 receives the fractal dimension reference interval, the current window fractal dimension, the rate of change of the fractal dimension, and the trend of change. If the current window fractal dimension exceeds the fractal dimension reference interval adjusted by the preset relaxation coefficient, the amplitude deviation criterion is triggered. If the rate of change of the fractal dimension exceeds the preset rate of change threshold and the trend of change remains in the same direction, the trend criterion is triggered. Based on the triggering results of the amplitude deviation criterion and the trend criterion, the applicability judgment result of the power prediction model to be discriminated is obtained.

[0127] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0128] Although the operations are described in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order or serial order shown, or requiring all of the operations shown to obtain the desired result. In certain environments, multitasking and parallel processing may be advantageous.

[0129] The methods and systems of this application can be implemented using standard programming techniques, utilizing rule-based logic or other logic to implement various method steps. It should also be noted that the terms "system" and "module" as used herein and in the claims are intended to include implementations using one or more lines of software code and / or hardware implementations and / or devices for receiving input.

[0130] Any step, operation, or procedure described herein may be performed or implemented using one or more hardware or software modules, either alone or in combination with other devices. In one embodiment, the software module is implemented using a computer program product comprising a computer-readable medium containing computer program code, which is executable by a computer processor to perform any or all of the described steps, operations, or procedures.

[0131] The foregoing description of implementations of this application has been provided for illustrative and descriptive purposes. The foregoing description is not exhaustive and is not intended to limit this application to the exact forms disclosed. Various modifications and variations may exist in accordance with the foregoing teachings, or may arise from practice of this application. These embodiments were chosen and described to illustrate the principles of this application and its practical application, enabling those skilled in the art to utilize this application in various implementations and modifications to suit the specific purpose of the concept.

[0132] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0133] It can be further understood that, unless otherwise specified, "connection" includes both direct connections where no other components exist between the two parties and indirect connections where other components exist between them.

[0134] It is further understood that although the operations are described in a specific order in the accompanying drawings in the embodiments of this application, this should not be construed as requiring these operations to be performed in the specific order or serial order shown, or requiring all the operations shown to be performed to obtain the desired result. In certain environments, multitasking and parallel processing may be advantageous.

[0135] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the field of this application that are not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0136] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

[0137] The above 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 method for judging the applicability of a photovoltaic power prediction model, characterized in that, include: Obtain historical measured power time series data of the target photovoltaic power station in the training set of the power prediction model to be judged, calculate the fractal dimension of the historical measured power time series data in each sliding window, obtain the training set fractal dimension sequence, and determine the fractal dimension benchmark interval according to the quantile of the training set fractal dimension sequence. The current power signal of the target photovoltaic power station is collected in real time to obtain the current window signal, and the fractal dimension of the current window is calculated based on the fractal dimension calculation method and window length consistent with the calculation of the fractal dimension sequence of the training set. Calculate the rate of change of the fractal dimension and determine the trend of change based on the fractal dimension of the current window at multiple recent consecutive moments; If the fractal dimension of the current window exceeds the reference range of the fractal dimension adjusted by the preset relaxation coefficient, the amplitude deviation criterion is triggered. If the rate of change of the fractal dimension exceeds a preset rate of change threshold and the trend of change remains in the same direction, then the trend criterion is triggered. Based on the triggering results of the amplitude deviation criterion and the trend criterion, the applicability determination result of the power prediction model to be determined is obtained.

2. The method for judging the applicability of the photovoltaic power prediction model according to claim 1, characterized in that, The acquisition of historical measured power time-series data of the target photovoltaic power station in the training set of the power prediction model to be judged includes: Obtain the original measured power time series data of the target photovoltaic power station covering multiple weather types within a historical period; Data from the outage and power restriction periods in the original measured power time series data are removed to obtain data for normal power generation periods; Linear interpolation is used to fill in the missing data in the normal power generation period data to obtain the historical measured power time series data.

3. The method for judging the applicability of the photovoltaic power prediction model according to claim 1, characterized in that, The calculation of the fractal dimension of the historical measured power time series data within each sliding window yields a training set fractal dimension sequence, including: Multiple window signals are sequentially obtained from the historical measured power time series data based on the sliding window length and sliding step size; The amplitude dimension of each window signal is normalized to obtain a normalized signal curve; The normalized signal curve is covered with boxes of multiple side length scales, and the number of boxes required to cover the normalized signal curve at each side length scale is counted. For each window signal, the logarithm of the reciprocal of each side length scale is used as the x-axis, and the logarithm of the corresponding number of boxes is used as the y-axis to obtain multiple regression data points; Linear regression is performed on the multiple regression data points, and the slope of the resulting fitted line is determined as the fractal dimension of the window signal; The fractal dimensions of the signals of each sliding window are arranged according to their order in the historical measured power time series data to obtain the training set fractal dimension sequence.

4. The method for judging the applicability of the photovoltaic power prediction model according to claim 3, characterized in that, The normalization process for each window signal in terms of amplitude dimension to obtain a normalized signal curve includes: Determine the maximum and minimum amplitude of each window signal; For each power sample value in each of the aforementioned window signals, the difference between the power sample value and the minimum amplitude is calculated, and the power sample value is normalized based on the difference between the maximum amplitude and the minimum amplitude to obtain a normalized power sample value. The normalized power sample values ​​are used as curve points to generate the normalized signal curve.

5. The method for judging the applicability of the photovoltaic power prediction model according to claim 1, characterized in that, The step of determining the fractal dimension baseline interval based on the quantiles of the fractal dimension sequence of the training set includes: Set the tail portion ratio used to determine the fractal dimension reference interval; According to the tail quantile ratio, the first quantile and the second quantile are calculated from the low-value side and the high-value side of the fractal dimension sequence of the training set, respectively. The first quantile is determined as the lower limit of the fractal dimension reference interval, and the second quantile is determined as the upper limit of the fractal dimension reference interval.

6. The method for judging the applicability of the photovoltaic power prediction model according to claim 1, characterized in that, The real-time acquisition of the current power signal of the target photovoltaic power station to obtain the current window signal includes: The current power signal is sampled in real time according to a fixed time resolution to obtain multiple current power sampling points arranged according to the acquisition time; For the plurality of current power sampling points, select the most recently consecutively collected current power sampling points whose number is equal to the window length; According to the acquisition time sequence, the selected current power sampling points are stored in the sliding buffer to obtain the current window signal.

7. The method for judging the applicability of the photovoltaic power prediction model according to claim 1, characterized in that, The step of calculating the rate of change of fractal dimension and determining the trend of change based on the fractal dimension of the current window at multiple recent consecutive time points includes: According to the calculation time order, the current window fractal dimension sequence is formed by the preset number of the most recently calculated consecutive fractal dimensions; Calculate the moving average of the current fractal dimension sequence, and determine the difference between the first current window fractal dimension and the second current window fractal dimension as the rate of change of the fractal dimension; wherein, the first current window fractal dimension is the current window fractal dimension at the last moment in the current fractal dimension sequence, and the second current window fractal dimension is the current window fractal dimension at the first moment in the current fractal dimension sequence; The trend of change is determined based on the moving average, the rate of change of the fractal dimension, and the direction of change of the fractal dimension of each current window in the current fractal dimension sequence.

8. The method for judging the applicability of the photovoltaic power prediction model according to claim 1, characterized in that, The fractal dimension change rate exceeding a preset change rate threshold includes: The rate of change of the fractal dimension is greater than the threshold for the rate of change and continues to rise; or The rate of change of the fractal dimension is less than the threshold for the rate of decrease and continues to decrease. The rising rate of change threshold and the falling rate of change threshold are determined based on the standard deviation of the rate of change of the fractal dimension sequence of the training set.

9. The method for judging the applicability of the photovoltaic power prediction model according to claim 1, characterized in that, The applicability determination result of the power prediction model to be judged, obtained based on the triggering results of the amplitude deviation criterion and the trend criterion, includes: When both the amplitude deviation criterion and the trend criterion are triggered, the applicability determination result is determined to be model misfit, and the confidence level of the applicability determination result is high confidence. When the amplitude deviation criterion is triggered but the trend criterion is not triggered, the applicability determination result is determined to be that the model may not fit, and the confidence level of the applicability determination result is moderate confidence. When the trend criterion is triggered and the magnitude deviation criterion is not triggered, the applicability determination result is determined to be that the model is in a trend drift, and the confidence level of the applicability determination result is low confidence. When neither the amplitude deviation criterion nor the trend criterion is triggered, the applicability determination result is determined to be model fit.

10. A photovoltaic power prediction model applicability judgment system, characterized in that, include: The benchmark interval determination module acquires the historical measured power time series data of the target photovoltaic power station in the training set of the power prediction model to be judged, calculates the fractal dimension of the historical measured power time series data in each sliding window, obtains the training set fractal dimension sequence, and determines the fractal dimension benchmark interval based on the quantile of the training set fractal dimension sequence. The current window fractal dimension calculation module collects the current power signal of the target photovoltaic power station in real time, obtains the current window signal, and calculates the current window fractal dimension based on the fractal dimension calculation method and window length that are consistent with the fractal dimension sequence of the training set calculated by the benchmark interval determination module. The trend determination module calculates the rate of change of the fractal dimension and determines the trend of change based on the fractal dimension of the current window at multiple recent consecutive moments. The applicability determination module receives the fractal dimension reference interval, the current window fractal dimension, the fractal dimension change rate, and the change trend. If the current window fractal dimension exceeds the fractal dimension reference interval adjusted by a preset relaxation coefficient, the amplitude deviation criterion is triggered. If the rate of change of the fractal dimension exceeds a preset rate of change threshold and the trend of change remains in the same direction, then the trend criterion is triggered. Based on the triggering results of the amplitude deviation criterion and the trend criterion, the applicability determination result of the power prediction model to be determined is obtained.