A real-time power prediction method for a distributed photovoltaic power station

Abstract

The application discloses a kind of real-time power prediction methods of distributed photovoltaic power station, it is related to electric power prediction technical field, including the following steps: from basic data, the first dynamic feature and the second dynamic feature are extracted synchronously, mapping database of historical feature-historical power data is constructed based on historical data, the first dynamic feature and the second dynamic feature are carried out similarity calculation with corresponding historical data in mapping database and obtain the first similarity and the second similarity, historical data is split into multiple historical feature and historical power data's associated entry according to time period and working condition, each associated entry corresponds a historical sample, select the historical sample of similarity higher than preset threshold in the first similarity and the second similarity, extract historical power data distribution characteristics corresponding to historical sample, predict based on historical power data distribution characteristics, extract double dynamic feature, after sample processing, clustering and association adjustment, improve prediction accuracy, provide support for safe and efficient operation of power grid.

Description

Technical Field

[0001] This invention relates to the field of power forecasting technology, and in particular to a real-time power forecasting method for distributed photovoltaic power plants. Background Technology

[0002] In recent years, with the global energy structure transitioning towards clean and low-carbon energy, distributed photovoltaic (PV) power stations have rapidly gained popularity due to their flexible deployment and proximity to load centers. Their output stability directly impacts the power balance and dispatch efficiency of regional power grids. Therefore, real-time power forecasting technology has become a key support for ensuring PV integration and improving grid reliability. Especially in scenarios with a high proportion of renewable energy connected to the grid, PV power is dynamically affected by multiple factors, including solar irradiance, ambient temperature, cloud cover changes, and equipment operating status. Accurate short- to ultra-short-term power forecasting not only helps grid dispatching departments to formulate power generation plans in advance and avoid curtailment, but also reduces the impact of PV output fluctuations on the voltage and frequency stability of the distribution network, thus having significant economic and social benefits for promoting the large-scale development of the new energy industry.

[0003] However, existing distributed photovoltaic power prediction technologies still have many shortcomings: most methods rely only on single-dimensional static meteorological data or isolated equipment parameters, failing to fully explore the dynamic characteristics of meteorological factors and the correlation between equipment operating status, resulting in feature representations that are difficult to reflect the power impact mechanism under complex operating conditions; some schemes do not standardize or discretize multi-dimensional features to unify the dimensions when matching historical samples, and the sample selection uses a single similarity threshold, which easily introduces low-quality samples. At the same time, cluster analysis often ignores the impact of seasonal and intraday time dimensions on power patterns, and the prediction parameters do not adaptively adjust the power fluctuation range in combination with real-time dynamic features, ultimately leading to a large deviation between the prediction results and the actual output. Summary of the Invention

[0004] The technical problem addressed by this invention is that existing distributed photovoltaic power prediction technologies still have several shortcomings: most methods rely solely on single-dimensional static meteorological data or isolated equipment parameters, failing to fully explore the dynamic characteristics of meteorological factors and the correlation between equipment operating status, resulting in feature representations that are difficult to reflect the power impact mechanisms under complex operating conditions; some schemes do not standardize or discretize multi-dimensional features to unify the dimensions when matching historical samples, and sample selection uses a single similarity threshold, which easily introduces low-quality samples; at the same time, cluster analysis often ignores the influence of seasonal and intraday time dimensions on power patterns, and prediction parameters do not adaptively adjust the power fluctuation range in conjunction with real-time dynamic features, ultimately leading to a large deviation between the prediction results and the actual output.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a real-time power prediction method for distributed photovoltaic power stations, comprising the following steps: Step S100: Simultaneously extract the first dynamic feature and the second dynamic feature from the basic data; Step S200: Construct a mapping database of historical features and historical power data based on historical data, and calculate the similarity between the first dynamic feature and the second dynamic feature and the corresponding historical data in the mapping database to obtain the first similarity and the second similarity. Step S300: The historical data is divided into multiple related entries of historical features and historical power data according to time period and operating condition, and each related entry corresponds to a historical sample; Historical samples with similarity higher than a preset threshold are selected from the first and second similarity samples, and the historical power data distribution features corresponding to the historical samples are extracted. Step S400: Make a prediction based on the distribution characteristics of the historical power data.

[0006] As a preferred embodiment of the real-time power prediction method for distributed photovoltaic power stations according to the present invention, the method includes: acquiring basic data for power prediction of the photovoltaic power station, wherein the basic data includes real-time weather forecasts and equipment status data of the photovoltaic power station; The real-time weather forecast includes solar irradiance data, ambient temperature data, and cloud cover data; The photovoltaic power station equipment status data includes inverter operating parameter data and photovoltaic module temperature data; Extracting the first dynamic feature from the basic data specifically includes: Set a time interval and extract the variation trend of the solar irradiance data for three consecutive time intervals; The criteria for determining the trend of change include: Set the amplitude threshold 'a' and fluctuation threshold 'b' for the solar irradiance data; If, in the solar irradiance data of three consecutive time intervals, the two adjacent increases of the value of the later time interval relative to the value of the previous time interval are both greater than 'a', then it is determined to be an increasing magnitude. The increasing magnitude is the ratio of the cumulative increase value of the three time intervals to the initial time interval value, and the ratio is the magnitude value corresponding to the increasing trend. If, in the solar irradiance data of three consecutive time intervals, the two adjacent decreases of the value of the later time interval relative to the value of the previous time interval are both greater than 'a', then it is determined to be a decreasing amplitude. The decreasing amplitude is the ratio of the cumulative decrease value of the three time intervals to the initial interval value, and the ratio is the amplitude value corresponding to the decreasing trend. If the fluctuation range of adjacent interval values ​​of solar irradiance data in three consecutive time intervals is less than or equal to b, it is determined to be a stable state. The trend of change includes increasing magnitude, decreasing magnitude, and stable state, where a and b are positive numbers, and the magnitude is expressed as the absolute value of the difference between the solar irradiance data of the later time interval and the solar irradiance data of the previous time interval. Record the extreme temperature difference that occurs within the time interval and the duration of the extreme temperature difference. Monitor the abrupt change in cloud cover level of the cloud cover data, and extract the instantaneous response characteristics of solar irradiance data from a preset time period before the abrupt change to a preset time period after the abrupt change. The instantaneous response characteristics include response delay duration and the percentage change in solar irradiance data; The first dynamic feature is formed by combining the changing trend of the solar irradiance data, the extreme temperature difference of the ambient temperature data, the duration of the extreme temperature difference, and the instantaneous response characteristics of the solar irradiance data at the moment of abrupt change in cloud cover data.

[0007] As a preferred embodiment of the real-time power prediction method for distributed photovoltaic power stations described in this invention, the extraction of a second dynamic feature from the basic data specifically includes: Using a preset time as the sampling unit, the fluctuation frequency of the inverter's operating parameter output current data within a fixed time period sliding window is extracted; The criteria for determining the fluctuation include: A current fluctuation data threshold is set. When the absolute value of the difference between the current fluctuation data of adjacent sampling units is greater than the current fluctuation data threshold, it is determined to be a fluctuation. Calculate the direction of change of the difference between the photovoltaic module temperature data and the ambient temperature data over a continuous period of time; The criteria for determining the direction of change include: Set a threshold for temperature difference changes; If, within a continuous time period, the difference between the final value and the initial value of the difference is greater than the temperature difference change threshold, it is determined to be a positive increase. If, within a continuous time period, the difference between the final value and the initial value of the difference is less than the temperature difference change threshold, it is determined to be a negative decrease. If, within a continuous period of time, the absolute value of the difference between the final value and the initial value is less than or equal to the temperature difference change threshold, it is determined to be basically stable; The directions of change include positive growth, negative decrease, and basic stability; The time synchronization characteristics of the changing trends of current fluctuation data and solar irradiance data associated with the inverter operating parameter data; The second dynamic feature is formed by combining the fluctuation frequency of the current fluctuation data, the direction of the difference between the photovoltaic module temperature data and the ambient temperature data, and the time synchronization characteristics of the current fluctuation data and the solar irradiance data in the inverter operating parameter data.

[0008] As a preferred embodiment of the real-time power prediction method for distributed photovoltaic power stations according to the present invention, the method includes: calculating the similarity between the first dynamic feature and the historical data of the mapping database to obtain a first similarity, specifically including: Standardization of the first dynamic feature includes: The amplitude of the solar irradiance data change trend, the extreme temperature difference of the ambient temperature data, the duration corresponding to the extreme temperature difference of the ambient temperature data, the response delay of the solar irradiance data at the moment of abrupt change in cloud cover data, and the change ratio of the solar irradiance data are all converted to the same preset value range, specifically including: The min-max standardization algorithm is adopted, with the minimum value of the historical first dynamic feature in the mapping database as the lower limit of the benchmark and the maximum value of the historical first dynamic feature as the upper limit of the benchmark. By calculating the difference between the original value of each first dynamic feature and the lower limit of the benchmark, and the proportional relationship between the difference between the upper limit of the benchmark and the lower limit of the benchmark, the original value of each first dynamic feature is scaled to a preset numerical range, thereby achieving the dimensional unification of multi-dimensional features. The original values ​​of the first dynamic feature include the amplitude of the trend of the solar irradiance data, the extreme temperature difference of the ambient temperature data, the duration of the extreme temperature difference of the ambient temperature data, the response delay of the solar irradiance data at the moment of abrupt change in cloud cover data, and the change ratio of the solar irradiance data. Using the Euclidean distance algorithm, the spatial distance between the standardized first dynamic feature and the historical first dynamic feature stored in the mapping database is calculated one by one. Then, c minus the normalized distance value is used as the first similarity, and a first reasonable numerical range of the first similarity is set. The first reasonable numerical range is a closed interval. The normalized distance value is the ratio of the spatial distance to the maximum spatial distance between the historical first dynamic features in the mapping database.

[0009] As a preferred embodiment of the real-time power prediction method for distributed photovoltaic power stations according to the present invention, the method includes: calculating the similarity between the second dynamic feature data and the historical data of the mapping database to obtain a second similarity, specifically including: Performing discretization encoding on the second dynamic feature includes: The frequency of inverter operating parameter data and output current data fluctuation within a preset time sliding window is divided into multiple levels according to preset value ranges, specifically including: Based on the minimum and maximum values ​​of the fluctuation frequency of the current data in the historical data of the mapping database, a first interval is set as a first proportional threshold where the fluctuation frequency is less than or equal to the average value of the current data, corresponding to the first level; The second interval is defined as a first proportional threshold where the fluctuation frequency is greater than the average current data value and less than or equal to the second proportional threshold where the fluctuation frequency is greater than or equal to the average current data value, corresponding to the second level; The third interval is defined as a second proportional threshold where the fluctuation frequency is greater than the average value of the current data, corresponding to the third level; The direction of change of the difference between photovoltaic module temperature data and ambient temperature data within a preset time period is expressed numerically, including: The positive growth corresponds to a first preset value in the temperature direction, the basically stable corresponds to a second preset value in the temperature direction, and the negative decrease corresponds to a third preset value in the temperature direction. The time synchronization characteristics of the current fluctuation data and the solar irradiance data of the inverter operating parameter data are represented by numerical values, and the time difference corresponding to the characteristics is used as the basis for judgment. The time difference is the absolute value of the difference between the time when the inverter current fluctuation data occurs and the time when the solar irradiance change trend is triggered, and the preset condition is that the time difference is less than or equal to a first time threshold. When the time difference meets the preset conditions, the feature corresponds to the synchronization - first preset value; When the time difference does not meet the preset conditions, the feature corresponds to the second preset value of synchronization, and the first preset value of synchronization is different from the second preset value of synchronization. The Hamming distance algorithm is used to calculate the number of bit differences between the discretized second dynamic feature data and the historical second dynamic features stored in the mapping database. Then, the ratio of the number of bit differences to the total number of bits of the feature is subtracted from b as the second similarity. A second reasonable numerical range of the second similarity is set, which is a closed interval.

[0010] As a preferred embodiment of the real-time power prediction method for distributed photovoltaic power stations described in this invention, the establishment of a first similarity or second similarity verification mechanism specifically includes: If the first similarity or second similarity of any historical sample exceeds the corresponding first reasonable value range or second reasonable value range, then feature data of the same type as the historical sample within a preset time range in the mapping database are called to recalculate and obtain the first similarity or second similarity. The historical samples include historical first dynamic features, historical second dynamic features, and associated historical power data.

[0011] As a preferred embodiment of the real-time power prediction method for distributed photovoltaic power stations according to the present invention, the method includes: extracting historical samples with similarity higher than a preset threshold from the first similarity and the second similarity, and extracting the historical power data distribution features corresponding to the historical samples, specifically including: Setting the preset threshold includes setting a third preset threshold and a fourth preset threshold. The value range of the third preset threshold is from the median to the upper limit of the first reasonable numerical range of the first similarity. The value range of the fourth preset threshold is from the median to the upper limit of the second reasonable numerical range of the second similarity. The historical samples in the mapping database are subjected to dual screening, and the dual screening conditions include a first similarity higher than a third preset threshold and a second similarity higher than a fourth preset threshold. If the dual screening conditions are met, the historical samples are retained to form a candidate sample set.

[0012] As a preferred embodiment of the real-time power prediction method for distributed photovoltaic power stations described in this invention, the method involves: extracting historical power data corresponding to each historical sample from the candidate sample set, calculating the distribution characteristics of the historical power data, and the distribution characteristics including the power mean, the occurrence time of the power peak, the power attenuation coefficient over time, and the power fluctuation variance. Clustering of the historical power data distribution characteristics in the candidate sample set according to the time dimension, specifically including: The K-means clustering algorithm was used, with the season and intraday time period of the historical samples as the basis for clustering and grouping by time dimension. The intraday time period includes morning, noon and afternoon. First, the historical power data distribution characteristics in the candidate sample set are initially divided into several groups according to the time dimension; Then, the distribution characteristics of the power mean, power peak occurrence time, power decay coefficient over time, and power fluctuation variance of each historical sample in each group are used as clustering input variables. The number of clusters is set to de. Through iterative calculation, the sum of squared Euclidean distances from each historical sample in each group to the corresponding cluster center is minimized to complete the clustering. The cluster center is the core feature virtual sample point of the corresponding cluster during the clustering process in each group, including feature values ​​with the same dimension as the clustering input variable. The feature values ​​include the power mean, the occurrence time of the power peak, the power decay coefficient over time, and the power fluctuation variance. Each feature value is the mean of the feature values ​​corresponding to all historical samples in the cluster. One or more power feature clusters are obtained, and each power feature cluster corresponds to a typical power change pattern.

[0013] As a preferred embodiment of the real-time power prediction method for distributed photovoltaic power stations described in this invention, the prediction based on the distribution characteristics of the historical power data specifically includes: Determine the time attribute parameters of the current time period to be predicted, including the season to which the time period belongs, the intraday time period, and the weather type label; From the power feature clusters, the target cluster with the highest matching degree with the time attribute parameter is selected. The criteria for determining the highest matching degree include: If the historical time period attribute parameters corresponding to the power feature cluster completely overlap with the seasonal, intraday time period, and weather type labels of the current time attribute parameters, it is determined to be the highest matching degree; If there is no historical time period attribute parameter corresponding to the power feature cluster that completely overlaps with the seasonal, intraday time period and weather type label of the current time attribute parameter, then the cluster with the most overlapping dimensions among the seasonal, intraday time period and weather type label is selected as the candidate with the highest matching degree. If the number of overlapping dimensions is the same, clusters that overlap in both season and intraday time period are preferred. Extract the typical power change pattern of the target cluster. The typical power change pattern refers to a standardized template that can represent the common law of power change of the cluster after statistically aggregating the power distribution characteristics of all historical samples in the target cluster. The average power, peak power occurrence time, power attenuation coefficient over time, and power fluctuation variance of the typical power change pattern are used as the basic prediction parameters.

[0014] As a preferred embodiment of the real-time power prediction method for distributed photovoltaic power stations described in this invention, the method involves calculating the correlation coefficient between the variance of historical power data in the target cluster and the amplitude of the changing trend of solar irradiance data in the current first dynamic feature, specifically including: The Pearson correlation coefficient algorithm is adopted, with the historical power data fluctuation variance of all historical samples in the target cluster as the first variable set, and the amplitude value of the solar irradiance data change trend in the historical first dynamic feature corresponding to each historical sample as the second variable set. The basic correlation coefficient at the historical level is obtained by calculating the ratio of the covariance of the first set of variables and the second set of variables to the product of the standard deviations of the first set of variables and the second set of variables. Then, substitute the magnitude of the solar irradiance data change trend in the current first dynamic feature into the fitting relationship corresponding to the basic correlation coefficient to determine the final correlation coefficient; The fitting relationship is constructed based on the statistical regularities of the first and second variable sets within the target cluster; Adjusting the power fluctuation range in the basic prediction parameters based on the correlation coefficient includes: The power fluctuation range is determined based on the power fluctuation variance in the basic prediction parameters; Using the initial power fluctuation range corresponding to the power fluctuation variance in the basic prediction parameters as a benchmark, the adjustment direction is determined based on the positive or negative value of the correlation coefficient, specifically including: If the correlation coefficient is positive, the initial power fluctuation range is adjusted by expanding it proportionally to the absolute value of the correlation coefficient; If the correlation coefficient is negative, the initial power fluctuation range is adjusted by reducing it proportionally to the absolute value of the correlation coefficient; Among them, the magnitude of the adjustment is positively correlated with the absolute value of the correlation coefficient. The larger the absolute value of the correlation coefficient, the greater the proportion of expansion or contraction of the fluctuation range. Based on the adjusted basic prediction parameters and power fluctuation range, a power prediction curve for the current prediction period is generated. The curve includes the predicted power value at each time node and the power fluctuation range corresponding to that node.

[0015] The beneficial effects of this invention are as follows: By simultaneously extracting and integrating the first dynamic feature of meteorological dynamic characteristics and the second dynamic feature of associated equipment operating status, a comprehensive and accurate capture of power influencing factors is achieved, breaking through the limitations of traditional single static data. In sample processing, min-max standardization and discretization coding are used to unify the dimensions for different features. Combined with a similarity verification mechanism and a "double threshold" dual screening strategy, low-quality and abnormal samples are effectively eliminated, ensuring the reliability of the candidate sample set. The clustering stage incorporates seasonal and intraday time dimensions, and the K-means algorithm is used to form typical power change patterns adapted to different time scenarios. During prediction, the Pearson correlation coefficient is used to establish the correlation between historical power fluctuations and real-time irradiance changes, adaptively adjusting the power fluctuation range. This ensures the prediction accuracy of core parameters such as power mean and peak times, and accurately reflects the power fluctuation range under real-time operating conditions. Ultimately, this method significantly improves the accuracy and stability of real-time power prediction, providing a reliable decision-making basis for power grid dispatch, helping to reduce curtailment of solar power and reduce the impact of photovoltaic output fluctuations on the voltage and frequency stability of the distribution network. It provides strong support for the safe and efficient operation of the power grid under high proportion of new energy grid connection, with outstanding technical value and economic and social benefits. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of a real-time power prediction method for a distributed photovoltaic power station, as provided in one embodiment of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Example, refer to Figure 1 As an embodiment of the present invention, a real-time power prediction method for a distributed photovoltaic power station is provided, comprising the following steps: Step S100: Simultaneously extract the first dynamic feature and the second dynamic feature from the basic data; Step S200: Construct a mapping database of historical features and historical power data based on historical data, and calculate the similarity between the first dynamic feature and the second dynamic feature and the corresponding historical data in the mapping database to obtain the first similarity and the second similarity. Step S300: The historical data is divided into multiple related entries of historical features and historical power data according to time period and operating condition, and each related entry corresponds to a historical sample; Historical samples with similarity higher than a preset threshold are selected from the first and second similarity scores, and the historical power data distribution features corresponding to the historical samples are extracted. Step S400: Make a prediction based on the distribution characteristics of historical power data.

[0019] In one embodiment, step S100 is first executed to synchronously extract the first dynamic feature and the second dynamic feature from the basic data. Then, step S200 is performed to construct a mapping database corresponding to historical features and historical power data based on historical data. The previously extracted first and second dynamic features are then compared with the corresponding historical data in the mapping database to calculate the first and second similarities. Subsequently, step S300 is advanced, where the historical data is first divided into multiple associated entries including historical features and historical power data according to time periods and operating conditions. (Specific splitting rules include: time periods are divided into a two-layer division of "season + intraday sunshine duration," and the season is divided into spring (March-May)). The solar irradiance is divided into four periods: summer (June-August), autumn (September-November), and winter (December-February). Intraday sunshine hours are categorized according to photovoltaic power variation: morning start-up period (7:00-9:00, irradiance gradually increases), midday peak period (9:00-15:00, stable irradiance and high power), and evening decline period (15:00-18:00, irradiance gradually decreases). Operating conditions are categorized by "weather type + photovoltaic power station equipment operating status." Weather type is determined based on historical meteorological data's solar irradiance thresholds: sunny (irradiance ≥ 800W / ㎡), cloudy (irradiance ≤ 300W / ㎡ < 800W / ㎡), overcast (irradiance ≤ 300W / ㎡), and rainy. (With precipitation and irradiance less than or equal to 200W / ㎡), the operating status of photovoltaic power station equipment is determined based on historical inverter current fluctuations and module temperature differences: normal operating condition (current fluctuation less than or equal to 0.5A, module-ambient temperature difference less than or equal to 5℃), slightly abnormal operating condition (0.5A less than current fluctuation less than or equal to 1A, 5℃ less than temperature difference less than or equal to 8℃), and fault operating condition (current fluctuation greater than or equal to 1A, temperature difference greater than or equal to 8℃). When breaking down the equipment, first form basic groups based on "season-intraday time period" (e.g., "summer-midday peak period," "winter-morning start-up period"). Then, within each basic group, further subdivide according to "weather type-photovoltaic power station equipment operating status" to ultimately form... The unique dimension of the associated entries, such as the entry "Summer - Midday Peak Period - Sunny Day - Normal Photovoltaic Power Station Equipment", corresponds to the packaged historical first dynamic feature, historical second dynamic feature and historical power data of the same period under this scenario (each entry is an independent historical sample), and each associated entry corresponds to a historical sample. Then, historical samples with both first similarity and second similarity higher than the preset threshold are selected from these historical samples (abnormal samples that still exceed the reasonable range after recalculation are simultaneously removed during the screening). The historical power data distribution features corresponding to these selected samples are extracted. Finally, step S400 is implemented to complete the prediction of the real-time power of the distributed photovoltaic power station based on the extracted historical power data distribution features.

[0020] To obtain the basic data for power prediction of photovoltaic power plants, the basic data includes real-time weather forecasts and equipment status data of photovoltaic power plants; Real-time weather forecasts include solar irradiance data, ambient temperature data, and cloud cover data; Photovoltaic power plant equipment status data includes inverter operating parameter data and photovoltaic module temperature data; Extracting the first dynamic feature from the basic data specifically includes: Set time intervals and extract the changing trends of solar irradiance data over three consecutive time intervals. The criteria for determining a trend of change include: Set the amplitude threshold 'a' and fluctuation threshold 'b' for solar irradiance data; If, in the solar irradiance data of three consecutive time intervals, the two adjacent increases of the value of the later time interval relative to the value of the previous time interval are both greater than 'a', then it is determined to be the increment. The increment is the ratio of the cumulative increase value of the three time intervals to the initial time interval value. The ratio is the magnitude value corresponding to the increment trend. If, in the solar irradiance data of three consecutive time intervals, the two adjacent decreases of the value of the later time interval relative to the value of the previous time interval are both greater than 'a', then it is determined to be a decreasing amplitude. The decreasing amplitude is the ratio of the cumulative decrease value of the three time intervals to the initial interval value. The ratio is the amplitude value corresponding to the decreasing trend. If the fluctuation range of adjacent interval values ​​of solar irradiance data in three consecutive time intervals is less than or equal to b, it is determined to be a stable state. The trend of change includes the magnitude of increase, the magnitude of decrease, and the stable state. a and b are positive numbers, and the magnitude is expressed as the absolute value of the difference between the solar irradiance data of the later time interval and the solar irradiance data of the previous time interval. Record the extreme temperature differences that occur within the time interval and the duration of the extreme temperature differences; Monitor the abrupt changes in cloud cover levels in cloud cover data, and extract the instantaneous response characteristics of solar irradiance data from a preset time period before the abrupt change to a preset time period after the abrupt change. Instantaneous response characteristics include response delay duration and the proportion of change in solar irradiance data; The first dynamic feature is formed by combining the changing trend of solar irradiance data, the extreme temperature difference of ambient temperature data, the duration of the extreme temperature difference, and the instantaneous response characteristics of solar irradiance data at the moment of abrupt change in cloud cover data.

[0021] In one embodiment, two types of core basic data need to be acquired first. One type is real-time weather forecast data (including solar irradiance data, in W / m², with sampling frequency consistent with subsequent time intervals, ambient temperature data, in °C, and cloud cover data, using the industry standard cloud cover level classification of 1-10, where level 1 is clear sky and level 10 is full cloud) acquired synchronously through the power plant data acquisition system (SCADA). The other type is photovoltaic power plant equipment status data (including inverter operating parameter data and photovoltaic module temperature data, in °C) also acquired through the system. It is necessary to ensure that the timestamps of the two types of data are consistent. Then, parameters are preset, setting the data extraction time interval to 5 minutes (based on the principle that it can capture significant changes in irradiance in a timely manner while avoiding data redundancy due to excessively short intervals, which is the mainstream application interval in the industry), and setting the amplitude threshold 'a' of solar irradiance data to 50 W / m² (based on long-term operational statistics). The fluctuation threshold b is 20W / ㎡ (based on the fact that fluctuations within this range are at the environmental noise level and do not change the power change pattern, so it can be judged as stable). The amplitude is defined as the absolute value of the difference between the irradiance data of the next time interval and the previous time interval. Then, the solar irradiance change trend is extracted. Based on the irradiance data of three consecutive intervals at 5-minute intervals, for example, when the data of three consecutive intervals are 800W / ㎡, 860W / ㎡, and 920W / ㎡, the adjacent increase amplitude is 60W / ㎡ greater than 50W / ㎡, which is judged as an increasing amplitude, with an amplitude value of (920-800) / 800=0.15. When the data are 800W / ㎡, 740W / ㎡, and 680W / ㎡, the adjacent decrease amplitude is 60W / ㎡ greater than 50W / ㎡, which is judged as a decreasing amplitude, with an amplitude value of (800-680) / 800=0.15. When the data is 800W / ㎡, 810W / ㎡, and 815W / ㎡, the adjacent fluctuation amplitude is less than or equal to 20W / ㎡, which is judged as a stable state. At the same time, the ambient temperature characteristics are extracted, and the extreme temperature difference and corresponding duration within a 5-minute interval are recorded. For example, when the temperature is 25℃, 28℃, and 26℃ within a certain interval, the extreme temperature difference is 3℃ and the duration is 5 minutes. Then, the instantaneous response characteristics of irradiance associated with cloud cover abrupt changes are extracted. The abrupt change time of cloud cover (such as 10:00 when the cloud cover changes from level 2 (partially cloudy) to level 6 (partially cloudy)) is monitored. Irradiance data are taken from a preset time period of 5 minutes before the abrupt change (9:55) to a preset time period of 5 minutes after the abrupt change (10:05). (The basis for the preset time period of 5 minutes before and after the abrupt change includes the fact that irradiance fluctuations caused by cloud movement often show minute-level characteristics.) The time window can fully cover the baseline irradiance state before the mutation and the entire response process after the mutation. It also matches the 5-minute data extraction interval, avoiding the omission of key response information due to an excessively short window or the introduction of irrelevant interference data due to an excessively long window, meeting the need for refined capture of cloud disturbances in ultra-short-term power forecasting. The response delay duration (e.g., cloud cover mutation at 10:00, irradiance starting to decrease at 10:01, delay of 1 minute) and change ratio (e.g., initial irradiance 800 W / m², stabilized at 500 W / m², ratio (800-500) / 800=0.375) are calculated. Finally, the solar irradiance change trend (including the magnitude of increase / decrease), the extreme temperature difference and duration of the environment, and the instantaneous irradiance response characteristics (response delay duration, change ratio) at the moment of cloud cover mutation are combined to form the first dynamic feature.

[0022] Extracting the second dynamic feature from the basic data specifically includes: Using a preset time as the sampling unit, extract the fluctuation frequency of the inverter's operating parameter output current data within a fixed time period sliding window; The criteria for determining fluctuations include: Set a current fluctuation data threshold. When the absolute value of the difference between the current fluctuation data of adjacent sampling units is greater than the current fluctuation data threshold, it is judged as a fluctuation. Calculate the direction of change of the difference between photovoltaic module temperature data and ambient temperature data over a continuous period of time; The criteria for determining the direction of change include: Set a threshold for temperature difference changes; If the difference between the final value and the initial value is greater than the temperature difference change threshold within a continuous period, it is determined to be a positive growth. If the difference between the final value and the initial value is less than the temperature difference change threshold within a continuous period, it is judged as a negative decrease. If, within a continuous period of time, the absolute value of the difference between the final value and the initial value is less than or equal to the temperature difference change threshold, it is determined to be basically stable. The directions of change include positive growth, negative decrease, and basic stability; The time synchronization characteristics of the changing trends of current fluctuation data and solar irradiance data in the related inverter operating parameter data; The second dynamic feature is formed by combining the fluctuation frequency of current fluctuation data in inverter operating parameter data, the direction of change of the difference between photovoltaic module temperature data and ambient temperature data, and the time synchronization characteristics of the change trend of current fluctuation data and solar irradiance data in inverter operating parameter data.

[0023] In one embodiment, the operation is carried out based on the acquired basic data (including inverter operating parameter data and photovoltaic module temperature data from photovoltaic power plant equipment status data, and solar irradiance data from real-time weather forecast data, and all data timestamps are consistent with the first dynamic feature extraction): First, a 1-minute sampling unit is preset (based on the fact that inverter output current responds quickly to changes in load and irradiance; 1-minute sampling can accurately capture short-term fluctuations and avoid data redundancy, conforming to the industry's conventional acquisition frequency). A 30-minute fixed-period sliding window is used (based on the fact that 30 minutes is a typical statistical period for short-term power changes in photovoltaic power plants, reflecting the phased current fluctuation pattern), and... A current fluctuation data threshold of 0.5A is set (based on the inverter's rated output current range and long-term data statistics; fluctuations below this value are considered normal equipment noise, with an impact on power of less than 3%). A valid fluctuation is defined as the absolute value of the current difference between adjacent 1-minute sampling units exceeding 0.5A. The total number of valid fluctuations within a 30-minute window is the fluctuation frequency. For example, if the current difference exceeding 0.5A occurs 6 times within a 30-minute window, the fluctuation frequency is 6 times. A 20-minute continuous time period is then selected (based on the time lag between the module and ambient temperature heat exchange; 20 minutes can fully reflect the trend of temperature difference changes and avoid short-term random fluctuation interference). A threshold of 1 is set. ℃ is the temperature difference threshold (set based on the fact that temperature differences below this value have negligible impact on the photoelectric conversion efficiency of the module). If the final value of the temperature difference between the module and the ambient temperature within 20 minutes (temperature difference at the 20th minute) is greater than 1℃ (e.g., initial 5℃, final 7℃), it is considered a positive increase. If the difference is less than -1℃ (e.g., initial 5℃, final 3℃), it is considered a negative decrease. If the absolute value of the difference is less than or equal to 1℃ (e.g., initial 5℃, final 5.5℃), it is considered basically stable. Then, the time synchronization characteristics of inverter current fluctuation data and solar irradiance change trends are correlated, using the trigger time of the irradiance change trend (e.g., the first time interval of the incremental increase) as the starting point. Using the start time as a reference, the time difference (absolute value) between the current fluctuation occurrence time and the reference time is calculated. A 2-minute threshold is set for synchronicity judgment (based on the fact that the physical delay of irradiance change transmitted to inverter current is usually less than or equal to 2 minutes; exceeding this time may indicate equipment malfunction). If the time difference is less than or equal to 2 minutes (e.g., irradiance triggers an increase at 10:00, current fluctuates at 10:01), it is judged as a synchronicity feature (marked as "1"). If it exceeds 2 minutes, it is a non-synchronicity feature (marked as "0"). Finally, the frequency of inverter current fluctuation, the direction of change of component and ambient temperature difference, and the time synchronicity feature of current fluctuation - irradiance change are combined to form the second dynamic feature.

[0024] The similarity between the first dynamic feature and the historical data in the mapping database is calculated to obtain the first similarity, specifically including: Standardization of the first dynamic feature includes: The amplitude of the trend in solar irradiance data, the extreme temperature difference in ambient temperature data, the duration corresponding to the extreme temperature difference in ambient temperature data, the response delay of solar irradiance data at the moment of abrupt change in cloud cover data, and the change ratio of solar irradiance data are all converted to the same preset numerical range, specifically including: The min-max standardization algorithm is adopted, which uses the minimum value of the historical first dynamic feature in the mapping database as the lower limit of the benchmark and the maximum value of the historical first dynamic feature as the upper limit of the benchmark. By calculating the difference between the original value of each first dynamic feature and the lower limit of the benchmark, and the proportional relationship between the difference between the upper limit of the benchmark and the lower limit of the benchmark, the original value of each first dynamic feature is scaled to a preset numerical range, thereby achieving the dimensional unification of multi-dimensional features. The raw values ​​of the first dynamic feature include the amplitude of the trend of solar irradiance data, the extreme temperature difference of ambient temperature data, the duration of the extreme temperature difference of ambient temperature data, the response delay of solar irradiance data at the moment of abrupt change in cloud cover data, and the change ratio of solar irradiance data. The Euclidean distance algorithm is used to calculate the spatial distance between the standardized first dynamic feature and the historical first dynamic feature stored in the mapping database one by one. Then, the normalized distance value is subtracted from c as the first similarity, and a first reasonable numerical range of the first similarity is set. The first reasonable numerical range is a closed interval. The normalized distance value is the ratio of the spatial distance to the maximum spatial distance between the first historical dynamic features in the mapping database.

[0025] In one embodiment, the operation is performed based on the extracted current first dynamic feature (including the amplitude of the solar irradiance change trend, the extreme temperature difference of the ambient temperature, the duration of the extreme temperature difference of the ambient temperature, the response delay and change ratio of solar irradiance at the moment of cloud cover change) and a pre-constructed historical feature-historical power data mapping database (the dimensions of the historical first dynamic feature stored in the database are completely consistent with the current first dynamic feature): First, the original value of the current first dynamic feature is processed using the min-max standardization algorithm, with the preset value range being [0,1] (the basis for setting: [0,1] is a commonly used standardization range in the industry, which can avoid calculation deviations caused by large differences in the values ​​of different dimension features, and facilitates subsequent similarity comparison). During the calculation, the minimum value of the corresponding dimension of the historical first dynamic feature in the mapping database is used as the lower limit of the benchmark, and the maximum value is used as the upper limit of the benchmark. The original value is scaled to [0,1] by "standardized value = (original value - lower limit of benchmark) / (upper limit of benchmark - lower limit of benchmark)". The interval achieves dimensional uniformity. For example, the baseline lower limit of the historical irradiance change trend amplitude value in the database is 0.05, and the baseline upper limit is 0.3. When the original value of the current amplitude value is 0.15, the standardized value is (0.15-0.05) / (0.3-0.05)=0.4. The baseline lower limit of the historical ambient temperature extreme value temperature difference is 1℃, and the baseline upper limit is 8℃. When the original value of the current temperature difference is 4℃, the standardized value is (4-1) / (8-1)≈0.43. Then, the Euclidean distance algorithm is used to calculate the spatial distance between the standardized current first dynamic feature and each historical first dynamic feature in the database. Then, the normalized distance value is processed according to "normalized distance value = current spatial distance / maximum spatial distance between all historical first dynamic features in the database" (the setting basis is: by normalizing the maximum spatial distance, the distance value can be constrained to the [0,1] interval, which is convenient for subsequent similarity calculation). For example, the maximum spatial distance between historical first dynamic features in the database is 2.5, and the current spatial distance is 1. At 25, the normalized distance value = 1.25 / 2.5 = 0.5. Then, c is set to 1 (the basis for setting c to 1 is that the first similarity and the normalized distance value have an inverse linear relationship, and the similarity result falls in the [0,1] interval, which conforms to the industry's similarity representation convention). The first similarity is obtained by calculating "first similarity = c - normalized distance value". For example, in the above example, the first similarity = 1 - 0.5 = 0.5. Finally, the first reasonable numerical interval of the first similarity is set to the closed interval [0.3,1] (the basis for setting is that based on the database's historical similarity statistics, when the similarity is lower than 0.3, the historical samples have weak correlation with the current features and insufficient reference value, while when it is greater than or equal to 0.3, the correlation is relatively strong). This is used to verify the validity of the similarity results. This method eliminates the bias caused by the difference in feature units by standardizing and unifying the units. Euclidean distance accurately measures the difference in feature space. The reasonable interval is used to screen effective results, laying the foundation for subsequent selection of highly correlated historical samples and improving the accuracy of power prediction.

[0026] The similarity between the second dynamic feature data and the historical data in the mapping database is calculated to obtain the second similarity, specifically including: Performing discretization encoding on the second dynamic feature includes: The frequency of inverter operating parameter data and output current data fluctuation within a preset time sliding window is divided into multiple levels according to preset value ranges, specifically including: Based on the minimum and maximum values ​​of the fluctuation frequency of the current data in the historical data of the mapping database, the first interval is set as the first proportional threshold where the fluctuation frequency is less than or equal to the average value of the current data, corresponding to the first level; The second interval is defined as the first proportional threshold where the fluctuation frequency is greater than the average current data and less than or equal to the second proportional threshold where the fluctuation frequency is less than or equal to the average current data, corresponding to the second level; The third interval is defined as the second proportional threshold where the fluctuation frequency is greater than the average current data, corresponding to the third level; The direction of change of the difference between photovoltaic module temperature data and ambient temperature data within a preset time period is expressed numerically, including: Positive growth corresponds to the first preset value in the temperature direction, basically stable corresponds to the second preset value in the temperature direction, and negative decrease corresponds to the third preset value in the temperature direction. The time synchronization characteristics of the current fluctuation data of inverter operating parameters and the changing trend of solar irradiance data are represented by numerical values, and the time difference corresponding to the characteristics is used as the basis for judgment. The time difference is the absolute value of the difference between the time when the inverter current fluctuation data occurs and the time when the solar irradiance change trend is triggered. The preset condition is that the time difference is less than or equal to the first time threshold. When the time difference meets the preset conditions, the feature corresponds to the synchronization - first preset value; When the time difference does not meet the preset conditions, the feature corresponds to the second preset value of synchronicity, and the first preset value of synchronicity is different from the second preset value of synchronicity. The Hamming distance algorithm is used to calculate the number of bit differences between the discretized second dynamic feature data and the historical second dynamic features stored in the mapping database. The ratio of the number of bit differences to the total number of bits of the feature is then used as the second similarity. A second reasonable numerical range for the second similarity is set, which is a closed interval.

[0027] In one embodiment, the operation needs to be carried out based on the extracted current second dynamic features (including inverter current fluctuation frequency, direction of change of photovoltaic module and ambient temperature difference, and time synchronization feature of current fluctuation-irradiance change) and a pre-built historical feature-historical power data mapping database (the dimensions of the historical second dynamic features in the mapping database are completely consistent with the current ones): First, the second dynamic features are discretized and encoded. The current fluctuation frequency is graded based on the minimum, maximum and average values ​​of the historical current fluctuation frequency in the database. Intervals are divided according to a first proportional threshold of 1 / 2 and a second proportional threshold of 3 / 2 (the setting basis: the average value reflects the normal fluctuation level and conforms to the actual operating law). The first interval is less than or equal to the historical average. Value × 1 / 2 corresponds to Level 1 (marked 1); Level 2 corresponds to the second interval where the historical average × 1 / 2 is less than the frequency and less than or equal to the historical average × 3 / 2 (marked 2); Level 3 corresponds to the third interval where the historical average × 3 / 2 is greater than the historical average (marked 3). For example, if the historical average is 6 times, the current frequency of 7 times is encoded as 2. The direction of temperature difference change is numerically set with the first preset value of 1 (positive growth), the second preset value of 0 (basically stable), and the third preset value of -1 (negative decrease) (setting basis: simple integer encoding is easy to calculate and clearly distinguish). If the current is positive growth, it is encoded as 1. The time synchronization characteristic is numerically set with the first time threshold of 2 minutes (setting basis: the physical delay of irradiance change being transmitted to current is usually less than or equal to 2 minutes). (The clock, consistent with the second dynamic feature extraction threshold), Synchronization - First preset value is set to 1 (time difference less than or equal to 2 minutes), Synchronization - Second preset value is set to 0 (time difference greater than 2 minutes) (Setting basis: First, to continue the unified logic of "simple integers" in discretization encoding, consistent with the integer encoding of temperature difference change direction (1, 0, -1) and current fluctuation frequency (1, 2, 3), avoiding the introduction of complex values ​​to increase computational redundancy; Second, the synchronization feature is a binary attribute (meets preset conditions / does not meet preset conditions), 1 and 0 are the industry's conventional symbols for "valid / compliant" and "invalid / non-compliant"; Third, the numerical range of 1 and 0 is adapted to the subsequent Hamming distance calculation, which can ensure the results when the bit difference number is statistically analyzed. The similarity is calculated to be an integer, ensuring the similarity result falls within the [0,1] interval (consistent with the first similarity logic). For example, a 1-minute time difference is encoded as 1. The Hamming distance algorithm is then used to calculate the number of bit differences between the current and historical features (the number of bits that differ in the corresponding dimension encoding). b is set to 1 (based on the principle that the similarity falls within the [0,1] interval, consistent with the first similarity logic). The second similarity is calculated using "second similarity = 1 - (number of bit differences / total number of bits in the feature)". For example, the similarity between the current encoding [2,1,1] and the historical encoding [2,0,1] (number of bit differences 1, total number of bits 3) is approximately 0.67. Finally, the second reasonable value interval is set to the closed interval [0.4,1] (based on historical statistics, below 0).(Since samples in group 4 have weak correlation and are difficult to reflect the impact of equipment on power, this method is used to verify the validity of the results. It solves the problem of heterogeneous feature calculation through discretization coding, accurately measures differences using Hamming distance, and selects valid samples within reasonable intervals, providing reliable support for improving the accuracy of power prediction at the photovoltaic power station equipment level.)

[0028] Establish a first or second similarity verification mechanism, specifically including: If the first similarity or second similarity of any historical sample exceeds the corresponding first reasonable value range or second reasonable value range, then feature data of the same type as the historical sample within a preset time range in the mapping database will be called to recalculate and obtain the first similarity or second similarity. Historical samples include historical first dynamic features, historical second dynamic features, and associated historical power data.

[0029] In one embodiment, the operation is carried out based on the calculated first and second similarity results of historical samples, a preset first reasonable numerical range (e.g., a closed interval [0.3,1]) and a second reasonable numerical range (e.g., a closed interval [0.4,1]), and a mapping database storing historical samples (including historical first dynamic features, historical second dynamic features, and associated historical power data). First, the verification triggering conditions are defined. If the first similarity of any historical sample exceeds the first reasonable numerical range (e.g., the calculated first similarity is 0.25, lower than the lower limit of [0.3,1]), or the second similarity exceeds the second reasonable numerical range (e.g., the calculated second similarity is 0.35, lower than the lower limit of [0.4,1]), the verification process is immediately triggered. Subsequently, the mapping database is called within a preset time range (the preset time range is set to the last 3 months, based on recent historical data and current power plant equipment operating conditions). (The regional meteorological conditions are closer, which can reduce the feature differences caused by the time span and improve the reliability of the recalculation results.) The feature data of the same type as the abnormal historical sample are used. The "same type" is defined as the season, intraday time, and weather type label of the historical sample being completely consistent with the abnormal sample (for example, if the abnormal sample is summer, noon, and sunny, then the historical feature data of the same type in the database is called). Then, according to the calculation logic of the first similarity (min-max normalization + Euclidean distance) or the second similarity (discretization encoding + Hamming distance), the first similarity or second similarity corresponding to the historical sample is recalculated using the called feature data of the same type (for example, if the anomaly is the first similarity, then the first dynamic feature of the same type is called, and after re-normalization, the Euclidean distance and the first similarity are calculated). Finally, the recalculated similarity result is used as the effective similarity value of the historical sample for subsequent screening of highly correlated historical samples.

[0030] Extract historical samples with similarity scores higher than a preset threshold from the first and second similarity scores, and extract the historical power data distribution features corresponding to the historical samples, specifically including: Setting preset thresholds includes setting a third preset threshold and a fourth preset threshold. The value range of the third preset threshold is from the median to the upper limit of the first reasonable numerical range of the first similarity, and the value range of the fourth preset threshold is from the median to the upper limit of the second reasonable numerical range of the second similarity. The historical samples in the mapping database are subjected to double screening. The conditions for double screening include a first similarity higher than a third preset threshold and a second similarity higher than a fourth preset threshold. If the dual screening criteria are met, historical samples are retained to form a candidate sample set.

[0031] In one embodiment, the operation is performed based on historical samples (including historical first dynamic features, historical second dynamic features, and associated historical power data) in the mapping database that have completed similarity calculation and verification, a preset first reasonable value interval (e.g., closed interval [0.3,1]) and a second reasonable value interval (e.g., closed interval [0.4,1]). First, a preset threshold is set, where the third preset threshold corresponds to the first similarity, and the value is taken from the median to the upper limit of the first reasonable value interval, specifically set to 0.65 (the setting basis is: the median of the first reasonable interval [0.3,1] is (0.3+1) / 2=0.65, and the first similarity above the median can ensure that the historical samples are strongly correlated with the current first dynamic features and exclude interference from low-correlation samples). The fourth preset threshold corresponds to the second similarity, and the value is taken from the median to the upper limit of the second reasonable value interval, specifically set to 0. 7 (Setting basis: The median of the second reasonable interval [0.4,1] is (0.4+1) / 2=0.7. The second similarity above the median can ensure that the historical samples are strongly correlated with the current second dynamic feature, which meets the high correlation requirements of the device dimension.) Then, a double screening is performed on the historical samples in the mapping database. The screening condition is that "the first similarity of the historical sample is higher than the third preset threshold of 0.65, and the second similarity is higher than the fourth preset threshold of 0.7". For example, if the first similarity calculation result of a historical sample is 0.72 (higher than 0.65) and the second similarity calculation result is 0.75 (higher than 0.7), then the double screening condition is met. Finally, all historical samples that meet the double screening condition are retained to form a candidate sample set, and the historical power data corresponding to each historical sample in the candidate sample set is extracted simultaneously to prepare for the subsequent extraction of the distribution features of historical power data.

[0032] Historical power data corresponding to each historical sample is extracted from the candidate sample set, and the distribution characteristics of the historical power data are calculated. The distribution characteristics include the power mean, the time of occurrence of the power peak, the power decay coefficient over time, and the power fluctuation variance. Clustering of the historical power data distribution characteristics in the candidate sample set along the time dimension, specifically including: The K-means clustering algorithm was used, with the season and intraday time period of the historical samples as the basis for clustering and grouping by time dimension. Intraday time period includes morning, noon and afternoon. First, the historical power data distribution characteristics in the candidate sample set are preliminarily divided into several groups according to the time dimension; Then, the distribution characteristics of the power mean, power peak occurrence time, power decay coefficient over time, and power fluctuation variance of each historical sample in each group are used as clustering input variables. The number of clusters is set to de. Through iterative calculation, the sum of squared Euclidean distances from each historical sample in each group to the corresponding cluster center is minimized to complete the clustering. The cluster center is the core feature virtual sample point of the corresponding cluster during the clustering process in each group, including feature values ​​with the same dimension as the clustering input variable. The feature values ​​include the power mean, the time of occurrence of the power peak, the power decay coefficient over time, and the power fluctuation variance. Each feature value is the mean of the feature values ​​corresponding to all historical samples in the cluster. One or more power feature clusters are obtained, and each power feature cluster corresponds to a typical power change pattern.

[0033] Prediction based on the distribution characteristics of historical power data specifically includes: Determine the time attribute parameters for the current time period to be predicted. The time attribute parameters include the season to which the time period belongs, the intraday time period, and the weather type label. From the power feature clusters, the target clusters with the highest matching degree with the time attribute parameters are selected. The criteria for determining the highest matching degree include: If the historical time attribute parameters of the power feature cluster completely overlap with the seasonal, intraday time period, and weather type labels of the current time attribute parameters, it is determined to be the highest matching degree. If there is no power feature cluster whose historical time attribute parameters completely overlap with the current time attribute parameters in terms of season, intraday time period, and weather type labels, then the cluster with the most overlapping dimensions in terms of season, intraday time period, and weather type labels is selected as the candidate with the highest matching degree. If the number of overlapping dimensions is the same, clusters that overlap in both season and intraday time period are preferred. Extract typical power change patterns of the target cluster. Typical power change patterns refer to standardized templates that represent the common laws of cluster power change after statistically aggregating the power distribution characteristics of all historical samples within the target cluster. The average power, peak power occurrence time, power attenuation coefficient over time, and power fluctuation variance of typical power change patterns are used as the basic prediction parameters.

[0034] In one embodiment, the operation is carried out based on the established candidate sample set (including historical samples and associated historical power data): First, the historical power data corresponding to each historical sample is extracted from the candidate sample set, and the distribution characteristics (including power mean, the time of occurrence of power peak, power attenuation coefficient over time and power fluctuation variance, such as a historical sample with a power mean of 750kW, peak time of 12:20, and attenuation coefficient of 0) are calculated.0.15 / h, fluctuation variance 45kW²), then these distribution characteristics are clustered by time dimension using K-means clustering. During clustering, the season (spring, summer, autumn, and winter) and intraday time (8:00-11:00 AM, 11:00-2:00 PM, 2:00-5:00 PM) of the historical samples are used as the initial grouping criteria (e.g., first dividing the samples into "summer noon" and "autumn morning" groups). Then, the four distribution characteristics of the samples within each group are used as clustering input variables. The number of clusters is set to a range of 3-8 (d=3, e=8, set based on the seasonality and intraday regularity of photovoltaic power station power changes; 3 clusters can cover " The system employs a basic model of "stable, peak, and low-peak" patterns, with eight clusters to further differentiate patterns under various weather conditions. This balances pattern discriminability with computational efficiency, avoiding inaccurate pattern merging due to too few clusters or overfitting due to too many. Clustering is achieved by iteratively minimizing the sum of squared Euclidean distances from each sample within a cluster to its corresponding cluster center. The cluster center is defined as the mean of the corresponding feature values ​​of all samples within the cluster (e.g., if a cluster contains 5 samples with mean power values ​​of 720kW, 740kW, 760kW, 750kW, and 730kW, then the mean power value at the cluster center is 740kW). This results in one or more power feature clusters, each corresponding to a typical power variation pattern (e.g., ...). The prediction process, which identifies a specific time period as a "summer midday sunny peak type," is based on historical power data distribution characteristics. First, the time attribute parameters for the current prediction period are determined (including the season, intraday time period, and weather type label, such as summer, 11:00-14:00, sunny). Then, the target cluster with the highest matching degree is selected from the power feature clusters. The matching criterion is: if the historical time attribute parameters (e.g., "summer midday sunny") of a cluster completely overlap with the current parameters, it is considered the highest matching degree. If no completely overlapping cluster exists, the cluster with the most overlapping dimensions among season, intraday time period, and weather type label is selected (e.g., the current value is "summer midday cloudy"). The process prioritizes "Summer Midday Cloudy" (overlapping two dimensions) over "Autumn Midday Overcast" (overlapping two dimensions), and then follows subsequent rules. If the number of overlapping dimensions is the same, clusters that overlap in both season and intraday time (e.g., "Summer Midday Cloudy" and "Autumn Midday Overcast" both overlap in two dimensions, so "Summer Midday Cloudy" is prioritized) are selected. Then, the typical power variation pattern of the target cluster is extracted (a standardized template formed by statistically aggregating the power distribution characteristics of all samples within the cluster, such as a mean power of 800kW, peak time of 12:30, attenuation coefficient of 0.02 / h, and fluctuation variance of 50kW²). These four features of the pattern are used as the basic prediction parameters.

[0035] Calculate the correlation coefficient between the variance of historical power data in the target cluster and the amplitude of the changing trend of solar irradiance data in the current first dynamic feature, specifically including: The Pearson correlation coefficient algorithm is adopted, with the historical power data fluctuation variance of all historical samples in the target cluster as the first variable set, and the amplitude value of the solar irradiance data change trend in the historical first dynamic feature corresponding to each historical sample as the second variable set. The basic correlation coefficient at the historical level is obtained by calculating the ratio of the covariance of the first set of variables and the second set of variables to the product of the standard deviations of the first set of variables and the second set of variables. Then, substitute the magnitude of the solar irradiance data change trend in the current first dynamic feature into the fitting relationship corresponding to the basic correlation coefficient to determine the final correlation coefficient; The fitting relationship is constructed based on the statistical regularities of the first and second variable sets within the target cluster; Adjusting the power fluctuation range in the basic forecast parameters based on the correlation coefficient includes: The power fluctuation range is determined based on the power fluctuation variance in the basic prediction parameters; Using the initial power fluctuation range corresponding to the power fluctuation variance in the basic forecast parameters as a benchmark, the adjustment direction is determined based on the positive or negative value of the correlation coefficient, specifically including: If the correlation coefficient is positive, the initial power fluctuation range is adjusted by increasing the proportion of the absolute value of the correlation coefficient. If the correlation coefficient is negative, adjust the initial power fluctuation range by reducing it proportionally to the absolute value of the correlation coefficient; Among them, the magnitude of the adjustment is positively correlated with the absolute value of the correlation coefficient. The larger the absolute value of the correlation coefficient, the greater the proportion of expansion or contraction of the fluctuation range. Based on the adjusted basic forecast parameters and power fluctuation range, a power forecast curve for the current forecast period is generated. The curve includes the forecast power value for each time node and the corresponding power fluctuation range.

[0036] In one embodiment, the operation is performed based on a determined target cluster (including historical power data of historical samples within the cluster and corresponding historical first dynamic features) and the current first dynamic feature: First, the Pearson correlation coefficient algorithm is used to calculate the correlation coefficient. The historical power data fluctuation variance of all historical samples within the target cluster is used as the first variable set (e.g., the fluctuation variances of 5 historical samples are 45kW², 50kW², 48kW², 52kW², and 46kW², respectively). The amplitude value of the solar irradiance data change trend in the historical first dynamic feature corresponding to each historical sample is used as the second variable set (corresponding amplitude values ​​are 0.15, 0.2, 0.18, 0.22, and 0.16, respectively). This is achieved through the "basic" method. The correlation coefficient is calculated as follows: "Correlation coefficient = Covariance of the first variable set and the second variable set / (Standard deviation of the first variable set × Standard deviation of the second variable set)". This yields the historical baseline correlation coefficient (e.g., a baseline correlation coefficient of 0.7). Then, a fitting relationship is constructed based on the statistical regularities of the first and second variable sets within the cluster (using linear fitting, e.g., a fitting relationship of y = 0.7x + 0.02, based on regression analysis of historical data within the cluster, ensuring a goodness of fit R² greater than or equal to 0.8 to guarantee reliable correlation fitting). The amplitude of the current trend in solar irradiance data (e.g., a current amplitude of 0.2) is substituted into the fitting relationship to obtain the final correlation coefficient: 0.7 × 0.2 + 0.02 = 0. 0.16. Next, based on this correlation coefficient, the power fluctuation range in the basic prediction parameters is adjusted. The initial power fluctuation range is determined by the power fluctuation variance of the basic prediction parameters (based on the following setting: the initial range is taken as ±1 standard deviation corresponding to the square root of the fluctuation variance, which conforms to the conventional distribution law of power fluctuation at the statistical level. For example, if the power fluctuation variance in the basic prediction parameters is 2500kW², the square root is 50kW, and the initial fluctuation range is ±50kW). The adjustment direction is determined by the positive or negative value of the correlation coefficient: if the correlation coefficient is positive (e.g., 0.16), the initial range is expanded proportionally to the absolute value of the correlation coefficient. The adjustment magnitude is positively correlated with the absolute value (the larger the absolute value, the larger the expansion ratio). In this example, the adjusted range = ±50×(1+0.16). 16) = ±58kW. If the correlation coefficient is negative (e.g., -0.12), the initial range is reduced proportionally to the absolute value (e.g., the adjusted range = ±50 × (1 - 0.12) = ±44kW). Finally, based on the adjusted basic prediction parameters (e.g., average power 800kW, peak time 12:30, attenuation coefficient 0.02 / h) and the adjusted power fluctuation range (±58kW), the power prediction curve for the current prediction period is generated. Each time node in the curve includes the predicted power value (e.g., predicted power 800kW at 12:00, predicted power 820kW at 12:30) and the corresponding fluctuation range (742-858kW in the 12:00 range, 762-878kW in the 12:30 range).

[0037] This invention achieves comprehensive and accurate capture of power influencing factors by simultaneously extracting and fusing first dynamic features of meteorological dynamics and second dynamic features of associated equipment operating status. It breaks through the limitations of traditional single static data. In sample processing, it employs min-max standardization and discretization coding to unify the dimensions for different features, combined with a similarity verification mechanism and a dual-threshold screening strategy to effectively eliminate low-quality and abnormal samples, ensuring the reliability of the candidate sample set. The clustering stage incorporates seasonal and intraday time-period dimensions, using the K-means algorithm to form typical power change patterns adapted to different time scenarios. During prediction, the Pearson correlation coefficient is used to establish a correlation between historical power fluctuations and real-time irradiance changes, adaptively adjusting the power fluctuation range. This ensures the prediction accuracy of core parameters such as power mean and peak times, while accurately reflecting the power fluctuation range under real-time operating conditions. Ultimately, this method significantly improves the accuracy and stability of real-time power prediction, providing reliable decision-making basis for grid dispatch, helping to reduce curtailment of solar power and mitigate the impact of photovoltaic output fluctuations on the voltage and frequency stability of the distribution network. It provides strong support for the safe and efficient operation of the grid under high-proportion renewable energy grid integration, demonstrating significant technical value and economic and social benefits.

[0038] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.

[0039] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A real-time power prediction method for a distributed photovoltaic power station, characterized in that, Includes the following steps: Step S100: Simultaneously extract the first dynamic feature and the second dynamic feature from the basic data; Step S200: Construct a mapping database of historical features and historical power data based on historical data, and calculate the similarity between the first dynamic feature and the second dynamic feature and the corresponding historical data in the mapping database to obtain the first similarity and the second similarity. Step S300: The historical data is divided into multiple related entries of historical features and historical power data according to time period and operating condition, and each related entry corresponds to a historical sample; Historical samples with similarity higher than a preset threshold are selected from the first similarity and the second similarity to form a candidate sample set, and the historical power data distribution features corresponding to the historical samples in the candidate sample set are extracted. The historical power data distribution characteristics in the candidate sample set are clustered according to the time dimension to obtain power feature clusters; Step S400: Make a prediction based on the distribution characteristics of the historical power data; The first dynamic feature is formed by combining the changing trend of solar irradiance data, the extreme temperature difference of ambient temperature data, the duration of the change, and the instantaneous response characteristics of solar irradiance data at the moment of abrupt change in cloud cover data. The second dynamic feature is formed by combining the fluctuation frequency of current fluctuation data in inverter operating parameter data, the direction of change of the difference between photovoltaic module temperature data and ambient temperature data, and the time synchronization characteristics of the change trend of current fluctuation data and solar irradiance data in inverter operating parameter data. Prediction based on the distribution characteristics of the historical power data specifically includes: Determine the time attribute parameters of the current time period to be predicted, including the season to which the time period belongs, the intraday time period, and the weather type label; The target cluster with the highest matching degree with the time attribute parameter is selected from the power feature clusters. The criteria for determining the highest matching degree include: If the historical time period attribute parameters corresponding to the power feature cluster completely overlap with the seasonal, intraday time period, and weather type labels of the current time attribute parameters, it is determined to be the highest matching degree; If there is no historical time period attribute parameter corresponding to the power feature cluster that completely overlaps with the seasonal, intraday time period and weather type label of the current time attribute parameter, then the cluster with the most overlapping dimensions among the seasonal, intraday time period and weather type label is selected as the candidate with the highest matching degree. If the number of overlapping dimensions is the same, clusters that overlap in both season and intraday time period are preferred. Extract the typical power change pattern of the target cluster. The typical power change pattern refers to the standardized template that can represent the common law of power change of the cluster after statistically aggregating the power distribution characteristics of all historical samples in the target cluster. The average power, peak power occurrence time, power attenuation coefficient over time, and power fluctuation variance of the typical power change pattern are used as the basic prediction parameters.

2. The real-time power prediction method for distributed photovoltaic power stations as described in claim 1, characterized in that: The basic data for power prediction of photovoltaic power plants are obtained, including real-time weather forecasts and equipment status data of photovoltaic power plants. The real-time weather forecast includes solar irradiance data, ambient temperature data, and cloud cover data; The photovoltaic power station equipment status data includes inverter operating parameter data and photovoltaic module temperature data; Extracting the first dynamic feature from the basic data specifically includes: Set a time interval and extract the variation trend of the solar irradiance data for three consecutive time intervals; The criteria for determining the trend of change include: Set the amplitude threshold 'a' and fluctuation threshold 'b' for the solar irradiance data; If, in the solar irradiance data of three consecutive time intervals, the two adjacent increases of the value of the later time interval relative to the value of the previous time interval are both greater than 'a', then it is determined to be an increasing magnitude. The increasing magnitude is the ratio of the cumulative increase value of the three time intervals to the initial time interval value, and the ratio is the magnitude value corresponding to the increasing trend. If, in the solar irradiance data of three consecutive time intervals, the two adjacent decreases of the value of the later time interval relative to the value of the previous time interval are both greater than 'a', then it is determined to be a decreasing amplitude. The decreasing amplitude is the ratio of the cumulative decrease value of the three time intervals to the initial interval value, and the ratio is the amplitude value corresponding to the decreasing trend. If the fluctuation range of adjacent interval values ​​of solar irradiance data in three consecutive time intervals is less than or equal to b, it is determined to be a stable state. The trend of change includes increasing magnitude, decreasing magnitude, and stable state, where a and b are positive numbers, and the magnitude is expressed as the absolute value of the difference between the solar irradiance data of the later time interval and the solar irradiance data of the previous time interval. Record the extreme temperature difference that occurs within the time interval and the duration of the extreme temperature difference. Monitor the abrupt change in cloud cover level of the cloud cover data, and extract the instantaneous response characteristics of solar irradiance data from a preset time period before the abrupt change to a preset time period after the abrupt change. The instantaneous response characteristics include the response delay duration and the proportion of change in solar irradiance data.

3. The real-time power prediction method for distributed photovoltaic power stations as described in claim 2, characterized in that: Extracting the second dynamic feature from the basic data specifically includes: Using a preset time as the sampling unit, the fluctuation frequency of the inverter's operating parameter output current data within a fixed time period sliding window is extracted; The criteria for determining the fluctuation include: A current fluctuation data threshold is set. When the absolute value of the difference between the current fluctuation data of adjacent sampling units is greater than the current fluctuation data threshold, it is determined to be a fluctuation. Calculate the direction of change of the difference between the photovoltaic module temperature data and the ambient temperature data over a continuous period of time; The criteria for determining the direction of change include: Set a threshold for temperature difference changes; If, within a continuous time period, the difference between the final value and the initial value of the difference is greater than the temperature difference change threshold, it is determined to be a positive increase. If, within a continuous time period, the difference between the final value and the initial value of the difference is less than the temperature difference change threshold, it is determined to be a negative decrease. If, within a continuous period of time, the absolute value of the difference between the final value and the initial value is less than or equal to the temperature difference change threshold, it is determined to be basically stable; The directions of change include positive growth, negative decrease, and basic stability; The time synchronization characteristics of the changing trends of current fluctuation data and solar irradiance data associated with the inverter operating parameter data; The second dynamic feature is formed by combining the fluctuation frequency of the current fluctuation data, the direction of the difference between the photovoltaic module temperature data and the ambient temperature data, and the time synchronization characteristics of the current fluctuation data and the solar irradiance data in the inverter operating parameter data.

4. The real-time power prediction method for distributed photovoltaic power stations as described in claim 3, characterized in that: The first dynamic feature is compared with the historical data in the mapping database to obtain a first similarity, specifically including: Standardization of the first dynamic feature includes: The amplitude of the solar irradiance data change trend, the extreme temperature difference of the ambient temperature data, the duration corresponding to the extreme temperature difference of the ambient temperature data, the response delay of the solar irradiance data at the moment of abrupt change in cloud cover data, and the change ratio of the solar irradiance data are all converted to the same preset value range, specifically including: The min-max standardization algorithm is adopted, with the minimum value of the historical first dynamic feature in the mapping database as the lower limit of the benchmark and the maximum value of the historical first dynamic feature as the upper limit of the benchmark. By calculating the difference between the original value of each first dynamic feature and the lower limit of the benchmark, and the proportional relationship between the difference between the upper limit of the benchmark and the lower limit of the benchmark, the original value of each first dynamic feature is scaled to a preset numerical range, thereby achieving the dimensional unification of multi-dimensional features. The original values ​​of the first dynamic feature include the amplitude of the trend of the solar irradiance data, the extreme temperature difference of the ambient temperature data, the duration of the extreme temperature difference of the ambient temperature data, the response delay of the solar irradiance data at the moment of abrupt change in cloud cover data, and the change ratio of the solar irradiance data. Using the Euclidean distance algorithm, the spatial distance between the standardized first dynamic feature and the historical first dynamic feature stored in the mapping database is calculated one by one. Then, c minus the normalized distance value is used as the first similarity, and a first reasonable numerical range of the first similarity is set. The first reasonable numerical range is a closed interval. The normalized distance value is the ratio of the spatial distance to the maximum spatial distance between the historical first dynamic features in the mapping database.

5. The real-time power prediction method for distributed photovoltaic power stations as described in claim 4, characterized in that: The second dynamic feature data and the historical data of the mapping database are compared to calculate the similarity to obtain a second similarity, specifically including: Performing discretization encoding on the second dynamic feature includes: The frequency of inverter operating parameter data and output current data fluctuation within a preset time sliding window is divided into multiple levels according to preset value ranges, specifically including: Based on the minimum and maximum values ​​of the fluctuation frequency of the current data in the historical data of the mapping database, a first interval is set as a first proportional threshold where the fluctuation frequency is less than or equal to the average value of the current data, corresponding to the first level; The second interval is defined as a first proportional threshold where the fluctuation frequency is greater than the average current data value and less than or equal to the second proportional threshold where the fluctuation frequency is greater than or equal to the average current data value, corresponding to the second level; The third interval is defined as a second proportional threshold where the fluctuation frequency is greater than the average value of the current data, corresponding to the third level; The direction of change of the difference between photovoltaic module temperature data and ambient temperature data within a preset time period is expressed numerically, including: The positive growth corresponds to a first preset value in the temperature direction, the basically stable corresponds to a second preset value in the temperature direction, and the negative decrease corresponds to a third preset value in the temperature direction. The time synchronization characteristics of the current fluctuation data and the solar irradiance data of the inverter operating parameter data are represented by numerical values, and the time difference corresponding to the characteristics is used as the basis for judgment. The time difference is the absolute value of the difference between the time when the inverter current fluctuation data occurs and the time when the solar irradiance change trend is triggered, and the preset condition is that the time difference is less than or equal to the first time threshold. When the time difference meets the preset conditions, the feature corresponds to the synchronization - first preset value; When the time difference does not meet the preset conditions, the feature corresponds to the second preset value of synchronization, and the first preset value of synchronization is different from the second preset value of synchronization. The Hamming distance algorithm is used to calculate the number of bit differences between the discretized second dynamic feature data and the historical second dynamic features stored in the mapping database. Then, the ratio of the number of bit differences to the total number of bits of the feature is subtracted from q as the second similarity. A second reasonable numerical range for the second similarity is set, which is a closed interval.

6. The real-time power prediction method for distributed photovoltaic power stations as described in claim 5, characterized in that: Establish a first similarity or second similarity verification mechanism, specifically including: If the first similarity or second similarity of any historical sample exceeds the corresponding first reasonable value range or second reasonable value range, then feature data of the same type as the historical sample within a preset time range in the mapping database are called to recalculate and obtain the first similarity or second similarity. The historical samples include historical first dynamic features, historical second dynamic features, and associated historical power data.

7. The real-time power prediction method for distributed photovoltaic power stations as described in claim 6, characterized in that: Extracting historical samples with similarity higher than a preset threshold from the first and second similarity scores, and extracting the historical power data distribution features corresponding to the historical samples, specifically including: Setting the preset threshold includes setting a third preset threshold and a fourth preset threshold. The value range of the third preset threshold is from the median to the upper limit of the first reasonable numerical range of the first similarity. The value range of the fourth preset threshold is from the median to the upper limit of the second reasonable numerical range of the second similarity. The historical samples in the mapping database are subjected to dual screening, and the dual screening conditions include a first similarity higher than a third preset threshold and a second similarity higher than a fourth preset threshold. If the dual screening conditions are met, the historical samples are retained to form a candidate sample set.

8. The real-time power prediction method for distributed photovoltaic power stations as described in claim 7, characterized in that: Historical power data corresponding to each historical sample is extracted from the candidate sample set, and the distribution characteristics of the historical power data are calculated. The distribution characteristics include the power mean, the time of occurrence of the power peak, the power decay coefficient over time, and the power fluctuation variance. Clustering of the historical power data distribution characteristics in the candidate sample set according to the time dimension, specifically including: The K-means clustering algorithm was used, with the season and intraday time period of the historical samples as the basis for clustering and grouping by time dimension. The intraday time period includes morning, noon and afternoon. First, the historical power data distribution characteristics in the candidate sample set are initially divided into several groups according to the time dimension; Then, the distribution characteristics of the power mean, power peak occurrence time, power decay coefficient over time, and power fluctuation variance of each historical sample in each group are used as clustering input variables. The number of clusters is set to de. Through iterative calculation, the sum of squared Euclidean distances from each historical sample in each group to the corresponding cluster center is minimized to complete the clustering. The cluster center is the core feature virtual sample point of the corresponding cluster during the clustering process in each group, including feature values ​​with the same dimension as the clustering input variable. The feature values ​​include the power mean, the occurrence time of the power peak, the power decay coefficient over time, and the power fluctuation variance. Each feature value is the mean of the feature values ​​corresponding to all historical samples in the cluster. A total of de power feature clusters are obtained, and each power feature cluster corresponds to a typical power change pattern.

9. The real-time power prediction method for distributed photovoltaic power stations as described in claim 1, characterized in that: Calculate the correlation coefficient between the variance of historical power data in the target cluster and the amplitude of the changing trend of solar irradiance data in the current first dynamic feature, specifically including: The Pearson correlation coefficient algorithm is adopted, with the historical power data fluctuation variance of all historical samples in the target cluster as the first variable set, and the amplitude value of the solar irradiance data change trend in the historical first dynamic feature corresponding to each historical sample as the second variable set. The basic correlation coefficient at the historical level is obtained by calculating the ratio of the covariance of the first set of variables and the second set of variables to the product of the standard deviations of the first set of variables and the second set of variables. Then, substitute the magnitude of the solar irradiance data change trend in the current first dynamic feature into the fitting relationship corresponding to the basic correlation coefficient to determine the final correlation coefficient; The fitting relationship is constructed based on the statistical regularities of the first and second variable sets within the target cluster; Adjusting the power fluctuation range in the basic prediction parameters based on the correlation coefficient includes: The power fluctuation range is determined based on the power fluctuation variance in the basic prediction parameters; Using the initial power fluctuation range corresponding to the power fluctuation variance in the basic prediction parameters as a benchmark, the adjustment direction is determined based on the positive or negative value of the correlation coefficient, specifically including: If the correlation coefficient is positive, the initial power fluctuation range is adjusted by expanding it proportionally to the absolute value of the correlation coefficient; If the correlation coefficient is negative, the initial power fluctuation range is adjusted by reducing it proportionally to the absolute value of the correlation coefficient; Among them, the magnitude of the adjustment is positively correlated with the absolute value of the correlation coefficient. The larger the absolute value of the correlation coefficient, the greater the proportion of expansion or contraction of the fluctuation range. Based on the adjusted basic prediction parameters and power fluctuation range, a power prediction curve for the current prediction period is generated. The curve includes the predicted power value at each time node and the power fluctuation range corresponding to that node.

Images