A photovoltaic climbing event recognition method and system based on improved SDA and multi-dimensional coupling discrimination

Abstract

The application relates to a photovoltaic climbing event recognition method and system based on improved SDA and multi-dimensional coupling discrimination, and belongs to the technical field of photovoltaic power. (1) Collecting historical meteorological and power data to construct a feature set, and adopting an adaptive SDA algorithm based on a bidirectional asymmetric gate width improvement to perform data compression and generate an initial trend sequence Q; (2) using an adaptive time window and a weather adaptation principle to merge and reconstruct trend segments, so that a reconstructed sequence Q' which more reflects a real climbing process is obtained; (3) based on the reconstructed sequence, a multi-dimensional coupling climbing discrimination index system is constructed, and a dynamic weight score function and continuity constraint are combined; (4) the events are classified by adopting One-Hot coding, a bidirectional dynamic coefficient objective function is defined, and the best climbing recognition interval is adaptively determined, so that the robustness and practicability of the method in different scenes are improved.

Description

Technical Field

[0001] This invention relates to a photovoltaic ramp-up event identification method and system based on improved SDA and multidimensional coupling discrimination, belonging to the field of photovoltaic power technology. Background Technology

[0002] As the scale of grid-connected photovoltaic power plants continues to expand, photovoltaic output is significantly affected by meteorological factors such as sunlight and cloud cover. Among them, "ramp-up events" (the phenomenon of a sharp increase or decrease in photovoltaic power in a short period of time) pose a severe challenge to the grid's reserve capacity configuration, dispatch plan optimization, and safe and stable operation. Existing technologies have core defects in identifying photovoltaic ramp-up events.

[0003] Traditional ramp identification methods often employ fixed-parameter algorithms, single-dimensional discrimination indicators, and empirical interval division rules, which present three major problems: First, fixed parameters cannot adapt to the fluctuations in photovoltaic output under extreme weather conditions such as sunny days, cloudy days, cold waves, and sandstorms. Power fluctuations are drastic under extreme weather conditions, and fixed parameters can easily lead to distorted trend extraction or excessive noise filtering. Second, the discrimination indicators do not take into account the intraday periodicity of photovoltaic output and the abnormal trend characteristics under extreme weather conditions, resulting in a high misjudgment rate over long time scales. Third, interval division relies on manually set thresholds, lacks adaptability, and cannot adapt to the ramp characteristics of different scenarios and extreme operating conditions, which is fundamentally different from the dynamic adaptation and multi-dimensional coupling approach of this invention. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a photovoltaic ramping event identification method and system based on improved SDA and multidimensional coupling discrimination. This addresses the deficiencies of traditional methods in terms of weather adaptability and accuracy in capturing ramping events. First, historical meteorological and power data are collected to construct a feature set, and an improved adaptive SDA algorithm based on bidirectional asymmetric gate width is used for data compression to generate an initial trend sequence Q. Then, the trend segments are merged and reconstructed using an adaptive time window and weather adaptation principle to obtain a reconstructed sequence Q' that more accurately reflects the actual ramping process, effectively suppressing over-segmentation caused by noise. Second, based on the reconstructed sequence, a multidimensional coupled ramping discrimination index system is constructed, and combined with a dynamic weighted scoring function and continuity constraints, accurate identification and intensity quantification of ramping events are achieved. Furthermore, One-Hot encoding is used to classify events, obtaining structured ramping type classification results. By defining a bidirectional dynamic coefficient objective function, the optimal ramping identification interval is adaptively determined, improving the robustness and practicality of the method in different scenarios.

[0005] This invention achieves refined processing of each process from data preprocessing, hill-climbing event identification, hill-climbing intensity quantification to result classification output, providing a new technical solution for photovoltaic hill-climbing event identification.

[0006] Terminology Explanation:

[0007] The Swing Door Algorithm (SDA) is essentially a data compression method based on linear trend mapping. Similar to a revolving door, it allows data to swing freely within a set "door width" (error band). Once a new data point exceeds the error range defined by the current "door," the previous point is recorded as the critical point, and a new "door" is started. This algorithm filters and removes some non-critical data points, retaining important nodes that effectively represent the changing trend. It boasts advantages such as fast response and ease of execution. Initially used for compression and trend recording of industrial process data, it has since been extended to areas such as wind power and photovoltaic power ramp-up event identification, energy storage signal processing, and data storage optimization.

[0008] The technical solution of the present invention is as follows:

[0009] A photovoltaic ramp-up event identification method based on improved SDA and multidimensional coupling discrimination, comprising the following steps:

[0010] (1) Collect historical meteorological and power data to construct a feature set, and use an adaptive SDA algorithm based on bidirectional asymmetric gate width improvement to compress the data and generate an initial trend sequence Q;

[0011] (2) The trend segments are merged and reconstructed by using the adaptive time window and weather adaptation principle to obtain a reconstructed sequence Q' that better reflects the real climbing process, effectively suppressing the over-segmentation phenomenon caused by noise.

[0012] (3) Based on the reconstructed sequence, a multi-dimensional coupled climbing discrimination index system is constructed, and the dynamic weight score function and continuity constraint are calculated to achieve accurate identification and intensity quantification of climbing events;

[0013] (4) One-Hot coding is used to classify events and obtain structured climbing type classification results. By defining a bidirectional dynamic coefficient objective function, the optimal climbing identification interval is adaptively determined to improve robustness and practicality in different scenarios.

[0014] According to a preferred embodiment of the present invention, in step (1), the adaptive SDA algorithm based on bidirectional asymmetric gate width improvement is as follows:

[0015] Let the upper boundary of the asymmetric gate width be n from the starting point. up The lower boundary is n from the starting point down Therefore, the actual upper and lower door widths are:

[0016] (1)

[0017] (2)

[0018] In the formula, For the actual on-site service width, P represents the actual bottom door width. N For photovoltaic installed capacity, to meet ;

[0019] The slope boundary is:

[0020] (3)

[0021] (4)

[0022] In the formula, k max,up The maximum permissible slope in the upward direction represents the upper limit of the slope that the current line segment can be allowed to slope; k min,down The minimum permissible slope in the downward direction is the lower limit of the current segment's slope. If the slope in the upward or downward direction exceeds the permissible value, the segment will be divided. P i P represents the photovoltaic power value at the current data point. s P represents the power value at the starting point of the trend line currently being constructed. i -P s The power difference between the current data point and the starting point; t i t represents the time of the current data point. s The time from the starting point of the trend line segment;

[0023] Compression conditions are:

[0024] When P i ≥P s When the trend is upward, the upward threshold width is used for judgment:

[0025] (5)

[0026] When P i <P s When the trend is downward, the lower threshold width is used for judgment:

[0027] (6)

[0028] In the formula, The slope of the current trend segment;

[0029] Based on the direction of photovoltaic power variation, different upper and lower allowable deviations are used to obtain the gate width.

[0030] (8)

[0031] The permissible slope range is:

[0032] (9)

[0033] And it satisfies:

[0034] (10)

[0035] According to a preferred embodiment of the present invention, in step (1), the initial trend sequence Q of photovoltaic power output output by the adaptive SDA algorithm with improved bidirectional asymmetric gate width is defined:

[0036] (11)

[0037] In the formula, For photovoltaic power output trend sequence, Let be the photovoltaic trend power value at time t, where t = 1, 2, ..., m.

[0038] According to a preferred embodiment of the present invention, in step (2), specifically:

[0039] (21) Based on weather type and power curve fluctuation characteristics, define extreme weather correction coefficient and relaxation coefficient. Extreme weather includes cold waves, sandstorms, etc., to provide quantitative parameters for subsequent steps. The definitions and values ​​of the two are as follows:

[0040] (12)

[0041] (13)

[0042] In the formula, K e This is an extreme weather correction factor used to adjust the discrimination threshold and score weights in subsequent steps; its value is 1.2 under extreme weather conditions. K s This is the relaxation coefficient, used to relax the criteria for judging the periodic consistency index under extreme weather conditions; its value is 0.6 under extreme weather conditions.

[0043] (22) Preliminary identification of bump events: A bump event is a power fluctuation with an amplitude of less than 0.1P. N For discrete fluctuations lasting no more than one hour and without a clear periodic pattern, traversing the trend sequence Q, the discrete intervals that satisfy the following conditions are defined as climbing events: and Record the start time t of each bump event. s End time t e and power fluctuation amplitude , where Q t Let Q be the photovoltaic power output trend value at time t. t+1 This represents the photovoltaic power output trend value at time t+1. The duration of the bump event, where h is the unit of time, i.e., "hours";

[0044] (23) Adaptive window parameter setting: Adjust the merging window width W and density threshold D based on the weather type. The parameters are determined by the extreme weather correction coefficient K. e Assist in making corrections;

[0045] Considering that sunny weather bump events are sparse, mostly single instantaneous events or measurement noise, a narrow window can avoid excessively expanding the merging range; cloudy and extreme weather bump events gradually become denser, and increasing the window width can cover the cluster of related events and reduce sequence fragmentation. The basic window width W0 based on weather type is defined as:

[0046] (14)

[0047] Based on this, the formula for calculating the combined window width under extreme weather conditions is as follows:

[0048] W=W0×K e (15)

[0049] After being corrected by an extreme weather correction factor, the window width is dynamically enlarged under extreme weather conditions to adapt to the need for merging cluttered bump events;

[0050] The density threshold is used to constrain the merging intensity. The density threshold D is set to D=2 on sunny days and D=3 on cloudy or extreme weather days.

[0051] (24) Density-aware merging execution: Using each extracted bump event as the core, construct a time window with a width of W. The number of bump events N within the count window is calculated. If N ≥ D, the events are considered related bump events and merged into a continuous trend segment using linear interpolation. The merged trend value is:

[0052] (16)

[0053] In the formula, Q represents the combined trend value. t+1 Q t-1 The main trend power value before and after the bump event (the stable trend value of the non-bump segment) serves as the "benchmark anchor point" for the merge segment, ensuring that the merge direction is consistent with the overall trend. The overall representative time proportion coefficient, where t is any time point within the merged segment, t s-1 , t e+1 These are the time points before and after the bump event, respectively; w represents the fluctuation retention weight, which is used to balance the trend smoothness and the correlation characteristics of the bump event, and to avoid trend distortion after merging.

[0054] If N < D, it is determined to be an isolated bump event, which is directly removed and filled with adjacent trend values ​​linearly to eliminate noise interference;

[0055] (25) Trend sequence reconstruction: Integrate the merged continuous trend segments and non-bump trend segments to generate a reconstructed sequence Q', ensuring the temporal continuity of the sequence and the integrity of the trend, and providing reliable data support for subsequent multi-dimensional climbing event identification.

[0056] According to a preferred embodiment of the present invention, in step (3), a multi-dimensional coupled climbing discrimination index system is constructed, and the interval (a, b) is defined, where a and b are time nodes, satisfying a < b. The power difference index, climbing rate index, period consistency index, and extreme weather adaptation index are all associated with the extreme weather coefficient output by the front-end preprocessing and the merged sequence Q'. The formulas for each index and the comprehensive discrimination result are defined as follows:

[0057] (17)

[0058] In the formula, K is the comprehensive discrimination result, when When K=0, the interval (a, b) is considered a valid climbing event; when K=0, the interval (a, b) is considered a non-climbing event or invalid fluctuation; K0 is the power difference index, K1 is the climbing rate index, K2 is the period consistency index, and K3 is the extreme weather adaptation index; the round() function rounds to the nearest integer to ensure that the output value is 0 or 1.

[0059] According to a further preferred embodiment of the present invention, in step (3), the power difference index K0: determines whether the power change range of the determination interval reaches the climbing threshold. The threshold is adaptively adjusted under extreme weather conditions. The discriminant of K0 is:

[0060] (18)

[0061] In the formula, The trend values ​​of the merged sequence at times b and a after step (2) are given by θ, which is the power difference threshold based on the selected data (set to 0.20P). N (This can be adjusted according to the actual scale of the power plant); K e This is a correction factor for extreme weather conditions.

[0062] Climbing rate index K1: Determines whether the rate of change of power in the interval meets the climbing requirements. The rate threshold is fine-tuned under extreme weather conditions. The K1 discriminant is:

[0063] (19)

[0064] In the formula, The minimum ramp rate threshold is used as the basis.

[0065] Periodic Consistency Index K2: Photovoltaic output is affected by the pattern of solar irradiance and has a significant intra-day periodicity. On sunny days, the power gradually increases after sunrise and gradually decreases before sunset. On cloudy days, there is also a relatively stable periodic fluctuation pattern. The traditional two-dimensional discrimination system of power difference-rate only focuses on the amplitude and speed of power change and does not utilize this inherent characteristic. It is easy to misjudge random fluctuations without periodicity (instantaneous equipment disturbances, residual measurement noise, etc.) as ramping events. Such random fluctuations may meet the power difference and rate thresholds for a short time, but they do not have fixed periodic characteristics and are not effective ramping events that the grid dispatch needs to focus on and that are predictable. Combining the intra-day periodicity of photovoltaics with the historical trend of the same period, random fluctuation interference is eliminated, and the correlation threshold is relaxed under extreme weather conditions.

[0066] (20)

[0067] In the formula, H a-b This is a historical photovoltaic power output trend sequence for the same period (excluding extreme weather data). is the Pearson correlation coefficient, used to measure the similarity of trends in power sequences, with a value range of [-1, 1]. The baseline periodic correlation threshold is calculated and set with different baseline values ​​based on correlation analysis of actual data; K s This represents the relaxation coefficient for extreme weather conditions.

[0068] Extreme Weather Adaptation Index K3: Under extreme weather conditions, the power output of photovoltaic power changes not only deviates from the normal intraday cycle, but also differs fundamentally from the normal fluctuations under non-extreme weather conditions. Traditional indicators can only determine whether the power change amplitude and rate meet the standards, and cannot distinguish whether the change is an effective ramp driven by extreme weather. Therefore, the core purpose of introducing the extreme weather adaptation index is to quantify the correlation between actual power changes and theoretical power changes under extreme weather conditions, so as to characterize and identify ramp events driven by extreme weather. At the same time, combined with the extreme weather coefficient system output in step (1), it avoids misjudging normal power changes under extreme weather conditions as invalid fluctuations, and also prevents abnormal fluctuations caused by non-extreme factors from being misjudged as extreme ramps. Finally, it improves the full-scene adaptation capability of the four-dimensional coupling discrimination system. The K3 discriminant is as follows:

[0069] (twenty one)

[0070] In the formula, The theoretical change in photovoltaic output under extreme scenarios is calculated based on the differentiated settings of different types of extreme weather. The calculation logic is accurately matched to the power impact mechanism of different extreme operating conditions. The value is 0 under non-extreme weather conditions. The interference of the extreme weather adaptation index K3 on the judgment of normal operating conditions is automatically shielded. Therefore, under non-extreme scenarios, K3=1, and the extreme scenario is extreme weather.

[0071] According to a preferred embodiment of the present invention, in step (3), the dynamic weighted score function and continuity constraint are calculated, and the interval (a, b) score function is adjusted for extreme weather conditions, resulting in the following formula:

[0072] (twenty two)

[0073] In the formula, α is the time weighting coefficient; β is the amplitude weighting coefficient; and K e This is a correction factor for extreme weather conditions. The average rate over the interval is represented by positive values ​​for uphill climbing and negative values ​​for downhill climbing. The larger the absolute value of F(a,b), the higher the climbing intensity, and the more obvious the score differentiation for high-intensity climbing under extreme weather conditions.

[0074] At the same time, the continuity constraint should be satisfied. For any time node a < c < b, the following condition must be met:

[0075] (twenty three)

[0076] In the formula, The preset baseline rate of change tolerance threshold is a reasonable upper limit of fluctuation fitted by a large amount of historical data. In extreme weather scenarios, K... e Appropriate amplification threshold;

[0077] Formula (23) is a constraint on the continuity of the climbing interval. By splitting the interval (a, b) into sub-intervals (a, c) and (c+1, b), the deviation between the total score of the whole interval and the sum of the scores of the two sub-intervals is required to meet the constraint range. If the deviation exceeds the range, it means that the score in the interval (a, b) changes drastically and the interval contains non-continuous climbing sections. The interval range needs to be adjusted and recalculated.

[0078] According to a preferred embodiment of the present invention, in step (4), the One-Hot encoding rule is as follows:

[0079] ① Encoding dimension definition: The vector dimension is 7, corresponding to no climbing, weak climbing-up, weak climbing-down, medium climbing-up, medium climbing-down, strong climbing-up, and strong climbing-down respectively. The 7 scenarios adopt sparse coding, with each scenario corresponding to a unique activation bit and the remaining bits being 0;

[0080] ②Intensity grading standard: Based on F(a, b), the formula for climbing intensity is defined as follows:

[0081] (twenty four)

[0082] Among them, in photovoltaic power ramp-up events, upward and downward are two basic types defined based on the direction and trend of power change. Upward represents an event in which photovoltaic power continuously increases within a certain period of time and the change exceeds a set threshold; downward represents an event in which photovoltaic power continuously decreases within a certain period of time and the change exceeds a set threshold.

[0083] ③ One-Hot climbing event encoding is performed based on climbing intensity and climbing direction, with each of the 7 scenarios corresponding to a unique encoding vector;

[0084] Traverse all valid intervals (K=1) that have passed the continuity constraint verification, and generate the corresponding One-Hot encoded vector set based on the strength and direction of F(a, b). Each of these corresponds one-to-one with the time nodes of the merged sequence Q', forming a three-dimensional feature group of time, encoding, and power trend.

[0085] By using the activation bits of the One-Hot encoded vector, the climbing direction and intensity level of the target section can be quickly determined.

[0086] According to a preferred embodiment of the present invention, in step (4), the optimal climbing interval is determined by using the bidirectional dynamic coefficient objective function G(a, b), the uplink / downlink calculation logic is switched based on the one-hot encoding recognition result, and the dynamic coefficient is optimized and adjusted by combining the climbing intensity and weather type.

[0087] (41) Uphill climbing:

[0088] (25)

[0089] (42) Downhill climbing:

[0090] (26)

[0091] In the formula, These are the uphill and downhill climbing dynamic coefficients, respectively, to adapt to the climbing acceleration / deceleration characteristics; w is the correction factor for extreme weather. s Assign values ​​to the climbing intensity weights based on One-Hot encoding;

[0092] The assignment principles are as follows:

[0093] (27)

[0094] Based on the two-way dynamic coefficient objective function formula, the range of the optimal climbing interval is obtained;

[0095] Upward climbing event: Take the sub-interval corresponding to the maximum value of the objective function G(a,b) as the optimal upward climbing interval (the interval with the best climbing intensity and rate has the most significant impact on power grid dispatch).

[0096] Downward climbing event: Take the sub-interval corresponding to the minimum value of the objective function G(a,b) as the optimal downward climbing interval;

[0097] Finally, the results are output, simultaneously showing the start and end times of the optimal interval, the one-hot encoding type (intensity + direction), the objective function value, and the extreme weather association markers, which can provide accurate feature inputs for subsequent slope event probability prediction.

[0098] A photovoltaic ramp-up event identification system based on improved SDA and multidimensional coupling discrimination includes:

[0099] The data compression module is used to collect historical meteorological and power data to construct a feature set, and uses an adaptive SDA algorithm based on bidirectional asymmetric gate width improvement to compress the data and generate an initial trend sequence Q.

[0100] The reconstruction module uses an adaptive time window and weather adaptation principle to merge and reconstruct trend segments, resulting in a reconstruction sequence Q' that better reflects the actual climbing process, effectively suppressing oversegmentation caused by noise.

[0101] The quantization module, based on the reconstructed sequence, constructs a multi-dimensional coupled climbing discrimination index system and calculates the dynamic weight score function and continuity constraints to achieve accurate identification and intensity quantification of climbing events.

[0102] The identification module (4) uses One-Hot encoding to classify events and obtains a structured classification result of climbing types. By defining a bidirectional dynamic coefficient objective function, the optimal climbing identification interval is adaptively determined to improve the robustness and practicality in different scenarios.

[0103] The beneficial effects of this invention are as follows:

[0104] 1. An integrated adaptive hill-climbing event identification framework was constructed, overcoming the limitations of traditional methods. This invention integrates meteorological data acquisition, power data compression, trend reconstruction, multi-dimensional discrimination, intensity quantification, and event coding into a unified process, avoiding the defects of disconnected links and fixed parameters in traditional methods. It realizes end-to-end automated processing from raw data to classification results, significantly improving the accuracy of photovoltaic hill-climbing event identification.

[0105] 2. This invention proposes a bidirectional asymmetric dynamic SDA data preprocessing method, achieving adaptive and high-fidelity extraction of photovoltaic power trends. By introducing weather type mapping and a bidirectional asymmetric gate width mechanism, the traditional SDA algorithm is improved. It can dynamically adjust the gate width according to different scenarios such as sunny, cloudy, and extreme weather, as well as the different physical characteristics of power rise and fall. This improvement makes trend extraction smoother in stable weather to highlight the main trend, and more sensitive in fluctuating weather to capture abrupt changes. Thus, while reducing the amount of data, it retains key information for subsequent accurate identification, solving the problems of poor adaptability, loss of detail, or excessive noise in traditional fixed-gate width SDA.

[0106] 3. This invention designs a bump event merging and trend reconstruction mechanism based on adaptive time window and weather adaptation, which effectively solves the problems of fragmentation and over-segmentation of the climbing trend. For the initially extracted climbing trend sequence, the correlation of continuous events is analyzed through adaptive time window, and the events are merged in combination with the weather adaptation principle to reconstruct a long trend sequence Q' with clearer physical meaning and better reflecting the real climbing process.

[0107] 4. This invention establishes a multi-dimensional coupled climbing discrimination index system and a dynamic weighted scoring function, achieving accurate quantitative characterization of climbing events and their intensity. This invention overcomes the shortcomings of single-threshold discrimination by introducing time consistency and extreme weather indicators on top of the basic standards of climbing rate and power difference. Furthermore, it introduces a dynamic weighted scoring function F(a,b) and continuity constraints, enabling refined and differentiated identification and quantitative scoring of climbing events of different forms and intensities.

[0108] 5. This invention integrates One-Hot encoding technology to achieve standardized and structured output of climbing event types. The identified climbing events are converted into one-hot encoded vectors, forming machine-readable standardized features that can be directly used for machine learning model training and event pattern statistical analysis, providing reliable support for subsequent probability prediction. Simultaneously, by defining a bidirectional dynamic coefficient objective function, the optimal climbing interval is automatically optimized, improving the algorithm's adaptability. Attached Figure Description

[0109] Figure 1 This is a schematic diagram of the process of the present invention;

[0110] Figure 2 This is a schematic diagram of the SDA triangle rule of the present invention;

[0111] Figure 3 This is a schematic diagram of the adaptive SDA algorithm with improved bidirectional asymmetric gate width according to the present invention.

[0112] Figure 4This is a preprocessing result of the improved bidirectional asymmetric gate width adaptive SDA algorithm of this invention under clear weather conditions;

[0113] Figure 5 This is a preprocessing result of the adaptive SDA algorithm with improved bidirectional asymmetric gate width in this invention under cloudy weather conditions;

[0114] Figure 6 This is a schematic diagram illustrating the principle of bump event merging processing in this invention.

[0115] Figure 7 This is a flowchart of the multidimensional coupled hill-climbing discrimination process of the present invention;

[0116] Figure 8 This is a diagram showing the classification results of hill climbing events based on One-Hot encoding according to the present invention. Detailed Implementation

[0117] The present invention will be further described below with reference to the embodiments and accompanying drawings, but is not limited thereto.

[0118] Example 1:

[0119] like Figure 1 As shown in the figure, this embodiment provides a photovoltaic ramp-up event identification method based on improved SDA and multidimensional coupling discrimination, and the steps are as follows:

[0120] (1) Collect historical meteorological and power data to construct a feature set, and use an adaptive SDA algorithm based on bidirectional asymmetric gate width improvement to compress the data and generate an initial trend sequence Q;

[0121] In the adaptive SDA algorithm based on bidirectional asymmetric gate width improvement:

[0122] When traditional fixed-gate width SDA is directly applied to identify ramp-up events in photovoltaic power plant power data, the following inherent drawbacks exist:

[0123] First, the fixed gate width leads to poor adaptability. Photovoltaic power output is affected by various meteorological factors such as sunlight, temperature, clouds, and dust storms, and its fluctuation characteristics vary significantly under different weather conditions. Traditional SDAs use a fixed gate width and cannot adapt to such dynamic changes. On sunny days with gentle fluctuations, the photovoltaic power curve has small fluctuations, and the overall curve trend is close to the "clear sky" curve, showing a trend of first rising and then falling, similar to the positive half-cycle of a sine function. In cloudy weather, due to the local movement and shading of cloud clusters, photovoltaic power will experience short-term fluctuations, but this effect will be eliminated when the cloud clusters move. Under extreme weather conditions (such as cold waves and dust storms), the photovoltaic power curve exhibits large fluctuations. The changes caused by extreme weather conditions are usually uncertain and highly persistent, resulting in a lower overall photovoltaic power output level. Based on the algorithm properties of SDA, if the fixed gate width is too large, it will lead to excessive data compression and loss of key details representing minor ramps; if the fixed gate width is too small, it will not be able to effectively filter high-frequency noise, and will instead misjudge normal power fluctuations as a large number of meaningless minor ramp events, resulting in data overfitting and fragmented trend segments, which will seriously affect the accuracy of subsequent event identification.

[0124] Second, it fails to consider the asymmetry in the direction of power change. The impact of photovoltaic ramping events on grid stability exhibits significant asymmetry in the directions of power increase and decrease. Typically, a sharp drop in power has a more severe impact on grid frequency and needs to be detected earlier and more sensitively. Traditional SDA uses a symmetrical threshold tolerance, applying the same acceptance threshold to both the rising and falling processes. This fails to reflect the difference in the impact of ramping direction on the power system, potentially leading to insufficient sensitivity in detecting the more damaging falling ramping events.

[0125] In summary, traditional fixed-gate SDAs have inherent defects such as insufficient adaptability, low identification accuracy, and unclear physical meaning when processing photovoltaic power data that has strong meteorological correlation, variable fluctuation patterns, and asymmetrical impact on the power grid. They are unable to meet the urgent needs of the power grid for accurate and rapid identification of ramping events under the high proportion of renewable energy access.

[0126] Addressing the two major shortcomings of traditional SDA, this invention fully considers the impact of power variation trends and weather types, and proposes an improved SDA algorithm based on bidirectional asymmetric gate width for trend compression extraction of photovoltaic power data. The specific implementation method is as follows:

[0127] The essence of the SDA algorithm is often described by the triangle principle or parallelogram principle. A schematic diagram of the SDA triangle rule is shown below. Figure 2 As shown, let the traditional SDA symmetrical gate width be... Then the distance from the upper and lower boundaries to the starting point is... Considering the number and accuracy of daily photovoltaic power output ramp-up data points, a fixed gate width of 0.01P is often selected.N P N The fixed parameters are for photovoltaic installed capacity, but considering the significant differences in the fluctuation of photovoltaic power curves under sunny, cloudy and extreme weather conditions, fixed parameters are prone to abnormal trend feature extraction, high misjudgment rate over long time scales, and lack of scene adaptability. Figure 2 In the diagram, a, b, c, d, and e are power points, and the power point variation trend is abcde. a is the starting point of compression segment 1 and also the ending point of the previous compression segment record. There are points a distance of [missing information] above and below point a. The door is wide. , It is the fulcrum for the rotation of the SDA "gate".

[0128] Let the upper boundary of the asymmetric gate width be n from the starting point. up The lower boundary is n from the starting point down Therefore, the actual upper and lower door widths are:

[0129] (1)

[0130] (2)

[0131] In the formula, For the actual on-site service width, P represents the actual bottom door width. N For photovoltaic installed capacity, to meet ;

[0132] The slope boundary is:

[0133] (3)

[0134] (4)

[0135] In the formula, k max,up The maximum permissible slope in the upward direction represents the upper limit of the slope that the current line segment can be allowed to slope; k min,down The minimum permissible slope in the downward direction is the lower limit of the current segment's slope. If the slope in the upward or downward direction exceeds the permissible value, the segment will be divided. P i P represents the photovoltaic power value at the current data point. s P represents the power value at the starting point of the trend line currently being constructed. i -P s The power difference between the current data point and the starting point; t i t represents the time of the current data point. s The time from the starting point of the trend line segment;

[0136] The compression condition for bidirectional asymmetric gatewidth SDA is:

[0137] When P i ≥P s When the trend is upward, the upward threshold width is used for judgment:

[0138] (5)

[0139] When P i <P s When the trend is downward, the lower threshold width is used for judgment:

[0140] (6)

[0141] In the formula, The slope of the current trend segment;

[0142] The fixed door width set by the traditional SDA is Then, for the starting point (t) s P s ) and data points (t) i P i The permissible slope range is:

[0143] (7)

[0144] The schematic diagram of a bidirectional asymmetric gate width SDA is as follows: Figure 3 As shown, compared to traditional SDA, different upper and lower tolerances are used according to the direction of photovoltaic power change to obtain the gate width.

[0145] (8)

[0146] Therefore, compared to the fixed door width of traditional SDA The allowable slope range of the asymmetric gate width-improved adaptive SDA algorithm is:

[0147] (9)

[0148] And it satisfies:

[0149] (10)

[0150] In summary, the improved gate width setting formula for asymmetric SDA is obtained. With these improvements, the method proposed in this embodiment can more accurately capture the trend of climbing events while reducing the shortcomings of excessive segmentation.

[0151] Basis for setting asymmetrical door width:

[0152] (1) A larger lower limit for the gate width: Considering that a rapid decrease in power has a more severe impact on frequency stability than an increase, downhill climbing needs to be identified and recognized earlier than uphill climbing. Furthermore, based on the different meteorological correlation scenarios of sunny, cloudy, and extreme weather, the power decrease caused by cloud cover, dust, etc., is usually more severe than the recovery and increase after dissipation. Therefore, the lower limit n of the gate width for each weather type is [not specified]. down Both are compared to the upper limit of the gate width n up Choose the larger value.

[0153] (2) The gate width is inversely proportional to the degree of data fluctuation: For scenarios with low fluctuation (sunny days), the main target for extraction is "steady state," while noise and minor disturbances are "secondary information." Using a large gate width can actively ignore this secondary information and focus on the main trend. For scenarios with high fluctuation (cloudy, extreme weather), minor fluctuations may be precursors to a climbing event. Using a small gate width can actively amplify these signals, ensuring that the algorithm is in a highly sensitive state and does not miss the beginning of any abnormal changes.

[0154] Based on tests of actual photovoltaic power plants, and considering the number and accuracy of daily photovoltaic output ramp-up data points, a set of threshold width settings based on weather type and their basis are given, as shown in Table 1.

[0155] Table 1: Basis for Door Width Setting Based on Weather Type

[0156]

[0157] The bidirectional asymmetric gate width SDA, based on weather type-specific gate width settings, can obtain power trend extraction results under different weather conditions. In sunny weather, the algorithm outputs an extremely smooth trend line consisting of only a few key points; other actual fluctuations deviating from this smooth curve can be explicitly analyzed in other steps. In cloudy weather, when a thick cloud obscures power and causes a distinct "U-shaped trough," a medium gate width can identify it as a complete downward-upward trend pair, rather than fragmented. The algorithm's sensitivity and output complexity fall between those of sunny and extreme weather modes. In extreme weather, the algorithm's sensitivity increases; for slow power declines caused by cold waves or severe fluctuations in sunlight during dust storms, a small gate width can clearly characterize them as a series of short trend segments, rather than a general downward straight line. The preprocessing results based on the bidirectional asymmetric gate width under sunny and cloudy weather conditions are as follows: Figure 4 , Figure 5 As shown.

[0158] Based on the above analysis, the initial trend sequence Q of photovoltaic power output output from the bidirectional asymmetric gate width improved adaptive SDA algorithm is defined as follows:

[0159] (11)

[0160] In the formula, For photovoltaic power output trend sequence, Let be the photovoltaic trend power value at time t, where t = 1, 2, ..., m.

[0161] (2) The trend segments are merged and reconstructed by using the adaptive time window and weather adaptation principle to obtain a reconstructed sequence Q' that better reflects the real climbing process, effectively suppressing the over-segmentation phenomenon caused by noise.

[0162] Even after dynamic preprocessing, the Q-sequence may still contain fragmented bump events, characterized by small, discrete power fluctuations. Therefore, this embodiment proposes an adaptive time window and weather-adaptive merging method to eliminate the interference of fragmented fluctuations on subsequent ramp-up determination. The principle of bump event merging is as follows: Figure 6 As shown.

[0163] A bump event is a power fluctuation amplitude of less than 0.1P. N Discrete fluctuations lasting no more than one hour and without obvious periodicity are often caused by momentary cloud cover or meteorological fluctuations under extreme weather conditions. They are not considered climbing events, but they are easily misjudged as climbing events. In order to identify and reduce misjudgments caused by bump events in the identification of photovoltaic climbing events, this embodiment adds a neighborhood bump event density constraint on the basis of the time window, and only merges discrete events with a density that meets the standard (≥2 events / window). This is suitable for dense bump scenarios under extreme weather conditions.

[0164] (21) Based on weather type and power curve fluctuation characteristics, define extreme weather correction coefficient and relaxation coefficient. Extreme weather includes cold waves, sandstorms, etc., to provide quantitative parameters for subsequent steps. The definitions and values ​​of the two are as follows:

[0165] (12)

[0166] (13)

[0167] In the formula, K e This is an extreme weather correction factor used to adjust the discrimination threshold and score weights in subsequent steps; its value is 1.2 under extreme weather conditions. K s This is the relaxation coefficient, used to relax the criteria for judging the periodic consistency index under extreme weather conditions; its value is 0.6 under extreme weather conditions.

[0168] (22) Preliminary identification of bump events: A bump event is a power fluctuation with an amplitude of less than 0.1P. N For discrete fluctuations lasting no more than one hour and without a clear periodic pattern, traversing the trend sequence Q, the discrete intervals that satisfy the following conditions are defined as climbing events: and Record the start time t of each bump event. s End time t e and power fluctuation amplitude , where Q t Let Q be the photovoltaic power output trend value at time t. t+1 This represents the photovoltaic power output trend value at time t+1. The duration of the bump event, where h is the unit of time, i.e., "hours";

[0169] (23) Adaptive window parameter setting: Adjust the merging window width W and density threshold D based on the weather type. The parameters are determined by the extreme weather correction coefficient K. e Assist in making corrections;

[0170] Considering that sunny weather bump events are sparse, mostly single instantaneous events or measurement noise, a narrow window can avoid excessively expanding the merging range; cloudy and extreme weather bump events gradually become denser, and increasing the window width can cover the cluster of related events and reduce sequence fragmentation. The basic window width W0 based on weather type is defined as:

[0171] (14)

[0172] Based on this, the formula for calculating the combined window width under extreme weather conditions is as follows:

[0173] W=W0×K e (15)

[0174] After being corrected by an extreme weather correction factor, the window width is dynamically enlarged under extreme weather conditions to adapt to the need for merging cluttered bump events;

[0175] The density threshold is used to constrain the merging intensity. The density threshold D is set to D=2 on sunny days and D=3 on cloudy or extreme weather days.

[0176] (24) Density-aware merging execution: Using each extracted bump event as the core, construct a time window with a width of W. The number of bump events N within the count window is calculated. If N ≥ D, the events are considered related bump events and merged into a continuous trend segment using linear interpolation. The merged trend value is:

[0177] (16)

[0178] In the formula, Q represents the combined trend value. t+1 Q t-1The main trend power value before and after the bump event (the stable trend value of the non-bump segment) serves as the "benchmark anchor point" for the merge segment, ensuring that the merge direction is consistent with the overall trend. The overall representative time proportion coefficient, where t is any time point within the merged segment, t s-1 , t e+1 These are the time points before and after the bump event, respectively; w represents the fluctuation retention weight, which is used to balance the trend smoothness and the correlation characteristics of the bump event, and to avoid trend distortion after merging.

[0179] If N < D, it is determined to be an isolated bump event, which is directly removed and filled with adjacent trend values ​​linearly to eliminate noise interference;

[0180] (25) Trend sequence reconstruction: Integrate the merged continuous trend segments and non-bump trend segments to generate a reconstructed sequence Q', ensuring the temporal continuity of the sequence and the integrity of the trend, and providing reliable data support for subsequent multi-dimensional climbing event identification.

[0181] (3) Based on the reconstructed sequence, a multi-dimensional coupled climbing discrimination index system is constructed, and the dynamic weight score function and continuity constraint are calculated to achieve accurate identification and intensity quantification of climbing events;

[0182] A multi-dimensional coupled climbing discrimination index system was constructed. After step (1) dynamic gate width rotating gate preprocessing (outputting trend sequence Q) and step (2) event density perception merging processing (outputting continuous trend sequence Q'), noise and fragmented fluctuations in photovoltaic power output have been effectively filtered, providing a continuous basic data set for subsequent climbing event discrimination. However, existing climbing discrimination methods still have obvious limitations and cannot adapt to the climbing characteristics under long-term scales and extreme weather conditions. On the one hand, traditional methods mostly use the two-dimensional index of "power difference-rate", without combining the intraday periodic characteristics of photovoltaic power output, which easily misjudges random fluctuations as climbing events, resulting in a high misjudgment rate; on the other hand, they do not incorporate extreme weather correction mechanisms. The power mutation pattern under extreme weather conditions is significantly different from that under normal weather conditions, and fixed index thresholds are difficult to adapt, which easily leads to missed judgments and misjudgments.

[0183] Based on this, to further overcome the limitations of traditional identification models, this embodiment proposes a four-dimensional coupled discrimination system that integrates power difference, climbing rate, periodic consistency analysis, and extreme weather. This system overcomes the limitations of traditional single indicators and fixed thresholds, deeply linking with front-end preprocessing results. Multi-dimensional indicator calculations are performed based on the merged sequence Q', while also embedding extreme weather correction logic. Through the logical coupling of various indicators, accurate determination of climbing events is achieved, significantly reducing the misjudgment rate over long time scales and under extreme weather conditions. The specific discrimination process is as follows: Figure 7 As shown.

[0184] Define the interval (a, b), where a and b are time nodes, satisfying a < b. The power difference index, ramp rate index, cycle consistency index, and extreme weather adaptation index are all related to the extreme weather coefficient output by the front-end preprocessing and the merged sequence Q'. The formulas for each index and the comprehensive discrimination result are defined as follows:

[0185] (17)

[0186] In the formula, K is the comprehensive discrimination result, when When K=0, the interval (a, b) is considered a valid climbing event; when K=0, the interval (a, b) is considered a non-climbing event or invalid fluctuation; K0 is the power difference index, K1 is the climbing rate index, K2 is the period consistency index, and K3 is the extreme weather adaptation index; the round() function rounds to the nearest integer to ensure that the output value is 0 or 1.

[0187] Power difference index K0: Determines whether the power change range has reached the climbing threshold. The threshold is adaptively adjusted under extreme weather conditions. The K0 discriminant is:

[0188] (18)

[0189] In the formula, The trend values ​​of the merged sequence at times b and a after step (2) are given by θ, which is the power difference threshold based on the selected data (set to 0.20P). N (This can be adjusted according to the actual scale of the power plant); K e This is a correction factor for extreme weather conditions.

[0190] Climbing rate index K1: Determines whether the rate of change of power in the interval meets the climbing requirements. The rate threshold is fine-tuned under extreme weather conditions. The K1 discriminant is:

[0191] (19)

[0192] In the formula, The minimum ramp rate threshold is used as the basis.

[0193] Periodic Consistency Index K2: Photovoltaic output is affected by the pattern of solar irradiance and has a significant intra-day periodicity. On sunny days, the power gradually increases after sunrise and gradually decreases before sunset. On cloudy days, there is also a relatively stable periodic fluctuation pattern. The traditional two-dimensional discrimination system of power difference-rate only focuses on the amplitude and speed of power change and does not utilize this inherent characteristic. It is easy to misjudge random fluctuations without periodicity (instantaneous equipment disturbances, residual measurement noise, etc.) as ramping events. Such random fluctuations may meet the power difference and rate thresholds for a short time, but they do not have fixed periodic characteristics and are not effective ramping events that the grid dispatch needs to focus on and that are predictable. Combining the intra-day periodicity of photovoltaics with the historical trend of the same period, random fluctuation interference is eliminated, and the correlation threshold is relaxed under extreme weather conditions.

[0194] (20)

[0195] In the formula, H a-b This is a historical photovoltaic power output trend sequence for the same period (excluding extreme weather data). is the Pearson correlation coefficient, used to measure the similarity of trends in power sequences, with a value range of [-1, 1]. The baseline periodic correlation threshold is calculated and set with different baseline values ​​based on correlation analysis of actual data; K s This represents the relaxation coefficient for extreme weather conditions.

[0196] Extreme Weather Adaptation Index K3: Under extreme weather conditions, the power output of photovoltaic power changes not only deviates from the normal intraday cycle, but also differs fundamentally from the normal fluctuations under non-extreme weather conditions. Traditional indicators can only determine whether the power change amplitude and rate meet the standards, and cannot distinguish whether the change is an effective ramp driven by extreme weather. Therefore, the core purpose of introducing the extreme weather adaptation index is to quantify the correlation between actual power changes and theoretical power changes under extreme weather conditions, so as to characterize and identify ramp events driven by extreme weather. At the same time, combined with the extreme weather coefficient system output in step (1), it avoids misjudging normal power changes under extreme weather conditions as invalid fluctuations, and also prevents abnormal fluctuations caused by non-extreme factors from being misjudged as extreme ramps. Finally, it improves the full-scene adaptation capability of the four-dimensional coupling discrimination system. The K3 discriminant is as follows:

[0197] (twenty one)

[0198] In the formula, The theoretical change in photovoltaic output under extreme scenarios is calculated based on the differentiated settings of different types of extreme weather. The calculation logic is accurately matched to the power impact mechanism of different extreme operating conditions. The value is 0 under non-extreme weather conditions. The interference of the extreme weather adaptation index K3 on the judgment of normal operating conditions is automatically shielded. Therefore, under non-extreme scenarios, K3=1, and the extreme scenario is extreme weather.

[0199] Taking actual data from two types of events, cold waves and sandstorms, as examples for analysis, The cold wave weather was obtained by fitting the meteorological model. The assumption is that temperature (T) is the core factor affecting photovoltaic (PV) output during a cold wave; if the temperature drops by 1°C, the average PV power decreases by 0.03. Taking a temperature baseline of 5℃, the theoretical change in photovoltaic output under cold wave weather can be calculated. (Dust storm weather) The basis for this setting is that PM10 dust concentration (S) is the direct cause of decreased solar irradiance, which in turn reduces photovoltaic power. For every 1 mg / m³ increase in dust concentration, photovoltaic power decreases by an average of 0.02 mg / m³. Then it can be passed The formula quantifies the impact of sandstorm intensity on photovoltaic power, and the specific parameters are calculated and adjusted according to different regions and extreme weather scenarios.

[0200] By calculating various indicators and multiplying them together to obtain the K value, a comprehensive judgment result is obtained, enabling accurate identification of hill climbing events.

[0201] The previous section completed the preliminary qualitative judgment of "whether it is a climbing event" by calculating the dynamic weight score function and continuity constraints. However, it did not quantitatively characterize the intensity and stability of the climbing event, and could not distinguish the degree of impact of different climbing events on power grid dispatch. Moreover, the traditional identification method lacks interval continuity verification, which is prone to distortion of climbing interval division due to sudden changes in scores, thereby affecting the accuracy of subsequent clustering and the reliability of optimal interval positioning. Therefore, this step quantifies the climbing intensity through the score function and verifies the rationality of the interval through continuity constraints. This not only fills the technical gap between the qualitative judgment in the previous step and the quantitative application in the subsequent step, but also solves the defects of the traditional method in interval division being highly subjective and having vague intensity distinctions, providing accurate quantitative basis for subsequent processes.

[0202] This step proposes a dynamic weighted scoring function and a continuity constraint method, which not only enables accurate quantitative characterization of climbing intensity but also transforms climbing events from qualitative existence to quantitative intensity by integrating core factors such as time, amplitude, and extreme weather. The scoring function... The absolute value directly reflects the climbing intensity, meeting the differentiated needs of scheduling scenarios;

[0203] The core of this step is to optimize the static weight design of the traditional scoring function, introduce dynamic weight coefficients to distinguish between time and magnitude priorities, and add continuity constraints to avoid misjudgment of intervals caused by sudden changes in scores.

[0204] After incorporating extreme weather weighting into the interval (a, b) score function, the formula is:

[0205] (twenty two)

[0206] In the formula, α is the time weighting coefficient; β is the amplitude weighting coefficient; and K e This is a correction factor for extreme weather conditions. The average rate over the interval is represented by positive values ​​for uphill climbing and negative values ​​for downhill climbing. The larger the absolute value of F(a,b), the higher the climbing intensity, and the more obvious the score differentiation for high-intensity climbing under extreme weather conditions.

[0207] At the same time, the continuity constraint should be satisfied. For any time node a < c < b, the following condition must be met:

[0208] (twenty three)

[0209] In the formula, The preset baseline rate of change tolerance threshold is a reasonable upper limit of fluctuation fitted by a large amount of historical data. In extreme weather scenarios, K... e Appropriate amplification threshold;

[0210] Formula (23) is a constraint on the continuity of the climbing interval. By splitting the interval (a, b) into sub-intervals (a, c) and (c+1, b), the deviation between the total score of the whole interval and the sum of the scores of the two sub-intervals is required to meet the constraint range. If the deviation exceeds the range, it means that the score in the interval (a, b) changes drastically and the interval contains non-continuous climbing sections. The interval range needs to be adjusted and recalculated.

[0211] This step provides quantitative input for subsequent steps. The score results serve as the hierarchical label basis for subsequent adaptive density clustering. By mapping the scores to three levels of labels (weak, medium, and strong), hierarchical clustering of climbing events is achieved, improving the objectivity of climbing event classification.

[0212] (4) One-Hot coding is used to classify events and obtain structured climbing type classification results. By defining a bidirectional dynamic coefficient objective function, the optimal climbing identification interval is adaptively determined to improve robustness and practicality in different scenarios.

[0213] Based on the absolute value of the climb intensity score, adaptive identification of climb intervals is achieved, adapting to the characteristics of climbs of different intensities. This step introduces one-hot encoding to quantify multi-dimensional features of climb events, enabling adaptive and accurate segmentation of climb intervals, adapting to the differentiated characteristics of climbs of different intensities and directions, and avoiding the hierarchical correlation interference of traditional label encoding. The classification result of climb events using one-hot encoding is shown in the figure below. Figure 8 As shown.

[0214] One-hot encoding is a coding technique that converts categorical variables into binary vectors. Its core principle is that for a categorical feature with N different values, a binary vector of length N is constructed to represent each value, in which only one position is 1 (activated state) and the rest are 0 (inactive state).

[0215] Based on the absolute value and sign of the scoring function F(a, b), a 7-dimensional One-Hot encoding vector is constructed from the perspectives of climbing intensity and climbing direction. This enables independent quantitative representation of climbing features and provides an accurate feature carrier for subsequent calculation of the objective function of the optimal climbing interval. The One-Hot encoding rules are as follows:

[0216] ① Encoding dimension definition: The vector dimension is 7, corresponding to no climbing, weak climbing-up, weak climbing-down, medium climbing-up, medium climbing-down, strong climbing-up, and strong climbing-down respectively. The 7 scenarios adopt sparse coding, with each scenario corresponding to a unique activation bit and the remaining bits being 0;

[0217] ②Intensity grading standard: Based on F(a, b), the formula for climbing intensity is defined as follows:

[0218] (twenty four)

[0219] Among them, in photovoltaic power ramp-up events, upward and downward are two basic types defined based on the direction and trend of power change. Upward represents an event in which photovoltaic power continuously increases within a certain period of time and the change exceeds a set threshold; downward represents an event in which photovoltaic power continuously decreases within a certain period of time and the change exceeds a set threshold.

[0220] ③ One-Hot climbing event encoding is performed based on climbing intensity and climbing direction. Each of the seven scenarios corresponds to a unique encoding vector, as shown in Table 2.

[0221] Table 2: One-Hot Climbing Event Coding Based on Climbing Intensity and Climbing Direction

[0222]

[0223] Traverse all valid intervals (K=1) that have passed the continuity constraint verification, and generate the corresponding One-Hot encoded vector set based on the strength and direction of F(a, b). Each of these corresponds one-to-one with the time nodes of the merged sequence Q', forming a three-dimensional feature group of time, encoding, and power trend.

[0224] By using the activation bits of the One-Hot encoded vector, the climbing direction and intensity level of the target section can be quickly determined, providing a basis for dynamic coefficient adjustment and replacing the original repetitive determination based on the F value, thus improving computational efficiency.

[0225] As shown in Table 2, when the activation bit in the encoding vector is the 2nd, 4th, or 6th bit, it is determined to be an upward climb; when the activation bit is the 3rd, 5th, or 7th bit, it is determined to be a downward climb. By combining the activation bits of the encoding vector with the corresponding mean F value, the weak, medium, and strong levels are accurately matched, providing a reference for the coefficient adjustment range under extreme weather conditions.

[0226] The optimal climbing range is determined by using the bidirectional dynamic coefficient objective function G(a,b). The uplink / downlink calculation logic is switched based on the one-hot encoding recognition result. The dynamic coefficient is optimized and adjusted by combining the climbing intensity and weather type.

[0227] (41) Upward climbing (encoding activation bits 2, 4, 6):

[0228] (25)

[0229] (42) Downward climbing (encoding activation bits 3, 5, 7):

[0230] (26)

[0231] In the formula, These are the uphill and downhill climbing dynamic coefficients, respectively, to adapt to the climbing acceleration / deceleration characteristics; w is the correction factor for extreme weather. s Assign values ​​to the climbing intensity weights based on One-Hot encoding;

[0232] The assignment principles are as follows:

[0233] (27)

[0234] Based on the two-way dynamic coefficient objective function formula, the range of the optimal climbing interval is obtained;

[0235] Upward climbing event: Take the sub-interval corresponding to the maximum value of the objective function G(a,b) as the optimal upward climbing interval (the interval with the best climbing intensity and rate has the most significant impact on power grid dispatch).

[0236] Downward climbing event: Take the sub-interval corresponding to the minimum value of the objective function G(a,b) as the optimal downward climbing interval;

[0237] Finally, the results are output, simultaneously showing the start and end times of the optimal interval, the one-hot encoding type (intensity + direction), the objective function value, and the extreme weather association markers, which can provide accurate feature inputs for subsequent slope event probability prediction.

[0238] Example 2:

[0239] This embodiment provides a photovoltaic ramping event identification system based on improved SDA and multidimensional coupling discrimination, including:

[0240] The data compression module is used to collect historical meteorological and power data to construct a feature set, and uses an adaptive SDA algorithm based on bidirectional asymmetric gate width improvement to compress the data and generate an initial trend sequence Q.

[0241] The reconstruction module uses an adaptive time window and weather adaptation principle to merge and reconstruct trend segments, resulting in a reconstruction sequence Q' that better reflects the actual climbing process, effectively suppressing oversegmentation caused by noise.

[0242] The quantization module, based on the reconstructed sequence, constructs a multi-dimensional coupled climbing discrimination index system and calculates the dynamic weight score function and continuity constraints to achieve accurate identification and intensity quantification of climbing events.

[0243] The identification module (4) uses One-Hot encoding to classify events and obtains a structured classification result of climbing types. By defining a bidirectional dynamic coefficient objective function, the optimal climbing identification interval is adaptively determined to improve the robustness and practicality in different scenarios.

Claims

1. A photovoltaic hill event recognition method based on improved SDA and multi-dimensional coupling discrimination, characterized in that, The steps are as follows: (1) Collect historical meteorological and power data to construct a feature set, and use an adaptive SDA algorithm based on bidirectional asymmetric gate width improvement to compress the data and generate an initial trend sequence Q; In the adaptive SDA algorithm based on bidirectional asymmetric gate width improvement: Let the upper boundary of the asymmetric gate width be n up , the lower boundary be n down , then the actual upper gate width and lower gate width are obtained as follows: （1） （2） wherein is the actual upper door width, is the actual lower door width, P N is the installed photovoltaic capacity, satisfying ; The slope boundary is: （3） （4） wherein k max,up is the maximum allowed slope in the rising direction; k min,down is the minimum allowed slope in the falling direction, P i is the photovoltaic power value of the current data point, P s is the power value of the starting point of the trend line currently being constructed, P i -P s is the power difference between the current data point and the starting point; t i is the time of the current data point, t s is the time of the starting point of the trend line segment; Compression conditions are: When P i ≥ P s , an upward trend, using the upward gate width to determine: （5） When P i When P s When P s When P s When P s When P s When P s When P s （6） In the formula, is the slope of the current trend segment; According to the change direction of the photovoltaic power, different upper and lower allowable deviations are adopted to obtain a gate width ; （8） The permissible slope range is: （9） And it satisfies: （10）； (2) The trend segments are merged and reconstructed using the adaptive time window and weather adaptation principle to obtain a reconstructed sequence Q' that better reflects the actual climbing process; (3) Based on the reconstructed sequence, a multi-dimensional coupled climbing discrimination index system is constructed, and the dynamic weight score function and continuity constraint are calculated to achieve accurate identification and intensity quantification of climbing events; (4) One-Hot coding is used to classify events and obtain structured climbing type classification results. By defining a bidirectional dynamic coefficient objective function, the optimal climbing identification interval is adaptively determined to improve robustness and practicality in different scenarios.

2. The photovoltaic hill event recognition method based on improved SDA and multi-dimensional coupling discrimination of claim 1, wherein, In step (1), the initial trend sequence Q of photovoltaic power output is defined as the output of the adaptive SDA algorithm with improved bidirectional asymmetric gate width: （11） wherein is the photovoltaic power trend sequence, is the photovoltaic trend power value at time t, t = 1, 2,..., m.

3. The photovoltaic hill event recognition method based on improved SDA and multi-dimensional coupling discrimination of claim 2, wherein, In step (2), specifically: (21) Based on weather type and power curve fluctuation characteristics, define extreme weather correction coefficient and relaxation coefficient. Extreme weather includes cold waves and sandstorms. The definitions and values ​​of the two are as follows: （12） （13） In the formula, K e is an extreme weather correction factor; K s is an extreme weather relaxation factor; (22) Bump event preliminary identification: a bump event is a discrete fluctuation with power fluctuation amplitude less than 0.1P, duration not more than one hour and no obvious periodic law, traversing the trend sequence Q, the discrete interval satisfying the following conditions is defined as a bump event: N and and , the starting time t s , the ending time t e and the power fluctuation amplitude of each bump event are recorded, wherein Q t is the photovoltaic output trend value at time t, Q t+1 is the photovoltaic output trend value at time t+1, is the duration of the bump event, and h is the time unit; (23) Adaptive window parameter setting: Adjust the merging window width W and density threshold D based on the weather type, the parameters are corrected by the extreme weather correction coefficient K e Assisted correction; Define the base window width W0 based on weather type as: （14） Based on this, the formula for calculating the combined window width under extreme weather conditions is as follows: W = W0 x K e (15) After being corrected by an extreme weather correction factor, the window width is dynamically enlarged under extreme weather conditions to adapt to the need for merging cluttered bump events; The density threshold is used to constrain the merging intensity. The density threshold D is set to D=2 on sunny days and D=3 on cloudy or extreme weather days. (24) Density-aware merging: For each extracted bump event, a time window of width W is constructed , and the number of bump events N within the window is counted. If N ≥ D, the event is determined to be a correlated bump event, and a linear interpolation is used to merge it with the previous trend segment. The merged trend value is （16） In the formula, is the merged trend value, Q t+1 , Q t-1 is the main trend power value before and after the bump event; The overall representative time proportion coefficient, wherein t is any time point in the merged section, t s-1 , t e+1 are respectively a time point before the bump event actually occurs and a time point after the bump event ends; w is a fluctuation reserved weight, which functions to balance the trend smoothness and the bump event correlation characteristics, and avoid distortion of the merged trend. If N < D, it is determined to be an isolated bump event, which is directly removed and filled with adjacent trend values ​​linearly to eliminate noise interference; (25) Trend sequence reconstruction: Integrate the merged continuous trend segments and non-bump trend segments to generate a reconstructed sequence Q', ensuring the temporal continuity of the sequence and the integrity of the trend.

4. The photovoltaic hill event recognition method based on improved SDA and multi-dimensional coupling discrimination of claim 3, wherein, In step (3), a multi-dimensional coupled climbing discrimination index system is constructed, defining the interval (a, b), where a and b are time nodes, satisfying a < b. The power difference index, climbing rate index, period consistency index, and extreme weather adaptation index are all associated with the extreme weather coefficient output by the front-end preprocessing and the merged sequence Q'. The formulas for each index and the comprehensive discrimination result are defined as follows: （17） In the formula, K is a comprehensive discrimination result, when K=1, the discrimination interval (a, b) is an effective climbing event; when K=0, the discrimination interval (a, b) is a non-climbing event or an invalid fluctuation; K0 is a power difference index, K1 is a climbing rate index, K2 is a period consistency index, and K3 is an extreme weather adaptation index; the round() function is rounding, which ensures that the output value is 0 or 1.

5. The photovoltaic hill event recognition method based on improved SDA and multi-dimensional coupling discrimination of claim 4, wherein, In step (3), the power difference index K0 is used to determine whether the power change range in the interval reaches the climbing threshold. The threshold is adaptively adjusted under extreme weather conditions. The discriminant of K0 is: （18） In the formula, is the trend value of the combined sequence at time b, a after step (2), and θ is the basic power difference threshold; K e is the extreme weather correction coefficient; Climbing rate index K1: Determines whether the rate of change of power in the interval meets the climbing requirements. The rate threshold is fine-tuned under extreme weather conditions. The K1 discriminant is: （19） In the formula, is a base minimum ramp rate threshold; Periodic consistency index K2: Photovoltaic output is affected by the solar irradiance pattern and has a significant intraday periodicity. On sunny days, the power gradually increases after sunrise and gradually decreases before sunset. On cloudy days, there is also a relatively stable periodic fluctuation pattern. By comparing with the historical trend of the same period, random fluctuation interference is eliminated, and the correlation threshold is relaxed under extreme weather conditions. （20） where H a-b is the historical photovoltaic power output trend sequence of the same period, is the Pearson correlation coefficient, which is used to measure the similarity of power sequence trends, and the value range is [-1, 1]; is the basic period correlation threshold, and different basic values are set according to the correlation analysis of actual data; K s is the extreme weather relaxation coefficient; Extreme weather adaptation index K3: By quantifying the correlation between actual power mutations and theoretical power changes in extreme weather, it enables the characterization and identification of extreme weather-driven climbing events. At the same time, combined with the extreme weather coefficient system output in step (1), it avoids misjudging normal power mutations under extreme weather as invalid fluctuations and prevents abnormal fluctuations caused by non-extreme factors from being misjudged as extreme climbing events. Finally, it improves the full-scene adaptation capability of the four-dimensional coupled discrimination system. The K3 discriminant is as follows: （21） In the formula, K3 is the theoretical variation of photovoltaic output under extreme scenarios, and is 0 under non-extreme weather. The extreme weather adaptation index K3 automatically shields the interference of the conventional working condition discrimination, so K3 = 1 under non-extreme scenarios, and the extreme scenario is extreme weather.

6. The photovoltaic hill event recognition method based on improved SDA and multi-dimensional coupling discrimination of claim 5, wherein, In step (3), the dynamic weighted score function and continuity constraint are calculated. After incorporating the weight adjustment for extreme weather into the interval (a, b) score function, the formula is: （22） In the formula, α is a time weight coefficient; β is an amplitude weight coefficient, K e is an extreme weather correction coefficient; is an interval average speed, a positive value representing an uphill climb and a negative value representing a downhill climb; The larger the absolute value of F(a,b), the higher the climbing intensity, and the more obvious the score differentiation of high-intensity climbing under extreme weather conditions. At the same time, the continuity constraint should be satisfied. For any time node a < c < b, the following condition must be met: （23） In the formula, is a preset basic change rate tolerance threshold; Formula (23) is a constraint on the continuity of the climbing interval. By splitting the interval (a, b) into sub-intervals (a, c) and (c+1, b), the deviation between the total score of the whole interval and the sum of the scores of the two sub-intervals is required to meet the constraint range. If the deviation exceeds the range, it means that the score in the interval (a, b) changes drastically and the interval contains non-continuous climbing sections. The interval range needs to be adjusted and recalculated.

7. The photovoltaic hill event recognition method based on improved SDA and multi-dimensional coupling discrimination of claim 6, wherein, In step (4), the One-Hot encoding rules are as follows: ① Encoding dimension definition: The vector dimension is 7, corresponding to no climbing, weak climbing-up, weak climbing-down, medium climbing-up, medium climbing-down, strong climbing-up, and strong climbing-down respectively. The 7 scenarios adopt sparse coding, with each scenario corresponding to a unique activation bit and the remaining bits being 0; ②Intensity grading standard: Based on F(a, b), the formula for climbing intensity is defined as follows: （24） Among them, in photovoltaic power ramp-up events, upward and downward are two basic types defined based on the direction and trend of power change. Upward represents an event in which photovoltaic power continuously increases within a certain period of time and the change exceeds a set threshold; downward represents an event in which photovoltaic power continuously decreases within a certain period of time and the change exceeds a set threshold. ③ One-Hot climbing event encoding is performed based on climbing intensity and climbing direction, with each of the 7 scenarios corresponding to a unique encoding vector; Traverse all the valid intervals verified by continuity constraints, generate the corresponding One-Hot encoding vector set based on the intensity and direction of F(a, b) Corresponding to the time nodes of the merged sequence Q', form a three-dimensional feature group of time, encoding and power trend. The climbing direction and intensity level of the target section are determined by the activation bits of the One-Hot encoded vector.

8. The photovoltaic hill event recognition method based on improved SDA and multi-dimensional coupling discrimination of claim 7, wherein, In step (4), the optimal climbing range is determined by the bidirectional dynamic coefficient objective function G(a,b), and the uplink / downlink calculation logic is switched based on the one-hot encoding recognition result. The dynamic coefficient is optimized and adjusted by combining the climbing intensity and weather type. (41) Uphill climbing: （25） (42) Downhill climbing: （26） In the formula, are respectively the uplink and downlink climbing dynamic coefficients; is an extreme weather correction coefficient, w s is a climbing intensity weight, which is assigned based on One-Hot encoding; The assignment principle is as follows: （27） Based on the two-way dynamic coefficient objective function formula, the range of the optimal climbing interval is obtained; Upward climbing event: Take the sub-interval corresponding to the maximum value of the objective function G(a,b) as the optimal upward climbing interval; Downward climbing event: Take the sub-interval corresponding to the minimum value of the objective function G(a,b) as the optimal downward climbing interval.

9. A photovoltaic hill event recognition system based on improved SDA and multi-dimensional coupling discrimination, applied to the photovoltaic hill event recognition method based on improved SDA and multi-dimensional coupling discrimination in claim 1, characterized in that, include: The data compression module is used to collect historical meteorological and power data to construct a feature set, and uses an adaptive SDA algorithm based on bidirectional asymmetric gate width improvement to compress the data and generate an initial trend sequence Q. The reconstruction module uses an adaptive time window and weather adaptation principle to merge and reconstruct trend segments, resulting in a reconstruction sequence Q' that better reflects the actual climbing process. The quantization module, based on the reconstructed sequence, constructs a multi-dimensional coupled climbing discrimination index system and calculates the dynamic weight score function and continuity constraints to achieve accurate identification and intensity quantification of climbing events. The identification module uses One-Hot encoding to classify events and obtain structured climbing type classification results. By defining a bidirectional dynamic coefficient objective function, it adaptively determines the optimal climbing identification interval, thereby improving robustness and practicality in different scenarios.

Images