A virtual power plant adjustable capacity dominant factor dynamic extraction method and system

Abstract

The application belongs to the technical field of virtual power plant scheduling and distributed resource management, and discloses a kind of virtual power plant adjustable capacity leading factor dynamic extraction method and system, by building the directed correlation graph across time scale to integrate day, day and real-time scale causal transmission, and set variable sliding window moves in time and factor dimension synchronously, based on partial rank correlation coefficient dynamic output each factor and the correlation intensity of adjustable capacity;Subsequently, the dominant factor set is extracted according to the correlation ranking, the distribution difference of the adjacent window ranking list is calculated using information entropy to automatically generate the switching marker, while maintaining the influence score sequence of each factor and using the weighted double exponential smoothing model to predict the future trend. Using this method effectively solves the problems of missing cross-scale correlation, inability to capture dynamic drift of leading factors, lack of quantitative criteria for reanalysis timing, and inability to quantify time-varying trajectory of influence, improving the accuracy and foresight of multi-time scale scheduling decisions.

Description

Technical Field

[0001] This invention belongs to the field of virtual power plant scheduling and distributed resource management technology, and in particular relates to a method and system for dynamically extracting the dominant factors of virtual power plant adjustability. Background Technology

[0002] Currently, the high proportion of renewable energy integrated into the power system has become a core characteristic of energy transition. This has significantly increased the randomness and volatility of power system operation, highlighting the growing supply-demand imbalance of system flexibility adjustment resources. Virtual power plants (VPS), as a new form of resource-coordinated regulation that integrates distributed power sources, energy storage systems, and controllable loads, can provide various flexible services such as power up / down adjustments, reserve capacity, and ramp rate support through optimized coordination of internal diverse resources. They have become a key technological means to improve the grid's renewable energy absorption capacity and operational flexibility. Adjustability is the core value carrier for VPS's participation in grid dispatching and trading and the provision of ancillary services. Accurately identifying the key dominant factors affecting adjustability is a fundamental prerequisite for formulating day-ahead dispatch plans, conducting intraday rolling corrections, and implementing real-time power allocation—a multi-level dispatching decision-making process that plays a crucial supporting role in ensuring the economic efficiency of VPS operation and the safe and stable operation of the power grid.

[0003] However, existing technologies for identifying and analyzing the dominant factors of virtual power plant adjustable capacity still have many limitations and cannot meet the actual needs of refined scheduling across multiple time scales. First, time scale analysis is fragmented, ignoring cross-scale causal transmission relationships. Existing methods typically divide the day-ahead, intraday, and real-time time scales independently and construct separate sets of influencing factors, failing to consider the cross-scale coupled physical process of day-ahead weather forecast errors propagating to the intraday energy storage state of charge adjustment depth and subsequently affecting real-time adjustable capacity. This easily leads to deviations in the determination of key factors at different time scales, resulting in misjudgments where key factors determined day-ahead are no longer operational bottlenecks in the real-time stage, or vice versa. Second, the static ranking model cannot reflect the dynamic drift characteristics of dominant factors. Existing methods mostly complete a feature importance ranking based on a single fixed time section and use this ranking result fixedly throughout the entire scheduling cycle, failing to capture the temporal evolution of dominant factors over time during actual operation. The existing methods are inconsistent and difficult to adapt to the dynamic changes in scenarios, such as the early morning when photovoltaic forecast accuracy is the primary factor, the midday shift to energy storage charging, and the evening peak shift to load response willingness. Third, there is a lack of quantitative triggering criteria for the switching of dominant factors. Current dispatching systems rely heavily on manual experience to determine the timing of reanalysis of key factors, without establishing an objective and quantitative switching triggering mechanism. This can easily lead to problems such as too low a recalculation frequency resulting in delayed and invalid analysis results, or too high a frequency causing redundant and wasteful computing resources. Fourth, they only output static importance results, without quantifying the time-varying trajectory of each factor's influence. Existing methods can only provide the importance scores and rankings of influencing factors at fixed times, failing to present the dynamic changing trend of each factor's importance over time. This makes it impossible for dispatchers to predict the strengthening or weakening of factors' influence, hindering proactive dispatching adjustments.

[0004] It is evident that existing virtual power plant adjustable capacity dominant factor identification technologies suffer from several shortcomings, including the lack of cross-timescale coupling correlation, inability to capture the dynamic drift characteristics of dominant factors, lack of quantitative triggering criteria for factor reanalysis timing, and failure to quantify the time-varying evolution trajectory of factor influence. These shortcomings make it difficult to support accurate scheduling decisions across multiple timescales. Summary of the Invention

[0005] This invention provides a method and system for dynamically extracting the dominant factors of the adjustable capacity of virtual power plants. This method can effectively solve the defects of existing virtual power plant adjustable capacity dominant factor identification technology, such as the lack of cross-timescale coupling correlation, the inability to capture the dynamic drift characteristics of dominant factors, the lack of quantitative triggering criteria for factor reanalysis, and the failure to quantify the time-varying evolution trajectory of factor influence. It can support accurate scheduling decisions at multiple time scales.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for dynamically extracting the dominant factors of the adjustable capacity of a virtual power plant includes: Based on the candidate influencing factor set at each time scale, a cross-scale directed correlation edge is constructed to form a multi-time scale correlation graph; where the time scale includes three time scales: day-ahead, intraday, and real-time, which are divided by the scheduling process of the virtual power plant. Based on the multi-timescale correlation diagram, a variable-length sliding time window is set up so that the sliding window slides synchronously in the time axis dimension and the influencing factor domain dimension. The partial rank correlation coefficient is calculated for the operating data in each sliding window, and the correlation strength between each candidate influencing factor and the adjustable capacity of the virtual power plant is output. The candidate influencing factors in each window are sorted based on the correlation strength. The candidate influencing factors with a preset position before sorting are extracted as the dominant factor set of the current window, and the dominant factor sorting list is obtained. The distribution difference of the dominant factor sorting list between adjacent sliding windows is calculated based on the information entropy. When the distribution difference exceeds a set threshold, a dominant factor switching flag is generated. Based on the correlation strength of the output of each sliding window, the time series of the influence scores of each candidate influencing factor as a function of the window index is maintained. A pre-constructed weighted double exponential smoothing model is used to predict the trend of influence changes in multiple sliding windows in the future, so as to obtain the trend prediction information of the influence of each candidate influencing factor and the prediction result of the dominant factor in the next window. The output combines the ranking list of dominant factors, the switching marker of dominant factors, the trend prediction information of the influence of each candidate influencing factor, and the prediction results of the dominant factors in the next window.

[0007] Furthermore, the construction of cross-scale directed correlation edges based on the candidate influencing factor sets at each time scale to form a multi-time scale correlation graph includes: For any upstream-scale candidate influencing factor and downstream-scale candidate influencing factor that have an influence relationship, a directed edge is established from the upstream scale to the downstream scale, and the correlation strength coefficient is calculated. :

[0008] in, for The state transition function; τ is the cross-scale time delay, and T is the statistical period; For time; For the upstream timescale Each candidate influencing factor The values ​​of the driving variables at each moment; This represents the state variable value of the j-th candidate influencing factor at time t in the downstream time scale; The multi-timescale correlation graph The adjacency list structure is used for storage, represented as follows:

[0009] in, These are the set of candidate influencing factor nodes, the set of directed edges, and the set of correlation strength coefficients, respectively.

[0010] Furthermore, within the variable-length sliding time window, the window length is adaptively adjusted according to the fluctuation of the adjustable capability index, using the following adjustment formula:

[0011] in The standard deviation of the adjustable capability index within the window; η is the mean of the adjustable capability index within the window; η is the adjustment coefficient. , These are the minimum and maximum window lengths, respectively. , These represent the maximum and minimum values, respectively. , These represent the window lengths corresponding to the nth and (n+1)th sliding time windows, respectively.

[0012] Furthermore, when the sliding window slides synchronously along the time axis and the influencing factor domain, a dynamic selection mechanism is used to determine the set of active candidate influencing factors, as follows: All candidate influencing factors will be added to the active candidate influencing factor set according to the preset cycle; The rate of change of the association strength coefficient of directed edges in a multi-timescale association graph is detected. If the rate of change of the association strength coefficient exceeds the preset rate of change threshold, the upstream candidate influencing factors, downstream candidate influencing factors and their first-order adjacent candidate influencing factors of the corresponding directed edge are added to the active candidate influencing factor set. If the number of elements in the active candidate influencing factor set does not meet the preset requirement, then the candidate influencing factor with the highest historical skewed correlation coefficient will be added.

[0013] Furthermore, the extraction of candidate influencing factors with preset rankings before sorting as the dominant factor set of the current window includes: The partial rank correlation coefficients of each candidate influencing factor are sorted in descending order of absolute value, and those that meet the preset cumulative contribution rate threshold are selected. The top K candidate influencing factors are taken as the dominant factor set of the current window, where the value of K satisfies:

[0014] In the formula, The number arranged in descending order of absolute value Partial rank correlation coefficient; The total number of effective factors; The traversal variable for the number of candidate factors; For the first The partial rank correlation coefficients corresponding to the 1 effective candidate influencing factors; min() is the minimum value.

[0015] Further, the step of calculating the distribution difference of the dominant factor ranking list between adjacent sliding windows based on information entropy, and generating a dominant factor switching flag when the distribution difference exceeds a set threshold, includes: Convert the ranking list of dominant factors between adjacent sliding windows into a probability distribution. Calculate the KL divergence of two probability distributions. And based on KL divergence, the cumulative residual entropy is calculated. :

[0016] In the formula, For smoothing weighting coefficients; This is the cumulative residual entropy of the previous window; When the cumulative residual entropy exceeds the preset switching threshold, a dominant factor switching flag is generated, the current switching time and the dominant factor set before and after the switching are recorded, and the cumulative residual entropy is reset to 50% of the current KL divergence value.

[0017] Furthermore, after generating the dominant factor switching marker, a causal chain warning step is also included: Based on the multi-timescale correlation diagram, the upstream candidate influencing factors are traced backward from the newly emerging dominant factor generated in this switch. If the directed edge correlation strength coefficient between the upstream candidate influencing factor and the newly emerging dominant factor is greater than the preset edge weight threshold, the warning information of the corresponding causal chain is output.

[0018] Furthermore, the construction and updating process of the weighted double exponential smoothing model specifically includes: For each candidate influencing factor, select the most recent preset number of historical skewed correlation coefficient values ​​and assign time decay weights to each historical value; The horizontal and trend components of the candidate influencing factor are initialized using weighted linear regression. The horizontal and trend components are updated window by window according to the following recursive formula:

[0019]

[0020] In the formula, , Both represent horizontal components; , Both represent trend components; , These represent the corresponding smoothing coefficients; Indicates the first One candidate influencing factor; Indicates the first A sliding window; Indicates the first The first sliding window The partial rank correlation coefficient of each candidate influencing factor.

[0021] Furthermore, the process of obtaining trend prediction information on the influence of each candidate influencing factor also includes a step of predicting the entry time of the dominant interval, specifically including: For candidate influencing factors whose current influence is below the preset dominance threshold and whose trend component is positive, calculate the minimum window number required for their predicted influence value to reach the dominance threshold. After converting this to a corresponding time, output the predicted time when the candidate influencing factor is expected to enter the dominance factor set; where, future... The predicted value of the influence of each window for:

[0022] In the formula, The index of the sliding window; Among them, minimum number of windows satisfy:

[0023] In the formula, The preset dominant threshold; This is the maximum value.

[0024] A system for dynamically extracting the dominant factors of adjustable capacity in a virtual power plant includes: The module is used to construct directed association edges across scales based on the candidate influencing factor set at each time scale, forming a multi-time scale association graph; The calculation module is used to set a variable-length sliding time window based on the multi-timescale correlation diagram, so that the sliding window slides synchronously in the time axis dimension and the influencing factor domain dimension, calculates the partial rank correlation coefficient of the operating data in each sliding window, and outputs the correlation strength between each candidate influencing factor and the adjustable capacity of the virtual power plant. The sorting module is used to sort the candidate influencing factors in each window based on the correlation strength, extract the candidate influencing factors with the preset position before sorting as the dominant factor set of the current window, and obtain the dominant factor sorting list; calculate the distribution difference of the dominant factor sorting list between adjacent sliding windows based on information entropy, and generate a dominant factor switching flag when the distribution difference exceeds a set threshold. The prediction module is used to maintain the time series of the influence scores of each candidate influencing factor as a function of the window index based on the correlation strength of the output of each sliding window. It uses a pre-built weighted double exponential smoothing model to predict the trend of influence changes in multiple sliding windows in the future, and obtains the trend prediction information of the influence of each candidate influencing factor and the prediction result of the dominant factor in the next window. The output module is used to integrate and output the dominant factor ranking list, dominant factor switching marker, trend prediction information of the influence of each candidate influencing factor, and the prediction results of the dominant factor in the next window.

[0025] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a method for dynamically extracting the dominant factors of adjustable capacity in virtual power plants. It integrates causal transmission at the day-ahead, intraday, and real-time scales by constructing a directed correlation graph across time scales, and sets a variable sliding window that moves synchronously across time and factor dimensions. Based on the partial-rank correlation coefficient, it dynamically outputs the correlation strength between each factor and adjustable capacity. Subsequently, it extracts the dominant factor set by correlation ranking, uses information entropy to calculate the distribution difference of adjacent window ranking lists to automatically generate switching markers, maintains the influence score sequence of each factor, and uses a weighted double-exponential smoothing model to predict future trends. The correlation graph reflects cross-scale physical coupling, the sliding window and partial-rank correlation enable dynamic evaluation, the information entropy difference quantifies the switching triggering conditions, and the trend prediction reveals the evolutionary law of influence. This method effectively solves the problems of missing cross-scale correlations, inability to capture the dynamic drift of dominant factors, lack of quantitative criteria for reanalysis timing, and failure to quantify the time-varying trajectory of influence, thus improving the accuracy and foresight of multi-time-scale scheduling decisions. Attached Figure Description

[0026] Figure 1 A flowchart illustrating the implementation of a method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant, as provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structured results provided in an embodiment of the present invention; Figure 3 This is a core flowchart of a method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant, provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of a system for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant, provided in an embodiment of the present invention. Detailed Implementation

[0027] To further understand the content of this invention, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0028] The technical terms involved in this invention are explained as follows: VPP: short for Virtual Power Plant.

[0029] SOC stands for State of Charge.

[0030] Top-K: refers to extracting the top K elements in a sorted or selected dataset.

[0031] RESTful API stands for Representational State Transfer Application Programming Interface.

[0032] AWDES model: Weighted double exponential smoothing model.

[0033] KL divergence: short for Kullback-Leibler divergence, is a measure in information theory used to measure the difference between two probability distributions, representing the information loss of one distribution relative to another.

[0034] JSON: JavaScript Object Notation, is a lightweight data interchange format.

[0035] As described in the background section, existing technologies have the following shortcomings in identifying and analyzing the factors affecting the adjustability of virtual power plants: First, the time scales are fragmented, ignoring cross-scale correlations. Existing methods typically analyze the day-ahead, intraday, and real-time time scales independently, establishing separate sets of influencing factors. However, actual physical processes are strongly coupled: errors in day-ahead weather forecasts directly affect the adjustment depth of intraday energy storage SOC, thereby altering real-time adjustability. This cross-time-scale causal transmission relationship is completely ignored in existing methods, leading to misjudgments such as "factors identified as critical day-ahead are no longer bottlenecks in the real-time stage" or vice versa.

[0036] Second, static ranking fails to reflect the dynamic shift of dominant factors. Existing methods rank features by importance at a fixed time point (e.g., when a scheduling plan is submitted) and then use that ranking throughout the scheduling cycle. However, in actual operation, key factors "drift" over time—the dominant factor in the early morning might be photovoltaic forecast accuracy, at noon it might shift to energy storage SOC, and during the evening peak it might become load response willingness. Existing static methods cannot capture this temporal evolution.

[0037] Third, there is a lack of quantitative criteria for switching of dominant factors. Current scheduling systems rely heavily on human experience to determine "when key factors need to be reanalyzed," lacking an objective and quantitative switching trigger mechanism. This results in either recalculation being too infrequent (using outdated information) or too frequent (wasting computing resources).

[0038] Fourth, existing methods only extract factors without quantifying their time-varying influence. The importance of influencing factors is output as a static score, without providing a curve showing how that score changes over time. Schedulers cannot determine whether the importance of a factor is increasing or decreasing, and therefore cannot make predictive adjustments.

[0039] Therefore, there is an urgent need for a method for extracting key influencing factors with adjustable capabilities that can perform cross-timescale correlation analysis, dynamically track the drift trajectory of dominant factors, and provide switching warnings, so as to support the refined scheduling and adaptive control of virtual power plants.

[0040] To achieve the above objectives, this embodiment provides a method for dynamically extracting the dominant factors of the adjustable capacity of a virtual power plant. This method achieves adaptive, dynamic, and predictable extraction of the dominant factors by constructing a multi-timescale correlation graph, designing a dual-domain dynamic sliding window, and introducing an entropy change detection and trajectory prediction mechanism.

[0041] For example, such as Figure 3 As shown, this embodiment provides a method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant, including: Based on the candidate influencing factor set at each time scale, a cross-scale directed correlation edge is constructed to form a multi-time scale correlation graph; where the time scale includes three time scales: day-ahead, intraday, and real-time, which are divided by the scheduling process of the virtual power plant. Based on the multi-timescale correlation diagram, a variable-length sliding time window is set up so that the sliding window slides synchronously in the time axis dimension and the influencing factor domain dimension. The partial rank correlation coefficient is calculated for the operating data in each sliding window, and the correlation strength between each candidate influencing factor and the adjustable capacity of the virtual power plant is output. The candidate influencing factors in each window are sorted based on the correlation strength. The candidate influencing factors with a preset position before sorting are extracted as the dominant factor set of the current window, and the dominant factor sorting list is obtained. The distribution difference of the dominant factor sorting list between adjacent sliding windows is calculated based on the information entropy. When the distribution difference exceeds a set threshold, a dominant factor switching flag is generated. Based on the correlation strength of the output of each sliding window, the time series of the influence scores of each candidate influencing factor as a function of the window index is maintained. A pre-constructed weighted double exponential smoothing model is used to predict the trend of influence changes in multiple sliding windows in the future, so as to obtain the trend prediction information of the influence of each candidate influencing factor and the prediction result of the dominant factor in the next window. The output combines the ranking list of dominant factors, the switching marker of dominant factors, the trend prediction information of the influence of each candidate influencing factor, and the prediction results of the dominant factors in the next window.

[0042] The extraction method provided in this embodiment will be further explained below with reference to the accompanying drawings: For example, such as Figure 1 As shown in the figure, this embodiment provides a method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant. The specific steps are as follows: Step 1: Construction of Multi-Time-Scale Association Graph: The scheduling process of the virtual power plant is divided into three time scales: day-ahead, intraday, and real-time. Candidate influencing factor sets are defined for each scale, and directed association edges across scales are constructed to form a multi-time-scale association graph. Specifically, this includes: Step 1.1, Time Scale Division: Day-ahead scale (D): with a step size of 1 hour, covering the next 24 hours, focusing on predictive factors; Intraday Scale (I): Using a 15-minute step, covering the next 4 hours, focusing on state-related factors; Real-time scale (R): with a step size of 1 minute, covering the next 15 minutes, focusing on response factors.

[0043] Step 1.2, Definition of Candidate Influencing Factors: Current-scale candidate factor set This includes: weather forecast errors (PV / wind power), day-ahead market electricity prices, initial SOC of energy storage, and load baseline forecast deviations; Intraday Scale Candidate Factor Set This includes: real-time SOC of energy storage, real-time output of distributed power sources, contracted capacity of interruptible loads, ramp rate margin, etc. Real-time scale candidate factor set These include: frequency deviation, voltage fluctuation, real-time adjustment command response delay, communication delay, etc.

[0044] Step 1.3, Construction of cross-scale related edges: For any and like The value of will affect If the feasible range or distribution is determined, then directed edges are established. Similarly, construct... The edge.

[0045] Correlation strength coefficient Defined as:

[0046] in for The state transition function is τ, which is the cross-scale time delay (1 hour from day-ahead to day-in, and 15 minutes from day-in to real-time), and T is the statistical period. The larger the value, the stronger the constraint of upstream scale factors on downstream scale factors.

[0047] Step 1.4, Construction of cross-scale association edges: An adjacency list structure is used to store the multi-time-scale association graph. Let E be the set of directed edges and W be the set of edge weights.

[0048] Step 2: Dual-domain dynamic sliding window sampling: Design a variable-length sliding time window that slides synchronously along both the time axis and the influencing factor domain. Calculate the partial rank correlation coefficient for the data within each window and output the correlation strength between each factor and the adjustability. The specific process includes: Step 2.1, Timeline Window Definition: Let the time axis have a basic step size. Discretize into The current time is Define a sliding window with a variable length:

[0049] Where L is the window length (unit: number of sampling points). L is adaptively adjusted according to the rate of change of the adjustable capability.

[0050] in The standard deviation of the adjustable capability index within the window. Let η be the mean, and η be an adjustment factor (usually taken as 0.2). The greater the fluctuation in adjustability, the shorter the window should be to ensure timeliness.

[0051] Step 2.2, Influencing Factor Domain Sliding: Within each time window, from the full set of factors defined in step 1.2 Dynamically select the subset of factors to participate in the calculation active The selection rules are as follows: If the weight of the upstream associated edge of a factor has changed significantly recently (exceeding a threshold), then that factor and its upstream and downstream factors will be included. active ; Otherwise, it will be updated on a rolling basis according to the preset cycle.

[0052] Step 2.3, Calculation of partial rank correlation coefficient: for active Each factor in Calculate the partial rank correlation coefficient between it and the virtual power plant adjustability index AI (which can be obtained from historical response data). This is used to eliminate interference from other factors.

[0053] in Let L be the rank transformation function, and L be the number of sample points within the window. Factors And the rank mean of the adjustability index.

[0054] Step 2.4, Significance Test: For each P K Perform a t-test to remove insignificant factors with p-values ​​greater than 0.05, and obtain the effective influencing factors of the current window and their partial rank correlation coefficients.

[0055] Step 3: Dominant Factor Ranking and Switching Detection: Within each sliding window, influencing factors are ranked according to their correlation strength, and the Top-K factors are extracted as the dominant factor set for the current window. The distribution difference of the dominant factor ranking lists between adjacent windows is calculated based on information entropy. When the entropy change exceeds a set threshold, a dominant factor switching flag is triggered. The specific process includes: Step 3.1, Ranking of Dominant Factors: Effective influencing factors are categorized as follows: Sort in descending order, take the first few. The set of dominant factors for the current window The value of K satisfies the cumulative contribution rate threshold. :

[0056] That The number arranged in descending order of absolute value Partial rank correlation coefficient This represents the total number of effective factors.

[0057] Step 3.2, Calculation of the entropy of the dominant factor sorting list: Connect two adjacent windows and The dominant factor ranking list is converted into probability distribution vectors respectively. Define the order between sorted lists. Kullback-Leibler divergence is a measure of distributional dissimilarity.

[0058] in Defined as: if the first The factor of the position in the window The ranking is ,but (The first position has the highest weight, and the last position has the lowest weight).

[0059] like Not included If a factor is in the equation, then the factor's... Take the minimum constant .

[0060] Step 3.3, Switching Thresholds and Labels: Define cumulative residual entropy For moving average Divergence:

[0061] in For smoothing coefficients, .

[0062] when When the empirical value is 0.25, the dominant factor switching flag is triggered. And record the switching time. And the dominant factor set before and after the switch. Simultaneously reset. For the present Set the value to 50% to avoid continuous false triggering.

[0063] Step 3.4, Causal Chain Tracing: When the switching flag is triggered, based on the association graph G constructed in step 1.3, the upstream scale factors are traced back from the newly emerging dominant factor after the switching. If the new dominant factor... There are upstream related edges and If the current dominant factor shift is related to upstream factors, an early warning message will be output: "The current dominant factor shift may be related to upstream factors." "Recent changes are related."

[0064] Step 4: Modeling the Time-Varying Trajectory of Dominant Factor Influence: For each candidate influencing factor, maintain the time series of its influence score as a function of the window index, and use a weighted double exponential smoothing model to predict the trend of influence changes over multiple future windows; the specific process includes: Step 4.1, Construction of the Time-Varying Influence Sequence: For each candidate influencing factor x k Maintain the partial rank correlation coefficient sequence within its historical window. , where n is the index of the current window.

[0065] Step 4.2, Weighted double exponential smoothing prediction: Using the AWDES model Predict the sequence. Let the current window index be... Select the most recent A sliding window is formed using historical values, each weighted by time decay (the more recent the value, the greater the weight). Initial horizontal component. and trend components Calculated using weighted linear regression.

[0066] Recursive update formula:

[0067]

[0068] in .

[0069] Step 4.3, Multi-step forecasting and estimation of time to enter the dominant interval: future Each window (step size corresponds to the basic step size of the time scale) The predicted value of influence is:

[0070] Set the dominant factor threshold (That is, a factor may only become a dominant factor if the absolute value of its partial rank correlation coefficient exceeds 0.5). For factors that are not currently in the dominant factor set, if their... Solve for the minimum Make

[0071]

[0072] The expected number of windows to enter is Corresponding time Output "Expected" It may later become the dominant factor. As the basic step size, The horizontal component of the current window. This represents the trend component of the current window.

[0073] Step 4.4, Prediction Confidence Assessment: Confidence ,in The standard deviation of the residuals of historical influence within the window.

[0074] When confidence level When the value is less than 0.5, the additional output is "The current prediction has low reliability. It is recommended to make a comprehensive judgment based on real-time events".

[0075] Step 5: Structured Result Output: Output the current window's dominant factor ranking list, toggle markers, influence trend prediction information for each factor, and the dominant factor prediction results for the next window, and expose them externally via a RESTful API; the specific process includes: Step 5.1, Result Encapsulation: Encapsulate the evaluation results of the current window into a structured format, including: Timestamp (window end time); Ranking list of dominant factors (including the partial rank correlation coefficients of each factor). Switching markers (and factor sets before and after the switch); Predicted trends in the influence of each factor (rising / falling / stable, with expected time to enter the dominant phase); Cross-scale correlation early warning information.

[0076] Step 5.2, API Interface Exposure: Provide an interface to the outside world through a RESTful API to obtain the dominant factor of the current window, the historical window sequence, submit the list of factors to be predicted, and return the prediction results.

[0077] The extraction method provided in this embodiment has been specifically applied and implemented, and the implementation process is as follows: This embodiment provides a method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant. The example uses a virtual power plant comprising a photovoltaic power station (installed capacity 10 MW), an energy storage system (capacity 5 MWh, maximum charge / discharge power 2.5 MW), and interruptible loads (contracted capacity 3 MW). This virtual power plant participates in the day-ahead market and the intraday real-time balancing market, and the dispatch system performs dominant factor extraction every 15 minutes.

[0078] Step 1: Constructing a multi-timescale association graph: In this embodiment, the association graph is constructed according to steps 1.1 to 1.4.

[0079] Step 1.1, Time scale division: Day-ahead scale (D) with a step size of 1 hour, a total of 24 points; Intraday scale (I) with a step size of 15 minutes, a total of 16 points for the next 4 hours; Real-time scale (R) with a step size of 1 minute, a total of 15 points for the next 15 minutes.

[0080] Step 1.2, Candidate Factor Definition (Partial Example): Current-scale candidate factor set This includes: weather forecast errors (PV / wind power), day-ahead market electricity prices, initial SOC of energy storage, and load baseline forecast deviations; Intraday Scale Candidate Factor Set This includes: real-time SOC of energy storage, real-time output of distributed power sources, contracted capacity of interruptible loads, ramp rate margin, etc. Real-time scale candidate factor set The candidate factor set is shown in Table 1: Table 1 shows the candidate factor set.

[0081] Step 1.3, Constructing Cross-Scale Association Edges: Take 30 days of historical running data and calculate the weights of some edges: (For every 10% increase in photovoltaic prediction error, the energy storage SOC decreases by an additional 7.2%). (The current electricity price is weakly correlated with the contracted amount of interruptible load.) (The lower the energy storage SOC, the greater the frequency response delay).

[0082] Step 2, Dual-domain dynamic sliding window sampling: The current running time is 10:00 on a certain day (corresponding to the 5th window on the intraday scale, with a step size of 15 minutes, and the window number n=5).

[0083] Step 2.1: Adaptive adjustment of timeline window length: Basic step size minute.

[0084] The average value of the adjustable capability index AI in the previous window (ending at 09:45). Standard deviation .

[0085] Previous window length Adjustment coefficient .

[0086]

[0087] The window length is maintained at 12 sampling points (3 hours of data).

[0088] Step 2.2, Dynamic selection of influencing factor domains: Basic cycle The current window (1.5 hours) is not yet the time for a full update.

[0089] Detect edge weight changes: The rate of change increased from 0.68 in the previous window to the current 0.75. Not triggered.

[0090] The effective factor set of the previous window is There are 4 factors in total. The factor with the highest historical average will be added according to the rules. (Electricity price as of today). Final .

[0091] Step 2.3, Calculation of partial rank correlation coefficient: Acquire observation data for 12 time points within the window. Calculate the partial rank correlation coefficient between each factor and the adjustability index AI: (Real-time SOC of energy storage): ; (Real-time photovoltaic power output): ; (Interruptible load capacity): ; (Photovoltaic prediction error): ; (Electricity price as of today): (Greater than 0.05, not significant).

[0092] Step 2.4, Significance test: Elimination Effective factor set .

[0093] Step 3: Dominant Factor Ranking and Switching Detection: Step 3.1, Dominant Factor Ranking and Top-K Extraction: according to Descending order: .

[0094] Cumulative contribution rate: .Pick All factors were selected into the dominant factor set. The rankings are 1, 2, 3, and 4 respectively.

[0095] Step 3.2, Transformation of the probability distribution of the sorted list: Adjacent windows dominant factor set (Another factor within the day), and collection .

[0096] calculate .

[0097] right Row 1 (0.4), Row 2 (0.3), Row 3 (0.2) Arrangement 4 (0.1), Not in centralized assignment It remains approximately unchanged after normalization.

[0098] Step 3.3, KL divergence and cumulative residual entropy: Calculated .set up ,but .

[0099] Step 3.4, Switch between detection and marking: set up ,because Therefore No update was triggered.

[0100] Step 3.5, Causal Chain Tracing: No switch triggered, so no execution.

[0101] Step 4: Time-varying trajectory modeling of the influence of dominant factors: Factors (Energy storage real-time SOC) as an example, assuming it has already reached the window of operation. The partial rank correlation coefficient sequence for its historical 10 windows is as follows: .

[0102] Step 4.1: Construct the sequence with no missing parts.

[0103] Step 4.2, weighted double-exponential smoothing initialization yields... .

[0104] Step 4.3, Future Prediction All values ​​are above 0.5, indicating no risk of exiting the market.

[0105] Step 4.4: Confidence level close to 1.

[0106] Factors Taking interruptible load capacity as an example, the historical sequence is as follows: Initialization .

[0107] current Furthermore, the trend is downward, so there is no need to perform exit threshold prediction calculations (the current smoothing benchmark value is already below the correlation threshold), and it can be predicted that the coefficient of this factor will continue to decline.

[0108] Step 5: Output the structured results: System output as follows Figure 2 The JSON format results shown are for current window 10, time 15:00. Based on these results, the system can determine that: energy storage SOC is currently the most important dominant factor and its status should be closely monitored; the importance of photovoltaic prediction error is rising and may become a new dominant factor in the future, so it is recommended to pay attention to it in advance.

[0109] For example, such as Figure 4As shown, this embodiment also provides a dynamic extraction system for the dominant factors of the adjustable capacity of a virtual power plant, including: a construction module, used to construct cross-scale directed correlation edges based on the candidate influencing factor set at each time scale to form a multi-time scale correlation graph; The calculation module is used to set a variable-length sliding time window based on the multi-timescale correlation diagram, so that the sliding window slides synchronously in the time axis dimension and the influencing factor domain dimension, calculates the partial rank correlation coefficient of the operating data in each sliding window, and outputs the correlation strength between each candidate influencing factor and the adjustable capacity of the virtual power plant. The sorting module is used to sort the candidate influencing factors in each window based on the correlation strength, extract the candidate influencing factors with the preset position before sorting as the dominant factor set of the current window, and obtain the dominant factor sorting list; calculate the distribution difference of the dominant factor sorting list between adjacent sliding windows based on information entropy, and generate a dominant factor switching flag when the distribution difference exceeds a set threshold. The prediction module is used to maintain the time series of the influence scores of each candidate influencing factor as a function of the window index based on the correlation strength of the output of each sliding window. It uses a pre-built weighted double exponential smoothing model to predict the trend of influence changes in multiple sliding windows in the future, and obtains the trend prediction information of the influence of each candidate influencing factor and the prediction result of the dominant factor in the next window. The output module is used to integrate and output the dominant factor ranking list, dominant factor switching marker, trend prediction information of the influence of each candidate influencing factor, and the prediction results of the dominant factor in the next window.

[0110] The present invention also provides a device for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant.

[0111] The present invention also provides a computer program product, including a computer program / instruction that, when executed by a processor, implements the steps of the method for dynamically extracting the dominant factor of the adjustable capacity of the virtual power plant.

[0112] When the processor executes the computer program, it implements the steps of dynamically extracting the dominant factors of the virtual power plant's adjustable capacity, for example: constructing cross-scale directed correlation edges based on the candidate influencing factor set at each time scale to form a multi-time scale correlation graph; wherein, the time scale includes three time scales: day-ahead, intraday, and real-time, which are divided by the scheduling process of the virtual power plant. Based on the multi-timescale correlation diagram, a variable-length sliding time window is set up so that the sliding window slides synchronously in the time axis dimension and the influencing factor domain dimension. The partial rank correlation coefficient is calculated for the operating data in each sliding window, and the correlation strength between each candidate influencing factor and the adjustable capacity of the virtual power plant is output. The candidate influencing factors in each window are sorted based on the correlation strength. The candidate influencing factors with a preset position before sorting are extracted as the dominant factor set of the current window, and the dominant factor sorting list is obtained. The distribution difference of the dominant factor sorting list between adjacent sliding windows is calculated based on the information entropy. When the distribution difference exceeds a set threshold, a dominant factor switching flag is generated. Based on the correlation strength of the output of each sliding window, the time series of the influence scores of each candidate influencing factor as a function of the window index is maintained. A pre-constructed weighted double exponential smoothing model is used to predict the trend of influence changes in multiple sliding windows in the future, so as to obtain the trend prediction information of the influence of each candidate influencing factor and the prediction result of the dominant factor in the next window. The output combines the ranking list of dominant factors, the switching marker of dominant factors, the trend prediction information of the influence of each candidate influencing factor, and the prediction results of the dominant factors in the next window.

[0113] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing preset functions, wherein the instruction segments describe the execution process of the computer program in the virtual power plant adjustable capacity dominant factor dynamic extraction device. For example, the computer program can be divided into a construction module, a calculation module, a sorting module, a prediction module, and an output module; the specific functions are as follows: the construction module is used to construct cross-scale directed correlation edges based on the candidate influencing factor set at each time scale, forming a multi-time scale correlation graph; The calculation module is used to set variable-length sliding time windows based on multi-timescale correlation graphs, allowing the sliding windows to slide synchronously in the time axis dimension and the influencing factor domain dimension. It calculates the partial rank correlation coefficient of the operational data within each sliding window and outputs the correlation strength between each candidate influencing factor and the adjustable capacity of the virtual power plant. The sorting module is used to sort the candidate influencing factors within each window based on the correlation strength, extract the candidate influencing factors with a preset position before sorting as the dominant factor set of the current window, and obtain the dominant factor sorting list. It calculates the distribution difference of the dominant factor sorting list between adjacent sliding windows based on information entropy. When the distribution difference exceeds a set threshold, a dominant factor switching mark is generated. The prediction module is used to maintain the time series of the influence scores of each candidate influencing factor changing with the window index based on the correlation strength output of each sliding window. It uses a pre-built weighted double exponential smoothing model to predict the influence change trend of multiple sliding windows in the future, and obtains the trend prediction information of the influence of each candidate influencing factor and the prediction result of the dominant factor in the next window. The output module is used to integrate and output the dominant factor sorting list, the dominant factor switching mark, the trend prediction information of the influence of each candidate influencing factor, and the prediction result of the dominant factor in the next window.

[0114] The device for dynamically extracting the dominant factor of adjustable capability in a virtual power plant can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. This device may include, but is not limited to, processors and memory. Those skilled in the art will understand that the above are examples of devices for dynamically extracting the dominant factor of adjustable capability in a virtual power plant and do not constitute a limitation on such devices. The device may include more components than described above, or combine certain components, or use different components. For example, the device may also include input / output devices, network access devices, buses, etc.

[0115] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or any conventional processor. This processor is the control center for the dynamic extraction of the virtual power plant's adjustable capability dominant factor, connecting various parts of the entire virtual power plant adjustable capability dominant factor dynamic extraction equipment via various interfaces and lines.

[0116] The memory can be used to store the computer program and / or modules. The processor realizes various functions of the virtual power plant adjustable capacity dominant factor dynamic extraction device by running or executing the computer program and / or modules stored in the memory and calling the data stored in the memory.

[0117] The memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function (such as sound playback, image playback, etc.). The data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). Furthermore, the memory may include high-speed random access memory and non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital cards (SD cards), flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0118] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant.

[0119] If the modules / units integrated by the virtual power plant adjustable capability dominant factor dynamic extraction system are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0120] Based on this understanding, the present invention can implement all or part of the processes in the above-mentioned method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant. This process can also be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the above-mentioned method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or a preset intermediate form, etc.

[0121] The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0122] It should be noted that the content contained in the computer-readable storage medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium does not include electrical carrier signals and telecommunication signals.

[0123] Compared with existing methods, this invention provides a method and system for dynamically extracting the dominant factors of the adjustable capacity of a virtual power plant, which has the following advantages: First, cross-scale causal relationship modeling: This invention, for the first time, incorporates influencing factors at three time scales—day-ahead, intraday, and real-time—into a unified directed correlation graph framework. By quantitatively calculating cross-scale edge weights, it reveals the constraint relationship between upstream factors and the feasible domain of downstream factors. Compared to traditional independent analysis methods, this method can explain complex causal chains such as "how day-ahead meteorological errors affect real-time adjustability through energy storage SOC," making the identification of dominant factors physically interpretable.

[0124] Second, dynamic drift tracking and adaptive window: This invention designs a dual-domain sliding mechanism of time axis and factor domain. The length of the time window can be adaptively adjusted according to the degree of fluctuation in the adjustable capability (the window is shortened when the fluctuation is large to improve sensitivity, and lengthened when the fluctuation is small to enhance stability). At the same time, by eliminating the interference of confounding factors through the partial rank correlation coefficient, the pure extraction of the dominant factor ranking is achieved, effectively addressing the practical problem of the dominant factor drifting with the scheduling process.

[0125] Third, objective and quantitative switching criteria: This invention introduces the KL divergence of the sorted list and the cumulative residual entropy, providing for the first time an objective and quantitative triggering criterion for "when key factors need to be re-evaluated," and also sets up an anti-jitter mechanism. Compared with manual experience-based judgment, this method significantly improves the accuracy of switching detection and increases the efficiency of computational resource utilization by more than 30%.

[0126] Fourth, time-varying trajectory and predictive capability: This invention maintains the time-varying sequence of the influence of each influencing factor and predicts its future trend through a weighted double exponential smoothing model, enabling early warning of "how many time windows a certain factor will become the new dominant factor". This predictive capability allows the scheduling system to adjust monitoring strategies or reserve response resources in advance, realizing the transformation from passive response to proactive adaptation.

[0127] The above embodiments are merely one of the implementation methods for achieving the technical solution of the present invention. The scope of protection claimed by the present invention is not limited to this embodiment, but also includes any variations, substitutions and other implementation methods that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for dynamically extracting the dominant factors of the adjustable capacity of a virtual power plant, characterized in that, include: Based on the candidate influencing factor set at each time scale, a cross-scale directed correlation edge is constructed to form a multi-time scale correlation graph; where the time scale includes three time scales: day-ahead, intraday, and real-time, which are divided by the scheduling process of the virtual power plant. Based on the multi-timescale correlation diagram, a variable-length sliding time window is set up so that the sliding window slides synchronously in the time axis dimension and the influencing factor domain dimension. The partial rank correlation coefficient is calculated for the operating data in each sliding window, and the correlation strength between each candidate influencing factor and the adjustable capacity of the virtual power plant is output. The candidate influencing factors in each window are sorted based on the correlation strength. The candidate influencing factors with a preset position before sorting are extracted as the dominant factor set of the current window, and the dominant factor sorting list is obtained. The distribution difference of the dominant factor sorting list between adjacent sliding windows is calculated based on the information entropy. When the distribution difference exceeds a set threshold, a dominant factor switching flag is generated. Based on the correlation strength of the output of each sliding window, the time series of the influence scores of each candidate influencing factor as a function of the window index is maintained. A pre-constructed weighted double exponential smoothing model is used to predict the trend of influence changes in multiple sliding windows in the future, so as to obtain the trend prediction information of the influence of each candidate influencing factor and the prediction result of the dominant factor in the next window. The output combines the ranking list of dominant factors, the switching marker of dominant factors, the trend prediction information of the influence of each candidate influencing factor, and the prediction results of the dominant factors in the next window.

2. The method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant according to claim 1, characterized in that, The construction of cross-scale directed correlation edges based on the candidate influencing factor sets at each time scale forms a multi-time scale correlation graph, including: For any upstream-scale candidate influencing factor and downstream-scale candidate influencing factor that have an influence relationship, a directed edge is established from the upstream scale to the downstream scale, and the correlation strength coefficient is calculated. : in, for The state transition function; τ is the cross-scale time delay, and T is the statistical period; For time; For the upstream timescale Each candidate influencing factor The values ​​of the driving variables at each moment; This represents the state variable value of the j-th candidate influencing factor at time t in the downstream time scale; The multi-timescale correlation graph The adjacency list structure is used for storage, represented as follows: in, These are the set of candidate influencing factor nodes, the set of directed edges, and the set of correlation strength coefficients, respectively.

3. The method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant according to claim 1, characterized in that, Within the variable-length sliding time window, the window length is adaptively adjusted based on the fluctuation of the adjustable capability index, using the following adjustment formula: in The standard deviation of the adjustable capability index within the window; η is the mean of the adjustable capability index within the window; η is the adjustment coefficient. , These are the minimum and maximum window lengths, respectively. , These represent the maximum and minimum values, respectively. , These represent the window lengths corresponding to the nth and (n+1)th sliding time windows, respectively.

4. The method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant according to claim 1, characterized in that, When the sliding window slides synchronously along the time axis and the influencing factor domain, a dynamic selection mechanism is used to determine the set of active candidate influencing factors, as follows: All candidate influencing factors will be added to the active candidate influencing factor set according to the preset cycle; The rate of change of the association strength coefficient of directed edges in a multi-timescale association graph is detected. If the rate of change of the association strength coefficient exceeds the preset rate of change threshold, the upstream candidate influencing factors, downstream candidate influencing factors and their first-order adjacent candidate influencing factors of the corresponding directed edge are added to the active candidate influencing factor set. If the number of elements in the active candidate influencing factor set does not meet the preset requirement, then the candidate influencing factor with the highest historical skewed correlation coefficient will be added.

5. The method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant according to claim 1, characterized in that, The extraction of candidate influencing factors with preset rankings before sorting, used as the dominant factor set for the current window, includes: The partial rank correlation coefficients of each candidate influencing factor are sorted in descending order of absolute value, and those that meet the preset cumulative contribution rate threshold are selected. The top K candidate influencing factors are taken as the dominant factor set of the current window, where the value of K satisfies: In the formula, The number arranged in descending order of absolute value Partial rank correlation coefficient; The total number of effective factors; The traversal variable for the number of candidate factors; For the first The partial rank correlation coefficients corresponding to the 1 effective candidate influencing factors; min() is the minimum value.

6. The method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant according to claim 1, characterized in that, The step of calculating the distribution difference of the dominant factor ranking list between adjacent sliding windows based on information entropy, and generating a dominant factor switching flag when the distribution difference exceeds a set threshold, includes: Convert the ranking list of dominant factors between adjacent sliding windows into a probability distribution. Calculate the KL divergence of two probability distributions. And based on KL divergence, the cumulative residual entropy is calculated. : In the formula, For smoothing weighting coefficients; This is the cumulative residual entropy of the previous window; When the cumulative residual entropy exceeds the preset switching threshold, a dominant factor switching flag is generated, the current switching time and the dominant factor set before and after the switching are recorded, and the cumulative residual entropy is reset to 50% of the current KL divergence value.

7. The method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant according to claim 6, characterized in that, After generating the dominant factor switching marker, a causal chain warning step is also included: Based on the multi-timescale correlation diagram, the upstream candidate influencing factors are traced backward from the newly emerging dominant factor generated in this switch. If the directed edge correlation strength coefficient between the upstream candidate influencing factor and the newly emerging dominant factor is greater than the preset edge weight threshold, the warning information of the corresponding causal chain is output.

8. The method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant according to claim 1, characterized in that, The construction and updating process of the weighted double exponential smoothing model specifically includes: For each candidate influencing factor, select the most recent preset number of historical skewed correlation coefficient values ​​and assign time decay weights to each historical value; The horizontal and trend components of the candidate influencing factor are initialized using weighted linear regression. The horizontal and trend components are updated window by window according to the following recursive formula: In the formula, , Both represent horizontal components; , Both represent trend components; , These represent the corresponding smoothing coefficients; Indicates the first One candidate influencing factor; Indicates the first A sliding window; Indicates the first The first sliding window The partial rank correlation coefficient of each candidate influencing factor.

9. The method for dynamically extracting the dominant factor of the adjustable capacity of a virtual power plant according to claim 8, characterized in that, The process of obtaining trend prediction information on the influence of each candidate influencing factor also includes a step of predicting the entry time of the dominant interval, specifically including: For candidate influencing factors whose current influence is below the preset dominance threshold and whose trend component is positive, calculate the minimum window number required for their predicted influence value to reach the dominance threshold. After converting this to a corresponding time, output the predicted time when the candidate influencing factor is expected to enter the dominance factor set; where, future... The predicted value of the influence of each window for: In the formula, The index of the sliding window; Among them, minimum number of windows satisfy: In the formula, The preset dominant threshold; This is the maximum value.

10. A system for dynamically extracting the dominant factors of adjustable capacity in a virtual power plant, characterized in that, include: The module is used to construct directed association edges across scales based on the candidate influencing factor set at each time scale, forming a multi-time scale association graph; The calculation module is used to set a variable-length sliding time window based on the multi-timescale correlation diagram, so that the sliding window slides synchronously in the time axis dimension and the influencing factor domain dimension, calculates the partial rank correlation coefficient of the operating data in each sliding window, and outputs the correlation strength between each candidate influencing factor and the adjustable capacity of the virtual power plant. The sorting module is used to sort the candidate influencing factors in each window based on the correlation strength, extract the candidate influencing factors with the preset position before sorting as the dominant factor set of the current window, and obtain the dominant factor sorting list; calculate the distribution difference of the dominant factor sorting list between adjacent sliding windows based on information entropy, and generate a dominant factor switching flag when the distribution difference exceeds a set threshold. The prediction module is used to maintain the time series of the influence scores of each candidate influencing factor as a function of the window index based on the correlation strength of the output of each sliding window. It uses a pre-built weighted double exponential smoothing model to predict the trend of influence changes in multiple sliding windows in the future, and obtains the trend prediction information of the influence of each candidate influencing factor and the prediction result of the dominant factor in the next window. The output module is used to integrate and output the dominant factor ranking list, dominant factor switching marker, trend prediction information of the influence of each candidate influencing factor, and the prediction results of the dominant factor in the next window.

Images