A method and apparatus for screening power sensitivity indicators based on multidimensional characteristics of net load.

By removing disturbance factors from the original power load to generate a net load signal, performing multi-scale mode decomposition and multi-dimensional feature extraction, and combining the collaborative relationships between objects to construct a comprehensive propagation edge weight, the problem of non-operational disturbance influence in existing technologies is solved, and the accuracy and comprehensiveness of power index screening are improved.

CN123117883APending Publication Date: 2026-11-17GUIZHOU POWER GRID CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611116544.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-27
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies in power load analysis fail to adequately isolate non-operational disturbance factors, resulting in the inclusion of external fluctuation components in load characteristics. This makes it impossible to accurately reflect the operational change characteristics of the analyzed object itself. Furthermore, the indicators are constructed with a single dimension, making it difficult to comprehensively capture multi-dimensional operational characteristics and affecting the accuracy of power indicator selection.

Method used

By acquiring raw power load and external disturbance data, removing disturbance loads to generate net load signals, performing multi-scale mode decomposition and extracting multi-scale features, and combining multi-dimensional operational characteristic indicators, the sensitivity of collaborative evolution, time delay relationships, prior relationships and power flow among the analysis objects are integrated to construct comprehensive propagation edge weights, calculate network diffusion contribution and aggregation sensitivity, and screen out core sensitivity indicators.

Benefits of technology

It effectively reduces interference from non-operational factors, accurately reflects the operational change characteristics of the analyzed object itself, identifies the collaborative changes and propagation paths between analyzed objects, and improves the accuracy and comprehensiveness of power index screening.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117883A_ABST
    Figure CN123117883A_ABST
Patent Text Reader

Abstract

This invention relates to the field of power index screening technology, and provides a method and apparatus for screening power sensitivity indicators based on multi-dimensional characteristics of net load. The method includes: removing disturbing loads determined by external disturbance factors from the original power load to generate a net load signal set; performing multi-scale modal decomposition on the net load signals and extracting multi-scale features, combining them with multi-dimensional operational characteristic indicators to construct a first candidate index set; based on the net load signal set, fusing and analyzing the sensitivity of co-evolution among analysis objects, time delay relationships, prior relationships, feature similarity, and power flow direction to construct a comprehensive propagation edge weight; calculating the network diffusion contribution and aggregation sensitivity based on the comprehensive propagation edge weight as a second candidate index set; merging the first and second candidate index sets to form a total candidate index set, and screening the core sensitivity index set to obtain a comprehensive sensitivity index. This application improves the accuracy and comprehensiveness of power index screening results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power index screening technology, and in particular to a method and apparatus for screening power sensitivity indicators based on multidimensional characteristics of net load. Background Technology

[0002] Electricity data is characterized by high collection frequency, wide coverage, and strong objectivity, reflecting changes in industrial production, commercial operations, and social consumption activities. Currently, indicators such as electricity consumption, load growth rate, year-on-year changes, and peak-valley characteristics are used as the main analytical basis. By constructing a multi-level analytical framework at the industry, sector, and regional levels, the operational climate, industry activity, and regional operational status are monitored. In terms of data infrastructure, existing technologies have the basic conditions to integrate electricity consumption collection systems, power grid marketing systems, dispatching and operation systems, meteorological data, calendar information, and macroeconomic statistical data. Through multi-source data fusion, load analysis is gradually shifting from simple electricity consumption statistics to operational information extraction. In terms of analytical modeling, methods such as time-frequency decomposition, machine learning, and graph neural networks are increasingly being applied to electricity analysis. By performing modal decomposition on load sequences to extract trend and periodic features, and using feature selection methods to compress high-dimensional indicators, the analysis is evolving from total quantity correlation analysis to industry structure identification and industrial synergy analysis.

[0003] The original power load is simultaneously affected by a combination of non-operational factors, including weather conditions, calendar schedules, electricity price adjustments, grid operation status, and industry seasonality. Existing technologies are insufficient in identifying and separating these non-operational disturbances, resulting in the extracted load characteristics still containing a large amount of fluctuation caused by external factors, failing to accurately reflect the operational changes of the analyzed object itself. Furthermore, existing technologies focus heavily on total volume and growth rate indicators during indicator construction, neglecting to adequately characterize the trends and fluctuations at different time scales in the load series, the structural distribution changes among different analyzed objects, and the interrelationships and propagation relationships between them. This leads to the constructed indicators having a single dimension in representing changes in operational status, making it difficult to comprehensively capture multi-dimensional operational characteristics, thus affecting the accuracy of power indicator selection. Summary of the Invention

[0004] This invention provides a method and apparatus for screening power sensitivity indicators based on multidimensional characteristics of net load, to address the problem of low accuracy in power indicator screening. The specific solution is as follows: Firstly, this application discloses a method for screening power sensitivity indicators based on multidimensional characteristics of net load, including: The original power load and external disturbance factor data are obtained, and the disturbance load determined by the external disturbance factor data is removed from the original power load. After constraint correction, a net load signal set is generated. Multi-scale mode decomposition is performed on the net load signals in the net load signal set, and multi-scale features are extracted. Combined with multi-dimensional operational feature indicators, a first candidate indicator set reflecting the change characteristics of the analysis object itself is constructed. Based on the net load signal set, the sensitivity of collaborative evolution among the analysis objects, time delay relationship, prior relationship, feature similarity and power flow direction are integrated to construct a comprehensive propagation edge weight; based on the comprehensive propagation edge weight, the network diffusion contribution and aggregation sensitivity are calculated, and the network diffusion contribution and aggregation sensitivity are used as a second candidate index set to reflect the linkage propagation characteristics among different analysis objects; The first candidate indicator set and the second candidate indicator set are merged to form a total candidate indicator set, and a core sensitivity indicator set is selected based on the total candidate indicator set to obtain a comprehensive sensitivity index.

[0005] Optionally, removing the disturbance load determined by the external disturbance factor data from the original power load includes: Based on the various types of data in the external disturbance factor data, corresponding load influence functions are constructed respectively. The outputs of each load influence function are superimposed to synthesize the non-operational disturbance counterfactual baseline load. The non-operational disturbance counterfactual baseline load is treated as a disturbance load and removed from the original power load.

[0006] Optionally, the step of performing multi-scale mode decomposition and extracting multi-scale features from the net load signal set includes: The net load signal is subjected to variational mode decomposition to obtain K modal components with different center frequencies, and the sum of the modal components is equal to the net load signal. Modal components with center frequencies below a first preset frequency threshold are considered long-term trend features, modal components with center frequencies above a second preset frequency threshold are considered short-term impact features, and modal components with center frequencies between the first preset frequency threshold and the second preset frequency threshold are considered medium-term periodic features; wherein, the second preset frequency threshold is greater than the first preset frequency threshold.

[0007] Optionally, the multidimensional operational characteristic indicators include: The entropy of electricity consumption structure change is determined based on the information entropy calculated from the proportion of electricity consumption of various analytical objects within a set period. It is used to reflect the uniformity of the distribution of electricity consumption structure. The stability of the operating rhythm is determined based on the degree of deviation of the net load signal of the analyzed object from the mean and peak-valley envelope of the net load signal within a set period. It is used to reflect the degree of change in the shape of the load curve and the activity rhythm of the analyzed object. The efficiency elasticity index is determined based on the ratio of the reference activity growth rate to the net electricity consumption growth rate of the analyzed object. It is used to distinguish whether changes in net load are caused by changes in activity or by changes in efficiency.

[0008] Optionally, the step of constructing a first candidate indicator set reflecting the change characteristics of the analyzed object by combining multidimensional operational characteristic indicators includes: The multi-scale features are combined with the entropy of the power consumption structure change, the stability of the operating rhythm, and the efficiency elasticity index to form a sensitive feature vector, which is then used as the first candidate index set. Analysis object exist The sensitive feature vector at time step is represented as: in: For the first to the second One modal component, The entropy of changes in electricity structure. To ensure operational rhythm stability, For efficiency elasticity indicators, superscript This represents the transpose of a vector.

[0009] Optionally, the step of fusing and analyzing the co-evolution sensitivity and time-delay relationship among the objects based on the net load signal set includes: For any two analysis objects, calculate the ratio of the product of the covariance and the standard deviation of the net load signal of each of the two analysis objects under multiple preset time delays, and use it as the co-evolution sensitivity under each time delay; The time delay at which the maximum co-evolution sensitivity is obtained is determined as the time delay relationship between the two analysis objects, and the maximum co-evolution sensitivity is taken as the co-evolution intensity corresponding to the time delay relationship.

[0010] Optionally, the fusion analysis of prior relationships, feature similarity, and power flow direction among the objects to construct a comprehensive propagation edge weight includes: Obtain the prior relationship strength between the analysis objects and the sensitive feature similarity determined by the sensitive feature vector; multiply the prior relationship strength and the sensitive feature similarity to generate the relationship coupling edge weight; The edge weights corresponding to the power flow direction, the leading and lagging edge weights determined based on the time delay relationship, the relationship coupling edge weights, and the similarity edge weights determined based on the sensitivity feature similarity are weighted and fused to obtain the comprehensive propagation edge weights.

[0011] Optionally, the calculation of network diffusion contribution based on the integrated propagation edge weights includes: Based on the comprehensive propagation edge weights, the lead-lag factors corresponding to the time lag relationships, and the sensitivity feature similarity, the sensitivity propagation intensity between the analyzed objects is calculated: In the formula: For a moment Analysis object Towards The sensitivity of the transmission intensity In order to comprehensively disseminate border rights, The leading-lag factor is determined based on the time-delay relationship. For sensitive feature similarity; For each analysis object, the ratio of the sum of the sensitivity output intensities of the analysis object as a propagation source to the sum of the sensitivity input intensities as a receiver is calculated as the network diffusion contribution of that analysis object. In the formula: For the analysis object At any moment The network diffusion contribution is represented by a value greater than 1, indicating an output node, and a value less than 1, indicating a receiving node.

[0012] Optionally, the aggregation sensitivity is calculated in the following manner: The analysis objects with prior associations are divided into object groups, and the object weights of each analysis object within the object group are obtained. Then, the sensitivity feature vectors of each analysis object within the object group are weighted and aggregated to obtain the aggregated sensitivity of the object group. In the formula: For the object group At any moment The aggregation sensitivity, For the analysis object Object weight, For sensitivity mapping function, This is for pooling operations.

[0013] Secondly, this application discloses a power sensitivity index screening device based on multidimensional characteristics of net load, comprising: The net load signal set generation module is used to acquire the original power load and external disturbance factor data, remove the disturbance load determined by the external disturbance factor data from the original power load, and generate the net load signal set after constraint correction. The first candidate index set construction module is used to perform multi-scale mode decomposition on the net load signals in the net load signal set and extract multi-scale features, and combine them with multi-dimensional operation feature indicators to construct a first candidate index set that reflects the change characteristics of the analysis object itself. The second candidate index set determination module is used to construct a comprehensive propagation edge weight based on the net load signal set, by integrating the sensitivity of collaborative evolution among the analysis objects, time delay relationship, prior relationship, feature similarity and power flow direction; and to calculate the network diffusion contribution and aggregation sensitivity based on the comprehensive propagation edge weight, and to use the network diffusion contribution and aggregation sensitivity as the second candidate index set reflecting the linkage propagation characteristics among different analysis objects. The comprehensive sensitivity index determination module is used to merge the first candidate index set and the second candidate index set to form a total candidate index set, and to screen the core sensitivity index set based on the total candidate index set to obtain the comprehensive sensitivity index.

[0014] As can be seen from the above technical solutions, the present invention has the following advantages: This invention effectively reduces the interference of non-operational factors such as weather conditions, calendar schedules, electricity price adjustments, grid operation status, and industry seasonality on load signals by removing disturbing loads determined by external disturbance factors from the original power load and generating a net load signal set through constraint correction. This allows the net load signal to more accurately reflect the operational change characteristics of the analyzed object itself. By performing multi-scale modal decomposition on the net load signal and extracting multi-scale features, and combining it with multi-dimensional operational characteristic indicators to construct a first candidate indicator set, it can characterize the change characteristics of the analyzed object from multiple dimensions such as long-term trends, medium-term cycles, short-term shocks, structural distribution, operating rhythm, and efficiency elasticity. By integrating the sensitivity of collaborative evolution among analyzed objects, time lag relationships, prior relationships, feature similarity, and power flow to construct a comprehensive propagation edge weight, and calculating the network diffusion contribution and aggregation sensitivity based on this comprehensive propagation edge weight, a second candidate indicator set reflecting the linkage propagation characteristics among different analyzed objects can be constructed, which can identify the collaborative changes, propagation paths, and diffusion roles among analyzed objects. By merging the first and second candidate indicator sets and screening the core sensitivity indicator set, information redundancy between indicators is reduced, and the accuracy and comprehensiveness of power indicator screening results are improved. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating a method for screening power sensitivity indicators based on multidimensional characteristics of net load in this application. Figure 2 This is a flowchart illustrating the process of determining the first candidate indicator set in this application; Figure 3 This is a flowchart illustrating the process of determining the second candidate indicator set in this application. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] Please see Figure 1 The power sensitivity index screening method based on multidimensional characteristics of net load provided in this application includes the following steps: S1. Obtain the original power load and external disturbance factor data, remove the disturbance load determined by the external disturbance factor data from the original power load, and generate the net load signal set after constraint correction; The raw power load refers to the power load values ​​of the analysis object at various times, recorded by the electricity information collection system, power grid marketing system, and dispatch operation system. In this embodiment, the analysis object is an industry or region, and the raw power load data is organized in time series form. External disturbance factor data includes meteorological data, calendar data, electricity price data, policy event data, power grid operation status data, and periodic data of the analysis object. These data are obtained from meteorological monitoring systems, calendar databases, power trading platforms, policy release systems, power grid dispatch logs, and historical load statistics records of the analysis object, respectively. This step involves removing disturbances caused by non-operational factors from the raw power load to obtain a set of net load signals that more accurately reflects the operational change characteristics of the analysis object itself.

[0019] In the original power load, changes in meteorological conditions can cause fluctuations in cooling or heating loads; calendar events such as holidays can alter the rhythm of production, operation, and consumption activities; electricity price adjustments may lead to shifts in electricity consumption periods; policy events may change electricity consumption behavior in the short term; grid maintenance or dispatch switching can affect local loads; and the periodic characteristics of the analyzed object itself can also cause the load to exhibit regular fluctuations. These factors-induced load changes are coupled with changes in operational activities themselves. If feature extraction is performed directly based on the original power load, load fluctuations caused by external factors can easily be misjudged as changes in operational status. Therefore, a counterfactual baseline load of non-operational disturbances is constructed to estimate the theoretical load formed solely by external disturbance factors when there are no substantial changes in economic activities themselves, and this load is then removed from the original power load. This includes the following: 1. Construct corresponding load influence functions based on various types of data in the external disturbance factor data, and superimpose the outputs of each load influence function to synthesize the non-operational disturbance counterfactual baseline load; Specifically, the data includes various types such as meteorological data, calendar data, electricity price data, policy event data, power grid operation status data, and periodic data of the analyzed object. Meteorological data includes elements such as temperature and humidity; calendar data includes holiday, weekday, and weekend markers; electricity price data includes time-of-use pricing or price adjustment information; policy event data includes event types and durations; power grid operation status data includes maintenance plans and dispatch switching records; and periodic data of the analyzed object includes seasonal and weekly periodic characteristics. The load impact functions constructed for each type of data are as follows: This represents the nonlinear response function of meteorological load, which is based on the analysis object. At any moment meteorological variables As input, the system uses a nonlinear mapping relationship pre-fitted based on historical meteorological data and historical load data for the corresponding time period to output an estimated value of load change caused by meteorological conditions. This represents the calendar effect correction function, which uses time... Calendar properties As input, the system uses the influence coefficient of calendar effect on load, which is obtained in advance based on historical calendar data and historical load data of the corresponding period, to output the estimated value of load change caused by calendar arrangement; This represents the electricity price response function, which is used to analyze the object. At any moment Electricity price information As input, the model uses a price elasticity model pre-fitted based on historical electricity price data and historical load data for the corresponding time period to output an estimated value of load change caused by electricity price factors. This represents the policy event impact attenuation function, which is based on the analysis object. At any moment Policy event information As input, and using a pre-defined impact intensity and attenuation coefficient, the output is an estimate of the load change caused by the policy event, which decays over time. This represents the power grid operating state correction function, which is based on the analysis object. At any moment Power grid operating status As input, based on the load impact rules corresponding to the power grid operating status, output the estimated load change caused by power grid factors; This represents the periodic baseline function of the object, which is used to analyze the object. At any moment Periodicity As input, the system extracts periodic fluctuation patterns based on historical load data of the analyzed object, and outputs estimated load changes caused by periodic patterns. The outputs of each load influence function are load values ​​with the same dimensions as the original power load. The modeling processes of each function are independent, with no interaction terms. Finally, the overall non-operational disturbance counterfactual baseline load is synthesized through superposition. The superposition of the outputs of the above load influence functions to synthesize the non-operational disturbance counterfactual baseline load is expressed as: in: Represents the object of analysis At any moment The non-operational disturbance counterfactual baseline load represents the theoretical load formed solely by external disturbances when there is no substantial change in economic activity itself.

[0020] 2. Treat non-operational disturbance counterfactual baseline loads as disturbance loads and remove them from the original power load.

[0021] Obtain the initial net load : In the formula: This represents the original electrical load.

[0022] After the above elimination, the initial net load needs to be constrained to further reduce the impact of residual disturbances. The constraint correction is achieved by solving the following optimization model: In the formula: For the collection of all objects being analyzed, For all moments within the entire observation period, This is the set of model parameters, containing the net load to be optimized. , , , These are the weight coefficients for the independence constraint, the stability constraint across the time window, and the consistency constraint for similar objects, respectively, used to adjust the relative importance of each constraint in the optimization objective; Let represent the set of external disturbance variables, where For meteorological variables, For calendar properties, For electricity price information, For policy event information, The power grid's operating status. This is a lifecycle property of the object.

[0023] The optimization model described above includes three constraint terms: reconstruction consistency constraint, independence constraint, and stability constraint. The expressions and meanings of each constraint term are as follows: Reconstructing consistency constraints is achieved through the first term of the above objective function. This term calculates the sum of squares of the differences between the original power load, the counterfactual baseline load, and the net load signal. By minimizing this term, the sum of the corrected net load signal and the counterfactual baseline load can stably reconstruct the original power load, ensuring the integrity of the load decomposition and the conservation of signal energy.

[0024] Independence constraints Indicates net load signal With the set of external disturbance variables The mutual information between them is expressed as: ;in, Net load signal and external disturbance variables The joint probability density, and These are the net load signal and the external disturbance variable, respectively. The marginal probability density. Mutual information measures the degree of statistical dependence between the net load signal and external disturbance variables. By minimizing this term, the statistical dependence between the net load signal and external disturbance variables such as weather, calendar, electricity price, policy events, grid operating status and object periodicity is reduced, making the net load signal as independent as possible from non-operational factors, and ensuring that the fluctuation of the net load signal mainly reflects the operational changes of the analyzed object itself.

[0025] Stability constraints include stability constraints across time windows. Consistency constraints with similar analysis objects Two items; stability constraints across time windows Represented as ;in, , Observation period Different time windows within; This describes the statistical distribution characteristics of the signal, which can be represented by statistical measures such as mean, variance, or quantiles. This constraint ensures the consistency of the statistical distribution of net load signals for the same analytical object across different time windows, making the corrected net load signals comparable across different time periods and preventing changes in indicator definitions due to time window switching. Consistency constraint for similar analytical objects. Represented as ; A set consisting of pairs of analytical objects belonging to the same category. and Representing the objects to be analyzed and Throughout the observation period The net load signal sequence within.

[0026] After constraint correction, the net load signal set of all analyzed objects over the entire observation period is obtained. , represented as: This set of net load signals serves as the input basis for subsequent multi-scale feature extraction and analysis of relationships between objects.

[0027] S2. Perform multi-scale mode decomposition on the net load signals in the net load signal set and extract multi-scale features. Combine with multi-dimensional operation feature indicators to construct a first candidate indicator set that reflects the change characteristics of the analysis object itself. Although the net load signal has eliminated non-operational disturbances, it still contains information across multiple time scales, including long-term trends, medium-term cyclical fluctuations, and short-term shocks. This step transforms the net load signal into multi-dimensional sensitivity characteristics through multi-scale modal decomposition and multi-dimensional operational characteristic index calculation. Please refer to [link to relevant documentation]. Figure 2 Specifically, it includes the following: S21. Perform variational mode decomposition on the net load signal to obtain K modal components with different center frequencies. The sum of the modal components is equal to the net load signal. Specifically, for each net load signal in the net load signal set The variational mode decomposition method is used for processing. Variational mode decomposition adaptively decomposes the net load signal into the following constrained variational optimization problem by constructing and solving it. Modal components with different center frequencies : The constraints are satisfied: in: Represents the object of analysis The One modal component, This represents the center frequency corresponding to the modal component, expressed in radians per second. The center angular frequency is the same as the ordinary frequency. The relationship between them is , Indicates time The partial derivative operator; For the Dirac function, Imaginary unit; This is a complex exponential demodulation operator used to demodulate the modal components to baseband. The above constraints ensure that the sum of all modal components can completely reconstruct the original payload signal, guaranteeing that the decomposition process does not lose the operational information in the original signal. The preset total number of modes is set according to the complexity of the net load signal and the analysis requirements.

[0028] Through the variational mode decomposition described above, the corresponding values ​​for each net load signal are obtained. The modal components have different center frequencies; the lower the center frequency, the longer the corresponding time scale.

[0029] S22. Modal components with center frequencies lower than a first preset frequency threshold are taken as long-term trend features, modal components with center frequencies higher than a second preset frequency threshold are taken as short-term impact features, and modal components with center frequencies between the first preset frequency threshold and the second preset frequency threshold are taken as medium-term periodic features; wherein, the second preset frequency threshold is greater than the first preset frequency threshold.

[0030] Variational mode decomposition obtained The modal components are arranged from low to high center frequency. To classify each modal component into a feature type with a clear operational meaning, a first preset frequency threshold is introduced. Second preset frequency threshold ,and First preset frequency threshold Second preset frequency threshold The preset frequency threshold is set according to the operational cycle characteristics and observation time scale of the analysis object. For example, the first preset frequency threshold is set to the frequency corresponding to the quarterly or annual cycle, and the second preset frequency threshold is set to the frequency corresponding to the weekly or daily cycle. For the center frequency... ,like The corresponding modal components serve as long-term trend characteristics, reflecting the long-term growth or contraction direction of the analyzed object's operating state; if The corresponding modal components serve as short-term shock characteristics, reflecting transient fluctuations caused by sudden events, temporary adjustments, etc.; if The corresponding modal components serve as medium-term cyclical features, reflecting medium-term fluctuations such as seasonal patterns, order cycles, or industry rotation. Therefore, multi-scale features are clearly classified into three categories: long-term trend features, medium-term cyclical features, and short-term shock features.

[0031] Based on the extraction of multi-scale features, multi-dimensional operational characteristic indicators are further calculated, including the entropy of electricity consumption structure changes, operational rhythm stability, and efficiency elasticity indicators, to supplement information that cannot be expressed by single-time-scale features from dimensions such as structural distribution, operational rhythm, and efficiency elasticity. Specifically: 1. Entropy of changes in electricity consumption structure, which is determined by the information entropy calculated based on the proportion of electricity consumption of various analytical objects within a set period, is used to reflect the uniformity of the distribution of electricity consumption structure; The entropy of electricity consumption structure change is used to quantify the uniformity of electricity consumption distribution among different analytical objects. When electricity consumption distribution tends to be more dispersed and the proportion of electricity consumption among various analytical objects is more balanced, the entropy of electricity consumption structure change increases; conversely, when electricity consumption distribution tends to be more concentrated and the proportion of electricity consumption among a few analytical objects increases, the entropy of electricity consumption structure change decreases. The specific calculation process is as follows: Within a set statistical period... First, calculate the electricity consumption percentage for each type of analysis object: In the formula: Indicates the first Class analysis objects in statistical periods The cumulative electricity consumption within the area This represents the total number of categories of the analyzed objects. Based on the electricity consumption percentage, the entropy of electricity consumption structure change is calculated: In the formula: For statistical period The entropy of changes in the internal electrical structure.

[0032] 2. Operational rhythm stability is determined based on the deviation of the net load signal of the analyzed object from the mean and peak-valley envelope of the net load signal within a set period. It is used to reflect the degree of change in the load curve shape and activity rhythm of the analyzed object. Operational rhythm stability is used to identify whether the load curve shape and activity rhythm of the analyzed object have changed. Even if the total electricity consumption remains stable, changes in the intraday peak-to-valley difference, load fluctuation amplitude, or electricity consumption period distribution may reflect adjustments in the operational rhythm. This indicator is measured by the degree of deviation of the net load signal from its mean and peak-to-valley envelope: In the formula: Represents the object of analysis In the statistical period Average net load within the area, Indicates the analysis object within the corresponding runtime cycle. At any moment The difference between peak and valley load envelopes. The lower the value of the operating rhythm stability index, the more stable the net load of the analyzed object is relative to the historical operating rhythm; a continuously rising value indicates that the operating rhythm has changed significantly.

[0033] 3. The efficiency elasticity index is determined based on the ratio of the reference activity growth rate to the net electricity consumption growth rate of the analyzed object. It is used to distinguish whether the change in net load is caused by the change in activity or by the change in efficiency.

[0034] The efficiency elasticity index is used to distinguish whether changes in net load are caused by changes in activity or changes in efficiency, thus avoiding misinterpreting improvements in efficiency as weakened operational activity, or increases in unit consumption as enhanced operational activity. The efficiency elasticity index is expressed as the ratio of the growth rate of reference activity to the growth rate of net electricity consumption: In the formula: Represents the object of analysis Based on the reference activity level in the previous statistical period, Indicates the increase in reference activity volume; Represents the object of analysis Net electricity consumption in the previous statistical period, This represents the net increase in electricity consumption. When... When this occurs, it indicates that the growth rate of activity is higher than the growth rate of net load, and utilization efficiency is improving; when This indicates that the growth rate of activity is lower than the growth rate of net load, and unit consumption increases.

[0035] S23. Combine the multi-scale features with the entropy of changes in electricity consumption structure, the stability of operating rhythm, and the efficiency elasticity index into a sensitive feature vector, and use the sensitive feature vector as the first candidate index set.

[0036] After completing multi-scale feature extraction and multi-dimensional operational characteristic index calculation, all the above features are combined to construct a first candidate index set that can comprehensively characterize the operational change characteristics of the analyzed object. Specifically, the long-term trend characteristics, medium-term cycle characteristics, and short-term shock characteristics are combined. Modal components to , with the entropy of changes in electricity consumption structure Stability of operating rhythm and efficiency elasticity indicators Combined into a structured sensitive feature vector. Analysis object. exist The sensitive feature vector at time step is represented as: in: For the first to the second One modal component, The entropy of changes in electricity structure. To ensure operational rhythm stability, For efficiency elasticity indicators, superscript This represents the transpose of a vector.

[0037] This sensitive feature vector uniformly describes the operational sensitivity of the analyzed object across multiple dimensions, including trend, cycle, impact, structure, rhythm, and efficiency. The sensitive feature vectors of all analyzed objects throughout the entire observation period are used as the first candidate indicator set, represented as: , For the collection of all objects being analyzed, This refers to all moments within the entire observation period; this first candidate indicator set, together with the second candidate indicator set constructed in step S3, constitutes the candidate input for the core sensitivity indicator screening in the subsequent step S4.

[0038] S3. Based on the net load signal set, integrate the sensitivity of collaborative evolution among the analysis objects, time delay relationship, prior relationship, feature similarity and power flow direction to construct a comprehensive propagation edge weight; calculate the network diffusion contribution and aggregation sensitivity based on the comprehensive propagation edge weight, and use the network diffusion contribution and aggregation sensitivity as the second candidate index set to reflect the linkage propagation characteristics among different analysis objects; The first candidate indicator set only describes the operational change characteristics of a single analytical object. However, there may be synchronous changes, leading-lagging changes, or propagation effects along specific paths between different analytical objects. If only the characteristics of a single analytical object are considered while ignoring the interconnected propagation relationships between them, it is difficult to identify the source, diffusion path, and scope of impact of operational changes. Therefore, this step, based on the net load signal set generated in S1, and based on the collaborative change relationships between analytical objects, sequentially identifies time-delay relationships, constructs comprehensive propagation edge weights, and calculates the network diffusion contribution and aggregation sensitivity based on the constructed network, forming a second candidate indicator set reflecting the interconnected propagation characteristics between different analytical objects. Please refer to [link to relevant documentation]. Figure 3 Specifically, it includes the following: S31. For any two analysis objects, calculate the ratio of the product of the covariance and the standard deviation of the net load signal of each analysis object under multiple preset time delays, and use it as the co-evolution sensitivity under each time delay. Define all analysis objects as a set ,for Any two different analysis objects and Obtain their net load signal sequences respectively. and A set of time delay parameters is preset. ,in The maximum time delay range is set according to the observation period and analysis requirements. For each preset time delay... Calculation and analysis object and Co-evolution sensitivity under this time lag: In the formula: This represents the covariance operation. This indicates the standard deviation calculation. Represents the object of analysis Changes in net load precede those of the analyzed object. , Represents the object of analysis The change in net load lags behind the analysis object. , This indicates that the two change synchronously. This co-evolution sensitivity reflects the degree of correlation between the net load signals of the two analyzed objects under different time delay conditions, and its value ranges between -1 and 1.

[0039] S32. The time delay at which the maximum co-evolution sensitivity is obtained is determined as the time delay relationship between the two analysis objects, and the maximum co-evolution sensitivity is taken as the co-evolution intensity corresponding to the time delay relationship; For the analysis object and In all the preset time delays traversed In the middle, the selected envoy Time delay to reach the maximum value : Will As the object of analysis and The time delay relationship between them represents the analysis object. Relative to the object of analysis The leading or lagging direction and duration. Simultaneously, this maximum co-evolution sensitivity. The co-evolution strength, corresponding to this time lag relationship, is used to measure the strength of the correlation between the two analyzed objects under this time lag. The closer the co-evolution strength is to 1, the more significant the co-evolution relationship between the two; the closer it is to 0, the weaker the correlation between the two.

[0040] S33. Obtain the prior relation strength between the analyzed objects and the sensitive feature similarity determined by the sensitive feature vector; multiply the prior relation strength and the sensitive feature similarity to generate the relation coupling edge weight; Prior relation strength The prior knowledge derived from the analysis objects regarding business collaboration, power exchange, spatial proximity, functional similarity, or supply cooperation can be obtained in advance from industry association databases, power grid topology records, geographic information systems, or historical statistical records. The strength of the prior relationship ranges between 0 and 1, where 0 indicates no known relationship exists between the analysis objects, and 1 indicates the existence of the strongest prior relationship. The sensitive feature similarity is determined by the sensitive feature vector constructed in step S2, expressed as: In the formula: The similarity calculation function is implemented using cosine similarity. and The objects of analysis and At any moment The sensitive feature vector is obtained. The prior relation strength is multiplied by the sensitive feature similarity to generate the relation coupling edge weights: The relational coupling edge weights integrate data-driven feature similarity and knowledge-driven prior relations, reducing the risk of false associations formed solely based on accidental similarity of load curves and improving the economic interpretability of relational edge weights between objects.

[0041] S34. The edge weights corresponding to the power flow, the leading and lagging edge weights determined based on the time delay relationship, the relationship coupling edge weights, and the similar edge weights determined based on the similarity of sensitive features are weighted and fused to obtain the comprehensive propagation edge weights.

[0042] Specifically, the integrated propagation of edge weights incorporates four types of edge weight information reflecting different dimensions of relationships between the analyzed objects: edge weights corresponding to power flow directions. The leading and lagging edge weights are determined based on the direction and intensity of power transmission between the analyzed objects in the power grid topology and power exchange records; and based on the time-delay relationship. The relationship coupling edge weights are determined based on the time delay relationship and co-evolution strength obtained in step S32. Generated by step S33; similar edge weights determined based on sensitive feature similarity. Based on the similarity of sensitive features Directly determined. The above four types of edge weights are weighted and fused to obtain the comprehensive propagation edge weights: In the formula: , , , The fusion coefficients for various edge weights satisfy... Configure according to the actual application scenario. Comprehensive propagation of edge rights. Characterizes the object of analysis and The strength of potential channels for operational sensitivity propagation between them.

[0043] Based on the construction of comprehensive propagation edge weights, the network diffusion contribution and aggregation sensitivity are further calculated to characterize the role of the analyzed object in the propagation network and the overall sensitivity of the associated object group.

[0044] 1. Calculate the network diffusion contribution: First, based on the comprehensive propagation edge weights, the lead-lag factors corresponding to the time lag relationship, and the similarity of sensitive features, the sensitivity propagation intensity between the analyzed objects is calculated: In the formula: For a moment Analysis object Towards The sensitivity of the transmission intensity The leading-lag factor is determined based on the time-delay relationship S32, when Analysis object Leading hour, The value is 1; when Analysis object Lagging behind hour, A value of 0 indicates Not to spread.

[0045] After obtaining the pairwise sensitivity propagation strength of all analyzed objects, for each analyzed object, the ratio of the sum of the sensitivity output strengths of the analyzed object as a propagation source to the sum of the sensitivity input strengths of the analyzed object as a receiver is calculated as the network diffusion contribution of that analyzed object. In the formula: For the analysis object At any moment The network diffusion contribution, when When the value is greater than 1, the characterization analysis object The output propagation strength is greater than the input propagation strength, making it more likely to be an output node with varying sensitivity; when When the value is less than 1, the characterization analysis object If the input propagation strength is greater than the output propagation strength, it is more likely to be a receiving node sensitive to changes; when When the value is close to 1, the characterization analysis object It plays an intermediary role in the network.

[0046] 2. Calculate aggregation sensitivity: Analytical objects with prior associations are grouped into object groups. These prior associations include belonging to the same industry chain, being located in the same region, or having functional collaboration. The object weights of each analytical object within a group are obtained; these weights can be determined based on the object's electricity consumption scale or its relative importance within the group. The sensitive feature vectors of each analytical object within a group are then weighted and aggregated to obtain the aggregated sensitivity of the object group. In the formula: For the object group At any moment The aggregation sensitivity, For the analysis object Object weight, For sensitivity mapping function, This is a pooling operation. Aggregate sensitivity weakens the impact of random fluctuations of a single analysis object by weighting and pooling the sensitive features of multiple analysis objects within a group of related objects, extracting common sensitive changes at the group level, and making the results closer to macroscopically significant interconnected changes.

[0047] The network diffusion contribution and aggregation sensitivity calculated above are used as the second candidate indicator set to reflect the characteristics of linkage propagation among different analytical objects. , represented as: The second candidate indicator set, together with the first candidate indicator set constructed in step S2, constitutes the candidate input for the core sensitivity indicator screening in the subsequent step S4.

[0048] S4. Merge the first candidate indicator set and the second candidate indicator set to form the total candidate indicator set, and select the core sensitivity indicator set based on the total candidate indicator set to obtain the comprehensive sensitivity index.

[0049] The first and second candidate indicator sets characterize the operational status from different perspectives, but the number of candidate indicators is large, and there may be issues such as information duplication, inconsistent directions, or insufficient stability among different indicators. Therefore, the two sets of candidate indicators are first merged into a single candidate indicator set. Then, the maximum relevance and minimum redundancy criteria and a multi-objective joint evaluation function are used for screening, eliminating redundant and unstable indicators, and retaining core sensitivity indicators with high representativeness, low redundancy, and strong anti-disturbance capabilities. Finally, through direction unification, standardization, and weighted synthesis, a comprehensive sensitivity index is obtained. The details are as follows: 1. The first candidate indicator set The second candidate index set output by step S3 Merge them to form a set of candidate indicators. : From the set of candidate indicators The process of selecting the core sensitivity index set is divided into two stages: initial screening with maximum relevance and minimum redundancy, and screening with multi-objective joint evaluation. In the first stage, the maximum relevance and minimum redundancy criterion is used for initial screening of candidate indicators. For each candidate indicator in the total set of candidate indicators... Calculate its maximum correlation and minimum redundancy evaluation value: In the formula: Indicate candidate indicators With target operating index The mutual information between them measures the correlation between candidate indicators and the target operating state, which is the target operating indicator. Used to characterize the overall operational status of the analyzed object during a specific period, and serving as a reference benchmark for candidate indicator selection, this target operational indicator... Based on the application scenario, specifically, the target operating index is determined by the weighted sum of multiple reference indicators related to the operating status. The weight of each reference indicator is determined based on its ability to represent the operating status. The observation targets are industries or regions, and the observation period is consistent with the sampling period of the net load signal; This represents the set of selected indicators. Indicate candidate indicators With selected indicators The mutual information between candidate indicators measures the degree of information redundancy. Based on this criterion, candidate indicators that are highly correlated with the target operating index and have low redundancy with already selected indicators are retained, while candidate indicators with low macroeconomic relevance and high information redundancy are eliminated.

[0050] The second stage involves a multi-objective joint evaluation of the candidate indicators that passed the initial screening. A multi-objective joint evaluation function is constructed as follows: In the formula: Indicate candidate indicators Overall score; To indicate the correlation between candidate indicators and the target operating index, candidate indicators are used. With target operating index Calculation of the absolute value of the Pearson correlation coefficient between them ; This indicates the time lead of the candidate indicator relative to the target operating index, calculated within a preset time lag range. With target operating index The time-delay correlation coefficient is used, and the time-delay value of the candidate index (indicated by a positive number) is normalized and used as the time-leading index; when the candidate index leads the target index... The value is relatively large; when both are synchronized or the candidate indicator lags behind, The value is relatively small; This indicates the representativeness or propagation explanatory power of the candidate indicators in the relational network, for candidate indicators (network diffusion contribution and aggregation sensitivity) from the second candidate indicator set. Take the higher benchmark value; for candidate indicators from the first candidate indicator set, The contribution of the node centrality or edge weight in the comprehensive propagation edge weight network constructed in step S3 is determined based on this indicator. This indicates stability across time windows, i.e., the consistency of statistical characteristics of candidate indicators across different time periods. The entire observation period is divided into multiple sub-time windows, and the mean of the candidate indicator within each sub-time window is calculated. The negative of the coefficient of variation of the mean of each sub-time window is used as the criterion. The smaller the coefficient of variation, the higher the stability. The larger the value; This indicates the effectiveness of efficiency correction, i.e., the ability of the candidate indicator to distinguish between changes in activity and changes in efficiency. When the candidate indicator is the efficiency elasticity indicator calculated in S2... Take the higher benchmark value; when the candidate index is the entropy of power consumption structure change or the stability of operating rhythm. Take a moderate benchmark value; when the candidate index is a multi-scale feature obtained from mode decomposition. The lower benchmark value should be chosen. The benchmark values ​​are set based on the degree of correlation between each candidate indicator and the efficiency elasticity indicator in terms of their definitions. The redundancy of candidate indicators, i.e., the degree of information overlap with other indicators, is represented by candidate indicators. With the candidate indicator set The average mutual information of all other candidate indicators is used as : The larger this value, the higher the degree of information overlap between the candidate indicator and other indicators, and the greater the redundancy. This indicates the sensitivity of the candidate indicator to external non-operational disturbances. The lower the value, the stronger the candidate indicator's anti-interference capability. Based on the original power load before the disturbance was removed in S1, the change range of the candidate indicator before and after the disturbance removal was recalculated. The ratio of this change range to the standard deviation of the candidate indicator was used as the threshold value. The smaller this ratio, the less the candidate index is affected by external non-operational disturbances, and the stronger its anti-interference ability. For candidate indices obtained from net load signals, its Typically lower; if some candidate indicators remain sensitive to external disturbances, then A higher score will result in a corresponding deduction in the overall evaluation. , , , , , , These are the weighting coefficients for each evaluation dimension. Through the above multi-objective joint evaluation, candidate indicators are comprehensively scored from multiple dimensions, including relevance, leadership, representativeness, stability, efficiency correction capability, redundancy, and robustness to disturbances. Screening constraints are set based on the comprehensive score to select indicators that meet the requirements. , , , The core sensitivity indicators constitute the core sensitivity indicator set. .in, This is the lower limit threshold for the comprehensive score. This is the upper limit threshold for redundancy. This is the upper threshold for perturbation sensitivity. 1. This is the lower limit threshold for stability. 2. After obtaining the core sensitivity index set, the indicators in the core sensitivity index set are subjected to directional unification and standardization processing so that they participate in the synthesis of the comprehensive index under a unified direction and scale.

[0051] First, based on core indicators and target operating indices The relevant directions determine the direction adjustment coefficient: In the formula For symbolic functions, Indicators With target operating index The correlation coefficient between them. Then, the indicators are standardized: In the formula As an indicator At any moment The standardized value, The original value of the indicator. and Representing indicators The mean and standard deviation. After unifying the directions, an increase in the values ​​of all core sensitivity indicators indicates an enhanced operational sensitivity.

[0052] The weights of each core sensitivity indicator are determined based on their comprehensive score normalization: In the formula Core indicators The weights of the standardized core sensitivity indicators are then weighted and synthesized to obtain the comprehensive sensitivity index: In the formula: For a moment The comprehensive sensitivity index comprehensively reflects multi-dimensional information about the changes in the analyzed object itself and the characteristics of the linkage and propagation between analyzed objects. It can characterize the direction, intensity, and potential risks of changes in the operational status as a whole. The comprehensive sensitivity index provides a quantifiable, interpretable, and dynamically updated indicator basis for subsequent operational status monitoring, trend judgment, and risk identification.

[0053] Furthermore, embodiments of the power sensitivity index screening device based on multidimensional characteristics of net load in this application include the following: The net load signal set generation module is used to acquire the original power load and external disturbance factor data, remove the disturbance load determined by the external disturbance factor data from the original power load, and generate the net load signal set after constraint correction. The first candidate index set construction module is used to perform multi-scale mode decomposition on the net load signals in the net load signal set and extract multi-scale features, and combine them with multi-dimensional operation feature indicators to construct a first candidate index set that reflects the change characteristics of the analysis object itself. The second candidate index set determination module is used to construct a comprehensive propagation edge weight based on the net load signal set, integrating the sensitivity of collaborative evolution among analysis objects, time delay relationship, prior relationship, feature similarity and power flow direction; calculate the network diffusion contribution and aggregation sensitivity based on the comprehensive propagation edge weight, and use the network diffusion contribution and aggregation sensitivity as the second candidate index set reflecting the linkage propagation characteristics among different analysis objects; The comprehensive sensitivity index determination module is used to merge the first candidate index set and the second candidate index set to form a total candidate index set, and to select the core sensitivity index set based on the total candidate index set to obtain the comprehensive sensitivity index.

[0054] For specific limitations regarding the device, please refer to the limitations of the method above, which will not be repeated here. Each module in the above device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0055] It is understood that those skilled in the art can combine various implementation methods in the above embodiments under the guidance of the above examples to obtain technical solutions with multiple implementation methods.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for screening power sensitivity indicators based on multidimensional characteristics of net load, characterized in that, include: The original power load and external disturbance factor data are obtained, and the disturbance load determined by the external disturbance factor data is removed from the original power load. After constraint correction, a net load signal set is generated. Multi-scale mode decomposition is performed on the net load signals in the net load signal set, and multi-scale features are extracted. Combined with multi-dimensional operational feature indicators, a first candidate indicator set reflecting the change characteristics of the analysis object itself is constructed. Based on the net load signal set, the sensitivity of collaborative evolution among the analysis objects, time delay relationship, prior relationship, feature similarity and power flow direction are integrated to construct a comprehensive propagation edge weight; based on the comprehensive propagation edge weight, the network diffusion contribution and aggregation sensitivity are calculated, and the network diffusion contribution and aggregation sensitivity are used as a second candidate index set to reflect the linkage propagation characteristics among different analysis objects; The first candidate indicator set and the second candidate indicator set are merged to form a total candidate indicator set, and a core sensitivity indicator set is selected based on the total candidate indicator set to obtain a comprehensive sensitivity index.

2. The method for screening power sensitivity indicators based on multidimensional characteristics of net load according to claim 1, characterized in that, The process of removing the disturbance load determined by the external disturbance factor data from the original power load includes: Based on the various types of data in the external disturbance factor data, corresponding load influence functions are constructed respectively. The outputs of each load influence function are superimposed to synthesize the non-operational disturbance counterfactual baseline load. The non-operational disturbance counterfactual baseline load is treated as a disturbance load and removed from the original power load.

3. The method for screening power sensitivity indicators based on multidimensional characteristics of net load according to claim 1, characterized in that, The step of performing multi-scale mode decomposition and extracting multi-scale features from the net load signal set includes: The net load signal is subjected to variational mode decomposition to obtain K modal components with different center frequencies, and the sum of the modal components is equal to the net load signal. Modal components with center frequencies below a first preset frequency threshold are considered long-term trend features, modal components with center frequencies above a second preset frequency threshold are considered short-term impact features, and modal components with center frequencies between the first preset frequency threshold and the second preset frequency threshold are considered medium-term periodic features; wherein, the second preset frequency threshold is greater than the first preset frequency threshold.

4. The method for screening power sensitivity indicators based on multidimensional characteristics of net load according to claim 3, characterized in that, The multidimensional operational characteristic indicators include: The entropy of electricity consumption structure change is determined based on the information entropy calculated from the proportion of electricity consumption of various analytical objects within a set period. It is used to reflect the uniformity of the distribution of electricity consumption structure. The stability of the operating rhythm is determined based on the degree of deviation of the net load signal of the analyzed object from the mean and peak-valley envelope of the net load signal within a set period. It is used to reflect the degree of change in the shape of the load curve and the activity rhythm of the analyzed object. The efficiency elasticity index is determined based on the ratio of the reference activity growth rate to the net electricity consumption growth rate of the analyzed object. It is used to distinguish whether changes in net load are caused by changes in activity or by changes in efficiency.

5. The method for screening power sensitivity indicators based on multidimensional characteristics of net load according to claim 4, characterized in that, The first candidate indicator set, which combines multidimensional operational characteristic indicators to reflect the changing characteristics of the analyzed object, includes: The multi-scale features are combined with the entropy of the power consumption structure change, the stability of the operating rhythm, and the efficiency elasticity index to form a sensitive feature vector, which is then used as the first candidate index set. Analysis object exist The sensitive feature vector at time step is represented as: in: For the first to the second One modal component, The entropy of changes in electricity structure. To ensure operational rhythm stability, For efficiency elasticity indicators, superscript This represents the transpose of a vector.

6. The method for screening power sensitivity indicators based on multidimensional characteristics of net load according to claim 1, characterized in that, The analysis of the sensitivity to co-evolution and time-delay relationships among the objects based on the net load signal set includes: For any two analysis objects, calculate the ratio of the product of the covariance and the standard deviation of the net load signal of each of the two analysis objects under multiple preset time delays, and use it as the co-evolution sensitivity under each time delay; The time delay at which the maximum co-evolution sensitivity is obtained is determined as the time delay relationship between the two analysis objects, and the maximum co-evolution sensitivity is taken as the co-evolution intensity corresponding to the time delay relationship.

7. The method for screening power sensitivity indicators based on multidimensional characteristics of net load according to claim 6, characterized in that, The fusion analysis of prior relationships, feature similarities, and power flow directions among the objects constructs a comprehensive propagation edge weight, including: Obtain the prior relationship strength between the analysis objects and the sensitive feature similarity determined by the sensitive feature vector; multiply the prior relationship strength and the sensitive feature similarity to generate the relationship coupling edge weight; The edge weights corresponding to the power flow direction, the leading and lagging edge weights determined based on the time delay relationship, the relationship coupling edge weights, and the similarity edge weights determined based on the sensitivity feature similarity are weighted and fused to obtain the comprehensive propagation edge weights.

8. The method for screening power sensitivity indicators based on multidimensional characteristics of net load according to claim 7, characterized in that, The calculation of network diffusion contribution based on the integrated propagation edge weights includes: Based on the comprehensive propagation edge weights, the lead-lag factors corresponding to the time lag relationships, and the sensitivity feature similarity, the sensitivity propagation intensity between the analyzed objects is calculated: In the formula: For a moment Analysis object Towards The sensitivity of the transmission intensity In order to comprehensively disseminate border rights, The leading-lag factor is determined based on the time-delay relationship. For sensitive feature similarity; For each analysis object, the ratio of the sum of the sensitivity output intensities of the analysis object as a propagation source to the sum of the sensitivity input intensities as a receiver is calculated as the network diffusion contribution of that analysis object. In the formula: For the analysis object At any moment The network diffusion contribution is represented by a value greater than 1, indicating an output node, and a value less than 1, indicating a receiving node.

9. The method for screening power sensitivity indicators based on multidimensional characteristics of net load according to claim 6, characterized in that, The aggregation sensitivity is calculated in the following manner: The analysis objects with prior associations are divided into object groups, and the object weights of each analysis object within the object group are obtained. Then, the sensitivity feature vectors of each analysis object within the object group are weighted and aggregated to obtain the aggregated sensitivity of the object group. In the formula: For the object group At any moment The aggregation sensitivity, For the analysis object Object weight, For sensitivity mapping function, This is for pooling operations.

10. A power sensitivity index screening device based on multidimensional characteristics of net load, characterized in that, The power sensitivity index screening method based on the multidimensional characteristics of net load as described in any one of claims 1 to 9 includes: The net load signal set generation module is used to acquire the original power load and external disturbance factor data, remove the disturbance load determined by the external disturbance factor data from the original power load, and generate the net load signal set after constraint correction. The first candidate index set construction module is used to perform multi-scale mode decomposition on the net load signals in the net load signal set and extract multi-scale features, and combine them with multi-dimensional operation feature indicators to construct a first candidate index set that reflects the change characteristics of the analysis object itself. The second candidate index set determination module is used to construct a comprehensive propagation edge weight based on the net load signal set, by integrating the sensitivity of collaborative evolution among the analysis objects, time delay relationship, prior relationship, feature similarity and power flow direction; and to calculate the network diffusion contribution and aggregation sensitivity based on the comprehensive propagation edge weight, and to use the network diffusion contribution and aggregation sensitivity as the second candidate index set reflecting the linkage propagation characteristics among different analysis objects. The comprehensive sensitivity index determination module is used to merge the first candidate index set and the second candidate index set to form a total candidate index set, and to screen the core sensitivity index set based on the total candidate index set to obtain the comprehensive sensitivity index.