A method, system and medium for adjustable resource time-varying sensitivity identification and ordering

Abstract

The present application relates to the technical field of power grid demand response evaluation, and particularly relates to a method, system and medium for identifying and sorting time-varying sensitivity of adjustable resources, which comprises the following steps: obtaining load response data, and constructing a response feature sequence and a response target sequence; inputting the response feature sequence into a time-varying sensitivity identification model for time sequence coding, and generating a price sensitivity coefficient and a temperature sensitivity coefficient at each time; determining a load change regression value based on the price sensitivity coefficient, the temperature sensitivity coefficient, and a price change amount and a temperature change amount, and training the time-varying sensitivity identification model according to the load change regression value and the response target sequence; constructing a two-dimensional sensitive feature based on the price sensitivity coefficient and the temperature sensitivity coefficient; determining a user level based on a non-dominant relationship, and determining a sorting result. Through the present application, the problem that the response relationship of user load to price and temperature has time-varying coupling characteristics, leading to difficulty in accurately identifying sensitivity and further causing distortion of user hierarchical sorting, is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of power grid demand response assessment technology, and in particular to a method, system and medium for identifying and ranking the time-varying sensitivity of adjustable resources. Background Technology

[0002] Demand response is a regulatory approach where the power grid guides users to adjust their electricity consumption behavior through electricity prices or incentive signals, thereby mobilizing adjustable resources on the load side to participate in supply and demand balance. Its implementation relies on accurately assessing the response sensitivity of different users and determining their participation order. In existing technologies, CN114065109A constructs an evaluation index system that includes electricity price sensitivity and temperature sensitivity, and prioritizes demand response users by combining index weights with an ideal solution; CN111582911A evaluates users' electricity price sensitivity and temperature sensitivity based on load, electricity price, and temperature data respectively, and forms user behavior profiles accordingly.

[0003] However, the aforementioned technologies mainly use statistical evaluation or index weighting methods to characterize user sensitivity, which makes it difficult to depict the hourly response relationship of user load to changes in electricity price and temperature. Furthermore, synthesizing different sensitivity dimensions into a comprehensive evaluation result can easily obscure the relative advantages of users in each sensitivity dimension, leading to inaccurate identification of users' time-varying sensitivity and making it difficult to provide a reliable basis for the objective hierarchical ranking of adjustable resources.

[0004] The information disclosed in this background section is intended only to enhance the understanding of the general background of this disclosure and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] This invention provides a method, system, and medium for identifying and ranking time-varying sensitivity of adjustable resources, which can effectively solve the problems in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for identifying and ranking the time-varying sensitivity of adjustable resources, the method comprising: Obtain load response data from multiple adjustable resource users, and construct a response feature sequence and a response target sequence based on the load response data; The response feature sequence is input into the time-varying sensitivity identification model for time-series encoding, and the electricity price sensitivity coefficient and temperature sensitivity coefficient are generated time-by-time based on the encoding results; The load change regression value is determined based on the electricity price sensitivity coefficient, the temperature sensitivity coefficient, and the corresponding electricity price change and temperature change, and the time-varying sensitivity identification model is trained based on the load change regression value and the response target sequence. The electricity price sensitivity coefficient and temperature sensitivity coefficient of each adjustable resource user are obtained according to the trained time-varying sensitivity identification model, and a two-dimensional sensitivity feature of each adjustable resource user is constructed based on the electricity price sensitivity coefficient and the temperature sensitivity coefficient. The user level of each adjustable resource user is determined based on the non-dominant relationship between the two-dimensional sensitive features, and the ranking result of each adjustable resource user is determined within the user level based on the two-dimensional sensitive features.

[0007] Furthermore, the load response data includes user attribute data, load data, electricity price data, temperature data, and time attribute data; The response feature sequence is constructed based on the changes in the load data, the electricity price data, and the temperature data, as well as the user attribute data and the time attribute data. The target response sequence is constructed based on the changes in the load data.

[0008] Furthermore, the structure of the time-varying sensitivity identification model includes: The feature embedding layer, temporal coding layer, time-varying coefficient generation layer, and physical regression layer are connected in sequence. The feature embedding layer performs feature mapping on the user attribute data and time attribute data in the load response data, and fuses the mapped features with the changes in electricity price and temperature to obtain a time-series feature sequence. The temporal coding layer encodes the temporal feature sequence to obtain the encoded features corresponding to the temporal feature sequence in the time dimension; The time-varying coefficient generation layer generates the electricity price sensitivity coefficient, the temperature sensitivity coefficient, and the intercept term on a time-by-time basis based on the encoding features; The physical regression layer determines the load change regression value based on the electricity price sensitivity coefficient, the temperature sensitivity coefficient, the intercept term, and the corresponding changes in electricity price and temperature at the time point.

[0009] Furthermore, the temporal coding layer uses an Informer encoder with the downsampling structure removed to encode the temporal feature sequence, so that the encoded features have the same number of time steps as the temporal feature sequence; The Informer encoder uses a sparse self-attention mechanism to select key temporal features representing the load response relationship from the temporal feature sequence, and obtains the encoded features through multi-head attention processing and feedforward processing; The time-varying coefficient generation layer decouples and maps the encoded features to obtain the electricity price sensitivity coefficient, the temperature sensitivity coefficient, and the intercept term corresponding to each time point.

[0010] Furthermore, the training of the time-varying sensitivity identification model includes: The response feature sequence and the response target sequence are used to construct training samples; The response feature sequence is input into the time-varying sensitivity identification model to obtain the load change regression value and the corresponding electricity price sensitivity coefficient, temperature sensitivity coefficient and intercept term; The model parameters of the time-varying sensitivity identification model are updated based on the error between the load change regression value and the response target sequence; When the error meets the preset convergence condition or the number of training iterations reaches the preset number, the trained time-varying sensitivity identification model is obtained.

[0011] Furthermore, the two-dimensional sensitive features are constructed, including: At the same time, each adjustable resource user is ranked according to the degree of electricity price response represented by the electricity price sensitivity coefficient, and the electricity price sensitivity score of each adjustable resource user is determined based on the corresponding ranking position. The adjustable resource users are sorted according to the temperature response degree represented by the temperature sensitivity coefficient, and the temperature sensitivity score of each adjustable resource user is determined based on the corresponding sorting position. The electricity price sensitivity score and the temperature sensitivity score of the same adjustable resource user are combined to obtain the two-dimensional sensitivity feature.

[0012] Further, determining the user hierarchy includes: At any given moment, if the electricity price sensitivity score and the temperature sensitivity score of any one of the adjustable resource users are not lower than those of another adjustable resource user, and at least one sensitivity score is higher than that of another adjustable resource user, then it is determined that any one of the adjustable resource users dominates the other adjustable resource user. Adjustable resource users who are not controlled by other adjustable resource users are classified as the current user level, and the adjustable resource users corresponding to the current user level are removed from the user set to be classified. Repeat the dominance relationship determination for the remaining adjustable resource users until the user hierarchy division of each adjustable resource user is completed; Map different numbers of user levels at each time point to a preset number of standardized user levels to obtain the standardized level corresponding to each adjustable resource user.

[0013] Further, determining the sorting result includes: Based on the difference between the electricity price sensitivity score and the temperature sensitivity score, the electricity price-dominant sensitivity ranking index and the temperature-dominant sensitivity ranking index are determined, and based on the synergistic relationship between the electricity price sensitivity score and the temperature sensitivity score, the dual-dominant sensitivity ranking index is determined. Based on the target indicators in the electricity price-dominated sensitivity ranking index, the temperature-dominated sensitivity ranking index, and the dual-dominated sensitivity ranking index, determine the electricity price-dominated, temperature-dominated, or dual-dominated sensitivity type corresponding to each adjustable resource user; The inter-layer ranking of each adjustable resource user is determined according to the standardized user level, and the intra-layer ranking of each adjustable resource user within the same standardized user level is determined based on the target index, thereby obtaining the ranking result.

[0014] An adjustable resource time-varying sensitivity identification and ranking system, the system comprising: The response data construction module acquires load response data from multiple adjustable resource users and constructs response feature sequences and response target sequences based on the load response data. The time-varying coefficient generation module inputs the response feature sequence into the time-varying sensitivity identification model for time-series encoding, and generates electricity price sensitivity coefficient and temperature sensitivity coefficient time by time based on the encoding results; The regression model training module determines the load change regression value based on the electricity price sensitivity coefficient, temperature sensitivity coefficient, and the corresponding changes in electricity price and temperature, and trains the time-varying sensitivity identification model based on the load change regression value and the response target sequence. The dual-dimensional feature construction module obtains the electricity price sensitivity coefficient and temperature sensitivity coefficient of each adjustable resource user based on the trained time-varying sensitivity identification model, and constructs the dual-dimensional sensitivity features of each adjustable resource user based on the electricity price sensitivity coefficient and temperature sensitivity coefficient. The user hierarchical ranking module determines the user level of each adjustable resource user based on the non-dominant relationship between two-dimensional sensitive features, and determines the ranking result of each adjustable resource user within the user level based on the two-dimensional sensitive features.

[0015] A computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, can implement the adjustable resource time-varying sensitivity identification and sorting method.

[0016] The technical solution of this invention can achieve the following technical effects: By acquiring load response data from adjustable resource users, response feature sequences and response target sequences are constructed. A time-varying sensitivity identification model is used to generate electricity price sensitivity coefficients and temperature sensitivity coefficients at each time step. Combined with load change regression values, the model is trained and a two-dimensional sensitivity feature of users is constructed. Users are stratified based on non-dominated relationships and ranked within the strata. This effectively solves the problem in existing power grid demand response assessments where the time-varying coupling characteristics of user load response to electricity price and temperature make it difficult to accurately identify sensitivity and further cause distortion in user stratification and ranking.

[0017] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0018] 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 some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the method for identifying and ranking the time-varying sensitivity of adjustable resources; Figure 2 This is a schematic diagram of a method for identifying and ranking the time-varying sensitivity of adjustable resources. Figure 3 A heatmap showing the average monthly electricity price coefficient for a specific user. Figure 4 A Pareto sensitivity sorting graph for industrial users at a given moment. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] Example 1; like Figure 1 As shown, this application provides a method for identifying and ranking the time-varying sensitivity of adjustable resources, the method including: S10: Obtain load response data from multiple adjustable resource users, and construct a response feature sequence and a response target sequence based on the load response data; S20: Input the response feature sequence into the time-varying sensitivity identification model for time-series encoding, and generate the electricity price sensitivity coefficient and temperature sensitivity coefficient time-by-time based on the encoding results; S30: Determine the load change regression value based on the electricity price sensitivity coefficient, temperature sensitivity coefficient, and the corresponding changes in electricity price and temperature, and train the time-varying sensitivity identification model based on the load change regression value and the response target sequence; S40: Obtain the electricity price sensitivity coefficient and temperature sensitivity coefficient of each adjustable resource user based on the trained time-varying sensitivity identification model, and construct the two-dimensional sensitivity features of each adjustable resource user based on the electricity price sensitivity coefficient and temperature sensitivity coefficient. S50: Determine the user level of each adjustable resource user based on the non-dominant relationship between the two-dimensional sensitive features, and determine the ranking result of each adjustable resource user within the user level based on the two-dimensional sensitive features.

[0023] Specifically, the system acquires user identifiers, historical loads, time-of-use electricity prices, meteorological temperatures, and month, date, time period, and weekday identifiers for each user within a continuous statistical period, and aligns the data according to a unified sampling time. Linear interpolation is used to complete missing data, sliding window filtering is used to smooth abnormal fluctuations, and Z-score standardization is performed, with electricity prices standardized uniformly by region, temperature standardized separately by city, and load standardized independently by user. Based on data from adjacent sampling times, load changes, electricity price changes, and temperature changes are calculated. User identifiers, time attributes, electricity price changes, temperature changes, and corresponding historical load characteristics are combined to form a response feature sequence, and load changes are combined to form a response target sequence. A time-varying sensitivity identification model is constructed, which includes a feature embedding layer, an improved Informer encoder, a time-varying coefficient generation layer, and a physical regression layer connected in sequence. The feature embedding layer maps user identifiers to user feature vectors and month, date, time period, and weekday identifiers to time semantic vectors. The user feature vectors, time semantic vectors, electricity price changes, and temperature changes are then concatenated and subjected to linear projection and positional encoding to form a high-dimensional feature sequence. The improved Informer encoder eliminates the convolutional downsampling structure, ensuring that the encoding result maintains the same number of time steps as the input sequence. ProbSparse self-attention, multi-head attention, and a feedforward neural network are used to extract long-term time-dependent features. The time-varying coefficient generation layer converts the encoded features of each time step into electricity price sensitivity coefficient, temperature sensitivity coefficient, and intercept term through a fully connected mapping. The physical regression layer calculates the load change regression value at the corresponding time point by summing the product of the electricity price sensitivity coefficient and the change in electricity price, the product of the temperature sensitivity coefficient and the change in temperature, and the intercept term. Thus, the electricity price sensitivity coefficient characterizes the degree of load response to changes in electricity price, and the temperature sensitivity coefficient characterizes the degree of load response to changes in temperature. Using the response feature sequence as model input and the response target sequence as supervision target, the mean square error between the load change regression value and the actual load change is used as the loss function. The parameters of the feature embedding layer, the improved Informer encoder, and the time-varying coefficient generation layer are updated through backpropagation. When the loss curve converges stably or the training times reach the preset number, the model parameters are saved to obtain the trained time-varying sensitivity identification model. The trained model is then used to process the data of each user to obtain the electricity price sensitivity coefficient sequence and the temperature sensitivity coefficient sequence corresponding to each sampling time. At any given time, the electricity price sensitivity coefficients of each user are arranged in ascending order and converted into electricity price sensitivity scores with consistent value ranges based on the ranking; the temperature sensitivity coefficients of each user are arranged in descending order and converted into temperature sensitivity scores with consistent value ranges based on the ranking; the electricity price sensitivity score and temperature sensitivity score of the same user are combined into a two-dimensional sensitivity feature, with higher scores indicating stronger response sensitivity in the corresponding dimension; For any two users at the same time, if one user's electricity price sensitivity score and temperature sensitivity score are both not lower than the other user's, and at least one sensitivity score is higher, the former is determined to dominate the latter. Users not dominated by other users are assigned to the first user level. After removing the first user level, the dominance relationship judgment is repeated for the remaining users to obtain the subsequent user levels in turn, until the level division of all users is completed. Then, the original levels formed at different times are mapped to a preset number of standardized levels so that the level results at each time have a unified expression form. Under user-level constraints, the following ranking criteria are applied: Electricity price-dominant sensitivity ranking index is calculated based on the degree to which the electricity price sensitivity score is higher and the temperature sensitivity score is lower; Temperature-dominant sensitivity ranking index is calculated based on the degree to which the temperature sensitivity score is higher and the electricity price sensitivity score is lower; and Dual-dominant sensitivity ranking index is calculated based on the degree to which both sensitivity scores are higher. The three categories of indicators are compared, and the type corresponding to the largest indicator is determined as the user's dominant sensitivity type. Inter-level ranking is then performed from front to back according to user level, and intra-level ranking is performed within the same user level according to the corresponding sensitivity ranking index. Finally, the user level, dominant sensitivity type, and ranking results for each adjustable resource user are output.

[0024] The technical solution of this invention obtains load response data of adjustable resource users to construct response feature sequences and response target sequences. It uses a time-varying sensitivity identification model to generate electricity price sensitivity coefficients and temperature sensitivity coefficients at each time step and combines them with load change regression values ​​to complete model training. This leads to the construction of two-dimensional sensitivity features of users. Based on non-dominated relationships, users are stratified and ranked within the strata. This effectively solves the problem in existing power grid demand response assessments where the time-varying coupling characteristics of user load response to electricity price and temperature make it difficult to accurately identify sensitivity and further cause distortion in user stratification and ranking.

[0025] Furthermore, load response data includes user attribute data, load data, electricity price data, temperature data, and time attribute data; The response characteristic sequence is constructed based on changes in load data, electricity price data, and temperature data, as well as user attribute data and time attribute data. The response target sequence is constructed based on the changes in load data.

[0026] As a preferred embodiment of the above, the data collected from users with adjustable resources is cleaned, normalized, and feature-constructed to form a user load response feature sequence, specifically including: Missing data is filled using linear interpolation, and outlier data is smoothed using sliding window filtering. The linear interpolation uses the following formula: ; In the formula, t1 and t2 represent the timestamps of two known data points; and These represent the known load values ​​corresponding to timestamps t1 and t2, respectively. A function representing the change of load value over time; This indicates the time point that needs to be filled in; Indicates t m The load value of missing time points; The sliding window filter uses the following formula: ; In the formula, The data at time k is filtered by a sliding window, where k represents the discrete-time index of the current time; i represents the relative position index of the data point within the window, ranging from 0 to N-1, where N is the window size; Represents the raw data at time ki; The data is standardized using the Z-score standardization method, and the formula can be expressed as: ; In the formula, z represents the standardized value; x represents the original data value; This represents the overall data mean; This represents the overall data standard deviation. Considering the differences in characteristics between different physical quantities, electricity price data adopts a unified regional standardization, temperature data is standardized separately for each city, and load data is standardized independently for each user.

[0027] Furthermore, the structure of the time-varying sensitivity identification model includes: The feature embedding layer, temporal coding layer, time-varying coefficient generation layer, and physical regression layer are connected in sequence. The feature embedding layer performs feature mapping on user attribute data and time attribute data in the load response data, and fuses the mapped features with electricity price changes and temperature changes to obtain a time-series feature sequence; The temporal coding layer encodes the temporal feature sequence to obtain the encoded features corresponding to the temporal feature sequence in the time dimension; The time-varying coefficient generation layer generates electricity price sensitivity coefficient, temperature sensitivity coefficient, and intercept term on a time-by-time basis based on coding features; The physical regression layer determines the load change regression value based on the electricity price sensitivity coefficient, temperature sensitivity coefficient, intercept term, and the corresponding changes in electricity price and temperature at the time point.

[0028] Furthermore, the temporal coding layer uses an Informer encoder with the downsampling structure removed to encode the temporal feature sequence, so that the encoded features have the same number of time steps as the temporal feature sequence; The Informer encoder employs a sparse self-attention mechanism to select key temporal features representing the load-response relationship from the temporal feature sequence, and obtains the encoded features through multi-head attention processing and feedforward processing; The time-varying coefficient generation layer decouples and maps the encoded features to obtain the electricity price sensitivity coefficient, temperature sensitivity coefficient and intercept term at each time point.

[0029] Furthermore, the training of the time-varying sensitivity identification model includes: The response feature sequence and the response target sequence are used to construct training samples; Input the response feature sequence into the time-varying sensitivity identification model to obtain the load change regression value and the corresponding electricity price sensitivity coefficient, temperature sensitivity coefficient and intercept term; The model parameters of the time-varying sensitivity identification model are updated based on the error between the load change regression value and the response target sequence; When the error meets the preset convergence condition or the number of training iterations reaches the preset number, the time-varying sensitivity identification model after training is obtained.

[0030] As a preferred embodiment of the above, the improved Informer time-varying coefficient regression model is composed of a feature embedding layer, an improved Informer encoder, a time-varying coefficient generation layer, and a physical regression layer connected sequentially, specifically including: The feature embedding layer is configured to map user IDs to user feature vectors through the UserEmbedding layer to capture inherent electricity consumption patterns, and to map month, date, time period, and weekday identifiers to high-dimensional temporal semantic vectors through the TimeFeatureEmbedding layer; and to embed the standardized electricity price difference sequence. Temperature difference sequence The algorithm concatenates user feature vectors and temporal semantic vectors, performs linear projection and positional encoding to generate a high-dimensional feature sequence that integrates user attributes, external driving factors, and temporal information. The improved Informer encoder is configured to remove the convolutional downsampling layer from the original Informer to maintain consistent input and output sequence lengths. A ProbSparse self-attention mechanism is used to filter key queries, and a multi-head attention and feedforward neural network output an encoded feature matrix containing long-term temporal dependencies. The time-varying coefficient generation layer is configured to decouple the encoded feature matrix and map it to a time-varying electricity price sensitivity coefficient through a fully connected layer. Temperature sensitivity coefficient and intercept term The physical regression layer is configured to calculate load change regression values ​​based on physical constraint formulas, achieving a fusion of deep features and physical interpretability. The regression formula is as follows: ; In the formula, This represents the regression value of the load change at time t; These represent the input electricity price and temperature difference, respectively. This represents the electricity price sensitivity coefficient; b(t) represents the temperature sensitivity coefficient; b(t) represents the intercept term. The training process of the time-varying coefficient regression model on the user load response feature sequence will be further explained as follows: Using the user load response feature sequence as model input and the actual load change as supervision target, a training sample set is constructed. The training sample set is then input into the improved Informer time-varying coefficient regression model, which outputs the load change regression value and the corresponding electricity price sensitivity coefficient, temperature sensitivity coefficient, and intercept term. The model parameters are updated by backpropagation with the goal of minimizing the error between the load change regression value and the actual load change. After the model learning curve has converged stably or reached the preset number of training iterations, the final time-varying coefficient regression model parameters are determined.

[0031] Furthermore, constructing two-dimensional sensitive features includes: At the same time, each adjustable resource user is ranked according to the degree of electricity price response represented by the electricity price sensitivity coefficient, and the electricity price sensitivity score of each adjustable resource user is determined based on the corresponding ranking position. Users of adjustable resources are ranked according to the degree of temperature response represented by the temperature sensitivity coefficient, and the temperature sensitivity score of each user is determined based on the corresponding ranking position. By combining the electricity price sensitivity score and temperature sensitivity score of the same adjustable resource user, a two-dimensional sensitivity feature is obtained.

[0032] As a preferred embodiment of the above, extracting the time-varying sensitivity features of user-adjustable resources specifically includes: The electricity price sensitivity coefficients of users at various times are used to construct an electricity price sensitivity coefficient sequence, and the temperature sensitivity coefficients of users at various times are used to construct a temperature sensitivity coefficient sequence. The electricity price sensitivity coefficient characterizes the degree of response of user load to changes in time-of-use electricity prices; the smaller the electricity price sensitivity coefficient, the stronger the user's tendency to reduce electricity consumption when electricity prices rise. The temperature sensitivity coefficient characterizes the degree of response of user load to changes in meteorological temperature; the larger the temperature sensitivity coefficient, the stronger the user's tendency to increase load when temperatures rise. Electricity price sensitivity scores and temperature sensitivity scores are constructed, specifically including: At any time t, let N be the number of users participating in the sorting. Sort the electricity price sensitivity coefficients of each user in ascending order to obtain user u. i Electricity price coefficient ranking And calculate the electricity price sensitivity score according to the following formula: ; In the formula, Indicates user u i The electricity price sensitivity fraction at time t, and ; Sort the temperature sensitivity coefficients of each user in descending order to obtain user u. i Temperature coefficient ranking The temperature sensitivity fraction is calculated according to the following formula: ; In the formula, Indicates user u i The temperature-sensitive fraction at time t, and ; Based on the electricity price sensitivity score and temperature sensitivity score, a two-dimensional sensitivity feature vector of the user UI at time t is constructed: ; The higher the score, the stronger the user's sensitivity in the corresponding dimension; The Pareto non-dominant relationship is used to hierarchically divide users, specifically including: For any two users u at the same time t i and u j If the formula is satisfied: ; And at least one of the following inequalities must be satisfied: ; Then user u i User u is controlled at time t j Users not dominated by other users are classified as the first Pareto level. After removing users from the first Pareto level from the user set, the non-dominance relationship determination is repeated for the remaining users to obtain the second Pareto level, the third Pareto level, and so on, until all users have completed the level division.

[0033] Furthermore, determining user tiers includes: At any given moment, if the electricity price sensitivity score and temperature sensitivity score of any adjustable resource user are both not lower than those of another adjustable resource user, and at least one sensitivity score is higher than that of another adjustable resource user, then it is determined that any adjustable resource user dominates the other adjustable resource user. Adjustable resource users that are not controlled by other adjustable resource users are classified as the current user level, and the adjustable resource users corresponding to the current user level are removed from the user set to be classified. Repeat the dominance relationship determination for the remaining adjustable resource users until the user hierarchy division for each adjustable resource user is completed; Map different numbers of user levels at different times to a preset number of standardized user levels to obtain the standardized level corresponding to each adjustable resource user.

[0034] As a preferred embodiment of the above, the standardization of the user hierarchy division results through hierarchical mapping is further elaborated as follows: Let user u be at time t i The original Pareto hierarchy is The maximum original Pareto level at time t is If the preset number of standardized layers is L, then user u i At time t, the normalization level It can be represented as: ; In the formula, This represents the floor operation; through standardized hierarchical mapping, the different original Pareto levels that may exist at different times are uniformly compressed into L fixed levels, where the first level is the optimal frontier level and the Lth level is the level with the weakest relative advantage.

[0035] Furthermore, determining the sorting result includes: Based on the difference between electricity price sensitivity scores and temperature sensitivity scores, electricity price-dominant sensitivity ranking indicators and temperature-dominant sensitivity ranking indicators are determined, and based on the synergistic relationship between electricity price sensitivity scores and temperature sensitivity scores, dual-dominant sensitivity ranking indicators are determined. Based on the target indicators among the electricity price-dominated sensitivity ranking indicators, temperature-dominated sensitivity ranking indicators, and dual-dominated sensitivity ranking indicators, determine the electricity price-dominated, temperature-dominated, or dual-dominated sensitivity type corresponding to each adjustable resource user. The inter-layer ranking of adjustable resource users is determined according to the standardized user tier, and the intra-layer ranking of adjustable resource users within the same standardized user tier is determined based on the target indicators, thus obtaining the ranking results.

[0036] As a preferred embodiment of the above, the preferred ranking of users at the same level and the output of the preferred ranking result of adjustable resources include: Under Pareto hierarchy constraints, users are sorted from front to back according to their standardized hierarchy. For users within the same standardized hierarchy, the maximum value among the calculated electricity price-dominant sensitivity ranking index, temperature-dominant sensitivity ranking index, and dual-dominant sensitivity ranking index is taken. ; In the formula, Indicates user u i The maximum sensitivity ranking index at time t; Indicates user u i Electricity price is the dominant sensitive ranking indicator at time t; Indicates user u i Temperature at time t is the dominant sensitivity ranking indicator; Indicates user u i The dual-dominant sensitive ranking metric at time t; i represents the user index; t represents the evaluation time; if This indicates that user u i At that moment, electricity prices are dominant; when This indicates that the user is temperature-driven at that moment; when This indicates that the user is a dual-dominant user at that moment; Therefore, based on the electricity price sensitivity coefficient and temperature sensitivity coefficient extracted from the improved Informer time-varying coefficient regression model, electricity price sensitivity scores and temperature sensitivity scores of users are constructed respectively, forming a two-dimensional sensitivity feature; then, based on the Pareto non-dominant relationship, the hierarchical position of users in the two-dimensional sensitivity space is determined, and the original hierarchy is mapped to a standardized hierarchy; finally, by combining the three types of sensitivity ranking indicators of electricity price dominance, temperature dominance, and dual dominance, users at the same level are ranked to obtain the optimal ranking results and dominant sensitivity types of adjustable resource users.

[0037] Example 2; Based on the same inventive concept as the adjustable resource time-varying sensitivity identification and ranking method in the foregoing embodiments, the present invention also provides an adjustable resource time-varying sensitivity identification and ranking system, the system comprising: The response data construction module acquires load response data from multiple adjustable resource users and constructs response feature sequences and response target sequences based on the load response data. The time-varying coefficient generation module inputs the response feature sequence into the time-varying sensitivity identification model for time-series encoding, and generates electricity price sensitivity coefficient and temperature sensitivity coefficient time by time based on the encoding results; The regression model training module determines the load change regression value based on the electricity price sensitivity coefficient, temperature sensitivity coefficient, and the corresponding changes in electricity price and temperature, and trains the time-varying sensitivity identification model based on the load change regression value and the response target sequence. The dual-dimensional feature construction module obtains the electricity price sensitivity coefficient and temperature sensitivity coefficient of each adjustable resource user based on the trained time-varying sensitivity identification model, and constructs the dual-dimensional sensitivity features of each adjustable resource user based on the electricity price sensitivity coefficient and temperature sensitivity coefficient. The user hierarchical ranking module determines the user level of each adjustable resource user based on the non-dominant relationship between two-dimensional sensitive features, and determines the ranking result of each adjustable resource user within the user level based on the two-dimensional sensitive features.

[0038] The system described above in this invention can effectively implement a method for identifying and ranking the time-varying sensitivity of adjustable resources, and the technical effects it can achieve are as described in the above embodiments, and will not be repeated here.

[0039] Example 3; Based on the same inventive concept as the adjustable resource time-varying sensitivity identification and sorting method in the foregoing embodiments, the present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, can implement the adjustable resource time-varying sensitivity identification and sorting method.

[0040] Example 4; In this embodiment, the implementation of the inventive method is further described in detail; Combination Figure 1 process and Figure 2 The model's principle and structure are illustrated by taking the electricity consumption data, time-of-use electricity price data, and temperature data of 100 industry users in a certain province from May to October 2023 as an example. First, data preprocessing is performed: missing values ​​in user historical load data, time-of-use electricity price data, and meteorological temperature data are filled by linear interpolation, and abnormal fluctuation data are smoothed by sliding window filtering. Then, load data, electricity price data, and temperature data are standardized respectively, and load change, electricity price change, and temperature change are constructed to form a user load response characteristic sequence. After data preprocessing, user information, time features, historical load sequences, electricity price changes, and temperature changes are used as model inputs and fed into the improved Informer time-varying coefficient regression model. The model extracts user attributes and time semantic information through the feature embedding layer, captures long-term time-series dependencies through the improved Informer encoder, and outputs the electricity price sensitivity coefficient, temperature sensitivity coefficient, and intercept term at each time step through the time-varying coefficient generation layer. Figure 3 A heat map of the average monthly electricity price coefficient for users in a certain industry is provided. Furthermore, electricity price sensitivity scores and temperature sensitivity scores are constructed based on electricity price sensitivity coefficients and temperature sensitivity coefficients to form a two-dimensional sensitivity characteristic of users; Finally, Pareto non-dominant relationship analysis is performed based on the two-dimensional sensitivity features to realize the hierarchical division of users in the two-dimensional sensitivity space; and by combining the three types of sensitivity ranking indicators of electricity price dominance, temperature dominance and dual dominance, users in the same level are ranked in the best order to obtain the ranking results of adjustable resource users and the dominant sensitivity type. Figure 4 The graph shows the ranking results of 100 industry users at a certain moment.

[0041] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for identifying and ranking the time-varying sensitivity of adjustable resources, characterized in that, The method includes: Obtain load response data from multiple adjustable resource users, and construct a response feature sequence and a response target sequence based on the load response data; The response feature sequence is input into the time-varying sensitivity identification model for time-series encoding, and the electricity price sensitivity coefficient and temperature sensitivity coefficient are generated time-by-time based on the encoding results; The load change regression value is determined based on the electricity price sensitivity coefficient, the temperature sensitivity coefficient, and the corresponding electricity price change and temperature change, and the time-varying sensitivity identification model is trained based on the load change regression value and the response target sequence. The electricity price sensitivity coefficient and temperature sensitivity coefficient of each adjustable resource user are obtained according to the trained time-varying sensitivity identification model, and a two-dimensional sensitivity feature of each adjustable resource user is constructed based on the electricity price sensitivity coefficient and the temperature sensitivity coefficient. The user level of each adjustable resource user is determined based on the non-dominant relationship between the two-dimensional sensitive features, and the ranking result of each adjustable resource user is determined within the user level based on the two-dimensional sensitive features.

2. The adjustable resource time-varying sensitivity identification and ranking method according to claim 1, characterized in that, The load response data includes user attribute data, load data, electricity price data, temperature data, and time attribute data; The response feature sequence is constructed based on the changes in the load data, the electricity price data, and the temperature data, as well as the user attribute data and the time attribute data. The target response sequence is constructed based on the changes in the load data.

3. The adjustable resource time-varying sensitivity identification and ranking method according to claim 1, characterized in that, The structure of the time-varying sensitivity identification model includes: The feature embedding layer, temporal coding layer, time-varying coefficient generation layer, and physical regression layer are connected in sequence. The feature embedding layer performs feature mapping on the user attribute data and time attribute data in the load response data, and fuses the mapped features with the changes in electricity price and temperature to obtain a time-series feature sequence. The temporal coding layer encodes the temporal feature sequence to obtain the encoded features corresponding to the temporal feature sequence in the time dimension; The time-varying coefficient generation layer generates the electricity price sensitivity coefficient, the temperature sensitivity coefficient, and the intercept term on a time-by-time basis based on the encoding features; The physical regression layer determines the load change regression value based on the electricity price sensitivity coefficient, the temperature sensitivity coefficient, the intercept term, and the corresponding changes in electricity price and temperature at the time point.

4. The adjustable resource time-varying sensitivity identification and ranking method according to claim 3, characterized in that, The temporal coding layer uses an Informer encoder with the downsampling structure removed to encode the temporal feature sequence, so that the encoded features have the same number of time steps as the temporal feature sequence; The Informer encoder uses a sparse self-attention mechanism to select key temporal features representing the load response relationship from the temporal feature sequence, and obtains the encoded features through multi-head attention processing and feedforward processing; The time-varying coefficient generation layer decouples and maps the encoded features to obtain the electricity price sensitivity coefficient, the temperature sensitivity coefficient, and the intercept term corresponding to each time point.

5. The adjustable resource time-varying sensitivity identification and ranking method according to claim 3, characterized in that, The training of the time-varying sensitivity identification model includes: The response feature sequence and the response target sequence are used to construct training samples; The response feature sequence is input into the time-varying sensitivity identification model to obtain the load change regression value and the corresponding electricity price sensitivity coefficient, temperature sensitivity coefficient and intercept term; The model parameters of the time-varying sensitivity identification model are updated based on the error between the load change regression value and the response target sequence; When the error meets the preset convergence condition or the number of training iterations reaches the preset number, the trained time-varying sensitivity identification model is obtained.

6. The adjustable resource time-varying sensitivity identification and ranking method according to claim 1, characterized in that, Constructing the two-dimensional sensitive features includes: At the same time, each adjustable resource user is ranked according to the degree of electricity price response represented by the electricity price sensitivity coefficient, and the electricity price sensitivity score of each adjustable resource user is determined based on the corresponding ranking position. The adjustable resource users are sorted according to the temperature response degree represented by the temperature sensitivity coefficient, and the temperature sensitivity score of each adjustable resource user is determined based on the corresponding sorting position. The electricity price sensitivity score and the temperature sensitivity score of the same adjustable resource user are combined to obtain the two-dimensional sensitivity feature.

7. The adjustable resource time-varying sensitivity identification and ranking method according to claim 6, characterized in that, Determining the user hierarchy includes: At any given moment, if the electricity price sensitivity score and the temperature sensitivity score of any one of the adjustable resource users are not lower than those of another adjustable resource user, and at least one sensitivity score is higher than that of another adjustable resource user, then it is determined that any one of the adjustable resource users dominates the other adjustable resource user. Adjustable resource users who are not controlled by other adjustable resource users are classified as the current user level, and the adjustable resource users corresponding to the current user level are removed from the user set to be classified. Repeat the dominance relationship determination for the remaining adjustable resource users until the user hierarchy division of each adjustable resource user is completed; Map different numbers of user levels at each time point to a preset number of standardized user levels to obtain the standardized level corresponding to each adjustable resource user.

8. The adjustable resource time-varying sensitivity identification and ranking method according to claim 7, characterized in that, Determining the sorting result includes: Based on the difference between the electricity price sensitivity score and the temperature sensitivity score, the electricity price-dominant sensitivity ranking index and the temperature-dominant sensitivity ranking index are determined, and based on the synergistic relationship between the electricity price sensitivity score and the temperature sensitivity score, the dual-dominant sensitivity ranking index is determined. Based on the target indicators in the electricity price-dominated sensitivity ranking index, the temperature-dominated sensitivity ranking index, and the dual-dominated sensitivity ranking index, determine the electricity price-dominated, temperature-dominated, or dual-dominated sensitivity type corresponding to each adjustable resource user; The inter-layer ranking of each adjustable resource user is determined according to the standardized user level, and the intra-layer ranking of each adjustable resource user within the same standardized user level is determined based on the target index, thereby obtaining the ranking result.

9. A system for identifying and ranking the time-varying sensitivity of adjustable resources, characterized in that, The system includes: The response data construction module acquires load response data from multiple adjustable resource users and constructs response feature sequences and response target sequences based on the load response data. The time-varying coefficient generation module inputs the response feature sequence into the time-varying sensitivity identification model for time-series encoding, and generates electricity price sensitivity coefficient and temperature sensitivity coefficient time by time based on the encoding results; The regression model training module determines the load change regression value based on the electricity price sensitivity coefficient, temperature sensitivity coefficient, and the corresponding changes in electricity price and temperature, and trains the time-varying sensitivity identification model based on the load change regression value and the response target sequence. The dual-dimensional feature construction module obtains the electricity price sensitivity coefficient and temperature sensitivity coefficient of each adjustable resource user based on the trained time-varying sensitivity identification model, and constructs the dual-dimensional sensitivity features of each adjustable resource user based on the electricity price sensitivity coefficient and temperature sensitivity coefficient. The user hierarchical ranking module determines the user level of each adjustable resource user based on the non-dominant relationship between two-dimensional sensitive features, and determines the ranking result of each adjustable resource user within the user level based on the two-dimensional sensitive features.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which includes program instructions that, when executed by a processor, can implement the adjustable resource time-varying sensitivity identification and sorting method as described in any one of claims 1-8.

Images