Multi-dimensional adjustment modeling and dynamic aggregation method considering user electricity price perception

Abstract

The invention discloses a multi-dimensional adjustment modeling and dynamic aggregation method considering user electricity price perception, and the method comprises the steps: collecting historical load data of multiple types of power users, carrying out the clustering analysis based on load characteristics, and obtaining a typical load sample; calculating each adjustment capability index for the typical load sample, and performing normalization processing; establishing a power consumer response willingness index model based on multi-cycle electricity price perception, and training to obtain a willingness prediction model; carrying out simulation prediction under a set electricity price period length, price and excitation strategy, and carrying out normalization to obtain a response willingness degree; integrating the regulation capability index and the response willingness to obtain a multi-dimensional regulation capability vector data set; and carrying out dynamic grouping on the user adjustment capability by adopting DEC to obtain an optimal aggregation structure. The method can describe behavior differences of the users under different motivation and electricity price period lengths, has high expansibility and adaptability, and is suitable for application scenes such as virtual power plant user grouping, load scheduling and demand response optimization.

Description

Technical Field

[0001] This invention belongs to the field of power system demand response and user load modeling, and specifically relates to a multi-dimensional adjustment modeling and dynamic aggregation method that considers user electricity price perception. Background Technology

[0002] With the development of new power systems and the advancement of "dual-carbon" goals, user-side load regulation resources are gradually becoming an important component of power system flexibility. Power demand response mechanisms, as a crucial means to improve system operating efficiency, promote clean energy consumption, and alleviate grid operation pressure, are being widely applied in various scenarios such as virtual power plants, active distribution networks, and integrated energy services. Especially given the high uncertainty and volatility on both the source and load sides, how to efficiently identify, model, and aggregate user load resources with response potential has become one of the core issues in current demand response research and engineering practice.

[0003] However, current technologies primarily focus on modeling user responsiveness based on historical characteristics of load curves or rule-driven modeling, lacking a systematic quantification of multidimensional indicators of actual user adjustability. Key physical characteristics such as response capacity, response time, duration, response rate, and response energy are often not comprehensively considered. Furthermore, user response behavior to electricity prices and incentive mechanisms exhibits significant individual differences and periodic variations. This "subjective willingness" level of response potential is often simplified or even ignored in existing models, making it difficult to accurately depict their true response performance under different strategies.

[0004] In addition, traditional aggregation methods often employ static clustering or partitioning strategies based on a single feature, which are difficult to adapt to the dynamic, multi-scale, and multi-modal characteristics of users' adjustment capabilities and behavioral preferences as they change over time. This limits the flexibility and effectiveness of aggregation strategies in multi-business, multi-cycle, and multi-objective scenarios. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-dimensional adjustment capability modeling and dynamic aggregation method that considers users' electricity price perception. By integrating users' physical response characteristics and behavioral preferences under the electricity price incentive cycle, a unified multi-dimensional adjustment capability vector representation is established. Combined with a deep embedded clustering method, the dynamic identification and optimal aggregation of user group structures are realized, effectively improving the accuracy of user response identification and the adaptability of aggregation scheduling.

[0006] To achieve the above objectives, the solution of the present invention is:

[0007] A multidimensional adjustment modeling and dynamic aggregation method considering users' perception of electricity prices includes the following steps:

[0008] Step 1: Collect historical load data from multiple types of electricity users within a preset time period, clean the data, and construct a structured load dataset.

[0009] Step 2: Perform cluster analysis on load data of different types of users based on load characteristics to obtain typical load samples;

[0010] Step 3: Obtain regulation capability indicators including response capacity, response time, duration, response rate, and response energy;

[0011] Step 4: Calculate the values ​​of the aforementioned indicators for the typical load sample and perform normalization processing;

[0012] Step 5: Establish a power user response willingness index model based on multi-cycle electricity price perception, and obtain the willingness prediction model through the load dataset constructed in Step 1.

[0013] Step 6: Under the set electricity price cycle length, price and incentive strategy, perform simulation prediction based on the willingness prediction model, and normalize the results to obtain the willingness to respond.

[0014] Step 7: Combine the adjustment ability index and the willingness to respond to construct a four-dimensional vector set that reflects the user's adjustment ability, and obtain a multi-dimensional adjustment ability vector dataset.

[0015] Step 8: Based on the constructed multidimensional adjustment capability vector dataset, a deep embedded clustering algorithm is used to dynamically group the user's adjustment capabilities to obtain the optimal aggregation structure.

[0016] In step 1 above, data cleaning includes cleaning historical load data using time alignment, missing value imputation, and Z-score normalization methods, as well as anomaly detection and holiday data filtering.

[0017] The specific process of step 2 above is as follows:

[0018] Step 21: Extract the load change characteristics of users within a typical day using a sliding time window, and construct a load feature vector set for each user;

[0019] Step 22: Use hierarchical clustering to initially classify all users, forming multiple initial clusters;

[0020] Step 23: Within each initial cluster, a secondary clustering method using fuzzy C-means clustering is performed to obtain typical load samples.

[0021] In step 3 above, response capacity refers to the maximum regulating power that the user can achieve within a set time period; response time refers to the delay between receiving the control command and the user starting to respond; duration is the maximum duration for which the user maintains the regulating power; response rate is the rate of power change per unit time; and response energy is the maximum regulating energy that can be achieved.

[0022] All of the above indicators are modeled and normalized based on historical typical load data.

[0023] In step 5 above, the following model for the electricity user response willingness index based on multi-cycle electricity price perception is established:

[0024] ,

[0025] in, This represents the maximum preference value. In response to the steepness coefficient, Electricity purchase price for users,

[0026] This represents the perceived midpoint electricity price. The midpoint parameter of the benchmark electricity price. This is the periodic adjustment coefficient.

[0027] The time period of the price signal is used to reflect the user's behavior threshold adjustment in response to electricity price changes over different time periods.

[0028] In step 5 above, based on the electricity user response willingness index model, the following is introduced: For the mean, Let be the normal density function of the standard deviation.

[0029] ,

[0030] The upper and lower envelopes are defined as follows:

[0031] ,

[0032] in, The upper envelope of the interval of uncertainty in the willingness of electricity users to respond. The lower envelope of the interval of uncertainty in the willingness of electricity users to respond. The amplitude coefficient is an uncertain disturbance.

[0033] In step 6 above, the user's willingness to respond under different combinations of electricity prices and periodic incentives will be assessed. Normalization is performed.

[0034] ,

[0035] in, and These represent the maximum and minimum willingness to respond among all user samples, respectively.

[0036] This represents the normalized response preference value.

[0037] In step 7 above, the user four-dimensional vector of adjustment capability This is a multi-indicator fusion expression, where each dimension is obtained by a weighted linear combination of response capacity, response time, duration, response rate, response energy, and response willingness, as defined below.

[0038] ,

[0039] Among them, the first Dimensions The calculation formula is as follows:

[0040] ,

[0041] ,

[0042] in, Indicates user Response capacity, Indicates user Response time Indicates user Duration, Indicates user Response rate, Indicates user Response energy, Indicates user The willingness to respond.

[0043] The specific process of step 8 above is as follows:

[0044] Step 81: Construct an autoencoder network, and convert each modulation capability into a four-dimensional vector. Mapping to low-dimensional latent space embedding representation ,

[0045] ,

[0046] in, For encoder mapping functions, For network parameters;

[0047] Step 82, introduce cluster centers into the latent space. The deviation between the sample distribution and the target distribution is updated by minimizing the KL divergence.

[0048] ,

[0049] in, For the sample Belongs to cluster center The soft assignment probability, For the target distribution, The loss is for the embedded spatial clustering.

[0050] Compared with the prior art, the beneficial effects of the present invention after adopting the above scheme are as follows: The present invention takes into account users' electricity price perception and combines an adaptive dynamic clustering mechanism to achieve accurate classification and efficient aggregation of multiple types of electricity users, so as to support more intelligent, differentiated and economical flexible load management and collaborative optimization scheduling strategies in the power system.

[0051] This invention can characterize the behavioral differences of users under different incentives and electricity price cycle lengths, and has strong scalability and adaptability. It is suitable for application scenarios such as virtual power plant user grouping, load dispatching and demand response optimization. Attached Figure Description

[0052] Figure 1 This is a flowchart of the present invention;

[0053] Figure 2 This is a schematic diagram of the electricity user response willingness index model based on multi-cycle electricity price perception in this invention. Detailed Implementation

[0054] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0055] like Figure 1 As shown, the present invention provides a multi-dimensional adjustment capability modeling and dynamic aggregation method that considers users' perception of electricity prices, including steps 1 to 8.

[0056] Step 1: Collect historical load data from various types of electricity users, clean the data, and build a structured dataset.

[0057] In one embodiment of the present invention, historical electricity consumption data at 15-minute intervals are collected from various types of electricity users (including residents, industrial and commercial users, public institutions, etc.) within a preset time period (e.g., 30 days, 60 days), and recorded as a load curve of 96 points / day. The original data is cleaned and processed by time alignment, missing value imputation, and Z-score normalization methods to construct a standardized load dataset.

[0058] Furthermore, to improve the applicability and robustness of the data, the cleaning process also includes anomaly detection (such as using box plots to remove extreme points) and holiday data filtering. The final output data structure is as follows: each user corresponds to one... 3D matrix This represents the number of days for sampling.

[0059] Step 2: Extract typical daily load characteristics, and combine systematic clustering and fuzzy C-means algorithm to achieve hierarchical clustering of load, and obtain representative typical load curves as samples for subsequent modeling.

[0060] Preferably, step 2, clustering different types of user load data, includes: extracting the load change characteristics of users within a typical day through a sliding time window, and constructing a load feature vector set for each user; based on this, using a hierarchical clustering method to initially classify all users, forming multiple initial clusters; subsequently, within each initial cluster, using a fuzzy C-means clustering method for secondary clustering, to achieve refined identification of intra-cluster differences and optimized expression of boundary fuzziness, thereby constructing a multi-level, hierarchical set of representative typical load patterns.

[0061] Furthermore, a membership threshold is introduced into fuzzy C-means clustering as a basis for inter-class adjustment. Marginal samples with membership below the set threshold are reclassified or retrained to improve the clarity of cluster boundaries. Finally, one or more representative samples are extracted from each cluster as the basis for subsequent modeling.

[0062] This invention segments the user load curve daily using a sliding window, extracts the daily load feature vector, and employs a hierarchical clustering algorithm for preliminary classification, dividing the samples into N initial clusters. Subsequently, fuzzy C-means (FCM) clustering is applied to each initial cluster for refinement, capturing intra-cluster variability and obtaining a more representative set of typical load samples as reference samples for subsequent modeling.

[0063] Step 3: Construct a regulatory capability index system that includes response capacity, response time, duration, response rate, and response energy.

[0064] Preferably, the different regulation capability indicators in step 3 include: response capacity refers to the maximum regulation power that the user can achieve within a set time period; response time refers to the delay between receiving the regulation command and the user starting to respond; duration is the maximum duration for which the user maintains the regulation power; response rate is the rate of power change per unit time; and response energy is the maximum achievable regulation energy. All of the above indicators are modeled and normalized based on the typical load sample data obtained in step 2.

[0065] For a typical load sample, extract and calculate five types of physical regulation capability indicators for users: response capacity ( ) refers to the maximum achievable power regulation range; response time ( The duration refers to the delay from when the command is issued to when the user responds; the duration ( ) refers to the longest time a user maintains a response state; response rate ( ) refers to the rate of change of load per unit time; response energy ( This refers to the energy that a user can release during the adjustment period. The above indicators are calculated using formulas.

[0066] In this embodiment, the response capacity RC is measured by the maximum load change, the response time RT is determined by the slope change of the fitted start point, the response rate RS is calculated by the maximum slope of the differential curve, and the response energy HSC is obtained by integration within the load adjustment range.

[0067] Step 4: Perform Min-Max normalization on the index values ​​calculated in Step 3 to unify the dimensions.

[0068] To standardize the scale of multidimensional indicators, the following normalization formula is adopted:

[0069]

[0070] in, These are the original index values. This is the normalized result. This operation applies to... , , , , The five categories of indicators are processed separately to form a standardized indicator set.

[0071] Furthermore, the normalization parameters (maximum and minimum values) are recorded synchronously during the normalization process for subsequent model evaluation and inverse transformation processing of policy mapping.

[0072] Step 5: Based on users' consumption psychology, establish a power user response willingness index model based on multi-cycle electricity price perception, and use historical data from the load dataset in Step 1 to train and optimize parameters, thereby obtaining a willingness prediction model.

[0073] Preferably, the electricity user response willingness index model established in step 5, taking into account the individual differences and response fluctuations among actual users at the same electricity price level, further introduces... The normal density function with mean and standard deviation :

[0074]

[0075] This function reflects the probabilistic characteristic that user responses to electricity prices fluctuate significantly in the central region (i.e., near the expected electricity price) and tend to stabilize in the peripheral region. Based on this, the upper and lower envelopes are defined as follows:

[0076]

[0077] in, The upper envelope of the interval of uncertainty in the willingness of electricity users to respond. The lower envelope of the interval of uncertainty in the willingness of electricity users to respond. The coefficient represents the amplitude of the uncertain disturbance. This structure enables continuous and smooth modeling of the uncertainty of user behavior, making the electricity price perception function both differentiable and behaviorally realistic.

[0078] The model parameters are obtained through training with historical data samples, which makes the prediction results both differentiable and realistic in terms of user behavior.

[0079] In this embodiment, the model is trained using historical response records of multi-period electricity prices (such as 1 day, 1 month, 6 months, etc.), and the user behavior preference function curve is fitted using the minimum mean square error loss function to obtain the final behavior response willingness prediction model.

[0080] Step 6: Combine electricity prices and incentive strategies to predict and normalize the user's willingness to respond.

[0081] Preferably, in step 6, the user's willingness to respond under different combinations of electricity prices and periodic incentives is considered. Normalization is performed, and the following standardized expression is adopted:

[0082]

[0083] in, and These represent the maximum and minimum willingness to respond among all user samples, respectively.

[0084] This represents the normalized response preference value.

[0085] Step 7: Integrate physical response and intentional behavior to construct a four-dimensional vector expression that reflects the user's adjustment ability, thereby realizing a multi-dimensional characterization of adjustment ability for multiple types of users.

[0086] Preferably, the user constructed in step 7 four-dimensional vector of adjustment capability This is a multi-indicator fusion expression, where each dimension is obtained by a weighted linear combination of six key user indicators—response capacity, response time, duration, response rate, response energy, and response willingness—defined as follows:

[0087]

[0088] Among them, the first Dimensions The calculation formula is:

[0089]

[0090]

[0091] in, :user The response capacity, i.e., the maximum adjustable power per unit time;

[0092] :user The response time is the delay from receiving the instruction to the start of the response.

[0093] :user The duration refers to the longest possible duration of the regulated state;

[0094] :user The response rate, i.e., the power change per unit time;

[0095] :user The response energy represents the available energy potential during load regulation.

[0096] :user The willingness to respond, that is, the psychological response level under the current electricity price / incentive scheme.

[0097] The weighting coefficients It can be flexibly set according to the clustering task objectives (such as emphasizing rapid response, stable output, etc.), and supports linear adjustment or adaptive learning based on genetic algorithms and entropy weight methods.

[0098] Step 8: Use the Deep Embedded Clustering (DEC) algorithm to dynamically cluster users, and combine it with visualization analysis methods to judge the rationality of the clustering structure, so as to realize adaptive identification and aggregation of response capabilities.

[0099] Preferably, in step 8, based on a four-dimensional vector dataset of adjustment capabilities of multiple user types, a deep embedded clustering algorithm (DEC) is used for dynamic clustering analysis. The DEC method introduces a deep autoencoder to learn the embedded representation of the original four-dimensional vectors, while simultaneously optimizing the clustering assignment objective function, thereby achieving feature compression and adaptive updating of the clustering structure. The process includes the following two stages:

[0100] (1) Pre-training stage: Construct an autoencoder network and combine each adjustment capability four-dimensional vector Mapping to low-dimensional latent space embedding representation :

[0101]

[0102] in For encoder mapping functions, For network parameters;

[0103] (2) Clustering optimization stage: Introducing cluster centers into the latent space. The deviation between the sample distribution and the target distribution is updated by minimizing the KL divergence:

[0104]

[0105] in, For the sample Belongs to cluster center The soft assignment probability, based on the definition of the student distribution,

[0106] The target distribution is used to enhance cluster discriminative power.

[0107] The loss is for the embedded spatial clustering.

[0108] In this invention, a periodic retraining or incremental training mechanism is used to support continuous learning and cluster structure updates for new user access or changes in user behavior, which has good adaptability and scalability.

[0109] Figure 2 This is a schematic diagram of the electricity user response willingness index model based on multi-cycle electricity price perception in the multi-dimensional adjustment capability modeling and dynamic aggregation method that considers user electricity price perception of the present invention.

[0110] Consumer psychology research shows that individuals are influenced not only by current benefits but also significantly by the stability and predictability of expected benefits during the decision-making process. In the context of electricity demand response, users tend to prefer sustainable, long-term, and controllable savings incentives, while exhibiting strong hesitation and behavioral inertia towards short-term, volatile price signals. This inertia stems from users' subjective resistance to the "switching costs" involved in adjusting their electricity consumption behavior, such as changes in plans, disruption of lifestyles, operational costs, and information processing burdens.

[0111] Therefore, when faced with short-term electricity price fluctuations (such as hourly dynamic pricing), it is difficult to effectively trigger user responses unless the price change reaches a certain threshold. In other words, the price response curve under short-term conditions has a higher behavioral initiation threshold, and users are more inclined to maintain their original load patterns, only exhibiting immediate response behavior when the price signal is sufficiently significant and the incentive is adequate.

[0112] In contrast, long-term electricity price adjustments (such as weekly or monthly electricity packages) are more likely to encourage users to think proactively and adjust their pricing based on stable expectations of future returns. Users will weigh their energy consumption patterns against the benefits of energy savings in their long-term planning, demonstrating a stronger "planned load reconfiguration capability." At this point, users are more sensitive to price changes, and the electricity price threshold required to trigger a response is significantly lower.

[0113] From a behavioral modeling perspective, users' response thresholds show a decreasing trend with increasing cycle length. This psychological pattern provides theoretical support for electricity price design and willingness modeling: a "cycle adjustment factor" should be introduced into the response willingness function to reflect the dynamic changes in user behavioral thresholds at different time scales. Based on this, this invention introduces a cycle weight parameter in electricity price perception response modeling to adjust the midpoint electricity price in the response, thereby accurately characterizing users' response tendencies when faced with different cycle incentives.

[0114] like Figure 2 As shown, the established electricity user response willingness index model based on multi-period electricity price perception shows that, overall, user electricity consumption preference is negatively correlated with electricity price, and there is a certain degree of uncertainty. As the electricity price period lengthens, the user's electricity consumption preference-electricity price curve gradually shifts to the right. Based on the above theory, an electricity price perception function based on the Sigmoid structure is constructed, and a normal perturbation term is introduced into it to form the uncertainty interval of response behavior.

[0115] In summary, the multi-dimensional adjustment capability modeling and dynamic aggregation method of this invention, which considers user electricity price perception, can achieve multi-dimensional quantitative modeling of the adjustment capabilities of various types of users. It also combines users' willingness to respond to electricity prices and incentive cycles for unified expression. Through deep clustering, it achieves dynamic identification and accurate grouping of user aggregation structures, thereby improving the accuracy of user response modeling and the adaptability of aggregation scheduling. This provides more reliable and efficient technical support for flexible load management and demand response optimization in power systems.

[0116] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0117] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0118] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0119] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0120] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0121] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A multi-dimensional adjustment modeling and dynamic aggregation method considering user electricity price perception, characterized in that... Includes the following steps: Step 1: Collect historical load data from multiple types of electricity users within a preset time period, clean the data, and construct a structured load dataset. Step 2: Perform cluster analysis on load data of different types of users based on load characteristics to obtain typical load samples; Step 3: Obtain regulation capability indicators including response capacity, response time, duration, response rate, and response energy; Step 4: Calculate the values ​​of the aforementioned indicators for the typical load sample and perform normalization processing; Step 5: Establish a power user response willingness index model based on multi-cycle electricity price perception, and obtain the willingness prediction model by training the load dataset constructed in Step 1. Step 6: Under the set electricity price cycle length, price and incentive strategy, perform simulation prediction based on the willingness prediction model, and normalize the results to obtain the willingness to respond. Step 7: Combine the adjustment ability index and the willingness to respond to construct a four-dimensional vector set that reflects the user's adjustment ability, and obtain a multi-dimensional adjustment ability vector dataset. Step 8: Based on the constructed multidimensional adjustment capability vector dataset, a deep embedded clustering algorithm is used to dynamically group the user's adjustment capabilities to obtain the optimal aggregation structure.

2. The method as described in claim 1, characterized in that: In step 1, data cleaning includes cleaning historical load data by time alignment, missing value imputation and Z-score normalization, as well as anomaly detection and holiday data filtering.

3. The method as described in claim 1, characterized in that: The specific process of step 2 is as follows: Step 21: Extract the load change characteristics of users within a typical day using a sliding time window, and construct a load feature vector set for each user; Step 22: Use hierarchical clustering to initially classify all users, forming multiple initial clusters; Step 23: Within each initial cluster, a secondary clustering method using fuzzy C-means clustering is performed to obtain typical load samples.

4. The method as described in claim 1, characterized in that: In step 3, response capacity refers to the maximum adjustment power that the user can achieve within a set time period; response time refers to the delay between receiving the control command and the user starting to respond; and duration is the maximum duration for which the user maintains the adjustment power. The response rate is the rate of power change per unit time; the response energy is the maximum achievable regulation energy. All of the above indicators are modeled and normalized based on historical typical load data.

5. The method as described in claim 1, characterized in that: In step 5, the following electricity user response willingness index model based on multi-cycle electricity price perception is established. ， in, This represents the maximum preference value. In response to the steepness coefficient, Electricity purchase price for users, This represents the perceived midpoint electricity price. The midpoint parameter of the benchmark electricity price. This is the periodic adjustment coefficient. The time period of the price signal is used to reflect the user's behavior threshold adjustment in response to electricity price changes over different time periods.

6. The method as described in claim 5, characterized in that: In step 5, based on the electricity user response willingness index model, the following is introduced: For the mean, Let be the normal density function of the standard deviation. ， The upper and lower envelopes are defined as follows: ， in, The upper envelope of the interval of uncertainty in the willingness of electricity users to respond. The lower envelope of the interval of uncertainty in the willingness of electricity users to respond. The amplitude coefficient is an uncertain disturbance.

7. The method as described in claim 1, characterized in that: In step 6, the user's willingness to respond under different combinations of electricity prices and periodic incentives will be assessed. Normalization is performed. ， in, and These represent the maximum and minimum willingness to respond among all user samples, respectively. This represents the normalized response preference value.

8. The method as described in claim 1, characterized in that: In step 7, the user four-dimensional vector of adjustment capability This is a multi-indicator fusion expression, where each dimension is obtained by a weighted linear combination of response capacity, response time, duration, response rate, response energy, and response willingness, as defined below. ， Among them, the first Dimensions The calculation formula is as follows: ， ， in, Indicates user Response capacity, Indicates user Response time Indicates user Duration, Indicates user Response rate, Indicates user Response energy, Indicates user The willingness to respond.

9. The method as described in claim 1, characterized in that: The specific process of step 8 is as follows: Step 81: Construct an autoencoder network, and convert each modulation capability into a four-dimensional vector. Mapping to low-dimensional latent space embedding representation , ， in, For encoder mapping functions, For network parameters; Step 82, introduce cluster centers into the latent space. The deviation between the sample distribution and the target distribution is updated by minimizing the KL divergence. ， in, For the sample Belongs to cluster center The soft assignment probability, For the target distribution, The loss is for the embedded spatial clustering.

Images