A Differentiated Load Guideline Generation Method Based on Load Response Characteristics

Abstract

This invention discloses a differentiated load profile generation method based on load response characteristic clustering, relating to the field of power system operation and dispatching technology. It addresses the problem of inaccurate load profile generation in existing methods. The method includes the following steps: acquiring historical load data and extracting electricity consumption patterns from the historical load data to obtain electricity consumption pattern curves; acquiring response characteristic parameters; decomposing the daily load curves of each node using a non-negative matrix factorization method based on the electricity consumption pattern curves to obtain the load composition coefficients of each node; constructing multi-dimensional feature vectors and clustering the nodes based on these feature vectors to divide them into multiple node groups; and for each node group, constructing a load profile optimization model and generating a load profile curve adapted to that node group by solving the model. This invention obtains accurate load profiles by constructing a complete process scheme for load characteristic analysis, user clustering, and profile generation.

Description

Technical Field

[0001] This invention relates to the field of power system operation and dispatching technology, and in particular to a method for generating differentiated load baselines based on load response characteristic grouping. Background Technology

[0002] With the increasing proportion of renewable energy integration and the growing complexity of power load structure, power system operation faces challenges such as insufficient flexibility in source-load interaction and widening peak-valley differences. Demand Response (DR), as an important load-side management technology, guides electricity consumption behavior to optimize and adjust over time by publishing load benchmarks, thus helping to improve grid regulation capabilities and renewable energy absorption levels. Load benchmarks refer to the baseline curves published by grid operators to guide load-side adjustments during specific time periods; the scientific validity and applicability of their formulation methods directly affect the effectiveness of demand response implementation.

[0003] In existing technologies, load guidelines are often established based on cluster analysis of the morphological similarity of historical load curves of nodes, thereby setting a unified guideline for nodes of the same type. For example, the published paper "Load Pattern Extraction and Cluster Analysis Based on Daily Load Curves" proposes a method that includes: extracting typical electricity consumption patterns from load curves through fuzzy C-means clustering; decomposing the load of each node into a linear combination of these patterns using the least squares method; clustering the nodes based on the coefficients obtained from the decomposition, and establishing a guideline accordingly.

[0004] However, such existing methods still have the following significant shortcomings in terms of technical implementation: The load decomposition technique has limitations: when using least squares decomposition, the load composition coefficient often becomes negative, which contradicts the actual physical meaning and affects the accuracy of subsequent node classification and guideline determination.

[0005] Clustering criteria fail to fully reflect the adjustment potential of nodes: Existing methods mostly rely on load patterns or composition coefficients for clustering, without integrating objective and quantifiable adjustment characteristics such as the adjustable load ratio of nodes and historical response performance, resulting in weak applicability of classification results in actual regulation.

[0006] The model has low computational efficiency and is difficult to scale: Existing methods usually adopt a node-by-node modeling and optimization approach, which has high computational complexity and poor scalability in large-scale node scenarios, and is not suitable for the implementation needs of real-time or near-real-time demand response systems.

[0007] Therefore, it is necessary to improve the load profile formulation method from multiple technical aspects, such as load decomposition methods, feature construction mechanisms, clustering techniques and optimization models, so as to improve the engineering practicality of demand response technology. Summary of the Invention

[0008] To overcome the shortcomings of the prior art, one of the objectives of this invention is to provide a differentiated load profile generation method based on load response characteristic grouping, which uses a non-negative matrix decomposition method to decompose the load curve, thereby obtaining a load composition vector with physical meaning.

[0009] One of the objectives of this invention is achieved through the following technical solution: A method for generating differentiated load profiles based on load response characteristic clustering includes the following steps: Historical load data of multiple load nodes are acquired, and the power consumption patterns of the historical load data are extracted to obtain typical power consumption pattern curves for multiple types of loads. Obtain the response characteristic parameters of each load node, including the adjustable load ratio and the response willingness coefficient; Based on the typical electricity consumption pattern curve, the daily load curve of each node is decomposed using the non-negative matrix decomposition method to obtain the load composition coefficient of each node. By integrating the load composition coefficient, adjustable load ratio, and response willingness coefficient of each node, a multi-dimensional feature vector is constructed, and the nodes are clustered based on the multi-dimensional feature vector to divide them into multiple node groups. For each node group, a load guideline optimization model based on the comprehensive response characteristics of the node group is constructed, and a load guideline curve adapted to the node group is generated by solving the model.

[0010] Furthermore, historical load data from multiple load nodes is acquired, and electricity consumption patterns are extracted from the historical load data to obtain typical electricity consumption pattern curves for various types of loads, including: The daily load curves for industrial, commercial, and residential loads were normalized respectively. Each load type is clustered using the fuzzy C-means clustering algorithm, and representative electricity consumption pattern curves are extracted.

[0011] By processing the three types of loads separately, the mutual interference of different power consumption characteristic data is avoided, making the extracted typical power consumption patterns more representative of the real power consumption patterns of the loads; and providing high-quality basis vectors for subsequent decomposition.

[0012] To improve clustering quality, fuzzy C-means clustering algorithm was used to cluster each load class, and representative electricity consumption pattern curves were extracted, including: Calculate the clustering effectiveness index under different numbers of clusters, and select the number of clusters that maximizes the effectiveness index as the optimal number of clusters; A fuzzy C-means clustering algorithm is used to cluster load curves of different categories based on the optimal number of clusters. During clustering, the distance between sample points and cluster centers is calculated using weighted Euclidean distance, where the weights are determined based on the variability of load at different time points among different samples. The weighted Euclidean distance is calculated using the following formula: , in, For the first The sample and the first The weighted Euclidean distance between cluster centers For time point The weight, Cluster center At time The value, Indicates the first Each sample at time point Average normalized load, This represents the total number of time points on the daily load curve. Each cluster center obtained after clustering is used as a typical power consumption pattern curve for that type of load.

[0013] The optimal number of clusters is automatically determined by the clustering effectiveness index, avoiding the subjectivity and inaccuracy caused by manual setting, and ensuring that the extracted pattern curves have high representativeness and discriminative power.

[0014] Furthermore, to provide a quantitative basis for subsequent precise clustering and differentiation benchmark formulation, the adjustable load ratio is determined based on the ratio of the rated power of adjustable equipment within a node to the total rated power of all equipment; its calculation satisfies: , in, Represents a node The percentage of users whose load can be adjusted in demand response scenarios. For the first The node in the node The rated power of the adjustable equipment For nodes The number of adjustable devices, For the first The node in the node Rated power of all types of equipment For nodes The total number of all devices; The response willingness coefficient is determined based on the ratio of the number of successful responses to historical demand response events of a node to the total number of excitation signals; its calculation satisfies: ,in, Represents a node The degree to which the user group accepts and actively participates in demand response incentive signals. For the first The number of successful responses from each node in history. For the first The total number of excitation signals received by each node.

[0015] To improve the stability of load composition, NMF (Non-negative Matrix Factorization) is used instead of least squares. Based on the typical electricity consumption pattern curve, the daily load curve of each node is decomposed using the non-negative matrix factorization method to obtain the load composition coefficient of each node, including: The extracted typical electricity consumption pattern curves are used to construct a basis matrix; Using the aforementioned base matrix as a reference, nonnegative matrix decomposition is performed on multiple daily load curves for each load node to obtain multiple corresponding load composition coefficient vectors, with the following constraints satisfied: , , in, This represents the load composition coefficient matrix. Representation matrix The elements in Represents a node. An index representing typical electricity consumption patterns. Represents the load matrix. Represents the basis matrix. Denotes the Frobenius norm; The average value of the multiple load composition coefficient vectors is taken as the load composition coefficient of the node.

[0016] Furthermore, by integrating the load composition coefficient, adjustable load ratio, and response willingness coefficient of each node, a multi-dimensional feature vector is constructed. Based on this multi-dimensional feature vector, the nodes are clustered into multiple node groups, including: The load composition coefficient, adjustable load ratio, and response willingness coefficient of each node are used as vector elements, and after normalization, they are concatenated to form the multi-dimensional feature vector. The normalization process adopts the min-max normalization method to map each feature value to the interval [0,1]. Fuzzy C-means clustering algorithm is used, with Euclidean distance as the distance metric between sample points and cluster centers, to cluster nodes and obtain multiple node groups; For each group of nodes that has been divided, calculate the comprehensive characteristic parameters of that group.

[0017] Minimum-maximum normalization maps features of different dimensions to the same scale, balancing the contribution of all features to the clustering results and providing a precise classification basis for setting the guideline.

[0018] To achieve feature aggregation from individuals to groups, the comprehensive characteristic parameters include the average load composition vector of the node group, the comprehensive adjustable ratio of the node group, and the comprehensive response willingness coefficient of the node group, the calculation of which satisfies: , , , in, This represents the vector of average load composition of the node group. This indicates the overall adjustable ratio of the node group. This represents the overall response willingness coefficient of the node group. For the first The total number of nodes contained in a node group. Represents a node The degree to which the user group accepts and actively participates in demand response incentive signals. Represents a node The percentage of users whose load can be adjusted in demand response scenarios. Indicates the first The average composition coefficient of each node.

[0019] To achieve a balanced optimization of the interests of all parties, the objective function of the load guideline optimization model is to minimize the sum of system scheduling cost and user adjustment cost, and its calculation satisfies: ,in, Indicates system scheduling cost, This indicates the user's adjustment cost.

[0020] To ensure the calculation of the baseline curve, the system scheduling cost is calculated using the load adjustment unit value coefficient, and the user adjustment cost is calculated using the penalty coefficient. The calculation of the system scheduling cost and the user adjustment cost satisfies the following: , , , in, This represents the system-level unified weighting coefficient. Indicates the penalty coefficient. This represents the unit value coefficient for load adjustment. This indicates the overall adjustable ratio of the node group. This represents the overall response willingness coefficient of the node group. This indicates the expected electricity consumption behavior of a cluster of nodes when they are not stimulated by a demand response signal. Represents a group of nodes The load guideline.

[0021] Furthermore, the constraints of the load baseline optimization model include: the offset of the load value of the baseline at each time point relative to the baseline load curve does not exceed the maximum adjustable range of the node group at the corresponding time point; wherein, the maximum adjustable range is determined based on the comprehensive adjustable load ratio of the node group and the baseline load value, and the baseline load curve is constructed based on the average load composition coefficient and typical electricity consumption pattern curve of the node group.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows: The non-negative matrix factorization method employed in this invention mathematically guarantees that the load composition coefficients obtained from the decomposition are always positive. This ensures that the decomposition results accurately reflect the actual contribution ratio of different power consumption patterns to the total load, providing more accurate and physically meaningful input features for subsequent node group partitioning. Furthermore, this invention integrates load composition characteristics, physical adjustability potential, and behavioral response reliability into a multi-dimensional feature vector, providing the most accurate classification basis for subsequent guideline setting and greatly improving the targeting and effectiveness of guideline setting. This invention formulates guidelines through node groups, ensuring a high degree of matching between the guidelines and the group's adjustability while significantly reducing computational complexity in large-scale scenarios. This invention proposes a complete solution covering load characteristic analysis, user grouping, and guideline generation, solving the problem of low matching degree between guidelines and actual needs in existing technologies. Attached Figure Description

[0023] Figure 1 This is a flowchart of the differentiated load guideline generation method based on load response characteristic grouping in Embodiment 1; Figure 2 This is a flowchart of the FCM algorithm in Example 1. Detailed Implementation

[0024] The present invention will now be described in more detail with reference to the accompanying drawings. It should be noted that the following description of the present invention with reference to the accompanying drawings is merely illustrative and not restrictive. Various embodiments can be combined with each other to form other embodiments not shown in the following description.

[0025] Example 1 Example 1 provides a method for generating differentiated load baselines based on load response characteristic clustering, aiming to address the problem of low matching degree between existing baselines and users' actual adjustment capabilities. First, typical electricity consumption patterns are extracted from historical load data through weighted fuzzy clustering, and the adjustable load ratio and response willingness of each node are quantified. Second, non-negative matrix factorization is used to obtain physically meaningful load composition coefficients. Then, load composition, adjustable ratio, and response willingness are fused to construct a multi-dimensional feature vector, clustering nodes into groups with similar response characteristics. Finally, an optimization model is constructed for each group, and a differentiated load baseline matching the adjustable capability of that group is automatically generated by setting a penalty coefficient inversely proportional to the group's response capability. This achieves precision, differentiation, and scalability in baseline setting.

[0026] The existing technical solutions mainly include the following three steps: Step 1: Electricity Consumption Pattern Extraction: For the daily load curves of industrial, commercial, and residential electrical equipment, the fuzzy C-means clustering algorithm (FCM) is used to extract representative electricity consumption pattern curves.

[0027] Step 2: Load characteristic decomposition steps: Using the electricity consumption pattern curve extracted in step (1) as the basis vector, the least squares algorithm is used to decompose the load for the daily load curves of different load nodes or areas.

[0028] , in, To predict the obtained daily load curve, This is the power consumption mode curve. The load composition factor represents different electricity consumption patterns. The decomposition objective is to minimize the deviation between the predicted load curve and the actual load curve.

[0029] Step 3: Clustering of Nodes or Regions: Using the load composition coefficients obtained from Step 2, FCM is used to perform clustering analysis of load nodes or regions.

[0030] Existing technologies are mainly used for load pattern recognition or load model updating. Their decomposition and clustering processes employ standard FCM and least squares methods, failing to distinguish the regulatory value of different time periods and unable to avoid the appearance of physically meaningless negative coefficients in least squares decomposition. Therefore, while existing methods can describe load patterns, they are insufficient to support fine-grained modeling of adjustable potential and response capabilities in demand response scenarios. This invention improves upon key steps: firstly, it introduces weighted Euclidean distance, enabling the clustering process to highlight load differences during key periods such as peak and trough times, better aligning with the regulatory requirements of demand response; secondly, it replaces least squares decomposition with nonnegative matrix factorization (NMF), ensuring that the load composition coefficients satisfy nonnegativity constraints, thereby obtaining physically meaningful load composition vectors and providing a reliable basis for subsequent node clustering based on response characteristics.

[0031] Based on the above principles, this embodiment identifies typical electricity consumption patterns, decomposes the load composition of nodes, and quantifies users' response capabilities to construct a node group with similar adjustment characteristics. On this basis, targeted load guidelines are formulated. The method of this embodiment can improve the matching degree between the guidelines and the actual adjustable capabilities of the nodes, achieving more refined and effective demand response management.

[0032] Specifically, the method in this embodiment employs nonnegative matrix factorization (NMF) to obtain load composition coefficients that satisfy nonnegativity constraints and have physical meaning, used to accurately characterize the load composition of nodes. A multi-feature vector is constructed, comprising load composition coefficients, adjustable load proportions, and response willingness coefficients. Weighted Euclidean distance is used to emphasize differences during critical periods, achieving node group partitioning that simultaneously reflects structural characteristics and adjustment capabilities. Based on the load composition and response capabilities of the groups, a group-level guideline optimization model is constructed. Guidelines are defined on a group-by-group basis, reducing the computational cost of node-by-node optimization while ensuring consistency between the guidelines and the group's adjustability.

[0033] Based on the above explanation, please refer to Figure 1 As shown, a method for generating differentiated load profiles based on load response characteristic clustering includes the following steps: S1. Obtain historical load data from multiple load nodes, and extract the power consumption patterns from the historical load data to obtain typical power consumption pattern curves for multiple types of loads; For industrial, commercial, and residential loads, their equipment combinations and user behaviors differ significantly. Industrial loads are mainly affected by production plans, commercial loads by business hours and environmental controls, while residential loads show a clear dependence on lifestyle habits. To avoid interference from heterogeneous data, highlight shape differences, and focus clustering on key time periods, this embodiment first performs normalization and weighted fuzzy clustering on the collected daily load curves of the three types. Specifically, in S1, historical load data of multiple load nodes is acquired, and electricity consumption patterns are extracted from the historical load data to obtain typical electricity consumption pattern curves for multiple load types, including: The daily load curves for industrial, commercial, and residential loads were normalized respectively. All samples are normalized to the [0,1] interval using min-max normalization, specifically as follows: ,in, The normalized load value, i.e., the first... The sample at the th New values ​​at each point in time; In the original daily load curve, the first The sample at the th Load values ​​at specific points in time; The minimum value of all samples at all time points; It represents the maximum value of all samples at all time points.

[0034] For the three types of loads after normalization, this embodiment uses the fuzzy C-means (FCM) clustering algorithm to extract the clustering data respectively. A representative typical electricity consumption pattern curve The fuzzy membership mechanism of FCM can more accurately represent situations where the same user may have multiple electricity consumption habits simultaneously, effectively handling the fuzziness in load response characteristics. Each load class is clustered using the fuzzy C-means clustering algorithm, and representative electricity consumption pattern curves are extracted. The FCM algorithm uses the value function shown in the following optimization formula. Obtain the membership degree of each sample point to all class centers: , , in, This represents the total number of cluster centers (typical patterns); The total number of daily load curve samples input; For weighted index (usually) =2); For the first The sample belongs to the first The membership degree of each cluster center; For the first The sample and the first The weighted Euclidean distance between cluster centers.

[0035] Specifically, fuzzy C-means clustering algorithm is used to cluster each load type and extract representative electricity consumption pattern curves, including: Calculate the clustering effectiveness index under different numbers of clusters, and select the number of clusters that maximizes the effectiveness index as the optimal number of clusters; A fuzzy C-means clustering algorithm is used to cluster load curves of different categories based on the optimal number of clusters. During clustering, the distance between sample points and cluster centers is calculated using weighted Euclidean distance, where the weights are determined based on the variability of load at different time points among different samples. The weighted Euclidean distance is calculated using the following formula: , in, For the first The sample and the first The weighted Euclidean distance between cluster centers For time point The weight, Cluster center At time The value, Indicates the first Each sample at time point Average normalized load, This represents the total number of time points on the daily load curve. This embodiment will use weights The value is set to be related to the cross-sample volatility (standard deviation) of the load during that period. They show a positive correlation, which can be calculated. The proportion of total variability at all points in time was used to implement the weighting. The adaptive allocation is defined as follows: , , , in, for The average normalized load at any given time. With this setting, The larger the value, the more significant the difference in load among different samples during that period, and the higher its corresponding weight. The larger the value, the more automatically the clustering process will focus on critical periods with drastic load changes and greater sensitivity to pattern differentiation.

[0036] Each cluster center obtained after clustering is used as a typical power consumption pattern curve for that type of load.

[0037] Please refer to Figure 2 The calculation process shown below, specifically the iterative steps of the FCM algorithm, are as follows: Initialize the membership matrix U: Randomly generate U, satisfying: , , .

[0038] Calculate cluster centers : .

[0039] Update the membership matrix U: Update the membership based on the new cluster centers to obtain the corresponding membership degrees. ,in, This represents the summation index when calculating membership degrees across all cluster centers, ..., with values ​​of 1, 2, ... .

[0040] Iteration Termination: The iteration termination condition of this invention is as follows: The change is less than the preset threshold of 10 -5 , The change is defined as the value function of the current iteration step. Compared with the previous step The absolute value of the difference, i.e. , If the condition is met, the iteration stops.

[0041] After the iteration is complete, the output yields the respective loads for industrial, commercial, and residential categories. Typical electricity consumption pattern curve To determine the optimal number of clusters. This invention uses a clustering effectiveness index. : , The number of clusters is represented as The minimum maximum membership degree for each sample. In practical applications, this is determined through testing. Increase from 1 to The result, choose The maximum number of clusters is determined to ensure that the extracted pattern curves have the highest representativeness and pattern discrimination.

[0042] S2. Obtain the response characteristic parameters of each load node, including the adjustable load ratio and the response willingness coefficient; When setting the node load baseline, S2 considers not only the load composition characteristics of the nodes, but more importantly, quantifies the node's demand response capability. This embodiment introduces two core parameters to quantify this capability: adjustable load ratio. and response willingness coefficient .

[0043] Adjustable load ratio characterizes the nodes The proportion of user load that can be adjusted in demand response scenarios reflects the physical adjustability potential of the node. This is calculated by statistically analyzing the rated power of adjustable equipment for different types of users within the node. , in, Represents a node The percentage of users whose load can be adjusted in demand response scenarios. For the first The node in the node The rated power of adjustable equipment (such as air conditioners, energy storage, etc.). For nodes The number of adjustable devices, For the first The node in the node Rated power of all types of equipment (including adjustable and non-adjustable). For nodes The total number of all devices.

[0044] The response willingness coefficient characterizes the nodes The degree to which the user group accepts and actively participates in demand response incentive signals reflects the reliability of the node's behavioral response. This is determined based on the node's historical demand response event records. ,in, Represents a node The degree to which the user group accepts and actively participates in demand response incentive signals. For the first The number of successful responses from each node in history. For the first The total number of excitation signals received by each node. For users newly connected to the system or those who have not received an excitation signal within the statistical period (i.e., ...). (For cases where the response rate is 0), a default initial response willingness coefficient can be set (e.g., the average level of this type of user). As subsequent participation records accumulate, this coefficient will be automatically updated to the actual statistical value.

[0045] S3. Based on the typical electricity consumption pattern curve, the daily load curve of each node is decomposed using the non-negative matrix decomposition method to obtain the load composition coefficient of each node. Existing technologies often use the least squares method to perform the decomposition step in S3, which can lead to negative coefficients that lack physical meaning. Therefore, this embodiment uses the non-negative matrix factorization (NMF) method to decompose the daily load curves of each node, ensuring that all the decomposed "load composition coefficients" are non-negative values, thus giving them a clear physical meaning, that is, representing the proportion of a certain typical electricity consumption pattern that the user includes.

[0046] S3 specifically includes: The extracted typical electricity consumption pattern curves are used to construct a basis matrix; Typical power consumption pattern curves obtained from S1 and S2 Using the basis vectors, form the basis matrix. : .

[0047] For any daily load curve It can be seen as A linear combination of typical electricity consumption patterns, load matrix It is from all load nodes 1 to Daily load curve The matrix formed : .

[0048] This embodiment employs the Non-negative Matrix Factorization (NFF) algorithm for load decomposition. The core advantage of NMF lies in its introduced non-negativity constraint, ensuring that the resulting load composition coefficients are... All values ​​are non-negative, thus ensuring the physical interpretability and practical significance of the decomposition results. The goal of NMF is to solve for the load composition coefficient matrix. , making the matrix With matrix The error between them is the smallest, of which each line That is, a node The load constitutes the coefficient vector.

[0049] Using the aforementioned base matrix as a reference, nonnegative matrix decomposition is performed on multiple daily load curves for each load node to obtain multiple corresponding load composition coefficient vectors, with the following constraints satisfied: , , in, This represents the load composition coefficient matrix. Representation matrix The elements in the text represent nodes. In the load curve, the first The composition coefficient of a typical electricity consumption pattern Represents a node. An index representing typical electricity consumption patterns. Represents the load matrix. Represents the basis matrix. Denotes the Frobenius norm; The average value of the multiple load composition coefficient vectors is taken as the load composition coefficient of the node.

[0050] Specifically, the NMF model described above is solved using an iterative algorithm to obtain the results for each load node. Load composition coefficient vector on a specific day , This represents a specific day (day index) from multiple historically collected daily load curves. Because load composition coefficients fluctuate with seasons, weather, etc., the decomposition results of a single day cannot serve as a stable representation of the nodal load composition. This invention addresses the issue of nodes... Multiple daily load curves collected historically ( (Data from the previous day), and perform NMF decomposition on each of them to obtain... Coefficient vectors Then, regarding this The mean vector is obtained by averaging the elements of each vector. ,Right now: .

[0051] S4. Integrate the load composition coefficient, adjustable load ratio and response willingness coefficient of each node to construct a multi-dimensional feature vector, and cluster the nodes based on the multi-dimensional feature vector to divide them into multiple node groups. S4 constructs multiple feature vectors to combine the inherent characteristics of the load (load composition) with the regulation potential (response characteristics), and performs a second FCM fine clustering of load nodes, providing a highly matched node group division for subsequent targeted baseline setting.

[0052] After clustering is completed, for each node group Calculate its comprehensive characteristics, including the average load composition vector of the node group. Node group integrated adjustable ratio Node group comprehensive response willingness coefficient , which serves as the input parameter for subsequent optimization of the directrix solution.

[0053] Specifically, the comprehensive characteristic parameters include the average load composition vector of the node group, the comprehensive adjustable ratio of the node group, and the comprehensive response willingness coefficient of the node group, the calculation of which satisfies: , , , in, This represents the vector of average load composition of the node group. This indicates the overall adjustable ratio of the node group. This represents the overall response willingness coefficient of the node group. For the first The total number of nodes contained in a node group. Represents a node The degree to which the user group accepts and actively participates in demand response incentive signals. Represents a node The percentage of users whose load can be adjusted in demand response scenarios. Indicates the first The average composition coefficient of each node.

[0054] S5. For each node group, construct a load guideline optimization model based on the comprehensive response characteristics of the node group, and generate a load guideline curve that is adapted to the node group by solving the model.

[0055] After dividing the node group, this invention uses the node group as the basic unit for defining the guideline. The nodes within the node group have a high degree of consistency in terms of load composition characteristics, adjustable load ratio, and response willingness. Therefore, constructing a guideline model based on the comprehensive characteristics of the group can reduce model complexity while maintaining consistency.

[0056] For each node group Based on its average composition coefficient and typical patterns Construct a group-level baseline load curve: ,in, This indicates the expected electricity consumption behavior of the node group when it is not subject to a demand response stimulus signal. The overall adjustable load ratio of the node group. and response willingness coefficient To determine the group in Maximum adjustable amplitude constraint at any time : .

[0057] The objective of this optimization model is to minimize the sum of system scheduling cost and user adjustment cost. This represents the economic benefits gained by the power grid through load adjustment; maximizing these benefits is equivalent to minimizing their negative value; user adjustment costs. This cost quantifies the discomfort experienced by users when load adjustments deviate from the baseline. It is expressed as a squared term, reflecting the characteristic that the adjustment difficulty increases non-linearly with the deviation. Its calculation satisfies: ,in, Indicates system scheduling cost, This indicates the user's adjustment cost.

[0058] The system scheduling cost is calculated using a load adjustment unit value coefficient, and the user adjustment cost is calculated using a penalty coefficient. The calculation of the system scheduling cost and the user adjustment cost satisfies the following: , , , in, This represents a system-level unified weighting coefficient. Its function is to provide a unified benchmark for the adjustment costs of all node groups, thereby balancing user adjustment costs. and system scheduling costs Relative priority in optimization Indicates the penalty coefficient. The overall response capability of the node group This is transformed into adjustment difficulty pricing in the optimization model, and its form is based on adjustment difficulty pricing in economics. This refers to the load adjustment unit value coefficient, which is the load adjustment unit value coefficient set by the grid side. It is an external input parameter set (or issued) by the grid side or the dispatch center. This indicates the overall adjustable ratio of the node group. This represents the overall response willingness coefficient of the node group. This indicates the expected electricity consumption behavior of a cluster of nodes when they are not stimulated by a demand response signal. Represents a group of nodes The load guideline.

[0059] The above formula reflects the inverse relationship between adjustment difficulty and overall response capability. Low response group ( Small) The value is huge, and the optimizer will face extremely high adjustment penalties. To ensure that the published guideline is feasible in practice, the optimization model will automatically limit the deviation range of the guideline, keeping it within the range allowed by the true adjustability of the population. High-response population ( big) When the value is extremely small, the population possesses stronger regulatory potential and higher responsiveness, allowing the published baseline to have a larger deviation during critical periods. The optimized model can adjust the baseline within a wider adjustable range, thus publishing a baseline with greater regulatory value to fully leverage the population's demand response capability. To prevent an infinitely large penalty coefficient due to excessively low responsiveness, a minimum threshold is set. ,when At that time, Set to the preset upper limit value.

[0060] The constraints of the load baseline optimization model include: the offset of the load value of the baseline at each time point relative to the baseline load curve shall not exceed the maximum adjustable range of the node group at the corresponding time point; wherein, the maximum adjustable range is determined based on the comprehensive adjustable load ratio of the node group and the baseline load value, and the baseline load curve is constructed based on the average load composition coefficient and typical electricity consumption pattern curve of the node group. The baseline must satisfy the maximum adjustable range constraint: , By solving the above optimization model, the targeted directrix curve of the node group can be obtained. Because users within a node group exhibit high consistency in load composition, adjustability, and response behavior, the obtained... It can be directly used as the standard for unified execution of the node group, without the need for individual allocation of nodes within the group.

[0061] In summary, the non-negativity constraint employed in this embodiment mathematically guarantees that the load composition coefficients obtained from the decomposition are always positive. This improvement allows the decomposition results to accurately reflect the actual contribution ratio of different electricity consumption patterns to the total load, providing more accurate and physically meaningful input features for subsequent node group partitioning. Furthermore, this embodiment's method integrates load composition characteristics, physical adjustability potential, and behavioral response reliability into a multi-dimensional feature vector, and uses weighted Euclidean clustering for partitioning. The resulting node groups are no longer simple sets of "similar electricity consumption," but rather highly consistent categories of "similar demand response potential." This provides the most accurate classification basis for subsequent guideline setting, greatly improving the targeting and effectiveness of guideline setting. The comprehensive response capability of the node group is transformed into a differentiated penalty factor in the optimization model. By linking adjustment costs to the response difficulty of the node group, differentiated guideline setting based on response potential is achieved, enabling guideline deployment to accurately schedule high-response groups to maximize grid benefits and effectively protect low-response groups to ensure the long-term sustainability of the demand response system. Ultimately, it achieved precise, differentiated, and cost-effective load alignment settings, solving the problem of low matching degree between the alignment and actual needs in existing technologies.

[0062] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

Claims

1. A method for generating differentiated load guidelines based on load response characteristic clustering, characterized in that, Includes the following steps: Historical load data of multiple load nodes are acquired, and the power consumption patterns of the historical load data are extracted to obtain typical power consumption pattern curves for multiple types of loads. Obtain the response characteristic parameters of each load node, including the adjustable load ratio and the response willingness coefficient; Based on the typical electricity consumption pattern curve, the daily load curve of each node is decomposed using the non-negative matrix decomposition method to obtain the load composition coefficient of each node. By integrating the load composition coefficient, adjustable load ratio, and response willingness coefficient of each node, a multi-dimensional feature vector is constructed, and the nodes are clustered based on the multi-dimensional feature vector to divide them into multiple node groups. For each node group, a load guideline optimization model based on the comprehensive response characteristics of the node group is constructed, and a load guideline curve adapted to the node group is generated by solving the model.

2. The method for generating differentiated load baselines based on load response characteristic clustering as described in claim 1, characterized in that, Historical load data from multiple load nodes is acquired, and electricity consumption patterns are extracted from the historical load data to obtain typical electricity consumption pattern curves for various types of loads, including: The daily load curves for industrial, commercial, and residential loads were normalized respectively. Each load type is clustered using the fuzzy C-means clustering algorithm, and representative electricity consumption pattern curves are extracted.

3. The method for generating differentiated load baselines based on load response characteristic clustering as described in claim 2, characterized in that, Each load class was clustered using the fuzzy C-means clustering algorithm, and representative electricity consumption pattern curves were extracted, including: Calculate the clustering effectiveness index under different numbers of clusters, and select the number of clusters that maximizes the effectiveness index as the optimal number of clusters; A fuzzy C-means clustering algorithm is used to cluster load curves of different categories based on the optimal number of clusters. During clustering, the distance between sample points and cluster centers is calculated using weighted Euclidean distance, where the weights are determined based on the variability of load at different time points among different samples. The weighted Euclidean distance is calculated using the following formula: ， in, For the first The sample and the first The weighted Euclidean distance between cluster centers For time point The weight, Cluster center At time The value, Indicates the first Each sample at time point Average normalized load, This represents the total number of time points on the daily load curve. Each cluster center obtained after clustering is used as a typical power consumption pattern curve for that type of load.

4. The method for generating differentiated load baselines based on load response characteristic clustering as described in claim 1, characterized in that, The adjustable load ratio is determined based on the ratio of the rated power of the adjustable equipment within the node to the total rated power of all equipment; its calculation satisfies: ， in, Represents a node The percentage of users whose load can be adjusted in demand response scenarios. For the first The node in the node The rated power of the adjustable equipment For nodes The number of adjustable devices, For the first The node in the node Rated power of all types of equipment For nodes The total number of all devices; The response willingness coefficient is determined based on the ratio of the number of successful responses to historical demand response events of a node to the total number of excitation signals; its calculation satisfies: ,in, Represents a node The degree to which the user group accepts and actively participates in demand response incentive signals. For the first The number of successful responses from each node in history. For the first The total number of excitation signals received by each node.

5. The method for generating differentiated load baselines based on load response characteristic clustering as described in claim 1, characterized in that, Based on the typical electricity consumption pattern curve, the daily load curve of each node is decomposed using the non-negative matrix decomposition method to obtain the load composition coefficient of each node, including: The extracted typical electricity consumption pattern curves are used to construct a basis matrix; Using the aforementioned base matrix as a reference, nonnegative matrix decomposition is performed on multiple daily load curves for each load node to obtain multiple corresponding load composition coefficient vectors, with the following constraints satisfied: ， ， in, This represents the load composition coefficient matrix. Representation matrix The elements in Represents a node. An index representing typical electricity consumption patterns. Represents the load matrix. Represents the basis matrix. Denotes the Frobenius norm; The average value of the multiple load composition coefficient vectors is taken as the load composition coefficient of the node.

6. The method for generating differentiated load baselines based on load response characteristic grouping as described in claim 1, characterized in that, By integrating the load composition coefficient, adjustable load ratio, and response willingness coefficient of each node, a multi-dimensional feature vector is constructed. Based on this multi-dimensional feature vector, the nodes are clustered into multiple node groups, including: The load composition coefficient, adjustable load ratio, and response willingness coefficient of each node are used as vector elements, and after normalization, they are concatenated to form the multi-dimensional feature vector. The normalization process adopts the min-max normalization method to map each feature value to the interval [0,1]. Fuzzy C-means clustering algorithm is used, with Euclidean distance as the distance metric between sample points and cluster centers, to cluster nodes and obtain multiple node groups; For each group of nodes that has been divided, calculate the comprehensive characteristic parameters of that group.

7. The method for generating differentiated load baselines based on load response characteristic grouping as described in claim 6, characterized in that, The comprehensive characteristic parameters include the average load composition vector of the node group, the comprehensive adjustable ratio of the node group, and the comprehensive response willingness coefficient of the node group, and their calculation satisfies: ， ， ， in, This represents the vector of average load composition of the node group. This indicates the overall adjustable ratio of the node group. This represents the overall response willingness coefficient of the node group. For the first The total number of nodes contained in a node group. Represents a node The degree to which the user group accepts and actively participates in demand response incentive signals. Represents a node The percentage of users whose load can be adjusted in demand response scenarios. Indicates the first The average composition coefficient of each node.

8. The method for generating differentiated load baselines based on load response characteristic clustering as described in claim 1, characterized in that, The objective function of the load alignment optimization model is to minimize the sum of system scheduling cost and user adjustment cost, and its calculation satisfies: ,in, Indicates system scheduling cost, This indicates the user's adjustment cost.

9. The method for generating differentiated load baselines based on load response characteristic grouping as described in claim 8, characterized in that, The system scheduling cost is calculated using a load adjustment unit value coefficient, and the user adjustment cost is calculated using a penalty coefficient. The calculation of the system scheduling cost and the user adjustment cost satisfies the following: ， ， ， in, This represents the system-level unified weighting coefficient. Indicates the penalty coefficient. This represents the unit value coefficient for load adjustment. This indicates the overall adjustable ratio of the node group. This represents the overall response willingness coefficient of the node group. This indicates the expected electricity consumption behavior of a cluster of nodes when they are not stimulated by a demand response signal. Represents a group of nodes The load guideline.

10. The method for generating differentiated load baselines based on load response characteristic clustering as described in claim 1 or 8, characterized in that, The constraints of the load baseline optimization model include: the offset of the load value of the baseline at each time point relative to the baseline load curve does not exceed the maximum adjustable range of the node group at the corresponding time point; wherein, the maximum adjustable range is determined based on the comprehensive adjustable load ratio of the node group and the baseline load value, and the baseline load curve is constructed based on the average load composition coefficient and typical electricity consumption pattern curve of the node group.

Images