Electric appliance load cluster aggregation method and system for smart power grid

Abstract

The invention discloses an electric appliance load cluster aggregation method and system for a smart power grid, and relates to the technical field of power load management. Multi-working-condition data of electric appliance loads are collected and preprocessed; carrying out clustering processing on the preprocessed multi-working-condition data by adopting an improved fuzzy C-means clustering algorithm; adopting a Monte Carlo method to carry out secondary aggregation processing on the multi-working-condition data after clustering processing; constructing a multi-working-condition classification model of the electrical load, performing working condition classification on the multi-working-condition data subjected to secondary aggregation processing, identifying and classifying electrical load modes; and based on the classification result of the electric appliance load modes, aggregating the electric appliance loads of the plurality of electric appliance areas in the building and the plurality of buildings, participating in the scheduling of the smart power grid, and evaluating the adjustable potential of the electric appliance loads. Independent models are established according to different working conditions, the electrical load requirements under different environment conditions can be more accurately met, and the self-adaptability and the energy-saving effect of the system are improved.

Description

Technical Field

[0001] This invention relates to the field of power load management technology, and more specifically to a method and system for clustering and aggregating electrical loads for smart grids. Background Technology

[0002] In recent years, with rapid economic development and continuous improvement in people's living standards, electricity load has been increasing, especially electrical load. In economically developed regions such as Shanghai, Jiangsu, and Zhejiang, the peak load ratio of electrical load even exceeds 50%, and it shows an upward trend year by year. Electrical loads have become an important demand response resource for the power system. However, due to their diverse types and dispersed access, dispatch centers find it difficult to directly obtain their aggregated power and carry out dispatch control, limiting their response potential. The large-scale grid connection of renewable energy has further strained system balancing resources. The rapid development of smart grid two-way communication technology and advanced metering systems (AMIs) has provided technical support for user-side load monitoring and control. Electrical loads are characterized by strong controllability and large dispatch potential. By aggregating massive electrical loads through load aggregators (LAs), coordinating and controlling their participation in system regulation, they demonstrate enormous potential in peak shaving, valley filling, maintaining power system stability, and providing ancillary services. The key to fully tapping and utilizing this potential lies in establishing an appropriate aggregation model for electrical loads.

[0003] Existing electrical load aggregation methods have the following shortcomings: 1. Traditional aggregation methods based on statistical characteristics mainly determine the general characteristics of the load, such as average load and peak load, by statistically analyzing and organizing historical data. However, they lack adaptability to complex dynamic environments and cannot accurately reflect load changes under different operating conditions. Therefore, they may have limited effectiveness when dealing with environments with large actual changes.

[0004] 2. Typical time series analysis-based aggregation methods rely on capturing the trend of load changes over time, including periodic and seasonal patterns. They have a weak response capability to sudden events and may have errors in predicting load changes under long-term steady-state conditions.

[0005] Therefore, how to provide a method and system for clustering and aggregating electrical loads for smart grids, and how to establish independent models for different operating conditions to more accurately adapt to the electrical load demands under different environmental conditions and improve the system's adaptability and energy-saving effect are problems that urgently need to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, this invention provides a method and system for clustering and aggregating electrical loads in smart grids. To address the problem of alternating operating conditions of building electrical loads, a solution based on a multi-condition model is proposed. The core idea of ​​the multi-condition model is to cluster training data according to the characteristics of operating conditions, and then establish a control model for each operating condition. The effectiveness of this method largely depends on the accuracy and efficiency of the clustering algorithm. By improving existing clustering algorithms, different operating conditions can be identified and classified more accurately, thereby providing more precise control strategies for each operating condition. This significantly improves the adaptability and energy-saving effect of the electrical system, ensuring that the electrical system can automatically adjust its operating mode according to changes in the environment and demand. The optimized electrical system can not only reduce energy consumption more effectively, but also improve comfort and overall efficiency. In addition, it can respond more flexibly to the needs of the power system, contributing to the stability of the power grid. Therefore, by improving the clustering algorithm and combining it with a multi-condition modeling strategy, the complex challenges of building electrical load management can be effectively addressed, contributing to sustainable energy development.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for clustering and aggregating electrical loads for smart grids, comprising: Collect multi-condition data of electrical load and preprocess the multi-condition data of electrical load; An improved fuzzy C-means clustering algorithm was used to cluster the preprocessed multi-condition data. The Monte Carlo method was used to perform a secondary aggregation process on the clustered multi-condition data; A multi-condition classification model for electrical loads is constructed, and the support vector machine algorithm is used to classify the multi-condition data after secondary aggregation, thereby identifying and classifying the electrical load patterns. Based on the classification results of electrical load patterns, the electrical loads of multiple electrical areas within a building and multiple buildings are aggregated to participate in the smart grid scheduling, and the adjustability potential of the electrical loads is evaluated.

[0008] Preferably, the adjustable potential of the electrical load is assessed, including: An assessment model for the aggregated response potential of electrical appliances considering multiple factors is established to quantitatively analyze the maximum response potential that an electrical appliance cluster can achieve through direct load control during demand response events.

[0009] Preferably, the improved fuzzy C-means clustering algorithm includes: The particle swarm optimization algorithm is used to optimize the fuzzy C-means clustering algorithm.

[0010] Preferably, the fuzzy C-means clustering algorithm randomly generates cluster centers in the initial iteration, groups samples that are close together into one class by measuring the Euclidean distance between samples, updates the cluster centers by calculating the mean of each class of samples, and then enters the next iteration. The particle swarm optimization algorithm optimizes the initial cluster centers generated by the fuzzy C-means clustering algorithm.

[0011] Preferably, the fuzzy C-means clustering algorithm includes: Set preprocessed Data samples ,in It is the first Given a load vector, the objective function J FCM Represented as: ; in, This indicates the number of categories to be divided into for the data samples. ; It is its similarity classification matrix. It is a sample for Membership degree; In the formula, ; Indicates the first Sample With the The distance between the center points of the class; It is the number of features of the sample; These are weighted parameters; A sample's membership values ​​to all clusters sum to 1 if: ; Calculate samples separately Membership degree of A1 and cluster center : ; set up For all kind, =0 ; .

[0012] Preferably, the particle swarm optimization algorithm obtains the optimal solution to the problem through information sharing among individuals in the swarm; In the particle swarm optimization algorithm, each particle records the best position it finds. Update the individual best position and the global best position based on the best positions found by all particles in the entire swarm. The algorithm stops when the predetermined maximum number of iterations is reached or the convergence condition is met.

[0013] Preferably, the update formulas for the velocity and position of each particle are as follows: ; ; in, It is a particle In time speed, It is a particle In time Location, It is inertial weight. and These are learning factors, also known as acceleration, representing the acceleration coefficients for approaching the individual's optimal position and the global optimal position, respectively. and Is Random numbers between Represents particles The optimal position for an individual.

[0014] Preferably, the Monte Carlo method is used to perform a secondary aggregation process on the clustered multi-condition data, including: Samples are drawn independently from the distribution of each random variable; for each sampling sequence, the electrical load of each sub-region is calculated: By statistically analyzing all simulation results, the estimated distribution, expected value, and variance of the electrical load are obtained. The total electrical load of the region is calculated based on the sum of the electrical loads of each sub-region.

[0015] Preferably, an electrical load aggregation system for smart grids includes: The data acquisition and preprocessing module is used to acquire multi-condition data of electrical loads and preprocess the multi-condition data of electrical loads. The clustering module is used to perform clustering processing on the preprocessed multi-condition data using an improved fuzzy C-means clustering algorithm; The secondary aggregation processing module is used to perform secondary aggregation processing on the clustered multi-condition data using the Monte Carlo method. The classification module is used to build a multi-condition classification model for electrical loads. It uses the support vector machine algorithm to classify the multi-condition data after secondary aggregation, and identifies and classifies the electrical load patterns. The smart grid dispatch module is used to aggregate the electrical loads of multiple electrical areas within a building and multiple buildings based on the classification results of electrical load patterns, participate in the dispatch of the smart grid, and evaluate the adjustable potential of electrical loads.

[0016] Preferably, the adjustable potential of the electrical load is assessed, including: An assessment model for the aggregated response potential of electrical appliances considering multiple factors is established to quantitatively analyze the maximum response potential that an electrical appliance cluster can achieve through direct load control during demand response events.

[0017] Preferably, the improved fuzzy C-means clustering algorithm includes: The particle swarm optimization algorithm is used to optimize the fuzzy C-means clustering algorithm; The fuzzy C-means clustering algorithm randomly generates cluster centers in the initial iteration. By measuring the Euclidean distance between samples, samples with similar distances are grouped into one class. The cluster centers are updated by calculating the mean of each class of samples, and then the algorithm enters the next iteration. The particle swarm optimization algorithm optimizes the initial cluster centers generated by the fuzzy C-means clustering algorithm.

[0018] Preferably, the fuzzy C-means clustering algorithm includes: Set preprocessed Data samples ,in It is the first Given a load vector, the objective function J FCM Represented as: ; in, This indicates the number of categories to be divided into for the data samples. ; It is its similarity classification matrix. It is a sample for Membership degree; In the formula, ; Indicates the first Sample With the The distance between the center points of the class; It is the number of features of the sample; These are weighted parameters; A sample's membership values ​​to all clusters sum to 1 if: ; Calculate samples separately Membership degree of A1 and cluster center : ; set up For all kind, =0 ; .

[0019] Preferably, the particle swarm optimization algorithm obtains the optimal solution to the problem through information sharing among individuals in the swarm; In the particle swarm optimization algorithm, each particle records the best position it finds. Update the individual best position and the global best position based on the best positions found by all particles in the entire swarm. The algorithm stops when the predetermined maximum number of iterations is reached or the convergence condition is met.

[0020] Preferably, the update formulas for the velocity and position of each particle are as follows: ; ; in, It is a particle In time speed, It is a particle In time Location, It is inertial weight. and These are learning factors, also known as acceleration, representing the acceleration coefficients for approaching the individual's optimal position and the global optimal position, respectively. and Is Random numbers between Represents particles The optimal position for an individual.

[0021] Preferably, the Monte Carlo method is used to perform a secondary aggregation process on the clustered multi-condition data, including: Samples are drawn independently from the distribution of each random variable; for each sampling sequence, the electrical load of each sub-region is calculated: By statistically analyzing all simulation results, the estimated distribution, expected value, and variance of the electrical load are obtained. The total electrical load of the region is calculated based on the sum of the electrical loads of each sub-region.

[0022] As can be seen from the above technical solutions, compared with the prior art, this invention discloses a method and system for clustering and aggregating electrical loads for smart grids, aiming to solve the shortcomings of existing electrical load management, especially the problems of electrical loads participating in the optimal scheduling of smart grids and power load forecasting under diverse and dynamically changing environmental conditions. Through improved fuzzy C-means clustering and particle swarm optimization algorithms, the accuracy of load type identification is effectively improved. An innovative approach combining Monte Carlo simulation is proposed to perform secondary aggregation of electrical loads in different areas, thereby more accurately reflecting the overall electrical load situation of the building area. Furthermore, the application of support vector machines enables accurate differentiation and modeling of electrical load patterns under different environmental conditions (such as temperature, humidity, and human activity). Through feature analysis of building electrical load data, electrical loads under different operating conditions are classified and managed, achieving more flexible power resource management and scheduling control, which provides strong support for demand response and scheduling control of smart grids. In practical applications, the method proposed in this invention can help power operators more accurately identify and utilize the demand response potential of building electrical loads, optimize energy use strategies, and reduce grid pressure. Through in-depth analysis and processing of multi-condition data, this method not only improves the adaptability and overall efficiency of electrical systems, but also provides new ideas for the future development of smart grids. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0024] Figure 1 A schematic diagram of the aggregation framework for multi-condition air conditioning loads participating in the smart grid, provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the Fuzzy C-means clustering process based on PSO optimization provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the support vector machine principle structure provided in an embodiment of the present invention; Figure 4 A schematic diagram of aggregated power of air conditioning load provided for an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] This invention discloses a method for clustering and aggregating electrical loads for smart grids, comprising: S100: Collect multi-condition data of electrical load and preprocess the multi-condition data of electrical load; The multi-condition data of the electrical load includes environmental data, equipment data, user behavior data, and power consumption data. After preprocessing, the collected data is divided into training set and test set in an 8:2 ratio. S200. An improved fuzzy C-means clustering algorithm is used to cluster the preprocessed multi-condition data; the Fuzzy C-means clustering algorithm model is optimized based on the particle swarm optimization (PSO) algorithm to improve the identification accuracy of electrical load types. S300. The Monte Carlo method is used to perform secondary aggregation processing on the clustered multi-condition data; considering the impact of different physical conditions (such as indoor temperature, humidity, building type, etc.) on electrical load, the Monte Carlo method is used to perform secondary aggregation of electrical load in different areas; S400. Construct a multi-condition classification model for electrical loads. Use the support vector machine algorithm to classify the multi-condition data after secondary aggregation, identify and classify the electrical load patterns. For the changes in electrical load under different conditions, perform multi-condition classification modeling. Based on the characteristics of different conditions, select the support vector machine (SVM) algorithm for condition classification. S500, based on the classification results of electrical load patterns, aggregates the electrical loads of multiple electrical areas within a building and multiple buildings to participate in smart grid dispatch, and assesses the adjustability potential of the electrical loads. By aggregating multiple electrical areas within a building and multiple buildings, the electrical loads participate in smart grid dispatch; the adjustability potential of the electrical loads is assessed, and an electrical aggregation response potential assessment model considering multiple factors is established. This quantitatively analyzes the maximum demand response potential that the electrical cluster can achieve under direct load control mode for temperature control during demand response events.

[0027] Specifically, in order to comprehensively acquire multi-condition data on electrical load, the data acquisition and preprocessing in step S100 are as follows: Firstly, for smart grids, Advanced Measurement Infrastructure (AMI) is used to acquire raw data. AMI systems provide high-frequency power consumption data, typically collected at minute or shorter intervals. AMIs also support two-way communication, allowing data to be transmitted not only to the power company but also to user equipment via control commands. Various sensors are deployed inside buildings to monitor the status of the environment and equipment. Temperature and humidity sensors monitor indoor and outdoor temperature and humidity to assess the impact of the environment on electrical loads. Power consumption monitoring sensors are installed in the circuits of electrical equipment to monitor their power consumption in real time. Human activity sensors, such as infrared sensors and door / window open / close sensors, monitor the activities of people inside the building to assess their impact on electrical loads. Solar radiation sensors measure the intensity of external solar radiation, which has a significant impact on electrical loads, especially in summer. Equipment status sensors monitor the on / off status, operating mode (e.g., cooling, heating, ventilation), and set temperature of electrical equipment. Data is categorized based on its source, primarily into environmental data, equipment data, user behavior data, and power consumption data. Different collection frequencies are recommended for different data types. For example, environmental and personnel activity data can be collected at 5-minute or 1-minute intervals, while equipment status data can be collected at higher frequencies (e.g., at the second level) to capture dynamic changes in electrical load. All collected data should be stored in an efficient and secure database (such as an SQL or NoSQL database) and indexed using timestamps. Data integrity and traceability should be ensured for subsequent analysis and model training.

[0028] Next, data cleaning is performed to identify and process outliers, such as extreme values ​​caused by sensor malfunctions or obvious errors. Missing values ​​may occur during data acquisition; spline interpolation is used to address this. Data from different sources may have different timestamp precision or sampling frequencies, thus requiring time alignment. Interpolation methods can be used to align timestamps and fill in missing values ​​in time series; for noisy data, smoothing filters can be used to reduce random fluctuations. Then, data feature extraction and construction are performed, directly extracting basic features such as temperature, humidity, power consumption, and operating time. For high-dimensional datasets, principal component analysis (PCA) is used to reduce data dimensionality and model complexity.

[0029] The main steps of PCA are as follows: Assume there is One sample, The number of indicators can form a size of Sample matrix .

[0030] ; Standardization: In order to eliminate the influence of different dimensions, the original variables need to be standardized.

[0031] Calculate the covariance matrix R of the standardized sample.

[0032] Calculate the eigenvalues ​​and eigenvectors of R.

[0033] Calculate the principal component contribution rate and cumulative contribution rate.

[0034] Calculate the principal component scores, perform weighted summation on the m principal components to calculate the scores, and extract the principal components.

[0035] Data compression is performed during storage and transmission. The preprocessed data undergoes quality assessment to ensure its integrity, consistency, and reliability. Assessment metrics include data coverage, invalid data ratio, and consistency of sampling frequency.

[0036] For a containing Data sets of historical data, ,in Indicates the first The load vector of the day, its dimensions It depends on the sampling frequency of the dataset.

[0037] First, the historical load dataset is normalized on a daily basis by dividing each load value for a given day by the maximum load value for that day: ; In the formula, This is the normalized load vector.

[0038] Secondly, the normalized dataset is randomly divided into two parts: a training set and a test set, respectively. and express, and These represent the number of days in the training set and the test set, respectively. The training set contains load data for 80% of the days, and the test set contains load data for the remaining 20% ​​of the days.

[0039] Clustering load data using clustering algorithms to analyze load types is a common and effective method to reduce the complexity of load data. However, due to the massive scale and complexity of the diverse and heterogeneous data on the load side, traditional clustering algorithms such as C-means and fuzzy C-means, commonly used in power load data preprocessing, suffer from local convergence problems caused by improper selection of initial cluster centers, affecting the effectiveness of power load preprocessing and consequently the accuracy of load forecasting. Therefore, this invention proposes an improved fuzzy C-means clustering algorithm for power load sample data feature preprocessing. By improving the selection of cluster centers in the traditional fuzzy C-means algorithm, the method improves the identification effect of air conditioning load types, laying the foundation for subsequent modeling.

[0040] Specifically, the fuzzy C-means clustering algorithm randomly generates cluster centers in the initial iteration, groups samples that are close together into one class by measuring the Euclidean distance between samples, updates the cluster centers by calculating the mean of each class of samples, and then enters the next iteration. The particle swarm optimization algorithm optimizes the initial cluster centers generated by the fuzzy C-means clustering algorithm.

[0041] As a preferred method for adjustable resource data mining in this embodiment of the invention, a Fuzzy C-means clustering algorithm model based on PSO optimization is adopted, and the algorithm flow is as follows: Figure 2 As shown.

[0042] To improve the clustering performance of traditional fuzzy C-means clustering algorithms on large-scale, massive datasets, and to assist neural networks in extracting the mapping relationship between factors such as temperature, humidity, power consumption, and operating duration and daily load, an IFCM clustering algorithm is proposed and applied to electrical load classification and identification. Considering the significant differences in load under different operating conditions, an improved fuzzy C-means clustering algorithm based on Particle Swarm Optimization (PSO) is proposed to classify load samples according to data characteristics.

[0043] FCM (Focused Clustering) is an unsupervised clustering algorithm that uses Euclidean distance for clustering. In the initial iteration, the algorithm randomly generates cluster centers. By measuring the Euclidean distance between samples, samples with similar distances are grouped into one class. The cluster centers are then updated by calculating the mean of each class, and the algorithm proceeds to the next iteration. The computational flow of FCM for updating cluster centers is as follows: Let the normalized Data samples ,in It is the first A load vector. This refers to the number of categories into which the data samples are divided. Indicates the corresponding Categories It is its similarity classification matrix, and the cluster centers of each category are: , It is a sample for Membership degree (abbreviated as) Then the objective function J FCM It can be represented as follows: ; In the formula, ; It is the Euclidean distance, the measure of the first... Sample With the The distance between the center points of the class; It is the number of features of the sample; It is a weighted parameter, and its value range is... ; The Fuzzy C-means algorithm seeks an optimal classification that produces the minimum function value J. FCM It requires that the sum of the membership values ​​of a sample to all clusters is 1, that is, it satisfies: ; Based on the above formula and the objective function J FCM The expressions calculate the samples respectively. Membership degree of A1 and cluster center : ; set up For all kind, =0 ; .

[0044] Specifically, since FCM randomly generates cluster centers in the initial iteration, it may cause the clusters to get stuck in local optima. To address this problem, this embodiment uses the PSO algorithm to optimize the initial cluster centers of FCM. PSO is a swarm intelligence-based optimization algorithm inspired by the foraging behavior of bird flocks. It is a swarm intelligence optimization method that simulates natural phenomena. It gradually approaches the optimal solution of the problem through information sharing among individuals in the swarm. In PSO, each possible solution is called a particle. Each particle has a position and velocity in the search space, and the position of the particle represents a potential solution. Each particle has a velocity vector and a fitness value. The velocity vector determines its movement direction and distance in the search space, and the fitness value measures the quality of the particle's position through a fitness function. Each particle records its best position found so far. The best positions found by all particles in the entire swarm are then recorded. The individual best position (pBest) and the global best position (gBest) are updated. The algorithm stops when the predetermined maximum number of iterations is reached or the convergence condition is met. The flowchart of the PSO optimization algorithm is shown below. Figure 2 As shown.

[0045] Specifically, the position vector X of each particle i Represents a set of candidate cluster centers: ; Where x k ∈R m Let be the center vector of the k-th class, and c be the preset number of clusters.

[0046] Using the FCM objective function J FCM The reciprocal of the value is used as the particle fitness value: ; In the formula, =10 -6 To prevent division by zero by extremely small constants, this design ensures that the fitness maximization process of PSO is equivalent to the FCM objective function minimization process.

[0047] The PSO optimization process is as follows: N particles are randomly generated, each containing a cxm-dimensional position vector and a velocity vector. The update formulas for the velocity and position of each particle are as follows: ; ; in, It is a particle In time speed, It is a particle In time Location, It is the inertial weight, which controls the influence of the particle's current velocity; and These are learning factors, also known as acceleration, representing the acceleration coefficients for approaching the individual's optimal position and the global optimal position, respectively. and Is Random numbers between Represents particles The optimal position for an individual.

[0048] Specifically, the Monte Carlo method is used to perform secondary aggregation processing on the clustered multi-condition data, including: The Monte Carlo method is a technique for numerical simulation and estimation using random sampling and statistical analysis, particularly suitable for solving system problems with stochasticity or complexity. For the simulation and prediction of electrical loads, the Monte Carlo method can consider the impact of different regional physical conditions (such as geographical location, building structure, and climate differences) on the load, thus more accurately reflecting the overall electrical load of the region.

[0049] Suppose that this embodiment of the invention aims to estimate the electrical load in a specific area at a certain time, denoted as . This load Due to the influence of various uncertain factors, it is necessary to consider factors such as temperature, building insulation performance, and user habits. These factors can all be regarded as random variables. Through Monte Carlo simulation, a large number of samples can be taken from these random variables, and statistical analysis can be performed on the sampling results to estimate... The distribution and expected value.

[0050] First, it is necessary to determine which factors affect electrical load. Let's assume there are... The main influencing factors are denoted as follows: each Both are random variables, representing the first... The possible values ​​of each factor.

[0051] These random variables are typically distributed based on historical data or other statistical models. Electrical load It can be viewed as a function of these random variables: ; In the formula, the function It can be determined by physical models or empirical formulas, and it represents the energy consumption of an electrical system given a specific set of physical conditions.

[0052] Samples are drawn independently from the distribution of each random variable; denoted as ,in , That is, the number of simulations. For each sampling sequence Calculate the corresponding electrical load. : ; By all Simulation results Statistical analysis was conducted to obtain the electrical load. The estimated distribution, expected value, and variance; ; ; The total electrical load of the region is calculated based on the sum of the electrical loads of each sub-region.

[0053] .

[0054] Because of the load in each sub-region The total load after merging is obtained through random sampling. It is also a random variable. Through the analysis of... Statistical analysis can more accurately reflect the electrical load situation of the entire region. By aggregating the loads of multiple sub-regions within the region, a more accurate and comprehensive regional load estimate can be obtained, thus providing strong support for energy management and planning.

[0055] Among them, the load L of each sub-region i The load value is calculated by establishing a power demand model based on electricity consumption characteristics, equipment types, and environmental factors. During the calculation process, the load value is accurately estimated by combining the types, quantities, and usage frequencies of electrical appliances in each sub-region, as well as the external environment (such as climate change). The calculation method also considers the impact of different time periods and seasons on the load.

[0056] Specifically, such as Figure 3 As shown, Support Vector Machines (SVMs) are supervised learning models used for classification and regression analysis, particularly well-suited for handling problems with small samples, nonlinearity, and high dimensionality. SVMs separate data points of different classes as accurately as possible by finding an optimal decision boundary. This characteristic makes SVMs highly effective in load pattern classification and recognition tasks, especially when dealing with complex problems such as electrical loads affected by various operating conditions. The expression for the hyperplane is: ,in, It is the normal vector of the hyperplane. This is the bias term. The optimal value is determined by maximizing the distance from the data points on either side of the classification boundary to the hyperplane. and .

[0057] The training process is achieved by solving an optimization problem: ,in, These are slack variables, used to allow some data points to be misclassified within the decision boundary. This is a penalty parameter used to control the degree of punishment for misclassification.

[0058] A trained SVM model can be used to classify new samples. Given a new feature vector... The category to which a value belongs is determined by calculating its distance to the decision boundary. : .

[0059] in, This represents the weighting coefficients obtained through simulation and statistical analysis.

[0060] This classification model can automatically identify and distinguish different electrical load modes based on the input operating conditions, thus enabling its application in electrical system management and optimization under different scenarios.

[0061] Based on the results of the above steps, the electrical loads of multiple electrical areas within a building and across multiple buildings are aggregated to form a unified load management unit, enabling it to better participate in smart grid dispatch. Through centralized management and dispatch, the use of power resources is optimized, the pressure on the power grid is reduced, and overall energy efficiency is improved.

[0062] Specifically, such as Figure 1 As shown, the assessment of the adjustability potential of electrical loads includes: An assessment model for the aggregated response potential of electrical appliances considering multiple factors is established to quantitatively analyze the maximum response potential that an electrical appliance cluster can achieve through direct load control during demand response events.

[0063] Specifically, by establishing an assessment model that considers multiple factors, the maximum response potential of an appliance cluster in demand response events, achievable through direct load control (such as temperature adjustment), is quantitatively analyzed. This assessment involves multiple factors, including temperature control range, user comfort, and equipment responsiveness, to estimate the overall demand response capability. Under the same outdoor temperature, the set temperature... The higher the setting, the greater the decrease in aggregated power, and thus the greater the scheduling potential. The temperature setpoint for large-scale electrical clusters is defined as follows: The average polymerization power at that time was the reference power. After adjusting the set temperature, the average polymerization power is: , regulatory potential The difference between the two: .

[0064] This invention addresses the characteristics of electrical load data in the context of smart grids, including finer granularity, greater quantity and dimensions, and more diverse formats and structures. It proposes an improved fuzzy C-means clustering algorithm, which demonstrates more accurate and efficient performance in data preprocessing and better extraction of data features. Based on the improved fuzzy C-means clustering method, it further combines Particle Swarm Optimization (PSO) to find the global optimum by simulating bird flock foraging behavior, adjusting and optimizing the parameters of the Integrated Flexible Control Model (IFCM), thereby improving the accuracy of electrical load type identification. Considering the impact of physical conditions such as geographical location, building structure, and climate differences on electrical loads, a secondary aggregation method for electrical loads in different regions based on Monte Carlo simulation is proposed, thus more accurately reflecting the overall electrical load situation of the building area. Through detailed analysis of the characteristics of electrical load data under different operating conditions, and combining the clustering results with the characteristics of different operating conditions, Support Vector Machine (SVM) is used for classification modeling to effectively identify and distinguish various load patterns. By aggregating multiple electrical areas within a building and electrical loads from multiple buildings, a unified load management unit is formed, and a demand response potential assessment model that considers multiple factors is established, enabling it to better participate in the scheduling and load balancing of the smart grid.

[0065] This invention provides an improved fuzzy C-means clustering method aimed at efficiently processing massive amounts of data and their high-dimensional features in the context of smart grids. Addressing the problem of local convergence in the initial cluster center selection of traditional C-means and fuzzy C-means clustering algorithms commonly used for data preprocessing, an improved scheme is proposed to enhance the accuracy and efficiency of data preprocessing and improve the clustering effect of electrical load data. The improved fuzzy C-means (IFCM) clustering method improves the clustering performance of traditional fuzzy C-means clustering algorithms on large-scale, massive, multi-dimensional, and heterogeneous load-side data through efficient data preprocessing and further refines data feature extraction. The IFCM clustering algorithm is proposed and applied to the classification of electrical load data.

[0066] An improved fuzzy C-means clustering method incorporates a particle swarm optimization (PSO) module into the traditional fuzzy C-means clustering algorithm, optimizing the cluster center selection process and addressing the shortcomings of traditional methods in handling multiple electrical load types under various operating conditions. This improved method effectively enhances the accuracy of electrical load type identification and adapts to the complexity of load data under diverse operating conditions.

[0067] An improved fuzzy C-means clustering method enhances clustering performance by optimizing model parameters. A secondary aggregation method based on Monte Carlo simulation is designed to consider the impact of different regional physical conditions, such as geographical location, building structure, and climate differences, on electrical loads, resulting in more stable and reliable aggregation results. The secondary aggregation method based on Monte Carlo simulation constructs electrical load category models for different operating conditions.

[0068] By utilizing the Support Vector Machine (SVM) algorithm to find the optimal separating hyperplane to maximize the classification margin, different load patterns are accurately identified and distinguished, significantly improving aggregation accuracy and model robustness. This method can accurately handle various load patterns and adapt to the scheduling requirements of electrical systems in smart grids.

[0069] Establish an evaluation model that considers multiple factors, and through centralized management and optimized scheduling methods, effectively integrate the electrical loads of various regions and buildings, optimize the use of power resources, reduce the pressure on the power grid, and improve overall energy efficiency.

[0070] In one specific embodiment of the present invention, an electrical load aggregation system for smart grids includes: The data acquisition and preprocessing module is used to acquire multi-condition data of electrical loads and preprocess the multi-condition data of electrical loads. The clustering module is used to perform clustering processing on the preprocessed multi-condition data using an improved fuzzy C-means clustering algorithm; The secondary aggregation processing module is used to perform secondary aggregation processing on the clustered multi-condition data using the Monte Carlo method. The classification module is used to build a multi-condition classification model for electrical loads. It uses the support vector machine algorithm to classify the multi-condition data after secondary aggregation, and identifies and classifies the electrical load patterns. The smart grid dispatch module is used to aggregate the electrical loads of multiple electrical areas within a building and multiple buildings based on the classification results of electrical load patterns, participate in the dispatch of the smart grid, and evaluate the adjustable potential of electrical loads.

[0071] Specifically, the adjustable potential of electrical loads is assessed, including: An assessment model for the aggregated response potential of electrical appliances considering multiple factors is established to quantitatively analyze the maximum response potential that an electrical appliance cluster can achieve through direct load control during demand response events.

[0072] Specifically, the improved fuzzy C-means clustering algorithm includes: The particle swarm optimization algorithm is used to optimize the fuzzy C-means clustering algorithm; The fuzzy C-means clustering algorithm randomly generates cluster centers in the initial iteration. By measuring the Euclidean distance between samples, samples with similar distances are grouped into one class. The cluster centers are updated by calculating the mean of each class of samples, and then the algorithm enters the next iteration. The particle swarm optimization algorithm optimizes the initial cluster centers generated by the fuzzy C-means clustering algorithm.

[0073] Specifically, the fuzzy C-means clustering algorithm includes: Set preprocessed Data samples ,in It is the first Given a load vector, the objective function J FCM Represented as: ; in, This indicates the number of categories to be divided into for the data samples. ; It is its similarity classification matrix. It is a sample for Membership degree; In the formula, ; Indicates the first Sample With the The distance between the center points of the class; It is the number of features of the sample; These are weighted parameters; A sample's membership values ​​to all clusters sum to 1 if: ; Calculate samples separately Membership degree of A1 and cluster center : ; set up For all kind, =0 ; .

[0074] Specifically, the particle swarm optimization algorithm obtains the optimal solution to the problem through information sharing among individuals in the swarm; In the particle swarm optimization algorithm, each particle records the best position it finds. Update the individual best position and the global best position based on the best positions found by all particles in the entire swarm. The algorithm stops when the predetermined maximum number of iterations is reached or the convergence condition is met.

[0075] Specifically, the formulas for updating the velocity and position of each particle are as follows: ; ; in, It is a particle In time speed, It is a particle In time Location, It is inertial weight. and These are learning factors, also known as acceleration, representing the acceleration coefficients for approaching the individual's optimal position and the global optimal position, respectively. and Is Random numbers between Represents particles The optimal position for an individual.

[0076] Specifically, the Monte Carlo method is used to perform secondary aggregation processing on the clustered multi-condition data, including: Samples are drawn independently from the distribution of each random variable; for each sampling sequence, the electrical load of each sub-region is calculated: By statistically analyzing all simulation results, the estimated distribution, expected value, and variance of the electrical load are obtained. The total electrical load of the region is calculated based on the sum of the electrical loads of each sub-region.

[0077] In one specific embodiment of the present invention, in a building air conditioning system, the air conditioning load exhibits diverse characteristics as time and environmental conditions change. These characteristics are influenced by various factors such as external temperature and humidity, building usage, and human activity. Traditional single control models may not be able to effectively cope with this diversity; therefore, it is necessary to establish corresponding control models for the different performance states of the air conditioning load under different environmental conditions.

[0078] This invention provides a method for analyzing the characteristics of air conditioning load-side data and aggregating air conditioning loads under a smart grid, including: This paper proposes a multi-condition cluster aggregation method for air conditioning load data from different operating conditions within a building complex. In the context of smart grids, air conditioning load data is becoming increasingly diverse and comprehensive, requiring effective preprocessing of massive amounts of data. Firstly, dimensionality reduction effectively reduces data complexity while preserving key features and minimizing redundant information. This process simplifies the construction of the air conditioning load aggregation model and reduces the difficulty of building an adjustable resource pool, facilitating in-depth analysis of users' air conditioning consumption behavior within buildings.

[0079] A method for classifying and predicting air conditioning loads under multiple operating conditions is proposed. Based on the characteristics of different operating conditions, air conditioning load data is classified, and a prediction model is established for each category. This method uses a clustering algorithm to preprocess the sample dataset, and then validates the model through cross-validation or hold-out methods to ensure that the model can accurately predict the air conditioning load response under different operating conditions, thereby improving the accuracy of the control strategy.

[0080] A quantitative calculation model for adjustable potential is proposed. This model helps operators optimize demand response strategies by quantifying the adjustment range of air conditioning load under different load conditions, thereby enhancing users' willingness and satisfaction to participate in demand response and improving the dispatching capability and flexibility of the smart grid.

[0081] A multi-condition air conditioning load aggregation method is proposed. By aggregating air conditioning loads from different buildings into a controllable virtual power plant, and utilizing smart grid technology for information exchange, integrated load management is achieved. The load aggregator centrally manages these loads and issues control commands to each load through a dispatch platform. This method enables demand response and dispatch control in smart grids, enhances the grid's load regulation capability, and improves the reliability and stability of grid operation.

[0082] In one specific embodiment of the present invention, the study focuses on split-type air conditioners used by small businesses and residents in different districts of a city during the summer. Besides the base load, the study includes 10,000 fixed-frequency air conditioners and a certain amount of load that can be moved, reduced, or transferred. The compensation electricity prices for the moveable, reduced, and transferred loads are RMB 0.25, RMB 0.3, and RMB 0.25 per kilowatt-hour, respectively. -1 The electricity price for air conditioning load compensation is 0.3 yuan per kilowatt-hour. -1 At any given time, the aggregated load of air conditioning in a single operating zone... The average operating conditions and power can be expressed as: ; ; ; In the formula, The number of air conditioners in the area. The operating status of each air conditioner in the area. For the first The rated power of the air conditioner For the first The operating status of an air conditioner within a cycle. For the first The time when the air conditioner is turned on. For the first The model calculates the air conditioning load aggregation time for a given day. For the vast majority of the time, the relative error between the aggregated power estimate and the actual value is less than 5%, and the aggregation accuracy meets the decision-making needs of the power grid dispatch center, significantly reducing the computational complexity of the day-ahead dispatch layer. Combining the outdoor temperature of a typical summer day, the aggregated power of the air conditioning load for that day is obtained through this model. Figure 4 As shown.

[0083] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0084] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for clustering and aggregating electrical loads for smart grids, characterized in that, include: Collect multi-condition data of electrical load and preprocess the multi-condition data of electrical load; An improved fuzzy C-means clustering algorithm was used to cluster the preprocessed multi-condition data. The Monte Carlo method was used to perform a secondary aggregation process on the clustered multi-condition data; A multi-condition classification model for electrical loads is constructed, and the support vector machine algorithm is used to classify the multi-condition data after secondary aggregation, thereby identifying and classifying the electrical load patterns. Based on the classification results of electrical load patterns, the electrical loads of multiple electrical areas within a building and multiple buildings are aggregated to participate in the smart grid scheduling, and the adjustability potential of the electrical loads is evaluated.

2. The method for clustering and aggregating electrical loads for smart grids according to claim 1, characterized in that, Assess the adjustability potential of electrical loads, including: An assessment model for the aggregated response potential of electrical appliances considering multiple factors is established to quantitatively analyze the maximum response potential that an electrical appliance cluster can achieve through direct load control during demand response events.

3. The method for clustering and aggregating electrical loads for smart grids according to claim 1, characterized in that, The improved fuzzy C-means clustering algorithm includes: The particle swarm optimization algorithm is used to optimize the fuzzy C-means clustering algorithm; The fuzzy C-means clustering algorithm randomly generates cluster centers in the initial iteration. By measuring the Euclidean distance between samples, samples with similar distances are grouped into one class. The cluster centers are updated by calculating the mean of each class of samples, and then the algorithm enters the next iteration. The particle swarm optimization algorithm optimizes the initial cluster centers generated by the fuzzy C-means clustering algorithm.

4. The method for clustering and aggregating electrical loads for smart grids according to claim 3, characterized in that, The fuzzy C-means clustering algorithm includes: Set preprocessed Data samples ,in It is the first Given a load vector, the objective function J FCM Represented as: ； in, This indicates the number of categories to be divided into for the data samples. ; It is its similarity classification matrix. It is a sample for Membership degree; In the formula, ; Indicates the first Sample With the The distance between the center points of the class; It is the number of features of the sample; These are weighted parameters; A sample's membership values ​​to all clusters sum to 1 if: ； Calculate samples separately Membership degree of A1 and cluster center : ； set up For all kind, =0 ; 。 5. The method for clustering and aggregating electrical loads for smart grids according to claim 3, characterized in that, The particle swarm optimization algorithm obtains the optimal solution to the problem through information sharing among individuals in the swarm; In the particle swarm optimization algorithm, each particle records the best position it finds. Update the individual best position and the global best position based on the best positions found by all particles in the entire swarm. The algorithm stops when the predetermined maximum number of iterations is reached or the convergence condition is met.

6. The method for clustering and aggregating electrical loads for smart grids according to claim 5, characterized in that, The formulas for updating the velocity and position of each particle are as follows: ； ； in, It is a particle In time speed, It is a particle In time Location, It is inertial weight. and These are learning factors, representing the acceleration coefficients for approaching the individual's optimal position and the global optimal position, respectively. and Is Random numbers between Represents particles The optimal position for an individual.

7. The method for clustering and aggregating electrical loads for smart grids according to claim 1, characterized in that, The Monte Carlo method is used to perform secondary aggregation processing on the clustered multi-condition data, including: Samples are drawn independently from the distribution of each random variable; for each sampling sequence, the electrical load of each sub-region is calculated: By statistically analyzing all simulation results, the estimated distribution, expected value, and variance of the electrical load are obtained. The total electrical load of the region is calculated based on the sum of the electrical loads of each sub-region.

8. An electrical load aggregation system for smart grids, characterized in that, include: The data acquisition and preprocessing module is used to acquire multi-condition data of electrical loads and preprocess the multi-condition data of electrical loads. The clustering module is used to perform clustering processing on the preprocessed multi-condition data using an improved fuzzy C-means clustering algorithm; The secondary aggregation processing module is used to perform secondary aggregation processing on the clustered multi-condition data using the Monte Carlo method. The classification module is used to build a multi-condition classification model for electrical loads. It uses the support vector machine algorithm to classify the multi-condition data after secondary aggregation, and identifies and classifies the electrical load patterns. The smart grid dispatch module is used to aggregate the electrical loads of multiple electrical areas within a building and multiple buildings based on the classification results of electrical load patterns, participate in the dispatch of the smart grid, and evaluate the adjustable potential of electrical loads.

9. An electrical load aggregation system for smart grids according to claim 8, characterized in that, Assess the adjustability potential of electrical loads, including: An assessment model for the aggregated response potential of electrical appliances considering multiple factors is established to quantitatively analyze the maximum response potential that an electrical appliance cluster can achieve through direct load control during demand response events.

10. An electrical load aggregation system for smart grids according to claim 8, characterized in that, The improved fuzzy C-means clustering algorithm includes: The particle swarm optimization algorithm is used to optimize the fuzzy C-means clustering algorithm; The fuzzy C-means clustering algorithm randomly generates cluster centers in the initial iteration. By measuring the Euclidean distance between samples, samples with similar distances are grouped into one class. The cluster centers are updated by calculating the mean of each class of samples, and then the algorithm enters the next iteration. The particle swarm optimization algorithm optimizes the initial cluster centers generated by the fuzzy C-means clustering algorithm.

Images