A distributed collaborative management system for discrete small loads on the power grid consumption side

Abstract

This invention relates to the field of smart grid technology, and more specifically, to a distributed collaborative management system for discrete small loads on the power grid consumption side. This system addresses the problems of existing technologies failing to accurately identify user electricity consumption patterns, accurately reflect users' actual response characteristics to changes in electricity prices and the environment, and provide a basis for developing personalized demand response strategies. This invention accurately identifies user electricity consumption patterns through a user behavior analysis module that performs normalization and cluster analysis on electricity consumption data. Combined with a response delay model and load adjustability analysis, it more accurately reflects users' actual response characteristics to changes in electricity prices and the environment. This module provides a basis for developing personalized demand response strategies, improves the accuracy of load forecasting and regulation, enhances the flexibility and energy efficiency of power grid dispatching, and also helps improve user participation and satisfaction.

Description

Technical Field

[0001] This invention relates to the field of smart grid technology, and more specifically, to a distributed collaborative management system for discrete small loads on the power grid consumer side. Background Technology

[0002] The large-scale integration of distributed energy resources and discrete small loads highlights the shortcomings of traditional centralized load management systems in real-time response, supply and demand balance, and resistance to single points of failure. The intermittency and volatility of loads increase the control burden of centralized architecture, leading to low resource utilization and serious energy curtailment problems. Current technologies have limited capabilities in handling dynamic coordination of electricity-side loads, demand response, and fault self-healing, making it urgent to develop distributed management systems to improve control efficiency and system resilience.

[0003] The patent application with publication number CN117913915A discloses a source-load collaborative management system in a smart active power distribution network, including a regional information acquisition module, an electricity consumption assessment module, an operation assessment module, and an execution management module. The modules are connected by signals. Based on the cluster analysis of the surge sub-regions, the load side is divided into assessment areas. The load side's power usage information and the electricity consumption behavior information provided by the regional information acquisition module are comprehensively weighted and calculated to determine the possible surge in electricity consumption in the assessment area. The power quality information and operation status information of the distributed power source are comprehensively weighted and calculated to obtain the scheduling status of the distributed power source. Based on the surge in electricity consumption area and the geographical location of the dispatchable distributed power source, the power scheduling of the distributed power source is realized. This invention helps to achieve more refined management of the region and helps to realize collaborative scheduling with the power source side in the case of a surge in electricity consumption on the load side.

[0004] However, while the aforementioned reference patents accurately identify electricity surges through load-side area division and weighted analysis of multi-source electricity consumption information, and achieve coordinated scheduling between the load side and the power supply side based on distributed power source classification and scheduling integration, they cannot accurately identify users' electricity consumption patterns, nor can they truly reflect users' actual response characteristics to changes in electricity prices and the environment, thus failing to provide a basis for formulating personalized demand response strategies. At the same time, they cannot coordinate load allocation while meeting global constraints, and cannot balance user comfort and control precision, thereby reducing regulation flexibility and resource utilization efficiency.

[0005] To address these issues, we propose a distributed collaborative management system for discrete small loads on the power grid's consumer side. Summary of the Invention

[0006] The purpose of this invention is to provide a distributed collaborative management system for discrete small loads on the power grid consumption side. This system solves the problems of existing technologies that cannot accurately identify users' electricity consumption patterns, cannot truly reflect users' actual response characteristics to changes in electricity prices and the environment, and cannot provide a basis for formulating personalized demand response strategies. At the same time, it cannot coordinate load allocation under the premise of meeting global constraints, cannot take into account both user comfort and control accuracy, and reduces the flexibility of regulation and resource utilization efficiency.

[0007] The objective of this invention is achieved through the following technical solution:

[0008] A distributed collaborative management system for discrete small loads on the power grid consumer side, comprising:

[0009] The data acquisition and processing module is used to collect electricity consumption data from each small load node in real time and perform preprocessing operations on the collected electricity consumption data.

[0010] The load forecasting module constructs a load forecasting model for the electricity consumption side that integrates historical electricity consumption data and historical meteorological data, and uses the model to forecast the electricity consumption side load for a future period of time.

[0011] The user behavior analysis module is used to analyze users' historical electricity consumption characteristics data and build a user behavior model that includes electricity consumption patterns, response delays, and load adjustability.

[0012] The distributed collaborative control module, based on load forecasting results and user behavior analysis results, uses a distributed optimization algorithm to make local control decisions at each node, and achieves collaborative regulation among multiple nodes through limited information interaction.

[0013] In a preferred embodiment of the present invention, the process by which the load forecasting module processes historical electricity consumption data and historical meteorological data includes:

[0014] Acquire historical electricity consumption data and historical meteorological data. The electricity consumption data includes active power, reactive power, voltage, current, frequency and power factor. The meteorological data includes ambient temperature, ambient humidity and solar radiation intensity. Generate a collection cycle and divide the collection cycle into multiple collection periods.

[0015] The active power change rate of each small load node is obtained in multiple collection periods. The active power change rate represents the ratio between the change in active power and the duration of the corresponding time period. A set A of active power change rates is constructed in this way, and the average of the differences between the largest and smallest subsets in set A is recorded as the active power change rate difference.

[0016] Similarly, the difference in the rate of change of reactive power, voltage, current, frequency, and power factor can be obtained using the same method as the difference in the rate of change of active power.

[0017] The ambient temperature of each small load node is obtained during multiple collection periods, and the arithmetic mean of the multiple ambient temperatures is calculated. The arithmetic mean of the multiple ambient temperatures is recorded as the average ambient temperature.

[0018] The average ambient humidity and average solar radiation intensity can be obtained by using the same method to calculate the average ambient temperature.

[0019] In a preferred embodiment of the present invention, the process by which the load forecasting module constructs an electricity-side load forecasting model that integrates meteorological data and uses the model to forecast the electricity-side load for a future period includes:

[0020] Electricity consumption features are extracted from historical electricity consumption data, including differences in active power change rate, reactive power change rate, voltage change rate, current change rate, frequency change rate, and power factor change rate. Meteorological features are extracted from historical meteorological data, including average ambient temperature, average ambient humidity, and average solar irradiance. The extracted electricity consumption features and meteorological features are combined to form an input feature matrix X, which is used as the input to the machine learning model. The historical future load value corresponding to each feature matrix X is used as the output label y. The machine learning model is trained with the goal of minimizing the loss function between the predicted value and the actual value until the loss function converges, resulting in a trained electricity load prediction model.

[0021] Real-time electricity consumption data and real-time meteorological data of each small load node are collected, the corresponding feature vectors are extracted and input into the trained electricity-side load prediction model, and the load prediction values ​​for the next H time steps are output. , where i=1,2,…,H, and H represents the prediction time step.

[0022] In a preferred embodiment of the present invention, the process by which the user behavior analysis module analyzes the user's historical electricity consumption characteristic data and identifies the electricity consumption pattern includes:

[0023] Obtain historical electricity consumption data for each user, including total energy consumption, instantaneous power, electricity price information, ambient temperature, and ambient humidity;

[0024] The electricity consumption characteristic data of each user is processed with uniform time granularity, set to once per hour. Then, the electricity consumption characteristic data is normalized. Key load features are extracted from the normalized electricity consumption characteristic data. Key load features include daily average load, peak-valley difference, peak period and load factor, forming a load feature vector.

[0025] The K-means clustering algorithm is used to classify users. The specific steps are as follows:

[0026] T1: Set the number of clusters K and initialize K cluster centers;

[0027] T2: Assign each user's load feature vector to the nearest cluster center;

[0028] T3: Update each cluster center to the feature mean of that user group;

[0029] T4: Repeat T2 and T3 until the classification is stable or converges;

[0030] Through cluster analysis, users were categorized into the following electricity consumption patterns: peak users, stable load users, and low load users.

[0031] In a preferred embodiment of the present invention, the process by which the user behavior analysis module constructs a user behavior model including electricity consumption patterns, response latency, and load adjustability includes:

[0032] Response delay refers to the lag in a user's response to external stimuli in their electricity consumption behavior.

[0033] Response delay can be modeled using a delay regression model, expressed by the following expression:

[0034] ;

[0035] in Let be the electricity consumption at the current time t. The electricity price at the current moment. The temperature at the current moment. All are regression coefficients. This is the error term;

[0036] Load adjustability refers to a user's ability to respond to external stimuli. If a user has multiple adjustable load devices, the load of each device can be expressed as:

[0037] ;

[0038] in The adjustable portion of the load at time t. Let N be the regulating load of the i-th device, and N be the total number of devices.

[0039] Taking into full account electricity consumption patterns, response delays, and load adjustability, a comprehensive user behavior model is constructed:

[0040] ;

[0041] in The total load or total power demand at time t. For the mode power or mode load at time t, The delayed power or delayed load at time t. The adjusted load or adjusted power is at time t.

[0042] In a preferred embodiment of the present invention, the process by which the distributed collaborative control module processes the load forecasting results and user behavior analysis results includes:

[0043] Obtain load forecasting results and user behavior analysis results. The load forecasting results are the load forecast values ​​for the next H time steps. The user behavior analysis results are and its three components , and ;

[0044] In a distributed power consumption system composed of small load nodes, if the system has N nodes, and each node i has its local load at each time step t:

[0045] ;

[0046] And a predicted load sequence for the next H time steps. , t=1,2,…,H.

[0047] In a preferred embodiment of the present invention, the process by which the distributed cooperative control module makes local control decisions at each node using a distributed optimization algorithm includes:

[0048] The global objective is to minimize the total cost function across all nodes, in the following form:

[0049] ;

[0050] ;

[0051] in It is the sequence of control variables of node i over H time steps. It is the local cost function for each node. The penalty term is used to control the adjustment range of the node. This is the deviation term used to control and predict errors;

[0052] Each node maintains a local copy and a global variable copy. Each node i uses its local behavioral model to decompose the current and predicted loads, obtaining... The local optimization constraints for each node i are as follows:

[0053] ;

[0054] Global power supply capacity is limited to Then the global constraints are:

[0055] .

[0056] In a preferred embodiment of the present invention, the process by which the distributed collaborative control module solves the distributed optimization algorithm and realizes collaborative regulation among multiple nodes through limited information interaction includes:

[0057] The ADMM distributed optimization algorithm is adopted, in which each node independently solves its own local problem and coordinates global consistency through limited information exchange. The specific steps are as follows:

[0058] S1: Local update, each node independently minimizes its own... Consider the current global estimate;

[0059] S2: Information exchange, nodes only exchange necessary information: each node uploads its prediction adjustment value at each time step. ;

[0060] The control center or neighboring nodes summarize and broadcast the updated global variables. ;

[0061] S3: Global coordination, updating global consistency variables, updating multipliers, and driving each node to gradually converge to a consistent solution.

[0062] Compared with the prior art, the advantages of this invention are:

[0063] (1) In this invention, the user behavior analysis module performs normalization processing and cluster analysis on electricity consumption data to accurately identify the user's electricity consumption pattern. Combined with the response delay model and load adjustability analysis, it can more realistically reflect the actual response characteristics of users to changes in electricity prices and the environment. This module provides a basis for formulating personalized demand response strategies, improves the accuracy of load forecasting and regulation, enhances the flexibility and energy efficiency of power grid dispatch, and also helps to improve user participation and satisfaction.

[0064] (2) In this invention, the distributed collaborative control module adopts a decentralized architecture, and each node makes independent decisions based on local information, which improves the robustness and scalability of the system. Through the ADMM algorithm, the load distribution is coordinated under the premise of satisfying global constraints, taking into account both user comfort and control accuracy. Limited information interaction reduces communication overhead, which is suitable for large-scale distributed systems and improves the flexibility of regulation and resource utilization efficiency. Attached Figure Description

[0065] Figure 1 This is a system block diagram of the present invention;

[0066] Figure 2This is a flowchart illustrating the steps involved in classifying users using the K-means clustering algorithm in this invention.

[0067] Figure 3 This is a flowchart illustrating the steps involved in solving the problem using the ADMM distributed optimization algorithm in this invention. Detailed Implementation

[0068] 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0069] Example 1: As Figure 1 and Figure 2 As shown, the present invention proposes a distributed collaborative management system for discrete small loads on the power grid consumption side, comprising:

[0070] The data acquisition and processing module is used to collect electricity consumption data of each small load node in real time and perform preprocessing operations on the collected electricity consumption data. The preprocessing operations include, but are not limited to, data cleaning, data filling and data normalization.

[0071] The data acquisition and processing module effectively improves the accuracy, completeness, and usability of electricity consumption data through preprocessing operations such as data cleaning, filling, and normalization. Cleaning removes noise and corrects errors to ensure data reliability; filling fills in missing data and maintains temporal continuity; and normalization unifies data scale and improves algorithm performance. These processes enhance system compatibility and real-time analysis capabilities, providing high-quality data support for load forecasting, user behavior analysis, and collaborative control, thus helping to achieve accurate demand response and efficient energy management.

[0072] The load forecasting module constructs a load forecasting model for the electricity consumption side that integrates historical electricity consumption data and historical meteorological data, and uses the model to forecast the electricity consumption side load for a future period of time.

[0073] The load forecasting module processes historical electricity consumption data and historical meteorological data in the following ways:

[0074] Acquire historical electricity consumption data and historical meteorological data. The electricity consumption data includes active power, reactive power, voltage, current, frequency and power factor. The meteorological data includes ambient temperature, ambient humidity and solar radiation intensity. Generate a collection cycle and divide the collection cycle into multiple collection periods.

[0075] The active power change rate of each small load node is obtained in multiple collection periods. The active power change rate represents the ratio between the change in active power and the duration of the corresponding time period. A set A of active power change rates is constructed in this way, and the average of the differences between the largest and smallest subsets in set A is recorded as the active power change rate difference.

[0076] Similarly, the difference in the rate of change of reactive power, voltage, current, frequency, and power factor can be obtained using the same method as the difference in the rate of change of active power.

[0077] The ambient temperature of each small load node is obtained during multiple collection periods, and the arithmetic mean of the multiple ambient temperatures is calculated. The arithmetic mean of the multiple ambient temperatures is recorded as the average ambient temperature.

[0078] The average ambient humidity and average solar radiation intensity can be obtained by using the same method to calculate the average ambient temperature.

[0079] The load forecasting module constructs an electricity load forecasting model that integrates meteorological data and uses the model to forecast the electricity load for a future period. This process includes:

[0080] Electricity consumption features are extracted from historical electricity consumption data, including differences in active power change rate, reactive power change rate, voltage change rate, current change rate, frequency change rate, and power factor change rate. Meteorological features are extracted from historical meteorological data, including average ambient temperature, average ambient humidity, and average solar irradiance. The extracted electricity consumption features and meteorological features are combined to form an input feature matrix X, which is used as the input to the machine learning model. The historical future load value corresponding to each feature matrix X is used as the output label y. The machine learning model is trained with the goal of minimizing the loss function between the predicted value and the actual value until the loss function converges, resulting in a trained electricity load prediction model.

[0081] Real-time electricity consumption data and real-time meteorological data of each small load node are collected, the corresponding feature vectors are extracted and input into the trained electricity-side load prediction model, and the load prediction values ​​for the next H time steps are output. , where i=1,2,…,H, and H represents the time step of the prediction (e.g., hourly prediction for the next 24 hours);

[0082] The load forecasting module integrates historical electricity consumption data and meteorological data, and constructs an input feature matrix using the difference in the rate of change of power parameters and meteorological characteristics, thereby improving forecast accuracy and model adaptability. This not only optimizes energy allocation and real-time decision-making, but also promotes the integrated utilization of renewable energy. In addition, accurate forecasting supports demand response strategies, helping users optimize their electricity consumption behavior, reduce electricity costs, and raise environmental awareness. Overall, this module improves energy management efficiency, enhances system flexibility, and improves user experience.

[0083] The user behavior analysis module is used to analyze users' historical electricity consumption characteristics data and build a user behavior model that includes electricity consumption patterns, response delays, and load adjustability.

[0084] The user behavior analysis module analyzes users' historical electricity consumption data and identifies their electricity consumption patterns. The process includes:

[0085] Obtain historical electricity consumption data for each user, including total energy consumption, instantaneous power, electricity price information, ambient temperature, and ambient humidity;

[0086] The electricity consumption characteristic data of each user is processed with a unified time granularity, set to once per hour (which can also be adjusted to 15 minutes or 30 minutes according to actual needs). Then, the electricity consumption characteristic data is normalized. Normalization is a conventional technique in existing technology and will not be elaborated on here. Through normalization, the load data of different users are converted to a unified dimension, eliminating the influence of absolute electricity consumption differences and highlighting relative electricity consumption behavior characteristics. Key load features are extracted from the normalized electricity consumption characteristic data. Key load features include daily average load, peak-valley difference, peak period and load factor, forming a load feature vector to enhance the discriminative power of clustering.

[0087] The K-means clustering algorithm is used to classify users. The specific steps are as follows:

[0088] T1: Set the number of clusters K and initialize K cluster centers;

[0089] T2: Assign each user's load feature vector to the nearest cluster center;

[0090] T3: Update each cluster center to the feature mean of that user group;

[0091] T4: Repeat T2 and T3 until the classification is stable or converges;

[0092] The optimal number of clusters K can be determined by the elbow method. The specific solution process is a common technique in the existing technology and will not be elaborated on here.

[0093] Through cluster analysis, users were categorized into the following electricity consumption patterns: peak-load users, stable-load users, and low-load users.

[0094] The process of building a user behavior model that includes electricity usage patterns, response latency, and load adjustability in the user behavior analysis module includes:

[0095] Response delay refers to the lag in how users respond to external stimuli (such as fluctuations in electricity prices, changes in ambient temperature, etc.). Modeling response delay is to capture the time lag in users' electricity consumption decisions and thus accurately predict their electricity consumption behavior.

[0096] Response delay can be modeled using a delay regression model, expressed by the following expression:

[0097] ;

[0098] in Let be the electricity consumption at the current time t. The electricity price at the current moment. The temperature at the current moment. All are regression coefficients. This is the error term;

[0099] Load adjustability refers to a user's ability to respond to external stimuli (such as electricity prices and climate change). Load adjustability mainly depends on the adjustable equipment in a user's home or business (such as air conditioners, water heaters, electric vehicles, etc.). If a user has multiple adjustable load devices (such as air conditioners and water heaters), the load of each device can be expressed as:

[0100] ;

[0101] in The adjustable portion of the load at time t. Let N be the regulating load of the i-th device, and N be the total number of devices.

[0102] The load regulation section of each device can be modeled using the following factors:

[0103] Electricity price response: If electricity prices are low, users may increase their use of air conditioning;

[0104] Temperature response: When the temperature is high, the air conditioning load may increase;

[0105] User habits: For example, some users are accustomed to using water heaters to heat water at night, or charging electric vehicles when electricity prices are low;

[0106] Taking into full account electricity consumption patterns, response delays, and load adjustability, a comprehensive user behavior model is constructed:

[0107] ;

[0108] in The total load or total power demand at time t. For model power or model load at time t, based on historical data and seasonal fluctuations, Let be the delayed power or delayed load at time t, considering the hysteresis effects of external stimuli such as electricity price and temperature. The adjusted load or adjusted power at time t reflects the user's responsiveness to power grid management;

[0109] The user behavior analysis module improves user classification accuracy through normalization and cluster analysis, effectively identifying different electricity consumption patterns. It also utilizes delayed regression models to accurately capture user response delays to electricity price fluctuations and environmental changes, thus more accurately predicting user electricity consumption behavior. Furthermore, considering load adjustability reflects users' actual responsiveness to electricity prices and climate change, especially for users with multiple adjustable devices, it enhances the flexibility of load management and the grid's optimized dispatch capabilities. This module supports the development of personalized demand response strategies, promoting efficient energy allocation and use, while also improving user experience and engagement, encouraging users to adjust their electricity consumption habits based on intelligent suggestions to reduce electricity costs. Overall, this module improves grid operational efficiency, promotes effective interaction between users and the grid, and achieves the goals of energy conservation and cost savings.

[0110] Example 2: The technical solution of this embodiment of the invention differs from that of Example 1 in that:

[0111] like Figure 1 and Figure 3 As shown, the distributed collaborative control module, based on load forecasting results and user behavior analysis results, uses a distributed optimization algorithm to make local control decisions at each node, and achieves collaborative regulation among multiple nodes through limited information interaction.

[0112] The process by which the distributed collaborative control module processes load forecasting results and user behavior analysis results includes:

[0113] Obtain load forecasting results and user behavior analysis results. The load forecasting results are the load forecast values ​​for the next H time steps. The user behavior analysis results are and its three components , and ;

[0114] In a distributed power consumption system composed of small load nodes, if the system has N nodes (users or load centers), and each node i (i=1, 2, ..., N) has its local load at each time step t:

[0115] ;

[0116] And a predicted load sequence for the next H time steps. , t=1,2,…,H;

[0117] The distributed collaborative control module employs a distributed optimization algorithm to make local control decisions at each node, including the following processes:

[0118] The global objective is to minimize the total cost function across all nodes, in the following form: ;

[0119] ;

[0120] in It is the sequence of control variables (such as load adjustment commands) of node i over H time steps. It is the local cost function for each node. A penalty item is added to control the magnitude of node adjustments and prevent over-adjustment. To control the deviation term from the prediction error and ensure tracking accuracy;

[0121] Each node maintains a local copy and a global variable copy. Each node i uses its local behavioral model to decompose the current and predicted loads, obtaining... The local optimization constraints for each node i are as follows:

[0122] ;

[0123] Global power supply capacity is limited to Then the global constraints are:

[0124] ;

[0125] The distributed collaborative control module solves the distributed optimization algorithm and achieves collaborative regulation among multiple nodes through limited information exchange. This process includes:

[0126] The ADMM distributed optimization algorithm is adopted, in which each node independently solves its own local problem and coordinates global consistency through limited information exchange. The specific steps are as follows:

[0127] S1: Local update, each node independently minimizes its own... Consider the current global estimate;

[0128] S2: Information exchange, nodes only exchange necessary information: each node uploads its prediction adjustment value at each time step. ;

[0129] The control center or neighboring nodes summarize and broadcast the updated global variables. ;

[0130] S3: Global coordination, updating global consistency variables (such as the total load of each node), updating multipliers, and driving each node to gradually converge to a consistent solution;

[0131] The following is an example of a distributed electricity consumption collaborative control system for a residential community.

[0132] The system consists of:

[0133] There are N=5 household users (i.e., 5 nodes), numbered i=1 to 5;

[0134] Each household is connected to a local distribution center and has limited communication capabilities;

[0135] The control objective is to minimize the cost of power consumption adjustments for all users (such as loss of comfort, cost of frequent equipment start-ups and shutdowns, etc.) while meeting the total power supply capacity limit.

[0136] With a time step of 1 hour, load forecasting and user behavior analysis for the next H=4 hours are available.

[0137] Input for load forecasting and user behavior analysis:

[0138] The local load of each node i at time t is represented as follows: ;

[0139] For example, the predicted load for family number 3 in the next 4 hours might be:

[0140] ;

[0141] User behavior analysis results include:

[0142] User preferences (whether or not they are willing to turn down the air conditioning).

[0143] Historical electricity usage patterns (peak hour usage);

[0144] Adjustable load types (such as water heaters, electric vehicle chargers);

[0145] Local cost function design (taking node i as an example):

[0146] The goal of each node is to minimize its own cost function:

[0147] ;

[0148] in It is the load adjustment instruction of node i at time t (positive number represents reducing load, negative number represents increasing load). It is an ideal adjustment value suggested by the user behavior model. It is the penalty coefficient:

[0149] : Prevent excessive adjustments (such as frequent switching of equipment);

[0150] : Ensure that the adjustments are as close as possible to the predicted values, and avoid excessive deviation;

[0151] Use ADMM to solve problems of the following form:

[0152] ;

[0153] Where C(t) is the upper limit of the system's global power supply capacity at time t;

[0154] ADMM specific steps (executed once every hour):

[0155] S1: Each node i independently minimizes its own cost function while considering the current global consistency variables and multipliers. In this step, each household calculates its own "optimal" load adjustment strategy based on its own user behavior model and the current system state.

[0156] S2: Each node uploads its load adjustment suggestions for the next 4 hours to the control center or neighboring nodes. After the control center summarizes the suggestions from all nodes, it calculates a new global consistency variable.

[0157] S3: Global coordination, updating global consistency variables (such as the total load of each node), updating multipliers, this process drives each node to gradually reach a consensus, ensuring that the overall power supply does not exceed the upper limit;

[0158] The distributed collaborative control module, through decentralized control, enables each node to make independent optimization decisions based on local information, reducing dependence on the central controller and improving system robustness and scalability. By utilizing a local cost function to comprehensively consider load adjustment magnitude and prediction deviation, a balance is achieved between user comfort and control accuracy. Employing the ADMM algorithm, while meeting global power supply capacity constraints, it coordinates load distribution among nodes through limited information interaction to gradually reach the overall optimal solution, reducing communication overhead and making it suitable for large-scale distributed systems. This not only improves the flexibility and response speed of load regulation but also enhances resource utilization efficiency.

[0159] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concept, should be covered within the scope of protection of the present invention.

Claims

1. A discrete small load distributed collaborative management system on the power grid power consumption side, characterized in that, include: The data acquisition and processing module is used to collect electricity consumption data from each small load node in real time and perform preprocessing operations on the collected electricity consumption data. The load forecasting module constructs a load forecasting model for the electricity consumption side that integrates historical electricity consumption data and historical meteorological data, and uses the model to forecast the electricity consumption side load for a future period of time. The user behavior analysis module is used to analyze users' historical electricity consumption characteristics data and build a user behavior model that includes electricity consumption patterns, response delays, and load adjustability. The distributed collaborative control module, based on load forecasting results and user behavior analysis results, uses a distributed optimization algorithm to make local control decisions at each node, and achieves collaborative regulation among multiple nodes through limited information interaction. The process by which the user behavior analysis module analyzes users' historical electricity consumption data and identifies electricity consumption patterns includes: Obtain historical electricity consumption data for each user, including total energy consumption, instantaneous power, electricity price information, ambient temperature, and ambient humidity; The electricity consumption characteristic data of each user is processed with uniform time granularity, set to once per hour. Then, the electricity consumption characteristic data is normalized. Key load features are extracted from the normalized electricity consumption characteristic data. Key load features include daily average load, peak-valley difference, peak period and load factor, forming a load feature vector. The K-means clustering algorithm is used to classify users. The specific steps are as follows: T1: Set the number of clusters K and initialize K cluster centers; T2: Assign each user's load feature vector to the nearest cluster center; T3: Update each cluster center to the feature mean of that user group; T4: Repeat T2 and T3 until the classification is stable or converges; Through cluster analysis, users were categorized into the following electricity consumption patterns: peak-load users, stable-load users, and low-load users. The process by which the user behavior analysis module constructs a user behavior model that includes electricity consumption patterns, response latency, and load adjustability includes: Response delay refers to the lag in a user's response to external stimuli in their electricity consumption behavior. Response delay can be modeled using a delay regression model, expressed by the following expression: ； in Let be the electricity consumption at the current time t. The electricity price at the current moment. The temperature at the current moment. All are regression coefficients. This is the error term; Load adjustability refers to a user's ability to respond to external stimuli. If a user has multiple adjustable load devices, the load of each device can be expressed as: ； in The adjustable portion of the load at time t. Let N be the regulating load of the i-th device, and N be the total number of devices. Taking into full account electricity consumption patterns, response delays, and load adjustability, a comprehensive user behavior model is constructed: ； in The total load or total power demand at time t. For the mode power or mode load at time t, The delayed power or delayed load at time t. The adjusted load or adjusted power is at time t.

2. The distributed collaborative management system for discrete small loads on the power grid consumer side according to claim 1, characterized in that, The process by which the load forecasting module processes historical electricity consumption data and historical meteorological data includes: Acquire historical electricity consumption data and historical meteorological data. The electricity consumption data includes active power, reactive power, voltage, current, frequency and power factor. The meteorological data includes ambient temperature, ambient humidity and solar radiation intensity. Generate a collection cycle and divide the collection cycle into multiple collection periods. The active power change rate of each small load node is obtained in multiple collection periods. The active power change rate represents the ratio between the change in active power and the duration of the corresponding time period. A set A of active power change rates is constructed in this way, and the average of the differences between the largest and smallest subsets in set A is recorded as the active power change rate difference. Similarly, the difference in the rate of change of reactive power, voltage, current, frequency, and power factor can be obtained using the same method as the difference in the rate of change of active power. The ambient temperature of each small load node is obtained during multiple collection periods, and the arithmetic mean of the multiple ambient temperatures is calculated. The arithmetic mean of the multiple ambient temperatures is recorded as the average ambient temperature. The average ambient humidity and average solar radiation intensity can be obtained by using the same method to calculate the average ambient temperature.

3. The distributed collaborative management system for discrete small loads on the power grid consumer side according to claim 2, characterized in that, The process by which the load forecasting module constructs an electricity-side load forecasting model that integrates meteorological data and uses the model to forecast the electricity-side load for a future period includes: Electricity consumption features are extracted from historical electricity consumption data, including differences in active power change rate, reactive power change rate, voltage change rate, current change rate, frequency change rate, and power factor change rate. Meteorological features are extracted from historical meteorological data, including average ambient temperature, average ambient humidity, and average solar irradiance. The extracted electricity consumption features and meteorological features are combined to form an input feature matrix X, which is used as the input to the machine learning model. The historical future load value corresponding to each feature matrix X is used as the output label y. The machine learning model is trained with the goal of minimizing the loss function between the predicted value and the actual value until the loss function converges, resulting in a trained electricity load prediction model. Real-time electricity consumption data and real-time meteorological data of each small load node are collected, the corresponding feature vectors are extracted and input into the trained electricity-side load prediction model, and the load prediction values ​​for the next H time steps are output. , where i=1,2,…,H, and H represents the prediction time step.

4. The distributed collaborative management system for discrete small loads on the power grid consumer side according to claim 1, characterized in that, The process by which the distributed collaborative control module processes the load forecasting results and user behavior analysis results includes: Obtain load forecasting results and user behavior analysis results. The load forecasting results are the load forecast values ​​for the next H time steps. The user behavior analysis results are and its three components , and ; In a distributed power consumption system composed of small load nodes, if the system has N nodes, and each node i has its local load at each time step t: ； And a predicted load sequence for the next H time steps. , t=1,2,…,H.

5. The distributed collaborative management system for discrete small loads on the power grid consumer side according to claim 4, characterized in that, The process by which the distributed collaborative control module makes local control decisions at each node using a distributed optimization algorithm includes: The global objective is to minimize the total cost function across all nodes, in the following form: ； ； in It is the sequence of control variables of node i over H time steps. It is the local cost function for each node. The penalty term is used to control the adjustment range of the node. Deviation term of control and prediction error; Each node maintains a local copy and a global variable copy. Each node i uses its local behavioral model to decompose the current and predicted loads, obtaining... The local optimization constraints for each node i are as follows: ； Global power supply capacity is limited to Then the global constraints are: 。 6. The distributed collaborative management system for discrete small loads on the power grid consumer side according to claim 5, characterized in that, The distributed collaborative control module solves the distributed optimization algorithm and achieves collaborative regulation among multiple nodes through limited information exchange. The process includes: The ADMM distributed optimization algorithm is adopted, in which each node independently solves its own local problem and coordinates global consistency through limited information exchange. The specific steps are as follows: S1: Local update, each node independently minimizes its own... Consider the current global estimate; S2: Information exchange, nodes only exchange necessary information: each node uploads its prediction adjustment value at each time step. ; The control center or neighboring nodes summarize and broadcast the updated global variables. ; S3: Global coordination, updating global consistency variables, updating multipliers, and driving each node to gradually converge to a consistent solution.

Images