A method and system for power system load forecasting based on agent simulation

Abstract

The present invention proposes a method and system for power system load forecasting based on intelligent agent simulation, which relates to the field of power systems. Based on real environmental information and a large language model, individual intelligent agents are constructed, and group categories are divided by clustering; the typical characteristics and behavior patterns of the group categories are input into the large language model to generate group electricity consumption behavior characteristics and patterns in different scenarios; the behavior probability distribution of each group category in different scenarios is estimated, and the probability distribution is sampled to obtain individual electricity consumption behavior; based on the individual electricity consumption behavior, the power system load forecast result is obtained through analysis. The present invention integrates LLMs into ABM, constructs millions of adaptive agents, simulates the diversity of user behavior and group effects, and combines environmental interactions to perform power load forecasting. While maintaining computational efficiency, it balances the complexity of agent behavior, so that detailed agent electricity consumption or charging behavior analysis can be achieved in large-scale simulations, providing support for power grid planning and management.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method and system for power system load forecasting based on intelligent agent simulation. Background Art

[0002] With the rapid development of smart grid technology and the widespread adoption of renewable energy, power system load forecasting faces new challenges and demands. In recent years, the widespread adoption of electric vehicles has also had a significant impact on power system load. The flexible and adjustable charging behavior of electric vehicles raises new challenges for grid operation and planning, posing new challenges to the temporal and spatial distribution of charging loads. Accurately forecasting the adjustable loads of electric vehicles and other loads is crucial for optimizing grid dispatch, increasing renewable energy integration, and reducing grid operating costs.

[0003] Traditional power system load forecasting methods mainly include time series analysis, regression analysis, neural network analysis, etc. These methods can predict load to a certain extent, but they often have some limitations when faced with complex and changing power systems and massive amounts of data:

[0004] (1) Difficulty in capturing the diversity of user behavior: Traditional methods usually assume that user behavior is fixed. For example, the invention patent with application number 2023109074762 focuses on power load forecasting by analyzing a large amount of historical power load data. It cannot consider the dynamic response of users to factors such as power consumption behavior, meteorological conditions, and economic activities.

[0005] (2) The calculation of individual electricity consumption behavior is complex: For example, in the invention application with application number 202410319798X, if it is necessary to accurately obtain the characteristics of individual electricity consumption behavior of users, it is necessary to systematically collect multi-dimensional data. Due to the large volume and diverse types of data, it not only consumes a lot of computing resources, but also has problems such as long processing cycle and low analysis efficiency, which greatly affects the timeliness and accuracy of electricity consumption behavior analysis.

[0006] (3) Lack of simulation of group behavior: Traditional methods have difficulty simulating the interactions between large-scale user groups and their combined impact on load.

[0007] (4) Insufficient adaptability to emerging technologies (such as electric vehicles): With the popularization of electric vehicles, the impact of charging behavior on electricity load is becoming increasingly significant, and traditional methods are difficult to accurately evaluate.

[0008] Agent-Based Modeling (ABM) is a method for studying complex systems by simulating the behavior and interactions of individual agents. In power system load forecasting, ABM can fully account for the complex interactions among various factors within the power system, such as electricity consumption behavior, meteorological conditions, and economic activities. By abstracting different electricity users in the power system (such as electric vehicle users, residential users, and agricultural users) into different agents and assigning them corresponding behavioral rules and decision-making mechanisms, it is possible to more accurately simulate electricity consumption habits and load trends. Furthermore, ABM offers excellent scalability and flexibility, enabling the easy introduction of new data sources and influencing factors to adapt to the ever-changing needs of the power system.

[0009] However, in real-world scenarios, ABMs are primarily used in applications such as epidemic spread and social media simulations, but are less commonly used in load forecasting for power system adjustable loads and electricity consumption. Even where such applications are used, ABMs rely on simplistic, rule-based agent behavior and fail to capture the nuances of individual real-world behaviors. Furthermore, the high computational cost of large-scale simulations limits the number of simulated populations, impacting forecasting effectiveness. Therefore, designing more rational large-scale agent simulation mechanisms to fully realize the potential of ABMs in power system load forecasting holds significant research and application value. Summary of the Invention

[0010] To address these issues, this paper proposes a power system load forecasting method and system based on agent-based simulation. By integrating large language models (LLMs) with ABM, millions of adaptive agents are constructed. By simulating the diversity and group effects of user (agent) behavior and integrating user-environment interactions, this method supports power system load forecasting for large user groups. This method balances the complexity of agent behavior while maintaining computational efficiency, enabling detailed analysis of agent power consumption or charging behavior in large-scale simulations, providing support for power grid planning and management.

[0011] In order to achieve the above object, the present invention adopts the following technical solutions:

[0012] In a first aspect, the present invention provides a method for power system load forecasting based on agent simulation, comprising:

[0013] Build an individual intelligent agent based on real environment information combined with a large language model; the real environment information includes user portraits and charging environment information;

[0014] Based on the behavioral characteristics simulated by individual agents and the real environment information in which they are located, a clustering algorithm is used to divide individual agents with similar characteristics into different group categories. Each group category has similar typical characteristics and behavioral patterns.

[0015] Input the typical characteristics and behavior patterns of group categories into the large language model to generate group electricity consumption behavior characteristics and patterns in different scenarios;

[0016] Use the Monte Carlo method to estimate the probability distribution of each group category's behavior in different scenarios, sample the probability distribution, and obtain individual electricity consumption behavior;

[0017] Based on individual electricity consumption behavior, the electricity load distribution at different time and space scales is analyzed to obtain the power system load forecast results.

[0018] In a second aspect, the present invention provides a power system load forecasting system based on agent simulation, comprising:

[0019] An individual agent building module is configured to build an individual agent based on real environment information combined with a large language model; the real environment information includes user portraits and charging environment information;

[0020] The large-scale agent construction and simulation module is configured to use a clustering algorithm to divide individual agents with similar characteristics into different groups based on the behavioral characteristics simulated by the individual agents and the real environment information in which they are located. Each group category has similar typical characteristics and behavioral patterns.

[0021] The individual electricity usage behavior mapping module is configured to input typical characteristics and behavior patterns of group categories into the large language model to generate group electricity usage behavior characteristics and patterns in different scenarios; use the Monte Carlo method to estimate the probability distribution of each group category's behavior in different scenarios, and sample the probability distribution to obtain individual electricity usage behavior;

[0022] The load forecasting module is configured to analyze the power load distribution at different time and space scales based on individual power consumption behavior and obtain the power system load forecast result.

[0023] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the power system load forecasting method based on intelligent agent simulation described in the first aspect.

[0024] In a fourth aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the method for power system load forecasting based on intelligent agent simulation described in the first aspect are implemented.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] (1) This invention constructs individual agents based on user profiles and charging environment information, and uses a clustering algorithm to classify individual agents with similar characteristics into groups. Relying on the powerful processing capabilities of large language models, group characteristics and patterns are input to generate group electricity usage behavior characteristics and patterns in different scenarios. The Monte Carlo method is then used to estimate and sample the behavior probability distribution to obtain individual electricity usage behavior. In this way, in large-scale simulations, the complexity of agent behavior and computational efficiency are successfully balanced, enabling detailed analysis of agent electricity or charging behavior. By analyzing the distribution of electricity load at different temporal and spatial scales, this provides solid support for grid planning and management, and improves the efficiency of power system load forecasting.

[0027] (2) The present invention can dynamically adjust the behavioral pattern of dynamic response to factors such as electricity consumption behavior, meteorological conditions, and economic activities according to different electricity consumption scenarios and environmental conditions, realize dynamic response to external environmental factors, capture the diversity of user behavior, and has strong adaptability.

[0028] (3) By using clustering methods to classify different groups, the present invention significantly reduces the number of calls to LLMs, thereby reducing the demand for computing resources and making it possible to simulate the behavior of large-scale electricity users. At the same time, by sampling the behavioral probability distribution, the complexity and diversity of individual behaviors can still be maintained.

[0029] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings, which constitute a part of the specification of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their description are used to explain the present invention but do not constitute a limitation of the present invention.

[0031] Figure 1 A main flow chart of a power system load forecasting method based on agent simulation provided by an embodiment of the present invention;

[0032] Figure 2 An overall framework diagram of a power system load forecasting method based on agent simulation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0034] Example 1

[0035] like Figure 1 As shown, this embodiment discloses a method for power system load forecasting based on agent simulation, comprising the following steps:

[0036] S1: Build an individual agent based on real-world environment information combined with a large language model; the real-world environment information includes user profiles and charging environment information;

[0037] S2: Based on the behavioral characteristics simulated by individual agents and the real environment information in which they are located, a clustering algorithm is used to divide individual agents with similar characteristics into different groups. Each group category has similar typical characteristics and behavioral patterns.

[0038] S3: Input the typical characteristics and behavior patterns of group categories into the large language model to generate group electricity consumption behavior characteristics and patterns in different scenarios;

[0039] S4: Use the Monte Carlo method to estimate the probability distribution of each group category's behavior in different scenarios, sample the probability distribution, and obtain individual electricity consumption behavior;

[0040] S5: Based on individual electricity consumption behavior, analyze the electricity load distribution at different time and space scales to obtain the power system load forecast results.

[0041] This embodiment discloses a method for power system load forecasting based on intelligent agent simulation, which can be used for forecasting adjustable loads such as electric vehicles, and can also be used for forecasting electricity consumption and loads of different categories such as residents and agriculture. This embodiment takes residential users as an example, and realizes electricity consumption and load forecasting at different time and space scales through three links: building a real environment, large-scale agent construction and simulation, and load forecasting. The electricity consumption behavior of residential users is affected by various factors such as daily living habits, seasonal changes, and electricity price policies. Through the ABM simulation method, the electricity consumption behavior patterns of residential users in different scenarios can be captured, and load forecasting can be performed accordingly. Next, combined with Figure 2 Provide detailed explanation.

[0042] 1. Real Environment Design

[0043] Construct a dynamic and real environment for electricity users, including building user portraits and collecting charging environment information.

[0044] 1. Building User Profiles

[0045] User portraits reflect the user's personal characteristics, preferences, and behavioral characteristics.

[0046] Among them, personal characteristics include age, gender, occupation, family size, economic status, and living habits (such as getting up early or going to bed late);

[0047] Preference characteristics include vehicle information (such as brand, model, battery capacity, etc.), charging habits (such as charging frequency, time, location preference, etc.), electricity usage habits (charging during the night when electricity prices are low), device usage habits (type and frequency of use of household appliances such as air conditioners, refrigerators, washing machines, etc.), electric vehicle charging attitude (sensitivity to electricity price fluctuations), etc.

[0048] Behavioral characteristics include users' historical electricity usage data.

[0049] These features can be collected from electric vehicle user registration information, historical charging records, vehicle management systems, and user questionnaires. For example, user A is 30 years old, male, an office worker, and has a medium income. He is sensitive to electric vehicle charging costs and usually charges at night when electricity prices are low.

[0050] For user profiles, at least one of the following characteristics can be extracted: personal characteristics, preference characteristics, and behavioral characteristics. For example, a 35-year-old office worker might leave home early and return late on weekdays, use the air conditioner and watch TV at night, use the washing machine and oven on weekends, and prefer to charge their phone between 10 pm and 2 am, often at home.

[0051] It should be understood that obtaining user portraits is achievable by those skilled in the art.

[0052] (2) Collecting charging environment information

[0053] The environment module provides the Agent with an interactive environment similar to the real world and provides a basis for decision-making.

[0054] The agent can perceive its own status (vehicle location, battery level) and the power usage environment (weather, electricity prices, economic activities, etc., the operating environment of electric vehicles (current traffic conditions), the operation and distribution of charging stations, etc.), the status of charging equipment, and interact with the environment through a series of actions.

[0055] Specifically, for adjustable load forecasting scenarios such as electric vehicle charging, the environment module mainly refers to the operating environment of electric vehicles and the operation of charging stations, including the energy consumption, battery status, charging efficiency of electric vehicles, as well as the location, number, price, and queue status of charging stations. Specific steps:

[0056] 1. Environment setup: Build an interactive environment similar to the real world, including the following information:

[0057] (1) Electricity price information: real-time updated time-of-use electricity prices, such as peak electricity prices, off-peak electricity prices, and possible seasonal electricity price adjustments.

[0058] (2) Weather information: Meteorological data such as temperature, humidity, rainfall, and wind speed, which affect residents’ living habits (e.g., the frequency of air conditioning use during high temperatures).

[0059] (3) Traffic conditions: For residential users who own electric vehicles, traffic congestion may affect their choice of charging time and location.

[0060] (4) Charging station and home appliance status: including the location, availability, charging standards, number of charging piles, operating status of home appliances, etc. of charging stations, providing a basis for the agent's decision-making.

[0061] 2. Agent Perception: The agent can perceive its own status (such as battery level and vehicle location) and environmental information (such as current electricity prices and charging station availability) in real time. For example, if the agent detects that electricity prices are low and there are idle charging stations nearby, it will consider whether to go there to charge.

[0062] In this embodiment, by interacting with real-world environmental information, the Agent can be provided with a basis for accurate decision-making. By constructing user profiles to collect personal, preference, and behavioral characteristics, and collecting charging environment information covering electricity prices, weather, and other aspects, the Agent can perceive and adapt to environmental changes. This solves the problem that existing technologies have difficulty capturing the diversity of user behavior and lack adaptability to emerging technologies. For example, users can adjust their electricity usage behavior based on real-time electricity prices and weather to achieve accurate load forecasting. At the same time, it lays the foundation for realizing power system load forecasting for large-scale user groups, assisting in grid planning and management, and improving the efficiency of power system operation.

[0063] 2. Large-scale Agent Construction and Simulation

[0064] (1) Individual Agent Construction

[0065] Individual agents based on large language models (LLMs) are constructed to simulate the electricity consumption behavior of electricity users. These agents involve various modules to create a robust dynamic system. Each module plays a key role in ensuring the agent can accurately predict power system load by modeling user profile information. These modules primarily include perception, dynamic decision-making, memory, action, feedback, and evaluation capabilities. Specifically, they are as follows:

[0066] 1. Perception

[0067] Collect and process real-time information related to electricity consumption and charging by electricity users, such as electricity prices, weather, economy, current traffic conditions, vehicle location, battery level, and availability of charging stations, to provide data support for the agent's decision-making.

[0068] Specifically, the agent collects real-time information through sensors or data interfaces. For example, the agent senses that the current battery charge is 20%, the vehicle is in the garage at home, the current electricity price is 0.3 yuan per kWh, the weather is clear, and traffic conditions are good.

[0069] 2. Dynamic decision-making capabilities

[0070] Obtain user portrait information from perception capability feedback, charging environment information, and historical information from memory capability, and input them into the LLMs model.

[0071] At each time step, based on its current status (such as battery level, location, time, etc.) and environmental information (such as electricity price, weather, availability of charging stations, etc.), combined with historical behavior patterns, and with the help of LLMs' natural language understanding and generation capabilities, the user's decision-making process is simulated to determine the individual's optimal electricity consumption decision.

[0072] Specifically, the agent combines user profiles (User A prefers charging at night during off-peak hours) with perception information and generates decisions through LLMs. For example, for a resident who is sensitive to electricity prices and owns an electric car, they can choose to charge their electric car during off-peak hours and decide whether to turn on the air conditioner based on weather conditions.

[0073] 3. Memory

[0074] Stores the user's historical electricity usage data and related environmental information, including short-term memory (such as electricity usage behavior in the past three days) and long-term memory (such as electricity usage behavior in the past week or longer), so that the agent can make more informed decisions based on historical experience.

[0075] 4. Ability to act

[0076] Based on the decision-making results of the dynamic decision-making capability, the agent executes corresponding power consumption behaviors and interacts with power-consuming devices, electric vehicles, charging stations, etc. in the environment, such as starting charging at a certain time and how long to turn on the air conditioner at a certain time. For example, the agent controls the vehicle to start charging at 10 pm and continue charging for 4 hours.

[0077] 5. Feedback and evaluation capabilities

[0078] After each electricity usage behavior is executed, the decision results are evaluated and reflected upon to analyze whether the decision meets the user's needs and expectations (such as whether charging is completed on time and whether costs are saved). If the user is satisfied with the charging result, the agent records this behavior as a success case; otherwise, based on the decision-making and planning capabilities of the large language model, other behaviors are executed.

[0079] By considering the user's satisfaction with the electricity consumption results and the consistency between the decision and the user's behavior, the agent's behavior strategy is continuously optimized.

[0080] It should be understood that the feedback and evaluation capabilities described above can be implemented by those skilled in the art based on existing technical means and logical methods. When evaluating decision results, clear quantitative indicators and judgment logic can be set to compare them with user needs and expectations.

[0081] In this embodiment, individual intelligent agents possess multiple capabilities that can simulate real-world user electricity usage. Perception acquires real-time information, dynamic decision-making leverages a large language model to combine multiple pieces of information to make optimal decisions, memory provides historical insights, action executes decisions, and feedback and evaluation optimizes behavioral strategies. This breaks through the limitations of traditional approaches that assume fixed user behavior and can consider users' dynamic responses to various factors. For example, residential users can flexibly adjust electricity usage according to different scenarios, accurately simulate complex electricity usage behaviors, improve load forecasting accuracy, and achieve precise analysis of power system loads, meeting grid planning and management requirements.

[0082] (2) Large-scale Agent Construction and Simulation

[0083] In power system load forecasting scenarios, it's often necessary to simulate the entire agent population to capture population effects and achieve load forecasting at multiple spatial and temporal scales. For different spatial dimensions, such as provinces, cities, and counties, each agent, time step, and specific behavior is used as a feature prompt to generate electricity usage behavior based on LLMs. This consumes a large amount of computing resources and cannot dynamically simulate the electricity usage behavior of the entire population.

[0084] Therefore, this example first uses a clustering method to cluster individual agents, generating statistically representative large-scale user groups. This allows the behavior of each agent category to be generated based on LLMs. For example, if electricity usage is determined solely by age and gender, then only different combinations of age and gender need to be considered as prompts. This approach maintains simulation scale while leveraging the LLMs' ability to capture adaptive agent behavior, balancing the expressiveness of individual agents with computational feasibility, thereby improving computational efficiency. The specific process is as follows:

[0085] 1. Typical characteristics and behavior construction of individual agents

[0086] Based on user portrait information, the characteristic dimensions of individual agents are extracted, including age, gender, occupation, charging habits, etc.

[0087] 2. Cluster analysis

[0088] Based on the aforementioned characteristic dimensions, a clustering algorithm is used to divide individual agents into several groups with similar characteristics. Each group has similar typical characteristics and behavior patterns. For example, users can be divided into groups such as office workers, taxi drivers, and freelancers.

[0089] 3. LLMs planning and decision-making

[0090] For each demographic, LLMs are used to generate patterns of electricity usage behavior in different scenarios. For example, for taxi drivers, LLMs can generate charging decisions during peak and off-peak hours, and at different charging station prices, based on factors such as their work patterns, charging needs, and price sensitivity. For office workers, LLMs can generate charging behavior patterns during weekday nights when electricity prices are low.

[0091] Monte Carlo methods are used to estimate the probability distribution of each class’s behavior under different scenarios. For example, for a certain class, LLMs can output a 70% probability of choosing to charge at home and a 30% probability of choosing to charge at a public charging station in a specific scenario.

[0092] Among them, the probability distribution is sampled to obtain individual electricity consumption behavior, including:

[0093] S401: Determine the number of individual agents, the number of group categories, the number of behavior queries, and the behavior set;

[0094] S402: Using the large language model and the Monte Carlo method, the same prompt query model is used multiple times to obtain behavior probability related results. The multiple results are aggregated and averaged to obtain the probability of each group category performing the behavior;

[0095] S403: Determine the specific electricity usage behavior of the individual using a random sampling method based on the group category to which the individual belongs and the corresponding group category behavior probability, and repeat the sampling continuously during the simulation process so that the individual behavior can be dynamically updated and consistent with the group probability distribution.

[0096] Specifically, for each category and each possible action, we need to estimate the probability of the action. Assume the following variables:

[0097] A: The total number of agents.

[0098] K: The total number of group categories.

[0099] M: The number of behavior queries for each group category.

[0100] Ψ: A set of behaviors, such as “whether to turn on the air conditioner” or “whether to charge the electric car”.

[0101] For each group category k and each behavior ψ, the probability of the behavior It can be estimated from the output of LLMs.

[0102] Use the Monte Carlo method to estimate probability: The Monte Carlo method estimates the probability distribution through multiple random sampling. The specific steps are as follows:

[0103] Initialization: For each group category k, initialize an empty sample set .

[0104] Multiple query LLMs: For each group category k, perform M query LLMs, each query uses the same prompt, but may get different outputs. Assume that the output of LLMs is a probability value p, which represents the probability of executing behavior ψ for the group category.

[0105] Collect samples: collect the probability value p obtained from each query into the sample set middle.

[0106] Calculate the average probability: for the sample set Take the average of all probability values ​​in and get the estimated probability of executing behavior ψ for this group category:

[0107] ;

[0108] in, is the probability value obtained from the i-th query.

[0109] During the simulation, each individual agent determines its own behavior based on the probability distribution of the behavior of the group category to which it belongs. The specific steps are as follows:

[0110] Determine the group category: According to the characteristics of individual agents, assign them to a certain category k.

[0111] Sampling behavior: Behavior probability distribution according to group category k , a random sampling method is used to assign specific behaviors to individual agents. For example, if the behavior is "whether to turn on the air conditioner", a random number is generated based on P(turn on air conditioner|k). If the random number is less than P(turn on air conditioner|k), the individual agent chooses not to turn on the air conditioner, otherwise it chooses to turn on the air conditioner.

[0112] At each step of the simulation, the above sampling process is repeated to determine the behavior of each individual agent at the current time step. In this way, the behavior of individual agents is dynamically updated during the simulation while maintaining consistency with the behavior probability distribution of the group class.

[0113] In this embodiment, the probability distribution and sampling method effectively balances computational efficiency and behavioral complexity. The Monte Carlo method is used to estimate the probability distribution of group category behavior. By querying a large language model multiple times and sampling, the powerful model is used to generate reasonable probabilities. Clustering is also used to reduce the number of model calls and computational costs. This solves the problem of high computational resource consumption in large-scale simulations, while maintaining computational efficiency while preserving the diversity of individual behaviors. For example, different groups have different probabilities of electricity consumption in different scenarios. Sampling to determine individual electricity consumption behavior makes large-scale simulations more realistic, thereby providing more reliable data for power system load forecasting.

[0114] 4. Individual Agent Behavior Sampling

[0115] During the simulation, each individual agent belonging to a specific category samples according to the probability distribution of behavior for that category, thereby determining its specific behavior. For example, a taxi driver belonging to a certain category decides whether to charge and which charging method to choose at a certain time based on the probability distribution of that category; or an office worker agent decides to start charging at 10 pm based on the group behavior probability distribution.

[0116] In this embodiment, a sampling strategy based on a group model is employed, significantly improving computational efficiency compared to traditional methods that calculate individual agents separately. Traditional methods are computationally complex and difficult to simulate large-scale group behavior. This sampling method, however, determines individual behavior based on the probability distribution of group behavior, reducing the need for repetitive computations. Individual behaviors are dynamically updated during the simulation process to align with the group probability distribution, preserving the complexity and diversity of individual behaviors while enabling large-scale simulation of electricity user behavior. This significantly addresses computational bottlenecks in existing technologies and improves prediction performance, providing efficient support for grid planning and management and enabling accurate prediction of power system loads.

[0117] 3. Load Forecasting

[0118] Based on the agent's behavioral decisions and interaction results, the power system load of the power user is predicted. By analyzing the power consumption behavior of each agent in different time periods, the power load curve and power consumption at different time and space scales are predicted, providing a basis for power grid load management and demand-side response. Specifically:

[0119] (1) Division of time and space scales

[0120] Time scale: Select a time scale such as hourly, daily, weekly, or monthly for load forecasting. For example, at the hourly scale, focus on changes in residential electricity consumption during different time periods (such as morning peak, lunch break, and evening peak); at the monthly scale, analyze changes in residential electricity load during different seasons (such as high temperatures in summer and heating in winter).

[0121] Spatial scale: Load forecasting is conducted based on cities, communities, households, etc. For example, at the city scale, the electricity consumption behavior of all residents is comprehensively considered to predict the electricity load of the entire city; at the community scale, the focus is on the changes in electricity load in a specific community.

[0122] (2) Load forecasting process

[0123] Based on the agent's behavioral decisions and interaction results, each agent's electricity usage behavior is analyzed over different time periods. For example, on an hourly scale, the frequency and duration of resident agents turning on devices like air conditioners and washing machines are counted each hour to calculate the resident user's electricity load for that hour.

[0124] like Figure 2 As shown in Figure 1, the electricity loads of different types of users are aggregated to obtain the total electricity load curves at different spatial scales. For example, at the city scale, the electricity loads of all residential users are added together to obtain the total electricity load curves of the city in different time periods.

[0125] It should be understood that the load curve is for illustration only, and its specific form and drawing can be achieved by those skilled in the art based on the principles described in this application and existing technologies, and does not affect the understanding of the technical solutions and principles of the present invention.

[0126] Load forecasting results provide a basis for grid load management and demand-side response. For example, based on predicted peak load periods, power generation plans can be adjusted in advance or demand-side response measures can be implemented, such as guiding residential users to use electricity during off-peak hours, to optimize grid operating efficiency and reduce costs.

[0127] For example, within a city, the charging load curve is predicted from 10 pm on weekdays to 2 am the following morning. The prediction results are provided to grid operators for load management and demand-side response. Based on the prediction results, grid operators adjust electricity pricing strategies, guide users to charge more efficiently, and optimize grid operation efficiency.

[0128] This specific embodiment addresses the limitations of existing agent-based modeling (ABM) in power system load forecasting and proposes an innovative solution. Specifically, an adaptive agent system is constructed by integrating large language models (LLMs) with dynamic environment interaction mechanisms. First, multi-dimensional data (such as user attributes, charging habits, and real-time electricity prices) is integrated through user profiling. This gives individual agents the ability to perceive, remember, and make dynamic decisions, forming a closed "perception-decision-action-feedback" loop. This dynamically optimizes agent behavior strategies. Furthermore, the natural language understanding and generation capabilities of LLMs are leveraged to simulate real-time user responses to electricity price fluctuations and weather changes, breaking through the limitations of traditional rule-based decision-making.

[0129] Secondly, to address the computational bottleneck of large-scale simulations, a group optimization strategy based on clustering and Monte Carlo methods is proposed: representative group categories are generated by clustering feature dimensions, group behavior probability distributions are generated using LLMs, and individual behavior diversification is achieved through sampling, thereby reducing computational costs while retaining individual differences.

[0130] Traditional methods struggle to simulate group behavior and its impact on load, nor can they accurately capture individual differences. This invention not only enables dynamic interaction between intelligent agents and the external environment (such as real-time electricity prices and charging station status), but also uses clustering and Monte Carlo methods to estimate the probability distribution of group behavior in different scenarios and then randomly sample individual electricity usage behaviors. This approach achieves computational efficiency while fully preserving the complexity and diversity of individual behaviors. This mapping from group behavior probabilities to individual behaviors overcomes traditional limitations and provides a more efficient and accurate approach to power system load forecasting.

[0131] Example 2

[0132] An individual agent building module is configured to build an individual agent based on real environment information combined with a large language model; the real environment information includes user portraits and charging environment information;

[0133] The large-scale agent construction and simulation module is configured to use a clustering algorithm to divide individual agents with similar characteristics into different groups based on the behavioral characteristics simulated by the individual agents and the real environment information in which they are located. Each group category has similar typical characteristics and behavioral patterns.

[0134] The individual electricity usage behavior mapping module is configured to input typical characteristics and behavior patterns of group categories into the large language model to generate group electricity usage behavior characteristics and patterns in different scenarios; use the Monte Carlo method to estimate the probability distribution of each group category's behavior in different scenarios, and sample the probability distribution to obtain individual electricity usage behavior;

[0135] The load forecasting module is configured to analyze the power load distribution at different time and space scales based on individual power consumption behavior and obtain the power system load forecast result.

[0136] Example 3

[0137] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the power system load forecasting method based on agent simulation as described in the first embodiment above are implemented.

[0138] Example 4

[0139] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of a method for power system load forecasting based on intelligent agent simulation as described in the first embodiment above are implemented.

[0140] The steps or modules involved in Examples 2 to 4 above correspond to those in Example 1. For detailed implementations, please refer to the relevant description of Example 1. The term "computer-readable storage medium" should be understood to mean a single medium or multiple media that includes one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and cause the processor to perform any method of the present invention.

[0141] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for power system load forecasting based on agent simulation, characterized in that: include: Build individual intelligent agents based on real environment information combined with large language models; The real environment information includes user portrait and charging environment information; Based on the behavioral characteristics simulated by individual agents and the real environment information in which they are located, a clustering algorithm is used to divide individual agents with similar characteristics into different group categories. Each group category has similar typical characteristics and behavioral patterns. Input the typical characteristics and behavior patterns of group categories into the large language model to generate group electricity consumption behavior characteristics and patterns in different scenarios; The Monte Carlo method is used to estimate the probability distribution of the behavior of each group category in different scenarios, and the probability distribution is sampled to obtain the individual electricity usage behavior. The sampling of the probability distribution to obtain the individual electricity usage behavior specifically includes: Determine the number of individual agents, the number of group categories, the number of behavior queries, and the set of behaviors; With the help of a large language model, using the Monte Carlo method, the same prompt query model is used multiple times to obtain results related to behavioral probabilities. The multiple results are aggregated and averaged to obtain the probability of each group category performing the behavior; Based on the group category to which the individual belongs and the corresponding group category behavior probability, the individual's specific electricity consumption behavior is determined by random sampling. The sampling is repeated continuously during the simulation process so that the individual behavior can be dynamically updated and consistent with the group probability distribution; Based on individual electricity consumption behavior, the electricity load distribution at different time and space scales is analyzed to obtain the power system load forecast results.

2. The method for power system load forecasting based on agent simulation according to claim 1, characterized in that: The user profile is constructed based on the user's electricity usage characteristics, which include personal basic information characteristics, preference characteristics and behavior characteristics; The charging environment information includes electricity price information, weather information, traffic conditions, charging station status and home appliance status.

3. The method for power system load forecasting based on agent simulation according to claim 1, characterized in that: The individual intelligent agent has the capabilities of perception, decision-making, memory, action, feedback and evaluation; among them, the perception capability is used to dynamically obtain real environmental information; the decision-making capability generates an individual preliminary electricity consumption decision by inputting real environmental information and historical individual electricity consumption behavior data into a large language model; the action capability is used to execute the preliminary electricity consumption decision; the memory capability is used to save the relevant electricity consumption behavior data for executing the preliminary electricity consumption decision as historical individual electricity consumption behavior; the feedback and evaluation capability is used to optimize the behavioral decision after executing the preliminary electricity consumption decision by considering the user's satisfaction with the electricity consumption results and the consistency between the decision and the user's behavior.

4. A power system load forecasting system based on agent simulation, characterized in that: include: The individual agent building module is configured to build an individual agent based on real environment information combined with a large language model; The real environment information includes user portrait and charging environment information; The large-scale agent construction and simulation module is configured to use a clustering algorithm to divide individual agents with similar characteristics into different groups based on the behavioral characteristics simulated by the individual agents and the real environment information in which they are located. Each group category has similar typical characteristics and behavioral patterns. The individual electricity usage behavior mapping module is configured to input typical characteristics and behavior patterns of group categories into the large language model to generate group electricity usage behavior characteristics and patterns in different scenarios; use the Monte Carlo method to estimate the probability distribution of each group category's behavior in different scenarios, and sample the probability distribution to obtain individual electricity usage behavior; wherein, sampling the probability distribution to obtain individual electricity usage behavior specifically includes: Determine the number of individual agents, the number of group categories, the number of behavior queries, and the set of behaviors; With the help of a large language model, using the Monte Carlo method, the same prompt query model is used multiple times to obtain results related to behavioral probabilities. The multiple results are aggregated and averaged to obtain the probability of each group category performing the behavior; Based on the group category to which the individual belongs and the corresponding group category behavior probability, the individual's specific electricity consumption behavior is determined by random sampling. The sampling is repeated continuously during the simulation process so that the individual behavior can be dynamically updated and consistent with the group probability distribution; The load forecasting module is configured to analyze the power load distribution at different time and space scales based on individual power consumption behavior and obtain the power system load forecast result.

5. The power system load forecasting system based on agent simulation according to claim 4, characterized in that: The user profile is constructed based on the user's electricity usage characteristics, which include personal basic information characteristics, preference characteristics and behavior characteristics; The charging environment information includes electricity price information, weather information, traffic conditions, charging station status and home appliance status.

6. The power system load forecasting system based on agent simulation according to claim 4, characterized in that: The individual intelligent agent has the capabilities of perception, decision-making, memory, action, feedback and evaluation; among them, the perception capability is used to dynamically obtain real environmental information; the decision-making capability generates an individual preliminary electricity consumption decision by inputting real environmental information and historical individual electricity consumption behavior data into a large language model; the action capability is used to execute the preliminary electricity consumption decision; the memory capability is used to save the relevant electricity consumption behavior data for executing the preliminary electricity consumption decision as historical individual electricity consumption behavior; the feedback and evaluation capability is used to optimize the behavioral decision after executing the preliminary electricity consumption decision by considering the user's satisfaction with the electricity consumption results and the consistency between the decision and the user's behavior.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the power system load forecasting method based on agent simulation according to any one of claims 1 to 3 are implemented.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the power system load forecasting method based on intelligent agent simulation according to any one of claims 1 to 3 are implemented.

Images