A multi-factor coupling considering agent electricity purchasing user load prediction method and system

Abstract

The application relates to a proxy power purchase user load prediction method and system considering multi-factor coupling. The method comprises the following steps: obtaining historical load data, electricity price data, meteorological data and calendar data of a region to be predicted, and preprocessing the data to obtain sample data and prediction input data; a basic load prediction model is constructed according to the sample data, and an initial load prediction value of a proxy power purchase user is obtained; a user structure correction coefficient is determined according to a power purchase mode selection utility and a reservation probability, and a unit load correction coefficient is determined according to a meteorological sensitive load change degree, a price elasticity modulation degree and a calendar correction result; and the initial load prediction value is jointly corrected by using the user structure correction coefficient and the unit load correction coefficient, so that a final load prediction value is obtained.

Description

Technical Field

[0001] This application relates to the field of power load forecasting technology, and in particular to a method and system for forecasting the load of electricity purchasers by agents, taking into account the coupling of multiple factors. Background Technology

[0002] With the advancement of electricity market reforms, industrial and commercial users who have not yet directly entered the electricity market are typically purchased by power grid companies on their behalf. For this agency-purchase business, it is necessary to predict the electricity load of the users being purchased by the agency during the forecast period in advance, during the formulation of monthly power purchase plans, load deviation management, and medium- and long-term electricity procurement. Therefore, the accuracy of the load forecast results for the agency-purchased users directly affects the rationality of the power purchase plan and the effectiveness of deviation risk control.

[0003] Existing load forecasting methods typically build time series models, regression models, or neural network models based on historical load data, and input external influencing factors such as meteorological data, calendar data, and electricity price data into the model to obtain load forecast results. These methods can achieve certain predictive effects in scenarios where the user base is relatively stable and external factors change gradually. However, for users who purchase electricity through agents, their user base is not always fixed. Some agents may be influenced by the difference between the average price of electricity purchased through agents and the average market reference price, choosing to switch to purchasing electricity from the market or electricity sales companies. Furthermore, these users typically do not immediately return to the original agent purchasing system, leading to changes in the size of the agent purchasing user base. If historical load data is still treated as a stable series for extrapolation, the forecast results are prone to deviating from the actual load level.

[0004] Furthermore, the load of electricity purchased through agents is also affected by meteorological, price, and calendar factors. Existing methods often treat meteorological, electricity price, and calendar data as independent input variables and overlay them, making it difficult to reflect the modulating effect of price factors on the response of meteorologically sensitive loads, and also failing to reflect the impact of calendar factors such as holidays on price sensitivity. When market price fluctuations, extreme temperatures, and holiday scenarios overlap, existing methods are prone to significant prediction biases due to insufficient characterization of the coupling relationships between multiple factors. Summary of the Invention

[0005] To at least partially overcome the problem in related technologies that do not fully consider the structural changes of the electricity purchaser group and the coupled effects of multiple factors such as weather, price, and calendar, resulting in low accuracy of load forecasting for electricity purchasers, this application provides a method and system for load forecasting for electricity purchasers that considers the coupling of multiple factors.

[0006] The proposed solution is as follows:

[0007] According to a first aspect of the embodiments of this application, a method for predicting the load of electricity purchased by agents, considering the coupling of multiple factors, is provided, including: Obtain electricity purchase-related data for the region to be predicted, and perform missing value imputation, outlier correction, and normalization on the electricity purchase-related data to obtain sample data and prediction input data; the electricity purchase-related data includes historical load data, electricity price data, meteorological data, and calendar data; A basic load forecasting model is constructed based on the sample data, and the forecast input data is input into the basic load forecasting model to obtain the initial load forecast value of the agent electricity purchaser during the forecast period. The utility of the electricity purchase method chosen by the agent electricity purchaser is determined, the retention probability of the agent electricity purchaser is calculated based on the utility of the choice, and the user structure correction coefficient is determined based on the retention probability; the electricity purchase method includes: remaining within the agent electricity purchase system, and transferring to the market electricity purchase or electricity sales company purchase; the user structure correction coefficient is used to characterize the degree of correction of the initial load forecast value of the agent electricity purchaser by the change in the size of the agent electricity purchaser group; The degree of change in weather-sensitive load is determined based on the meteorological data, the degree of price elasticity modulation corresponding to the weather-sensitive load is determined based on the electricity price data, and the degree of price elasticity modulation is corrected based on the calendar data. A unit load correction coefficient is determined based on the degree of change in weather-sensitive load and the corrected degree of price elasticity modulation. The unit load correction coefficient is used to characterize the degree of correction of the initial load forecast value of the agent electricity purchaser by the change in unit electricity intensity under the coupled influence of meteorological factors, price factors and calendar factors. Based on the user structure correction coefficient and the unit load correction coefficient, the initial load forecast value of the agent electricity purchase user is jointly corrected to obtain the final load forecast value of the agent electricity purchase user for the forecast period.

[0008] Preferably, the historical load data is divided into load data of users who purchase electricity through agents and load data of users who directly trade electricity, according to user type; The meteorological data includes daily average temperature data, maximum temperature data, and minimum temperature data; The calendar data includes date information with weekday tags, rest day tags, and holiday tags; The electricity price data includes the average price of electricity purchased through agents and the average market reference price.

[0009] Preferably, constructing a basic load forecasting model based on the sample data includes: Historical load characteristics, electricity price characteristics, meteorological characteristics, and calendar characteristics are extracted from the sample data; Using the historical load characteristics, electricity price characteristics, meteorological characteristics, and calendar characteristics as model inputs, and the load of the proxy electricity purchaser in the corresponding historical period as model output, the preset prediction model is trained to obtain the basic load prediction model. The preset prediction model includes any one of the following: statistical regression prediction model, time series prediction model, and neural network prediction model.

[0010] Preferably, determining the utility of the electricity purchase method for users who purchase electricity through an agent includes: To obtain price sensitivity coefficients and the non-price utility of electricity purchase agents within the electricity purchase agent system; Based on the average electricity purchase price through agents, the price sensitivity coefficient, and the non-price utility of electricity purchasers remaining within the electricity purchase system through agents, the retention option utility of electricity purchasers within the electricity purchase system through agents is calculated. Obtain the market reference average price, the non-price utility of agents purchasing electricity from switching to market-based electricity purchases or electricity sales companies, and the switching costs of agents purchasing electricity. Based on the aforementioned market reference average price, price sensitivity coefficient, non-price utility of the agent electricity purchaser for switching to market electricity purchase or electricity sales company purchase, and the switching cost of the agent electricity purchaser, calculate the switching choice utility of the agent electricity purchaser for switching to market electricity purchase or electricity sales company purchase.

[0011] Preferably, calculating the retention probability of the agent-purchased electricity user based on the selection utility includes: Based on the difference between the utility of the retention option and the utility of the transfer option, the utility difference of the electricity purchase method for the agent electricity purchaser is determined; The utility difference of the electricity purchase method is input into a preset binary discrete choice model to obtain the retention probability of the agent electricity purchase user to remain in the agent electricity purchase system; The retention probability is used to characterize the likelihood that a user who has not yet left the agent-purchased electricity system will continue to choose to remain in the agent-purchased electricity system during the current forecast period. When the retention option utility increases relative to the transfer option utility, the retention probability increases; when the transfer option utility increases relative to the retention option utility, the retention probability decreases.

[0012] Preferably, determining the user structure correction coefficient based on the retention probability includes: Obtain the cumulative retention rate of agent electricity purchase users in the initial forecast period, and use the cumulative retention rate of agent electricity purchase users in the initial forecast period as the benchmark rate in which all agent electricity purchase users are within the agent electricity purchase system; Obtain the cumulative retention rate of electricity purchasing agents from the previous forecast period; Based on the retention probability of the current forecast period, the cumulative retention ratio of the agent electricity purchaser in the previous forecast period is updated to obtain the cumulative retention ratio of the agent electricity purchaser in the current forecast period. The cumulative retention rate of agent-purchased electricity users in the current forecast period is determined as the user structure correction coefficient for the current forecast period.

[0013] Preferably, determining the degree of change in meteorological sensitive loads based on the meteorological data includes: Obtain the temperature forecast data and baseline temperature data for the period to be predicted, and calculate the predicted temperature deviation; The temperature rise response coefficient is determined by linear regression of historical load data and meteorological data, and the basic meteorological sensitive load is calculated based on the predicted temperature deviation and the temperature rise response coefficient. The basic meteorological sensitive load is used to characterize the degree of load change caused by changes in meteorological conditions during the forecast period.

[0014] Preferably, determining the price elasticity modulation level corresponding to weather-sensitive loads based on the electricity price data, and correcting the price elasticity modulation level based on the calendar data, includes: Obtain the benchmark electricity purchase price and the price elasticity coefficient of weather-sensitive loads; The degree of price elasticity modulation corresponding to the weather-sensitive load is determined based on the deviation of the average electricity purchase price by the agent from the benchmark electricity purchase price and the price elasticity coefficient of the weather-sensitive load. Determine the date type corresponding to the time period to be predicted based on the calendar data; When the date type indicates that the period to be predicted belongs to a holiday, the price elasticity modulation degree is attenuated and corrected according to the preset holiday elasticity attenuation coefficient to obtain the corrected price elasticity modulation degree. When the date type indicates that the period to be predicted does not fall within a holiday, the price elasticity modulation degree is used as the corrected price elasticity modulation degree.

[0015] Preferably, the unit load correction coefficient is determined based on the degree of change in the meteorologically sensitive load and the corrected degree of price elasticity modulation, including: Obtain the historical average meteorological sensitive load for the same period and the initial load forecast value of the agent electricity purchaser; Based on the basic meteorological sensitive load and the corrected price elasticity modulation degree, the corrected meteorological sensitive load for the forecast period is determined; The degree of change per unit load is determined based on the degree of change of the modified meteorological sensitive load relative to the historical average meteorological sensitive load for the same period. The unit load correction coefficient is determined based on the degree of unit load change and the initial load forecast value of the agent electricity purchaser; The unit load correction factor is used to correct the load component in the initial load forecast value of the agent electricity purchaser that is affected by the coupling of meteorological factors, price factors and calendar factors.

[0016] According to a second aspect of the embodiments of this application, a load forecasting system for agent-purchased electricity users that considers multi-factor coupling is provided, comprising: Processor and memory; The processor and memory are connected via a communication bus: The processor is used to call and execute the program stored in the memory; The memory is used to store a program, which is at least used to execute a method for predicting the load of electricity purchasers by agents, considering the coupling of multiple factors, as described in any of the above.

[0017] The technical solution provided in this application may include the following beneficial effects: This application first constructs a basic load forecasting model to obtain the initial load forecast values ​​for proxy electricity purchasers, and then introduces user structure correction coefficients and unit load correction coefficients for joint correction, so that the load forecasting process no longer relies solely on the extrapolation results of historical load sequences. On the one hand, the user structure correction coefficient can reflect the changes in the size of the proxy electricity purchaser group due to the shift to market electricity purchase or electricity sales company purchase, based on the utility and retention probability of proxy electricity purchasers in choosing different electricity purchase methods, thus reducing the forecasting bias caused by changes in the user base. On the other hand, the unit load correction coefficient can comprehensively consider the coupled effects of meteorological factors, price factors, and calendar factors on the unit electricity intensity of proxy electricity purchasers, reflecting the response differences of meteorologically sensitive loads under electricity price changes and different calendar scenarios. Therefore, this application can simultaneously correct the initial load forecast values ​​from two dimensions: changes in user group size and changes in unit electricity intensity, improving the accuracy and adaptability of the proxy electricity purchaser load forecast results under scenarios of price fluctuations, weather changes, and overlapping holidays.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] Figure 1 This is a flowchart illustrating a method for predicting the load of electricity purchased by agents, considering the coupling of multiple factors, provided in one embodiment of this application. Figure 2This is a schematic diagram of the structure of a load forecasting system for agent-purchased electricity users that considers the coupling of multiple factors, provided in one embodiment of this application.

[0021] Reference numerals: Processor-21; Memory-22. Detailed Implementation

[0022] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0023] Example 1 Figure 1 This is a flowchart illustrating a method for predicting the load of electricity purchased by an agent, considering the coupling of multiple factors, provided in one embodiment of this application. (Refer to...) Figure 1 A method for predicting the load of electricity purchased by agents, considering the coupling of multiple factors, includes: S11. Obtain electricity purchase-related data for the area to be predicted, and perform missing value filling, outlier correction, and normalization on the electricity purchase-related data to obtain sample data and prediction input data; the electricity purchase-related data includes historical load data, electricity price data, meteorological data, and calendar data; S12. Construct a basic load forecasting model based on sample data, and input the forecasting input data into the basic load forecasting model to obtain the initial load forecast value of the agent electricity purchaser during the forecasting period. S13. Determine the utility of the electricity purchase method for the agent-purchased electricity users, calculate the retention probability of the agent-purchased electricity users based on the selection utility, and determine the user structure correction coefficient based on the retention probability; the electricity purchase method includes: remaining within the agent-purchased electricity system, and transferring to the market or a power sales company for electricity purchase; the user structure correction coefficient is used to characterize the degree of correction of the initial load forecast value of the agent-purchased electricity users by the change in the size of the agent-purchased electricity user group. S14. Determine the degree of change in weather-sensitive loads based on meteorological data, determine the degree of price elasticity modulation corresponding to weather-sensitive loads based on electricity price data, and correct the degree of price elasticity modulation based on calendar data. Determine the unit load correction coefficient based on the degree of change in weather-sensitive loads and the corrected degree of price elasticity modulation. The unit load correction coefficient is used to characterize the degree of correction of the initial load forecast value of the agent electricity purchaser by the change in unit electricity intensity of the agent electricity purchaser under the coupled influence of meteorological factors, price factors and calendar factors. S15. Based on the user structure correction coefficient and the unit load correction coefficient, the initial load forecast value of the agent-purchased electricity user is jointly corrected to obtain the final load forecast value of the agent-purchased electricity user for the forecast period.

[0024] For ease of understanding, the following explains some key terms in this embodiment: Electricity purchase-related data: This refers to various raw data used for load forecasting, covering multiple dimensions of information influencing user electricity consumption behavior. This data typically requires preprocessing to ensure its quality and usability before being used for model training and prediction.

[0025] Basic load forecasting model: This refers to a model built based on historical data for preliminary forecasting of the load of customers purchasing electricity through agents. This model can capture the basic trends and periodic patterns of load changes, providing an initial forecast benchmark for subsequent refined adjustments.

[0026] Choice utility refers to the overall satisfaction or value perceived by electricity buyers when choosing between different electricity purchase methods. The magnitude of choice utility directly affects users' electricity purchase decisions.

[0027] Retention probability: This refers to the likelihood that users of the electricity purchasing agent system will continue to choose to remain within the system during the current forecast period. This probability reflects the dynamic trend of the user group size.

[0028] User structure correction factor: This refers to a multiplicative or additive factor used to adjust the initial load forecast value of agency-purchased electricity users. This factor is mainly used to reflect the changes in the size of the agency-purchased electricity user group due to user transfers out or in, thereby correcting the initial forecast value.

[0029] The degree of change in weather-sensitive load refers to the magnitude of load fluctuations caused by changes in weather conditions (such as temperature and humidity). This degree of change is a key indicator for assessing the impact of weather factors on electricity load.

[0030] Price elasticity modulation degree: refers to the degree to which weather-sensitive loads respond to changes in electricity prices. This modulation degree reflects the extent to which users adjust their electricity consumption behavior in response to changes in weather conditions under different electricity price levels.

[0031] Unit load correction factor: This refers to a multiplicative or additive factor used to adjust the unit electricity intensity variation in the initial load forecast of users who purchase electricity through an agent. This factor is mainly used to characterize the correction of the initial load forecast value by the change in the user's unit electricity intensity under the coupled influence of meteorological factors, price factors, and calendar factors.

[0032] Joint correction: This refers to the process of applying both the user structure correction factor and the unit load correction factor to the initial load forecast value for customers purchasing electricity through an agent. This correction allows for a more comprehensive consideration of changes in user group size and the combined effects of multiple factors, thereby improving the accuracy of the final load forecast.

[0033] This embodiment provides a method for load forecasting of electricity purchasers by agents, considering the coupling of multiple factors. First, it is necessary to acquire electricity purchase-related data for the area to be forecasted and preprocess this data. For example, electricity purchase-related data can be obtained from the power company's data platform or a third-party data service provider, including historical load data, electricity price data, meteorological data, and calendar data. During the data preprocessing stage, interpolation methods can be used to fill in missing values, such as linear interpolation or spline interpolation; statistical methods (such as the three-standard-deviation method) can be used to identify and correct outliers; and the max-min normalization method can be used to scale the data to a specific range to eliminate the influence of dimensions.

[0034] Secondly, a basic load forecasting model is constructed based on the processed sample data. For example, a multiple linear regression model can be used as the basic load forecasting model, employing historical load data, electricity price data, meteorological data, and calendar data as input variables, and historical load data of proxy electricity purchasers as output variables for model training. After training, the forecast input data for the period to be predicted is input into the model to obtain the initial load forecast values ​​for proxy electricity purchasers.

[0035] Furthermore, it is necessary to determine the utility of the electricity purchase method chosen by users of the agency-purchased electricity system, and calculate the retention probability based on this, thereby determining the user structure correction coefficient. For example, user preference information for different electricity purchase methods (such as remaining within the agency-purchased electricity system, transferring to the market, or purchasing from a retail electricity company) can be obtained through questionnaires or historical data analysis. A simple utility function can be set, for example, considering only price factors: when the average price of agency-purchased electricity is lower than the average market reference price, the retention utility is higher; conversely, the transfer utility is higher. Based on this utility difference, a simple logical judgment rule can be used to determine the retention probability; for example, when the retention utility is greater than the transfer utility, the retention probability is set to 1, otherwise it is 0. The user structure correction coefficient can be simply set as the ratio of the number of users expected to remain within the agency-purchased electricity system in the current forecast period to the initial number of users. This coefficient is used to reflect the impact of dynamic changes in the size of the agency-purchased electricity user group on the initial load forecast value.

[0036] Furthermore, the degree of change in weather-sensitive loads is determined based on meteorological data, the corresponding price elasticity modulation level is determined based on electricity price data, and this price elasticity modulation level is corrected based on calendar data to finally determine the unit load correction factor. For example, the impact of temperature changes on load can be estimated based on the simple linear relationship between historical temperature and load, thereby determining the degree of change in weather-sensitive loads. The price elasticity modulation level can be set to a fixed value; for example, when electricity prices rise, the response level of weather-sensitive loads decreases. Calendar data can be used to identify whether the forecast period is a holiday; for example, during holidays, the price elasticity modulation level can be simply reduced by a fixed percentage. The unit load correction factor can be calculated based on the degree of change in weather-sensitive loads and the corrected price elasticity modulation level; for example, the two can be simply multiplied or added together to reflect the combined impact of meteorological, price, and calendar factors on unit electricity intensity.

[0037] Finally, the initial load forecast values ​​for electricity purchased through agents are jointly corrected using the user structure correction factor and the unit load correction factor. For example, the initial load forecast value can be multiplied by the user structure correction factor, and then by the unit load correction factor, to obtain the final load forecast value for electricity purchased through agents during the forecast period. This joint correction method can comprehensively consider the changes in user group size and the coupling effects of multiple factors, making the forecast results closer to reality.

[0038] This application first constructs a basic load forecasting model to obtain the initial load forecast values ​​for proxy electricity purchasers, and then introduces user structure correction coefficients and unit load correction coefficients for joint correction, so that the load forecasting process no longer relies solely on the extrapolation results of historical load sequences. On the one hand, the user structure correction coefficient can reflect the changes in the size of the proxy electricity purchaser group due to the shift to market electricity purchase or electricity sales company purchase, based on the utility and retention probability of proxy electricity purchasers in choosing different electricity purchase methods, thus reducing the forecasting bias caused by changes in the user base. On the other hand, the unit load correction coefficient can comprehensively consider the coupled effects of meteorological factors, price factors, and calendar factors on the unit electricity intensity of proxy electricity purchasers, reflecting the response differences of meteorologically sensitive loads under electricity price changes and different calendar scenarios. Therefore, this application can simultaneously correct the initial load forecast values ​​from two dimensions: changes in user group size and changes in unit electricity intensity, improving the accuracy and adaptability of the proxy electricity purchaser load forecast results under scenarios of price fluctuations, weather changes, and overlapping holidays.

[0039] Example 2 In some embodiments, historical load data is divided into load data for users who purchase electricity through agents and load data for users who directly trade electricity, based on user type. Meteorological data includes daily average temperature data, maximum temperature data, and minimum temperature data; Calendar data includes date information with weekday labels, rest day labels, and holiday labels; Electricity price data includes the average price of electricity purchased through agents and the average market reference price.

[0040] Specifically, historical load data is subdivided into load data for agents purchasing electricity and load data for direct transaction users, aiming to more accurately identify and analyze the electricity consumption behavior patterns of agents purchasing electricity. Agents purchasing electricity and direct transaction users may differ significantly in terms of load characteristics, response to price signals, and policy impact. This distinction effectively avoids interference from the load characteristics of direct transaction users on load forecasting for agents purchasing electricity, allowing the basic load forecasting model to focus more on capturing the inherent load patterns of the agent purchasing electricity group, thereby improving the targeting and accuracy of forecasts. In practice, this classification can be identified and filtered through user registration information on the electricity trading platform, electricity purchase contract type, or historical transaction records.

[0041] Meteorological data is refined into daily average temperature data, maximum temperature data, and minimum temperature data to provide more comprehensive temperature information. Temperature is a key meteorological factor affecting electricity load, especially heating and cooling loads. Daily average temperature data reflects the overall temperature throughout the day, helping to capture persistent load variation trends; maximum and minimum temperature data can capture the instantaneous or peak-valley effects of extreme temperatures on the load, such as the surge in air conditioning load due to high summer temperatures or the increase in heating load due to low winter temperatures. By obtaining this refined temperature data, the impact of meteorological factors on the load can be quantified more precisely, especially when determining the degree of change in weather-sensitive loads, enabling a more accurate characterization of the load's response curve to temperature changes. This data is typically obtained from meteorological monitoring stations, meteorological service agencies, or historical meteorological databases.

[0042] Calendar data is defined as date information including weekday, rest day, and holiday labels. User electricity consumption patterns differ significantly across date types. For example, industrial and commercial loads are typically higher on weekdays, while residential load peaks at night; rest days and holidays may see higher residential loads throughout the day and lower industrial loads. Labeling date information helps models identify and learn load patterns across different date types, thus more accurately reflecting the impact of calendar factors on load during forecasting. This is crucial for subsequent adjustments to price elasticity modulation and for understanding user structure and unit load variations. These labels can be generated and maintained by querying national statutory holiday schedules, weekend definitions, and other relevant data.

[0043] Electricity price data is broken down into the average price paid by agents and the average market reference price. The average price paid by agents is the actual price paid by users purchasing electricity through agents, directly impacting their electricity costs and behavior. The average market reference price is the price that users purchasing electricity through agents refer to when considering switching to market-based or retail electricity purchases, serving as a key basis for their electricity purchase method decisions. By distinguishing between these two types of prices, the utility of purchasing electricity through agents can be calculated more accurately, thereby determining the retention probability and user structure correction coefficient. Simultaneously, this electricity price data is also an important input for determining the degree of price elasticity modulation corresponding to weather-sensitive loads, reflecting the impact of price factors on user electricity intensity. The average price paid by agents is usually published by the power grid company or government departments, while the average market reference price can be obtained from electricity trading platforms or market quotations.

[0044] Example 3 In some embodiments, a basic load forecasting model is constructed based on sample data, including: Historical load characteristics, electricity price characteristics, meteorological characteristics, and calendar characteristics are extracted from the sample data; Using historical load characteristics, electricity price characteristics, meteorological characteristics, and calendar characteristics as model inputs, and the load of proxy electricity purchasers in the corresponding historical period as model output, the preset prediction model is trained to obtain the basic load prediction model. The preset prediction model includes any one of the following: statistical regression prediction model, time series prediction model, and neural network prediction model.

[0045] Specifically, historical load characteristics, electricity price characteristics, meteorological characteristics, and calendar characteristics are extracted from the sample data to transform raw electricity purchase-related data into structured inputs that can be used by predictive models for learning. Historical load characteristics can include load values, load change rates, peak-valley load ratios, etc., for specific time periods in the past (such as the previous hour, day, or week). These characteristics reflect the time-series dependence and periodicity of the load itself. Electricity price characteristics can include historical average electricity purchase prices, market reference average prices, electricity price fluctuations, and deviations from benchmark prices, used to capture the impact of electricity prices on user electricity consumption behavior. Meteorological characteristics can include daily average temperature data, maximum temperature data, minimum temperature data, humidity, wind speed, precipitation, etc., which are closely related to users' electricity needs for heating and cooling. Calendar characteristics can include date type (such as weekday labels, rest day labels, holiday labels), season, month, day of the week, etc., used to reflect the differences in user electricity consumption patterns under different date attributes. The extraction of these features typically involves techniques such as data aggregation, time window sliding, and feature engineering to ensure that the features can comprehensively and effectively characterize the driving factors of load changes.

[0046] Subsequently, the historical load characteristics, electricity price characteristics, meteorological characteristics, and calendar characteristics are used as model inputs, and the load of proxy electricity purchasers for the corresponding historical periods is used as model output to train the preset prediction model, thereby obtaining the basic load prediction model. This process adopts a supervised learning paradigm, that is, the model establishes its predictive ability by learning the mapping relationship between historical input characteristics and actual load output. During training, the model adjusts its internal parameters according to the preset optimization objective (e.g., minimizing prediction error) until it reaches convergence or meets the preset performance indicators. In this way, the basic load prediction model can learn and generalize the inherent patterns of load changes of proxy electricity purchasers from a large amount of historical data.

[0047] The preset prediction model can include any one of the following: statistical regression prediction model, time series prediction model, and neural network prediction model. Statistical regression prediction models, such as multiple linear regression, ridge regression, or support vector regression, predict by establishing a linear or nonlinear functional relationship between input features and load. Their advantages include relatively simple model structure and strong interpretability. Time series prediction models, such as autoregressive integral moving average (ARIMA), seasonal ARIMA (SARIMA), or exponential smoothing models, focus on capturing the time dependence, trend, and seasonality of load data itself, and are suitable for data with obvious time-series characteristics. Neural network prediction models, such as multilayer perceptron (MLP), recurrent neural networks (RNN) and their variants long short-term memory networks (LSTM), and gated recurrent units (GRU), have strong nonlinear fitting capabilities and self-learning capabilities, and can handle complex nonlinear relationships and high-dimensional data, especially suitable for capturing complex coupling effects between multiple factors. The most suitable preset model type can be selected based on the characteristics and complexity of the actual data, as well as the requirements for model interpretability and prediction accuracy.

[0048] Example 4 In some embodiments, determining the utility of a user's choice of electricity purchase method through an agent includes: To obtain price sensitivity coefficients and the non-price utility of electricity purchase agents within the electricity purchase agent system; Based on the average electricity purchase price through agents, the price sensitivity coefficient, and the non-price utility of electricity purchasers remaining within the electricity purchase system through agents, the retention option utility of electricity purchasers within the electricity purchase system through agents is calculated. Obtain the market reference average price, the non-price utility of agents purchasing electricity from switching to market-based electricity purchases or electricity sales companies, and the switching costs of agents purchasing electricity. Based on the market reference average price, price sensitivity coefficient, non-price utility of agents purchasing electricity from switching to market-based or retail electricity purchases, and switching costs of agents purchasing electricity, the switching utility of agents purchasing electricity from switching to market-based or retail electricity purchases is calculated.

[0049] The price sensitivity coefficient is used to quantify the responsiveness of electricity purchase agents to changes in electricity prices. This coefficient can be obtained through statistical analysis of historical electricity purchase behavior data, regression modeling, or market research questionnaires. For example, the price sensitivity coefficient can be fitted by analyzing changes in user purchase volume or purchase method choices under different electricity price levels. This coefficient can be an average for all electricity purchase agents or can be segmented based on user type, electricity consumption, and other characteristics. The non-price utility of remaining within the electricity purchase agent system refers to the perceived value or satisfaction users gain from choosing to remain within the system, excluding price factors. This may include the convenience of the purchase process, the stability of power supply, trust in the electricity purchase agent, the quality of customer service, and avoiding the uncertainty of switching. This non-price utility can be determined through user satisfaction surveys, expert experience assessments, or analysis of the impact of non-price factors on user choices based on historical data.

[0050] Secondly, based on the average electricity purchase price through the agent, the price sensitivity coefficient, and the non-price utility of remaining within the agent electricity purchase system, the retention option utility for users is calculated. The average electricity purchase price refers to the average electricity price actually paid by users within the agent electricity purchase system. This data is directly derived from electricity price data in electricity purchase-related data and is one of the core price factors considered by users when evaluating electricity purchase methods. Retention option utility is the overall attractiveness or satisfaction of users choosing to remain within the agent electricity purchase system after comprehensively considering both price and non-price factors. Its calculation typically uses a utility function. For example, the non-price utility can be subtracted from the product of the price sensitivity coefficient and the average electricity purchase price to quantify the negative impact of price on utility. The specific function form can be adjusted based on actual data and model fitting results.

[0051] Secondly, we need to obtain the market reference average price, the non-price utility of agents purchasing electricity from switching to market-based or retail electricity purchases, and the switching costs for agents. The market reference average price refers to the average electricity price that agents purchasing electricity might face if they choose to switch to market-based or retail electricity purchases. This data also comes from electricity price data in electricity purchase-related data, providing an important price reference for users when comparing different electricity purchase methods. The non-price utility of agents purchasing electricity from switching to market-based or retail electricity purchases refers to the perceived value or satisfaction gained by agents purchasing electricity from choosing to switch to market-based or retail electricity purchases, besides price factors. This may include the flexibility of market-based electricity purchases, the diversity of choices, potential lower prices (even if uncertain), or value-added services offered by specific retail electricity companies. This utility can also be obtained through market research, user interviews, or analysis of market-based electricity purchaser behavior. The switching costs for agents purchasing electricity refer to the various costs incurred by agents purchasing electricity from the agent system to market-based or retail electricity purchases. This includes not only direct economic costs such as potential administrative fees and contract penalties, but also the time cost for users to learn the new electricity purchase process, the psychological cost of adapting to a new service provider, and the perceived risk of uncertainty regarding future electricity purchases. The quantification of switching costs can be determined through historical data analysis, user surveys, or expert assessments.

[0052] Finally, based on the market reference average price, price sensitivity coefficient, non-price utility of switching from market-based to market-based or retail-based electricity purchases for agent-purchased electricity users, and switching costs for agent-purchased electricity users, the switching-out utility for agent-purchased electricity users is calculated. Switching-out utility is the overall attractiveness or satisfaction of agent-purchased electricity users choosing to switch to market-based or retail-based electricity purchases after comprehensively considering price factors, non-price factors, and switching costs. Its calculation also uses a utility function, expressed as: UA = -β1 × λagent + SA; UB=-β1×λmarket+SB-Cswitch; Wherein, UA represents the retention utility of the agent-purchased electricity user within the agent-purchased electricity system; UB represents the transfer utility of the agent-purchased electricity user to purchase electricity from the market or a retail electricity company; λagent represents the average agent-purchased electricity price; λmarket represents the average market reference price; β1 represents the price sensitivity coefficient; SA represents the non-price utility of the agent-purchased electricity user for remaining within the agent-purchased electricity system; SB represents the non-price utility of the agent-purchased electricity user for transferring to purchase electricity from the market or a retail electricity company; and Cswitch represents the user's switching cost.

[0053] Furthermore, the retention probability of the agent-purchased electricity user is calculated based on the selection utility, including: The difference in utility between the retention option and the transfer option is determined based on the difference in utility between the electricity purchase methods of the agent-purchased electricity users; By inputting the difference in utility of electricity purchase methods into a preset binary discrete choice model, the retention probability of users who purchase electricity through agents remains within the agent electricity purchase system is obtained. The retention probability is used to characterize the likelihood that users who have not yet switched out of the agent-purchased electricity system will continue to choose to remain in the agent-purchased electricity system during the current forecast period. When the utility of the retention option increases relative to the utility of the switch-out option, the retention probability increases; when the utility of the switch-out option increases relative to the utility of the retention option, the retention probability decreases.

[0054] Specifically, this utility difference in electricity purchase methods aims to quantify the relative attractiveness of different electricity purchase methods for users of the agency-purchase system (i.e., remaining within the agency system or transferring to the market or a retail electricity company). Typically, this can be obtained by comparing the utility of the retention option with the utility of the transfer option, or by calculating their difference. For example, the utility of the retention option can be directly calculated by subtracting the utility of the transfer option from the utility of the retention option, yielding a value reflecting the user's net preference. The larger this difference value, the more inclined the user is to remain within the agency system; conversely, the smaller the difference, the more inclined they are to transfer.

[0055] Subsequently, the utility difference of the electricity purchase method is input into a preset binary discrete choice model to obtain the retention probability of the electricity purchaser remaining in the agent-purchase system. The preset binary discrete choice model is a statistical model whose core function is to transform the continuous or discrete input variable of the utility difference of the electricity purchase method into a probability value between 0 and 1, predicting the likelihood of a user choosing one of the electricity purchase methods. Typical binary discrete choice models include the Logit model or the Probit model. Taking the Logit model as an example, it can take the utility difference of the electricity purchase method as input and map it to a retention probability through a sigmoid function (such as the sigmoid function). This model is usually trained based on historical user behavior data to determine the model parameters, thereby accurately reflecting the user's choice tendency when faced with different utility differences. Through this model, the abstract utility difference can be transformed into a quantifiable retention probability, providing direct input for subsequent load forecasting.

[0056] The retention probability characterizes the likelihood that users who have not yet switched out of the proxy electricity purchase system will continue to choose to remain within the system during the current prediction period. This probability is dynamic and closely related to users' relative preferences for different electricity purchase methods. When the utility of the retention option increases relative to the utility of the switch-out option, it means that users find staying within the proxy electricity purchase system more attractive, and therefore their retention probability increases accordingly. Conversely, when the utility of the switch-out option increases relative to the utility of the retention option, users are more inclined to leave, and the retention probability decreases. This mechanism ensures that the retention probability output by the model accurately captures the dynamic changes in user behavior.

[0057] The probability that a user will choose to purchase electricity through the agent is: Pagent=e^UA / (e^UA+e^UB)=1 / (1+e^(UB-UA)) Poor simplification efficiency: ΔU=UB-UA=β1(λagent-λmarket)+(SB-SA)-Cswitch Let ΔS = SB - SA - Cswitch be the net switching cost parameter, then: Pagent=1 / (1+e^(β1(λagent-λmarket)+ΔS)).

[0058] Furthermore, user structure correction coefficients are determined based on retention probabilities, including: Obtain the cumulative retention rate of agent-purchased electricity users in the initial forecast period, and use the cumulative retention rate of agent-purchased electricity users in the initial forecast period as the benchmark rate for all agent-purchased electricity users to be within the agent-purchased electricity system; Obtain the cumulative retention rate of electricity purchasing agents from the previous forecast period; Based on the retention probability of the current forecast period, the cumulative retention ratio of the agent electricity purchaser in the previous forecast period is updated to obtain the cumulative retention ratio of the agent electricity purchaser in the current forecast period. The cumulative retention rate of electricity purchase agents in the current forecast period is determined as the user structure correction coefficient for the current forecast period.

[0059] Specifically, when obtaining the cumulative retention rate of proxy electricity purchasers for the initial forecast period and using this rate as a benchmark rate indicating that all proxy electricity purchasers are within the proxy electricity purchase system, this step aims to establish a baseline for changes in the user structure. The cumulative retention rate for proxy electricity purchasers in the initial forecast period is typically set to 100% or 1.0, meaning that at this initial point, all considered proxy electricity purchasers are considered to be retained within the proxy electricity purchase system. This benchmark rate serves as the starting point for subsequent calculations of user structure correction coefficients, ensuring the continuity and traceability of the correction process. For example, at the initial system startup or a historical baseline, this rate can be set to 1.0 to represent the complete user group.

[0060] Subsequently, the cumulative retention rate of proxy electricity purchasers in the previous forecast period is obtained. This step is used to obtain the user structure status of the previous time period. The cumulative retention rate of proxy electricity purchasers in the previous forecast period is calculated based on historical data and the retention probability of the previous forecast period. It reflects the actual size of the proxy electricity purchaser group relative to the initial baseline ratio as of the end of the previous period. This ratio serves as the input for the user structure correction calculation in the current period, reflecting the cumulative effect of user structure changes.

[0061] Based on this, the cumulative retention rate of proxy electricity purchasers in the previous forecast period is updated according to the retention probability of the current forecast period, resulting in the cumulative retention rate of proxy electricity purchasers in the current forecast period. This step is crucial for dynamically updating the user structure. The retention probability of the current forecast period represents the likelihood that existing proxy electricity purchasers will continue to remain within the proxy electricity purchase system in the current period. By multiplying the cumulative retention rate of proxy electricity purchasers in the previous forecast period by the retention probability of the current forecast period, the cumulative retention rate of proxy electricity purchasers in the current forecast period can be obtained. This multiplicative update mechanism accurately reflects the cumulative change in the user group size over time. For example, if the cumulative retention rate in the previous period was 0.9 and the retention probability in the current period is 0.95, then the cumulative retention rate in the current period is updated to 0.9. 0.95 = 0.855.

[0062] Finally, the cumulative retention rate of proxy electricity purchasers in the current forecast period is determined as the user structure correction factor for the current forecast period. This step directly uses the calculated cumulative retention rate of proxy electricity purchasers in the current forecast period as the user structure correction factor. This correction factor is a value between 0 and 1, quantifying the percentage of the proxy electricity purchaser group size relative to the initial baseline. This factor will be directly applied to the initial load forecast value of proxy electricity purchasers to correct for load deviations caused by changes in user structure.

[0063] It should be noted that, considering the unidirectional transfer characteristic under policy constraints, the structural correction coefficient βstruct reflects the cumulative retention ratio of the agent-purchased electricity user group. Let βstruct(0) = 1 at the initial time, indicating that all users are in the agent-purchased electricity system at the initial time.

[0064] For the t-th prediction period, the structural correction coefficient is recursively calculated using the following formula: βstruct(t)=βstruct(t-1)×Pagent(t) Here, Pagent(t) is the retention probability calculated based on the current electricity price in period t, representing the conditional probability that users who have not yet switched out will continue to choose to remain in the agent-purchased electricity system under the current market environment. Policy constraints require users who switch out to accept a higher punitive price if they switch back; therefore, βstruct(t) ≤ βstruct(t-1) holds, reflecting the monotonically non-increasing characteristic of the group size. Here, the individual retention probability from discrete choice theory is directly applied to the recursion of the cumulative retention ratio of the group. Its economic meaning is that in each period, all existing agent-purchased electricity users independently make retention decisions with the probability of Pagent(t), and users who switch out permanently leave the selection pool, thus forming a cumulative shrinkage effect of the group.

[0065] Example 5 In some embodiments, determining the degree of change in meteorological sensitive loads based on meteorological data includes: Obtain the temperature forecast data and baseline temperature data for the period to be predicted, and calculate the predicted temperature deviation; The temperature rise response coefficient is determined by linear regression of historical load data and meteorological data. The basic meteorological sensitive load is calculated based on the predicted temperature deviation and the temperature rise response coefficient. The basic meteorological sensitive load is used to characterize the degree of load change caused by changes in meteorological conditions during the forecast period.

[0066] Specifically, the process involves acquiring temperature forecast data and baseline temperature data for the period to be predicted, and then calculating the predicted temperature deviation. The temperature forecast data for the period to be predicted is typically obtained from professional third-party meteorological services or internal meteorological forecasting models, and its granularity can be consistent with the load forecast period, such as hourly or daily temperature forecasts. The baseline temperature data can be set as the average temperature for the same historical period (e.g., the same date over the past five years), or determined based on a threshold temperature that the load is insensitive to. The predicted temperature deviation is calculated as the difference between the temperature forecast data and the baseline temperature data for the period to be predicted; this difference directly reflects the degree of deviation of meteorological conditions from normal conditions during the forecast period.

[0067] Based on this, the temperature rise response coefficient is determined through linear regression of historical load data and meteorological data. The determination of the temperature rise response coefficient aims to quantify the sensitivity of the load of electricity purchased by agents to temperature changes. Specifically, historical load data (especially data from electricity purchased by agents) and meteorological data (such as daily average temperature data, maximum temperature data, and minimum temperature data) can be statistically analyzed. By constructing a linear regression model, with historical load as the dependent variable and temperature as the independent variable, a linear relationship between load and temperature is fitted. The slope of this linear regression model is the temperature rise response coefficient, which represents the amount of load change caused by each unit change in temperature. To improve accuracy, regression models can be established for different seasons, time periods, or temperature ranges to capture the nonlinear characteristics of the load's response to temperature.

[0068] Subsequently, the basic meteorological sensitive load is calculated based on the predicted temperature deviation and the temperature rise response coefficient. The basic meteorological sensitive load is calculated by multiplying the previously obtained predicted temperature deviation by the temperature rise response coefficient. For example, if the predicted temperature deviation is positive and the temperature rise response coefficient is positive (indicating that the load increases with temperature), the calculated basic meteorological sensitive load is positive, indicating that the load will increase due to the temperature rise. Conversely, if the predicted temperature deviation is negative, the basic meteorological sensitive load is negative, indicating that the load will decrease due to the temperature decrease. This calculation result directly quantifies the load increase or decrease caused solely by changes in meteorological conditions (mainly temperature) during the forecast period. The basic meteorological sensitive load characterizes the degree of load change caused by changes in meteorological conditions during the forecast period, ensuring that the impact of meteorological factors on the load can be accurately isolated and quantified during subsequent load forecast correction processes.

[0069] The basic meteorological sensitive load Qweather_base is calculated based on the difference between the temperature forecast value Tforecast and the base temperature Tbase for the period to be predicted. Qweather_base=γ×max(Tforecast-Tbase,0) Where γ is the temperature rise response coefficient, obtained through linear regression of temperature and load from historical data. Tbase is the reference temperature, which can be the average temperature of spring and autumn.

[0070] Furthermore, the degree of price elasticity modulation corresponding to weather-sensitive loads is determined based on electricity price data, and the degree of price elasticity modulation is corrected based on calendar data, including: Obtain the benchmark electricity purchase price and the price elasticity coefficient of weather-sensitive loads; The degree of price elasticity modulation corresponding to weather-sensitive loads is determined based on the deviation of the average electricity purchase price by the agent from the benchmark electricity purchase price and the price elasticity coefficient of weather-sensitive loads. Determine the date type corresponding to the time period to be predicted based on calendar data; When the date type indicates that the period to be predicted belongs to a holiday, the price elasticity modulation degree is attenuated and corrected according to the preset holiday elasticity attenuation coefficient to obtain the corrected price elasticity modulation degree. When the date type indicates that the period to be predicted does not fall within a holiday, the degree of price elasticity modulation is used as the corrected degree of price elasticity modulation.

[0071] Specifically, this includes: First, obtaining the benchmark electricity purchase price and the price elasticity coefficient for weather-sensitive loads. The benchmark electricity purchase price can be set as the average electricity purchase price over a historical period, such as the average agent purchase price over the past year or several years, or the standard purchase price determined by regulatory authorities. Its function is to provide a reference point for measuring the degree of deviation of the current electricity price. The price elasticity coefficient for weather-sensitive loads is a quantitative indicator used to describe the responsiveness of weather-sensitive loads to changes in electricity prices. For example, when the electricity price increases by 1%, the weather-sensitive load may decrease by 0.X%. This coefficient can be obtained by performing multiple regression analysis or constructing econometric models on historical load data, electricity price data, and meteorological data to reflect the sensitivity of users to electricity prices under different weather conditions.

[0072] Based on this, the degree of price elasticity modulation corresponding to weather-sensitive loads is determined according to the deviation of the average agent purchase price from the benchmark purchase price and the obtained price elasticity coefficient of weather-sensitive loads. Specifically, the percentage deviation between the average agent purchase price and the benchmark purchase price can be calculated, and then this deviation value can be multiplied by the price elasticity coefficient of weather-sensitive loads to obtain a percentage or proportion value, which represents the degree of adjustment or influence of the current electricity price on weather-sensitive loads. For example, if the average agent purchase price is higher than the benchmark purchase price and the price elasticity coefficient is negative, then the degree of price elasticity modulation will reflect a suppressive effect on weather-sensitive loads.

[0073] Simultaneously, the date type corresponding to the period to be predicted is determined based on calendar data. Calendar data includes labels indicating whether a date is a weekday, weekend, or public holiday. By querying or matching dates for the period to be predicted, the date type can be accurately identified. This step forms the basis for subsequent adjustments, as user electricity consumption patterns and price sensitivity differ significantly across different date types.

[0074] When the date type indicates that the forecast period falls within a holiday, the price elasticity modulation level is attenuated and corrected according to a preset holiday elasticity attenuation coefficient to obtain the corrected price elasticity modulation level. The holiday elasticity attenuation coefficient is a preset value between 0 and 1, such as 0.5 or 0.8, used to reduce the impact of price elasticity on load during holidays. During holidays, users may be more inclined to maintain comfort or engage in leisure activities, and their response to electricity price changes is less sensitive than on weekdays. Therefore, the attenuation coefficient can more accurately reflect this behavioral characteristic. For example, if the calculated price elasticity modulation level is -5% and the holiday attenuation coefficient is 0.6, the corrected modulation level becomes -3%. When the date type indicates that the forecast period does not fall within a holiday, the price elasticity modulation level is directly used as the corrected price elasticity modulation level, i.e., no attenuation processing is performed.

[0075] In this embodiment, the price elasticity of demand E for weather-sensitive loads (mainly referring to temperature-controlled loads such as air conditioning) is defined as (usually a negative value, such as E∈[-0.5,-0.1]), representing the percentage change in weather-sensitive loads for every 1% increase in electricity price. It is important to note that E here is not the self-price elasticity of total electricity demand at the macro level, but rather an elasticity coefficient specifically for the weather-sensitive component. This coefficient is obtained by quantile regression of temperature-sensitive loads and electricity price fluctuations in historical data or by a panel data model. Since temperature-controlled loads are significantly more sensitive to price than rigid loads such as industrial production lines and lighting, using a dedicated elasticity coefficient for weather-sensitive loads can more accurately characterize the moderating effect of electricity prices on temperature-driven electricity consumption behavior. The load adjustment ratio Rprice is calculated when the average agent purchase price deviates from the benchmark purchase price λbase: Rprice=1+E×(λagent-λbase) / λbase Wherein, λbase is the benchmark electricity purchase price, which can be the average agent purchase price over the past 12 months.

[0076] During holidays, users spend more time at home, increasing their essential electricity needs and reducing their sensitivity to electricity prices. A holiday elasticity attenuation coefficient θholiday∈(0,1) is defined, representing the percentage reduction in the absolute value of elasticity during holidays.

[0077] The corrected elasticity value is: Edaytype=E×θholiday (The day to be predicted is a public holiday); Edaytype=E (The day to be predicted is a working day); The price elasticity adjustment factor after considering calendar type is: Rprice_daytype=1+Edaytype×(λagent-λbase) / λbase.

[0078] Furthermore, the unit load correction factor is determined based on the degree of change in meteorologically sensitive loads and the corrected degree of price elasticity modulation, including: Obtain historical average weather-sensitive loads and initial load forecasts for electricity purchasers acting as agents; Based on the basic meteorological sensitive load and the modified price elasticity modulation degree, the modified meteorological sensitive load for the forecast period is determined; The degree of change per unit load is determined based on the degree of change of the modified meteorological sensitive load relative to the historical average meteorological sensitive load for the same period. The unit load correction factor is determined based on the degree of unit load variation and the initial load forecast of the electricity purchaser. The unit load correction factor is used to correct the load component in the initial load forecast of the electricity purchaser that is affected by the coupling of meteorological, price and calendar factors.

[0079] First, obtain the historical average weather-sensitive load for the same period and the initial load forecast for agency-purchased electricity users. The historical average weather-sensitive load refers to the average load change caused by changes in weather conditions on the same or similar dates in the past (e.g., the same day, month, or week last year). This data is obtained by analyzing historical load data and meteorological data, separating the impact of meteorological factors on the load, and averaging the data. Its purpose is to provide a benchmark for comparison of the weather-sensitive load for the current forecast period. The initial load forecast for agency-purchased electricity users is obtained from the basic load forecasting model based on the forecast input data and serves as the benchmark load for subsequent adjustments.

[0080] Secondly, the corrected meteorological sensitive load for the forecast period is determined based on the baseline meteorological sensitive load and the adjusted price elasticity modulation level. The baseline meteorological sensitive load characterizes the degree of load change caused by changes in meteorological conditions during the forecast period, while the adjusted price elasticity modulation level reflects the elasticity change of the meteorological sensitive load under the influence of current electricity prices and calendar factors. The process of determining the corrected meteorological sensitive load involves multiplicatively or additively adjusting the baseline meteorological sensitive load with a coefficient based on the adjusted price elasticity modulation level. For example, the baseline meteorological sensitive load can be amplified or reduced according to the price elasticity modulation level to obtain a more realistic corrected meteorological sensitive load that comprehensively considers the influence of meteorological, price, and calendar factors.

[0081] Next, the degree of change per unit load is determined based on the degree of change of the revised meteorological sensitive load relative to the historical average meteorological sensitive load for the same period. This step aims to quantify the difference between the revised meteorological sensitive load for the current forecast period and the historical baseline. The degree of change per unit load can be expressed as the difference between the revised meteorological sensitive load and the historical average meteorological sensitive load for the same period, or as the ratio of the two, or can be characterized by other mathematical relationships. For example, if the revised meteorological sensitive load is higher than the historical average meteorological sensitive load for the same period, the degree of change per unit load is positive, and vice versa.

[0082] Finally, a unit load correction factor is determined based on the degree of unit load variation and the initial load forecast of the electricity purchaser. The unit load correction factor is a key factor used to directly correct the initial load forecast of the electricity purchaser. This factor can be determined by directly using the degree of unit load variation as the correction factor, or by calculating it proportionally to the initial load forecast of the electricity purchaser to obtain a multiplicative or additive correction factor applicable to the initial load forecast. For example, the degree of unit load variation can be divided by the initial load forecast of the electricity purchaser to obtain a percentage-based correction factor. This unit load correction factor is used to correct the load component in the initial load forecast of the electricity purchaser that is affected by the coupling of meteorological, price, and calendar factors.

[0083] After price elasticity adjustment, the actual contribution of the basic weather-sensitive load Qweather_base to the weather-sensitive load is: ΔQweather=Qweather_base×Rprice_daytype; Since the base forecast value Pbase typically includes the total load level under average weather conditions, and the total load can be decomposed into the sum of the rigid base load Prigid and the average weather-sensitive load Qweather_avg, i.e., Pbase ≈ Prigid + Qweather_avg, when actual weather conditions deviate from the average, the correction for the weather-sensitive load should only apply to the elastic component. Therefore, the corrected total load should be Pbase ≈ Prigid + Qweather_base × Rprice_daytype. Expressed in the form of a unit load correction factor γweather, its strict definition is: γweather=(Pbase-Qweather_avg+Qweather_base×Rprice_daytype) / Pbase; Where Qweather_avg represents the historical average weather-sensitive load for the same period. In scenarios where the electricity purchaser is mainly a commercial and residential user, the proportion of weather-sensitive load is usually high. If Pbase has been trained based on average weather conditions and the proportion of weather-sensitive load is stable, the above formula can be simplified to a form that only scales the weather-sensitive component. In this case, the unit load correction factor is approximately: γweather≈1+(Qweather_base×Rprice_daytype-Qweather_avg) / Pbase.

[0084] Example 6 A load forecasting system for agent-purchased electricity users that considers multi-factor coupling, referring to Figure 2 ,include: Processor 21 and memory 22; Processor 21 and memory are connected via communication bus 22: The processor 21 is used to call and execute the program stored in the memory 22; The memory 22 is used to store a program, which is at least used to execute a method for predicting the load of a proxy electricity purchaser that considers multiple factors coupled, as in any of the above embodiments.

[0085] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0086] It should be noted that in the description of this application, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this application, unless otherwise stated, "a plurality of" means at least two.

[0087] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the function involved, as will be understood by those skilled in the art to which embodiments of this application pertain.

[0088] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0089] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0090] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0091] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0092] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0093] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for load forecasting of electricity purchasers by agents, considering the coupling of multiple factors, characterized in that, include: Obtain electricity purchase-related data for the region to be predicted, and perform missing value imputation, outlier correction, and normalization on the electricity purchase-related data to obtain sample data and prediction input data; the electricity purchase-related data includes historical load data, electricity price data, meteorological data, and calendar data; A basic load forecasting model is constructed based on the sample data, and the forecast input data is input into the basic load forecasting model to obtain the initial load forecast value of the agent electricity purchaser during the forecast period. The utility of the electricity purchase method chosen by the agent electricity purchaser is determined, the retention probability of the agent electricity purchaser is calculated based on the utility of the choice, and the user structure correction coefficient is determined based on the retention probability; the electricity purchase method includes: remaining within the agent electricity purchase system, and transferring to the market electricity purchase or electricity sales company purchase; the user structure correction coefficient is used to characterize the degree of correction of the initial load forecast value of the agent electricity purchaser by the change in the size of the agent electricity purchaser group; The degree of change in weather-sensitive load is determined based on the meteorological data, the degree of price elasticity modulation corresponding to the weather-sensitive load is determined based on the electricity price data, and the degree of price elasticity modulation is corrected based on the calendar data. A unit load correction coefficient is determined based on the degree of change in weather-sensitive load and the corrected degree of price elasticity modulation. The unit load correction coefficient is used to characterize the degree of correction of the initial load forecast value of the agent electricity purchaser by the change in unit electricity intensity under the coupled influence of meteorological factors, price factors and calendar factors. Based on the user structure correction coefficient and the unit load correction coefficient, the initial load forecast value of the agent electricity purchase user is jointly corrected to obtain the final load forecast value of the agent electricity purchase user for the forecast period.

2. The method according to claim 1, characterized in that, The historical load data is divided into load data for users who purchase electricity through agents and load data for users who directly trade electricity, according to user type. The meteorological data includes daily average temperature data, maximum temperature data, and minimum temperature data; The calendar data includes date information with weekday tags, rest day tags, and holiday tags; The electricity price data includes the average price of electricity purchased through agents and the average market reference price.

3. The method according to claim 1, characterized in that, Based on the sample data, a basic load forecasting model is constructed, including: Historical load characteristics, electricity price characteristics, meteorological characteristics, and calendar characteristics are extracted from the sample data; Using the historical load characteristics, electricity price characteristics, meteorological characteristics, and calendar characteristics as model inputs, and the load of the proxy electricity purchaser in the corresponding historical period as model output, the preset prediction model is trained to obtain the basic load prediction model. The preset prediction model includes any one of the following: statistical regression prediction model, time series prediction model, and neural network prediction model.

4. The method according to claim 2, characterized in that, Determine the utility of the electricity purchase method chosen by the agent-purchaser, including: Obtain price sensitivity coefficients and the non-price utility of electricity purchase agents within the electricity purchase agent system; Based on the average electricity purchase price through agents, the price sensitivity coefficient, and the non-price utility of electricity purchasers remaining within the electricity purchase system through agents, the retention option utility of electricity purchasers within the electricity purchase system through agents is calculated. Obtain the market reference average price, the non-price utility of agents purchasing electricity from switching to market-based electricity purchases or electricity sales companies, and the switching costs of agents purchasing electricity. Based on the aforementioned market reference average price, price sensitivity coefficient, non-price utility of the agent electricity purchaser for switching to market electricity purchase or electricity sales company purchase, and the switching cost of the agent electricity purchaser, calculate the switching choice utility of the agent electricity purchaser for switching to market electricity purchase or electricity sales company purchase.

5. The method according to claim 4, characterized in that, The retention probability of the agent-purchased electricity user is calculated based on the selected utility, including: Based on the difference between the retention option utility and the transfer option utility, the utility difference of the electricity purchase method for the agent electricity purchaser is determined; The utility difference of the electricity purchase method is input into a preset binary discrete choice model to obtain the retention probability of the agent electricity purchase user remaining in the agent electricity purchase system; The retention probability is used to characterize the likelihood that a user who has not yet left the agent-purchased electricity system will continue to choose to remain in the agent-purchased electricity system during the current forecast period. When the retention option utility increases relative to the transfer option utility, the retention probability increases; when the transfer option utility increases relative to the retention option utility, the retention probability decreases.

6. The method according to claim 5, characterized in that, Determining the user structure correction coefficient based on the retention probability includes: Obtain the cumulative retention rate of agent electricity purchase users in the initial forecast period, and use the cumulative retention rate of agent electricity purchase users in the initial forecast period as the benchmark rate in which all agent electricity purchase users are within the agent electricity purchase system; Obtain the cumulative retention rate of electricity purchasing agents from the previous forecast period; Based on the retention probability of the current forecast period, the cumulative retention ratio of the agent electricity purchaser in the previous forecast period is updated to obtain the cumulative retention ratio of the agent electricity purchaser in the current forecast period. The cumulative retention rate of agent-purchased electricity users in the current forecast period is determined as the user structure correction coefficient for the current forecast period.

7. The method according to claim 2, characterized in that, Determining the degree of change in meteorological sensitive loads based on the aforementioned meteorological data includes: Obtain the temperature forecast data and baseline temperature data for the period to be predicted, and calculate the predicted temperature deviation; The temperature rise response coefficient is determined by linear regression of historical load data and meteorological data, and the basic meteorological sensitive load is calculated based on the predicted temperature deviation and the temperature rise response coefficient. The basic meteorological sensitive load is used to characterize the degree of load change caused by changes in meteorological conditions during the forecast period.

8. The method according to claim 7, characterized in that, The degree of price elasticity modulation corresponding to weather-sensitive loads is determined based on the electricity price data, and the degree of price elasticity modulation is corrected based on the calendar data, including: Obtain the benchmark electricity purchase price and the price elasticity coefficient of weather-sensitive loads; The degree of price elasticity modulation corresponding to the weather-sensitive load is determined based on the deviation of the average electricity purchase price by the agent from the benchmark electricity purchase price and the price elasticity coefficient of the weather-sensitive load. Determine the date type corresponding to the time period to be predicted based on the calendar data; When the date type indicates that the period to be predicted belongs to a holiday, the price elasticity modulation degree is attenuated and corrected according to the preset holiday elasticity attenuation coefficient to obtain the corrected price elasticity modulation degree. When the date type indicates that the period to be predicted does not fall within a holiday, the price elasticity modulation degree is used as the corrected price elasticity modulation degree.

9. The method according to claim 8, characterized in that, The unit load correction factor is determined based on the degree of change in the meteorologically sensitive load and the corrected price elasticity modulation degree, including: Obtain the historical average meteorological sensitive load for the same period and the initial load forecast value of the agent electricity purchaser; Based on the basic meteorological sensitive load and the corrected price elasticity modulation degree, the corrected meteorological sensitive load for the forecast period is determined; The degree of change per unit load is determined based on the degree of change of the modified meteorological sensitive load relative to the historical average meteorological sensitive load for the same period. The unit load correction coefficient is determined based on the degree of unit load change and the initial load forecast value of the agent electricity purchaser; The unit load correction factor is used to correct the load component in the initial load forecast value of the agent electricity purchaser that is affected by the coupling of meteorological factors, price factors and calendar factors.

10. A load forecasting system for agent-purchased electricity users considering multi-factor coupling, characterized in that, include: Processor and memory; The processor and memory are connected via a communication bus: The processor is used to call and execute the program stored in the memory; The memory is used to store a program, which is at least used to execute the load forecasting method for agent-purchased electricity users that considers multi-factor coupling as described in any one of claims 1-9.

Images