Power transaction matching recommendation method and system facing user demand

Abstract

The application is suitable for the field of power demand analysis, and provides a power transaction matching recommendation method and system method facing user demand, which comprises the following steps: obtaining user power consumption data, generating a resonance group and a characteristic package with similar power consumption modes and risk tolerance; combining the resonance group characteristics with real-time market data to generate a power consumption comprehensive bidding plan; performing transaction matching based on the bidding plan, preferentially selecting the power generation resources closest in electrical distance, introducing a green energy premium coefficient, and outputting a transaction result; performing portfolio optimization analysis according to the transaction result, and generating a customized power package containing traditional energy, new energy and energy storage in combination with user risk preferences. Through multi-dimensional data integration and intelligent algorithm application, the application realizes accurate matching of power generation resources and user demand, improves power transaction efficiency and user satisfaction, and promotes new energy consumption and sustainable development of the power market.

Description

Technical Field

[0001] This invention belongs to the field of electricity demand analysis, and in particular relates to a method and system for matching and recommending electricity transactions based on user needs. Background Technology

[0002] With the marketization of the electricity market, users' demands for electricity services are gradually shifting from simply ensuring power supply to a focus on personalization, economy, and environmental friendliness. The rapid development of new energy and energy storage technologies has further enriched the energy types and trading models for electricity transactions. Currently, electricity trading matching is gradually transforming from the traditional "generation-side-driven" to "user-side demand-oriented." This requires combining multi-dimensional data such as user electricity consumption characteristics, price sensitivity, and risk preferences to achieve precise matching between power generation resources and user needs, thereby improving the efficiency of electricity resource allocation, promoting the consumption of new energy, and ensuring the stability of power supply.

[0003] In existing technologies, electricity trading matching methods are mostly based on clustering and grouping user basic electricity load data, achieving transaction matching through simple bidding strategies, and providing standardized electricity packages based on the matching results. These methods typically include steps such as user load data collection, clustering and grouping, bidding matching, and package generation. Clustering often relies on the overlap between peak and off-peak periods of the load curve, the bidding strategy is mainly based on the market benchmark electricity price, and the matching focuses on whether the capacity of generation resources meets user demand.

[0004] However, existing technologies do not fully integrate deep-seated characteristics such as users' price sensitivity, risk preference, and historical performance rates, resulting in insufficient homogeneity of user groups generated by clustering, making it difficult to reflect the differences in users' real needs; the electrical distance between power generation resources and users is not considered during transaction matching, which can easily lead to excessive transmission losses or insufficient power supply reliability; there is a lack of differentiated incentive mechanisms for green energy, making it difficult to effectively guide the consumption of new energy; and the generation of customized packages does not combine portfolio optimization and user risk preferences, making it impossible to balance the economy, stability, and environmental protection of the packages, and making it difficult to meet users' personalized needs. Summary of the Invention

[0005] The purpose of this invention is to provide a power trading matching and recommendation method and system oriented towards user needs, aiming to solve the technical problems existing in the prior art as identified in the background art.

[0006] This invention is implemented as follows: a power trading matching and recommendation method oriented towards user needs, the method comprising:

[0007] Acquire user electricity load characteristics, price sensitivity, and risk preference data, and integrate them into user electricity data;

[0008] The user electricity consumption data is aggregated in a spatiotemporal dimension to generate a resonance group with similar electricity consumption patterns and risk tolerance.

[0009] By using reinforcement learning bidding agents, a comprehensive electricity bidding plan is generated based on the aggregation scale of the resonance group, historical fulfillment rate, and real-time market electricity consumption data.

[0010] Based on the aforementioned comprehensive electricity consumption bidding plan, transaction matching is performed, prioritizing the matching of power generation resources with the closest electrical distance, and a green energy premium coefficient is introduced to output the transaction results;

[0011] Based on the transaction results, portfolio optimization analysis is performed, and customized electricity packages including traditional energy, new energy and energy storage are generated by combining risk preference data.

[0012] As a further aspect of the present invention, the acquisition of user electricity load characteristics, price sensitivity, and risk preference data includes:

[0013] The peak-valley distribution, standard deviation of load fluctuation rate, and transferable load identifiers of the load curve for the next 24 hours are collected by smart meters.

[0014] Based on the price elasticity matrix, the load response rate of users to the ±20% fluctuation of time-of-use electricity prices is calculated, and a price sensitivity coefficient vector is generated.

[0015] Determine users' tolerance for electricity price fluctuations of ±25% and power outages of more than 1 hour by using risk preference data, and output the risk preference level.

[0016] Extract historical performance rate curves of users from the power trading database and calculate the average performance rate.

[0017] As a further aspect of the present invention, the step of aggregating the user electricity consumption data in a spatiotemporal dimension to generate a resonance group with similar electricity consumption patterns and risk tolerance includes:

[0018] The overlap of electricity consumption periods among users is calculated based on the peak-valley time distribution of the load curve. When the overlap of time periods is >80%, it is marked as a spatiotemporally related user group.

[0019] For spatiotemporal correlation groups, the shape distance of the load curves of users within the group and the standard deviation of the risk preference level of users within the group are calculated. When the shape distance is <15%, it is determined that they have similar electricity consumption patterns, and when the standard deviation is <0.5, it is determined that they have similar risk tolerance.

[0020] Clustering optimization is performed on user groups that simultaneously meet the criteria of similar electricity consumption patterns and similar risk tolerance to generate resonance groups with similar electricity consumption patterns and risk tolerance, as well as resonance group feature packages containing the mean load volatility, risk preference level, and average performance rate.

[0021] As a further aspect of the present invention, the generation of the comprehensive electricity bidding plan includes:

[0022] The resonance group feature package and real-time market electricity consumption data are input into the state encoder to generate a six-dimensional state vector.

[0023] The DuelingDDQN network is used to process the six-dimensional state vector, and the output actions are obtained based on the price sensitivity coefficient, the proportion of transferable load, and the time period overlap, including time-sharing pricing strategy, standby capacity parameters, and ramp-up rate compensation scheme.

[0024] Risk hedging parameters are dynamically set based on historical performance rate curves, and a contract for difference (CFD) hedging mechanism is embedded.

[0025] Integrate output actions and risk hedging parameters to generate a comprehensive bidding plan.

[0026] As a further aspect of the present invention, the transaction matching prioritizes matching power generation resources with the closest electrical distance, and introduces a green energy premium coefficient, outputting the transaction results including:

[0027] Analyze the time-sharing pricing strategy in the tender plan and initiate the auction process. When the bid reaches the reserve price of the pricing strategy, submit a technical bid application that meets the standby capacity parameters and ramp-up rate compensation plan.

[0028] Calculate the electrical distance between the power generation resources and the resonance group, prioritize resources with a topological distance of <20km and that meet the ramp rate compensation scheme, and generate matching result data;

[0029] The green premium is calculated using a price sensitivity coefficient, and the transaction result is output, which includes matching result data, green premium, and performance terms.

[0030] As a further aspect of the present invention, the generation of customized electricity packages including traditional energy, new energy, and energy storage includes:

[0031] Using the matching results data as input, we construct an investment portfolio model, combine the user's risk preference level to solve for the optimal investment weight, simulate electricity price fluctuation scenarios, verify the stability of returns, and output the investment portfolio optimization parameters.

[0032] Map the energy type percentage data in the matching results to asset categories in the Black-Litterman model;

[0033] Based on the risk preference level data, set constraint data, and use the load volatility data to generate scenario optimization data. Combine the standby capacity parameters in the bidding plan to generate standby capacity service data.

[0034] Based on the transferable load identifier, the standard deviation of load volatility, and portfolio optimization parameters, the energy storage configuration capacity and time-slot scheduling strategy are determined, and energy storage configuration data is generated.

[0035] Based on energy storage configuration data, the system outputs a power package that includes energy allocation data, time-of-use pricing strategy data, standby capacity service data, and energy storage configuration data.

[0036] Another object of the present invention is to provide a power trading matching and recommendation system oriented towards user needs, the system comprising:

[0037] The data acquisition and integration module is used to acquire user electricity load characteristics, price sensitivity and risk preference data, and integrate them into user electricity data;

[0038] The resonance group generation module is used to aggregate the user's electricity consumption data in a spatiotemporal dimension to generate resonance groups with similar electricity consumption patterns and risk tolerance.

[0039] The bidding agent module is used to generate a comprehensive electricity bidding plan based on the aggregation scale of the resonance group, historical performance rate and real-time market electricity consumption data through reinforcement learning bidding agent.

[0040] The transaction matching module is used to perform transaction matching based on the comprehensive electricity bidding plan, prioritize matching the power generation resources with the closest electrical distance, introduce a green energy premium coefficient, and output the transaction results;

[0041] The portfolio optimization analysis module is used to perform portfolio optimization analysis based on the transaction results and generate customized electricity packages that include traditional energy, new energy and energy storage by combining risk preference data.

[0042] The beneficial effects of this invention are:

[0043] This invention achieves an upgrade in power trading from passive matching to proactive adaptation through a full-process design that integrates multi-dimensional data, performs precise clustering, intelligent bidding, optimized matching, and generates customized packages. By acquiring data on user electricity load characteristics, price sensitivity, risk preference, and historical performance rates, a comprehensive user electricity profile was constructed, providing a precise data foundation for subsequent stages. Resonance groups were generated through spatiotemporal aggregation, and combined with load curve shape similarity and risk preference consistency analysis, ensuring a high degree of homogeneity in user electricity consumption patterns and risk tolerance within each group, laying the foundation for unified trading strategy formulation. Reinforcement learning-based bidding agents were employed to generate comprehensive bidding plans, dynamically adjusting bids and reserve strategies based on characteristics such as price sensitivity and transferable load ratio, enhancing the targeting and flexibility of bidding. During transaction matching, priority was given to power generation resources with short electrical distances, reducing transmission losses and power supply fluctuations. The introduction of a green energy premium effectively incentivized renewable energy consumption, balancing economics and environmental protection. Finally, customized packages generated based on portfolio optimization integrated traditional energy, renewable energy, and energy storage, achieving a dynamic balance of return, risk, and environmental protection based on user risk preferences and load characteristics, meeting the personalized needs of different users. The overall process improved the efficiency, accuracy, and user satisfaction of electricity trading, promoting the development of the electricity market towards a more flexible, efficient, and sustainable direction. Attached Figure Description

[0044] Figure 1 A flowchart of a user-demand-oriented power transaction matching and recommendation method provided in an embodiment of the present invention;

[0045] Figure 2 A flowchart for obtaining user electricity load characteristics, price sensitivity, and risk preference data provided in an embodiment of the present invention;

[0046] Figure 3 A flowchart for generating a resonant group with similar power consumption patterns and risk tolerance, provided for embodiments of the present invention;

[0047] Figure 4 A flowchart for generating a comprehensive electricity consumption bidding plan is provided as an embodiment of the present invention;

[0048] Figure 5 A flowchart for outputting transaction results provided in an embodiment of the present invention;

[0049] Figure 6 A flowchart for generating customized electricity packages that include traditional energy, new energy and energy storage, provided for embodiments of the present invention;

[0050] Figure 7 This is a structural block diagram of a user-demand-oriented power trading matching and recommendation system provided in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0052] Figure 1 A flowchart of a user-demand-oriented power trading matching and recommendation method provided in an embodiment of the present invention is shown below. Figure 1 As shown, the method includes:

[0053] S100 acquires user electricity load characteristics, price sensitivity, and risk preference data, and integrates them into user electricity data;

[0054] The specific implementation of this step, which involves collecting data on the peak and valley periods of the load curve for the next 24 hours, the standard deviation of load fluctuation rate, and the identification of transferable loads using smart meters, relies on the real-time monitoring and high-frequency data recording functions of smart meters. To obtain the peak-valley time distribution, the 24 hours must first be divided into several standard time periods (e.g., one hour per unit). By statistically analyzing the proportion of load in each time period to the total daily load, the specific time periods of peak load (over 20%), average load (10%-20%), and valley load (below 10%) are determined. The purpose is to capture the temporal patterns of user electricity consumption behavior, providing a time-dimensional basis for subsequent time-of-use pricing strategies and energy storage dispatch. The standard deviation of load volatility is obtained by calculating the square root of the square of the deviation of each time period's load from the average load within 24 hours. This indicator effectively reflects the stability of user electricity consumption. High volatility indicates large fluctuations in electricity consumption, requiring more reserve capacity or energy storage support, while low volatility indicates stable electricity consumption, reducing the need for additional configuration. The determination of transferable load identifiers needs to be combined with the type of user's electrical equipment. Smart meters identify loads that can flexibly adjust their usage time at different times and mark their capacity proportion, providing clear targets for subsequent energy storage configuration and load transfer strategies, thereby improving the utilization efficiency of power resources.

[0055] Based on the price elasticity matrix, the load response rate of users to ±20% fluctuations in time-of-use (TOU) electricity prices is calculated, generating a price sensitivity coefficient vector. The price elasticity matrix is ​​a matrix reflecting the quantitative relationship between electricity price changes and load changes in different time periods. Its elements represent the proportion of load change in that time period and other related time periods when the electricity price changes by 1%. In practice, it is necessary to select time periods in historical user electricity consumption data where electricity prices fluctuate by ±20%, statistically analyze the corresponding load changes, calculate the response rate (load change rate / electricity price change rate), and then integrate the response rates of different time periods into a vector. The larger the absolute value of the value in the vector, the more sensitive the user is to the electricity price during that time period. The purpose of this step is to quantify the intensity of users' reaction to price signals. For example, highly sensitive users will significantly reduce their load when electricity prices rise, while less sensitive users will be less affected. Its advantage is that it makes subsequent TOU pricing strategies more targeted, avoiding user churn or revenue loss caused by uniform pricing, and ensuring that price adjustments match users' actual affordability.

[0056] By analyzing risk preference data, we determine users' tolerance for electricity price fluctuations of ±25% and power outages exceeding 1 hour, and output a risk preference level. We categorize users' tolerance for ±25% electricity price fluctuations into categories such as completely unacceptable, barely acceptable, basically acceptable, and completely acceptable, based on their historical electricity consumption behavior and responses to similar situations. Similarly, we classify their tolerance for power outages exceeding 1 hour. Finally, we combine these two tolerance levels to determine the risk preference level (a higher level indicates stronger tolerance). Classifying users from a risk perspective ensures that subsequent electricity packages strike a balance between benefit and risk. This approach avoids recommending highly volatile energy combinations to risk-averse users and overly conservative packages to risk-averse users, thus increasing user acceptance of the packages.

[0057] Historical performance rate curves for users are extracted from the electricity trading database, and the average performance rate is calculated. The historical performance rate curve is plotted by comparing time (horizontal axis) with the performance rate (number of performances / total number of transactions) on the vertical axis, based on records of users' timely and full completion of electricity consumption or payment in each past transaction. The average performance rate is the arithmetic mean of the curve's points. This step is used to assess users' creditworthiness and cooperation stability. For example, users with high average performance rates are more trustworthy and can serve as high-quality members in resonance group aggregation, reducing the risk of default in transactions. This provides a basis for setting subsequent risk hedging parameters, ensuring the rights and interests of both parties in the transaction are protected.

[0058] like Figure 2 As shown, the acquisition of user electricity load characteristics, price sensitivity, and risk preference data includes:

[0059] S110 collects the peak-valley period distribution, load fluctuation standard deviation, and transferable load identifier of the load curve for the next 24 hours through smart meters;

[0060] S120, calculates the user's load response rate to a ±20% fluctuation in time-of-use electricity price based on the price elasticity matrix, and generates a price sensitivity coefficient vector;

[0061] S130 determines users’ tolerance for electricity price fluctuations of ±25% and power outages of more than 1 hour by using risk preference data, and outputs the risk preference level.

[0062] S140: Extract the historical performance rate curves of users from the power trading database and calculate the average performance rate.

[0063] S200, perform spatiotemporal aggregation on the user electricity consumption data to generate a resonance group with similar electricity consumption patterns and risk tolerance.

[0064] The overlap of electricity consumption periods among users is calculated based on the peak-valley time distribution of load curves. When the overlap exceeds a certain threshold, users are marked as spatiotemporally related user groups. First, the peak-valley time periods of each user's 24-hour load curve are converted into standardized time intervals (peak hours are 8:00-12:00 and 18:00-22:00, and valley hours are 0:00-6:00). Then, the overlap is calculated by comparing the proportion of overlapping peak-valley time periods to the total time period for different users. For example, if two users' peak periods both include 18:00-22:00 and their valley periods both include 0:00-6:00, with only some differences in the flat periods, the overlap calculation result will be high. The core purpose of this step is to filter users with similar electricity consumption rhythms from a temporal perspective, while combining the spatial attributes of distribution substations (users in the same substation are geographically close and have close grid connections) to initially narrow down the aggregation range. Its advantage lies in avoiding including users with completely misaligned electricity consumption periods in the same group through spatiotemporal constraints, reducing the redundancy of subsequent data processing, and focusing on effective samples for similarity analysis.

[0065] For spatiotemporally related user groups, it is necessary to determine similar electricity consumption patterns and risk tolerance by calculating the load curve shape distance and the standard deviation of risk preference levels. The load curve shape distance is calculated using the Dynamic Time Warping (DTW) algorithm. This algorithm finds the optimal matching path between two load curves by stretching or compressing local segments of the time series, and is more suitable for handling load curves with time offsets or length differences than the traditional Euclidean distance.

[0066] During the calculation, the user's hourly load data needs to be used as a time series. The minimum cumulative distance is then solved iteratively using the DTW formula. The smaller the distance, the more similar the curve shapes, meaning the closer the electricity consumption patterns. This step goes beyond the superficial correlation of simple time period overlap, deeply capturing the inherent patterns of load fluctuations (even during the day, some users have stable loads, while others experience frequent load fluctuations), ensuring that users within a group not only have consistent electricity consumption times but also highly similar electricity consumption behavior patterns.

[0067] The calculation of the standard deviation of risk preference levels is based on the risk preference levels (1-5) of users within the group. The formula quantifies the dispersion of the level distribution. The smaller the standard deviation, the more consistent the tolerance of users within the group for electricity price fluctuations and power outages. This avoids aggregating users with excessively different risk tolerance, eliminates internal conflicts for subsequent unified risk hedging strategies and package design, ensures the consistency of group decision-making from a risk perspective, and reduces coordination costs in subsequent transactions.

[0068] Clustering optimization is performed on user groups that simultaneously meet similar electricity consumption patterns and risk tolerance capabilities to generate resonant groups and feature packages. Using the DTW distance and risk preference standard deviation within each group as constraints, user groups are continuously merged or split until all groups meet the termination condition of shape distance <15% and standard deviation <0.5. The final output resonant group feature package integrates the mean load volatility (reflecting overall electricity consumption stability), risk preference level (representing the group's risk attitude), and average performance rate (reflecting historical credit level) within the group. These feature data will directly serve as input parameters for subsequent bidding plan generation, further improving the homogeneity of the groups and ensuring that each resonant group becomes an organic whole with similar electricity consumption behavior, consistent risk attitude, and predictable credit level.

[0069] like Figure 3 As shown, the process of aggregating the user electricity consumption data in a spatiotemporal dimension to generate a resonance group with similar electricity consumption patterns and risk tolerance includes:

[0070] S210, calculate the overlap of electricity consumption periods among users based on the peak-valley time distribution of the load curve, and mark users as spatiotemporally related user groups when the overlap of time periods is >80%.

[0071] S220, for spatiotemporal correlation groups, calculate the distance between the load curve shapes of users within the group and the standard deviation of the risk preference levels of users within the group. When the shape difference is less than 15%, it is judged that they have similar power consumption patterns; when the standard deviation is less than 0.5, it is judged that they have similar risk tolerance.

[0072] ;

[0073] In the formula, , users respectively With users Load values ​​at the same time scale, Before the load curves of the two users , Minimum shape distance of data points;

[0074] ;

[0075] In the formula, For the first in the group Individual user risk preference level (1-5). This represents the average risk preference level of users within the group. The number of users in the group;

[0076] S230 performs cluster optimization on user groups that simultaneously meet the requirements of similar electricity consumption patterns and similar risk tolerance, generating resonance groups with similar electricity consumption patterns and risk tolerance, as well as resonance group feature packages containing the mean load volatility, risk preference level, and average performance rate.

[0077] S300, through reinforcement learning bidding agent, generates comprehensive electricity bidding plan based on the aggregation scale of the resonance group, historical fulfillment rate and real-time market electricity consumption data;

[0078] The mean load volatility in the resonance group feature package reflects the overall stability of the group's electricity consumption, the risk appetite level reflects the group's tolerance for market fluctuations, and the average performance rate represents the credit level of historical transactions. Real-time market electricity consumption data includes dynamic information such as the current market benchmark electricity price, supply and demand balance index, proportion of renewable energy generation, and transmission line load rate. The state encoder compresses these multi-source heterogeneous data into a six-dimensional vector through normalization (mapping feature values ​​of different magnitudes to the 0-1 range) and feature fusion. Each dimension corresponds to the group's risk level, electricity consumption stability, credit rating, market benchmark price, supply and demand relationship, and proportion of renewable energy, respectively. The purpose of this operation is to remove redundant information and retain key features, transforming high-dimensional complex data into low-dimensional structured data suitable for reinforcement learning network processing. The advantage lies in reducing the computational complexity of the network while ensuring the integrity of the input information.

[0079] By separating the state value function and the action advantage function, the DuelingDDQN network can more accurately evaluate the actual value of different actions in the current state, avoiding the bias in action value evaluation of traditional DQN networks. Specifically, the network first extracts features from the six-dimensional state vector, and then designs branches for calculation for different output actions: the generation of time-of-use pricing strategies is based on the price sensitivity coefficient. The formula couples the market benchmark electricity price with the user's price sensitivity. When the absolute value of the user's price sensitivity coefficient is large (i.e., the user responds strongly to price fluctuations), the price will move closer to the benchmark electricity price to increase the probability of a transaction. When the sensitivity is low, the price can be adjusted appropriately to pursue higher returns, so that the pricing strategy is highly matched with the user's price affordability, avoiding missed transactions due to excessively high prices or lost returns due to excessively low prices.

[0080] The determination of the reserve capacity parameter is related to the proportion of transferable load. The higher the proportion of transferable load, the more flexible the user's electricity use, and the required reserve capacity can be appropriately reduced. Conversely, the reserve capacity needs to be increased to cope with sudden load. The formula quantifies this relationship into a specific reserve capacity value by multiplying the reserve coefficient and the proportion of transferable load, ensuring that the allocation of reserve resources is neither excessively redundant nor insufficient.

[0081] The ramp rate compensation scheme is generated based on the time period overlap. User groups with high time period overlap exhibit strong synchronicity during peak load periods. When load fluctuations (peak-valley differences) are large, a higher ramp rate compensation is required to incentivize power generators to respond quickly to load changes. The formula uses the negative correlation between the compensation coefficient and the time period overlap to adapt the compensation scheme to the load fluctuation characteristics of the group. The advantage of this sub-step lies in its ability to dynamically optimize the bidding strategy through the autonomous decision-making capability of reinforcement learning algorithms, making it more adaptable to market changes and differences in group characteristics compared to traditional fixed strategies.

[0082] The historical performance rate curve records the performance of the resonance group in past transactions. If the curve is consistently high (the performance rate is stable at a high level), it indicates that the group has good credit. At this time, the risk hedging parameters can be appropriately reduced, and only a small number of CFDs are used to lock in the price of the core load. If the curve fluctuates greatly or is generally low, the hedging parameters need to be increased and the coverage of CFDs expanded to avoid losses caused by user defaults or market price crashes.

[0083] The embedding of Contracts for Difference (CFDs) is specifically manifested in the agreement of a fixed-price trading volume in the bidding plan, with the remaining portion participating in market bidding. Its purpose is to provide a safety net for price risk for both parties while pursuing market returns, transforming historical credit data into a quantifiable basis for risk control, and matching risk response strategies with the actual credit level of the group.

[0084] The integrated output actions and risk hedging parameters generate a comprehensive bidding plan. The plan not only includes core transaction terms such as time-of-use pricing, standby capacity, and ramp-up compensation, but also includes specific risk hedging schemes (trading volume of CFDs and price lock-in periods), forming a strategy-risk integrated bidding plan. This provides a complete and clear bidding basis for the transaction matching process, ensuring that the power generator can fully understand the needs and risk preferences of the resonance group and reducing information asymmetry between the two parties.

[0085] like Figure 4 As shown, the generated comprehensive electricity consumption bidding plan includes:

[0086] S310 inputs the resonance group feature package and real-time market electricity consumption data into the state encoder to generate a six-dimensional state vector;

[0087] The S320 uses the DuelingDDQN network to process the six-dimensional state vector and obtains output actions based on the price sensitivity coefficient, the proportion of transferable load, and the time period overlap, including time-sharing pricing strategy, standby capacity parameters, and ramp-up compensation scheme.

[0088] The formula for calculating time-sharing pricing strategy is:

[0089] ;

[0090] In the formula, for Time-of-day pricing for The time-of-use market benchmark electricity price This is the price adjustment factor (range: 0.1-0.3). This is the price sensitivity coefficient (calculated from the price elasticity matrix, with a value ranging from -1 to 1).

[0091] The formula for calculating standby capacity parameters is:

[0092] ;

[0093] In the formula, This is the reserve capacity value. This is a reserve factor (value between 0.15 and 0.25). The proportion of transferable load (the ratio of the load corresponding to the transferable load identifier to the total load). This represents the average load of the resonance group.

[0094] The formula for calculating the gradient compensation scheme is as follows:

[0095] ;

[0096] In the formula, The amount is the compensation for the slope rate. This is the compensation coefficient (values ​​range from 0.05 to 0.1). The overlap of time periods (value 0-1). This refers to load fluctuation (the difference between the peak and valley values ​​of the load curve).

[0097] S330 dynamically sets risk hedging parameters based on historical performance rate curves and embeds a contract for difference (CFD) hedging mechanism;

[0098] S340 integrates output actions and risk hedging parameters to generate a comprehensive bidding plan.

[0099] S400: Based on the comprehensive electricity bidding plan, the transaction matching is carried out, priority is given to matching the power generation resources with the closest electrical distance, and a green energy premium coefficient is introduced to output the transaction results;

[0100] The process involves analyzing the time-of-use pricing strategy in the tender proposal and initiating the auction process. When the bid reaches the reserve price of the pricing strategy, a technical bid application is submitted. This is the market-based pricing stage for transaction matching. The time-of-use pricing strategy includes reserve prices for different time periods (such as peak, off-peak, and low-peak periods). The analysis process requires extracting the price threshold and corresponding load for each time period. For example, the reserve price for the peak period from 8:00 AM to 8:00 PM may be higher than that for other time periods to match the higher electricity demand during that period.

[0101] The auction process relies on the trading platform's real-time price adjustment mechanism. The initial price is set slightly above the reserve price, then decreases in a fixed gradient (a certain percentage decrease every minute) until it reaches the reserve price for a specific time period. At this point, the system automatically triggers a technical bid application, submitting the corresponding standby capacity parameters and ramp-up compensation plan for that time period (such as the compensation standard for the maximum hourly load change rate). This market-based bidding mechanism discovers reasonable transaction prices, avoiding both excessively high bids leading to resource idleness and excessively low bids harming user interests. Its advantages lie in the transparency and efficiency of the price formation process, and the ability to adapt to different time-of-use electricity consumption characteristics through time-slot auctions, ensuring precise matching of supply and demand for each time period. For example, for residential users in the resonant group during the evening peak hours, the auction mechanism can quickly match power generation resources for that period, ensuring power supply stability.

[0102] The electrical distance is calculated using a formula: it is the sum of the products of the resistance and impedance values ​​of each segment of the transmission line from the power generation resource to the distribution substation where the resonance group is located. The core function of this formula is to quantify energy loss and voltage stability during power transmission. The smaller the electrical distance, the lower the transmission line loss, the smaller the voltage fluctuation, and the higher the power supply reliability. In practice, the transmission line parameters between the power generation resource and the distribution substation need to be retrieved from the power grid topology database, substituted into the formula to calculate the electrical distance value, and then combined with the topological distance (physical spatial distance) for screening: resources with the smallest electrical distance and a topological distance within 20km are given priority. At the same time, it is verified whether the ramp rate of the resource meets the load fluctuation requirements of the resonance group (if the load of the resonance group increases by 10% per hour during peak hours, the matching power generation resource must have a ramp rate capability of no less than that percentage). Finding the optimal solution among numerous power generation resources that is close in distance, has low loss, and responds quickly reduces transmission costs and ensures rapid response to load fluctuations in the group. The advantage lies in breaking through the limitations of simple physical distance and selecting resources from the perspective of grid operation characteristics. For example, a wind farm may be physically far away, but its electrical distance is smaller (due to better line parameters), and it can still be a priority option, achieving a balance between efficiency and cost.

[0103] Calculating the green premium using a price sensitivity coefficient is an incentive design step that balances environmental protection and user preferences. The formula for calculating the green premium combines the price sensitivity coefficient, the green energy coefficient, and the unit power generation cost of new energy sources. The price sensitivity coefficient reflects the user's sensitivity to price. When users are less price-sensitive, the green premium can be appropriately increased to incentivize more new energy sources to be integrated; conversely, for price-sensitive user groups, the premium needs to be controlled at a lower level to avoid excessive burden.

[0104] In practice, it is necessary to extract the price sensitivity coefficient vector of the resonance group, combine it with the actual power generation cost of new energy and the preset green energy coefficient, and calculate the green premium value for each time period. Economic incentives guide the priority matching of new energy resources, promoting the green transformation of the energy structure. The advantage lies in linking the green premium to users' price affordability, avoiding resource misallocation caused by one-size-fits-all subsidies.

[0105] The matching results clearly define the types of power generation resources traded, the proportion of each type, and the corresponding power supply periods; the green premium, as a supplement to the price of new energy transactions, is listed separately in the transaction price; the performance terms cover power supply reliability commitments, price adjustment mechanisms, and default handling methods. The specific implementation of this operation relies on the contract generation module of the trading platform, which integrates the above data into standardized transaction certificates.

[0106] like Figure 5 As shown, the transaction matching prioritizes matching power generation resources with the closest electrical distance and introduces a green energy premium coefficient. The output transaction results include:

[0107] S410: Analyze the time-sharing pricing strategy in the tender plan and initiate the auction process. When the bid reaches the reserve price of the pricing strategy, submit a technical bid application that meets the standby capacity parameters and ramp-up compensation scheme.

[0108] S420 calculates the electrical distance between the power generation resources and the resonance group, prioritizes resources with a topological distance of <20km and that meet the ramp rate compensation scheme, and generates matching result data;

[0109] ;

[0110] In the formula, For electrical distance, For the first Resistance value of a transmission line segment (unit: Ω). For the first Impedance value of a transmission line segment (unit: Ω). The number of transmission line segments from the power generation resources to the distribution area where the resonance group is located;

[0111] S430 uses a price sensitivity coefficient to calculate the green premium and outputs a transaction result that includes matching data, the green premium, and performance terms.

[0112] ;

[0113] In the formula, For the premium of green energy, The green energy coefficient (values ​​range from 1.1 to 1.3). This is the price sensitivity coefficient. Cost per unit of electricity generated from new energy sources;

[0114] S500 performs portfolio optimization analysis based on the transaction results and generates customized electricity packages that include traditional energy, new energy and energy storage by combining risk preference data.

[0115] The matching results data are used as input to construct an investment portfolio model. The optimal investment weights are calculated by combining user risk preference levels, and the stability of returns is verified by simulating electricity price fluctuations. The optimized parameters for the investment portfolio are then output. The matching results data includes the types of traditional and renewable energy transactions, the volume-price relationship of each type, and their proportions. This data constitutes the basic asset information of the investment portfolio. The investment portfolio model adopts a mean-variance model, the core of which is to find a balance between return and risk. The model incorporates the expected rate of return for each type of energy (estimated based on transaction prices and historical data), the risk aversion coefficient (determined by risk preference levels; for example, conservative users choose a higher value to reduce volatility), and the return covariance (reflecting the correlation of returns between different energy sources) into the calculation to find the investment weights (i.e., the proportions of traditional energy, renewable energy, and energy storage) that achieve the optimal return-risk balance. Simulating electricity price fluctuations involves generating a large number of possible electricity price trends to test the stability of the investment portfolio's returns under different scenarios, ensuring that the package maintains its expected performance despite market fluctuations.

[0116] The core value of the Black-Litterman model lies in combining market equilibrium returns with user subjective opinions (represented in this case, the energy preferences of the resonance group), correcting the excessive reliance of traditional mean-variance models on historical data. The mapping process requires transforming the actual energy types and their proportions into asset classes recognizable by the model, assigning corresponding weights and expected returns. This operation necessitates establishing a correspondence between energy types and financial asset classes. The aim is to make portfolio optimization more aligned with the actual characteristics of the electricity market. As a special commodity, electricity prices are significantly influenced by non-market factors. The introduction of the Black-Litterman model effectively integrates these subjective opinions, avoiding misjudgments of market shifts by purely data-driven models, and ensuring that asset allocation both respects historical patterns and adapts to real-world changes.

[0117] Risk appetite determines the stringency of constraints: groups with low risk tolerance require higher reserve capacity redundancy, while groups with high risk tolerance can have constraints relaxed to reduce costs. Load volatility data generates multi-scenario optimization data to ensure that reserve capacity services can cover electricity demand under different fluctuation conditions. For groups with high volatility, the scenario optimization data will include more extreme load growth scenarios, corresponding to stricter requirements for reserve capacity response speed.

[0118] The standby capacity parameters (from the tender proposal) are then translated into specific service terms, such as the activation conditions of the standby capacity (triggered when the actual load exceeds the planned load by 10%), duration, and compensation standards.

[0119] like Figure 6 As shown, the generation of customized electricity packages that include traditional energy, new energy, and energy storage includes:

[0120] S510 uses the matching results data as input to construct an investment portfolio model and combines it with the user's risk preference level to solve for the optimal investment weights:

[0121] The optimal investment weights are determined using a mean-variance model, and the calculation formula is as follows:

[0122] ;

[0123] ;

[0124] In the formula, For the first Investment weighting for energy-related categories (including traditional energy, new energy, and energy storage). For the first Expected rate of return for energy-like products The risk aversion coefficient (determined by the risk preference level, 1.5 for conservative, 1.0 for moderate, and 0.5 for aggressive) For the first Class and the Covariance of returns for energy-related products For the number of energy types;

[0125] It also simulates electricity price fluctuation scenarios, verifies return stability, and outputs portfolio optimization parameters;

[0126] S520, map the energy type percentage data in the matching result data to Black-Litterman model asset categories;

[0127] S530, Based on the risk preference level data, set constraint data, and use the load volatility data to generate scenario optimization data, and combine the standby capacity parameters in the tender plan to generate standby capacity service data;

[0128] S540 determines the energy storage configuration capacity and time-segment scheduling strategy based on the transferable load identifier, load volatility standard deviation, and portfolio optimization parameters, and generates energy storage configuration data.

[0129] The S550, based on energy storage configuration data, outputs a power package that includes energy allocation data, time-of-use pricing strategy data, standby capacity service data, and energy storage configuration data.

[0130] Figure 7 This is a structural block diagram of a user-demand-oriented power trading matching and recommendation system provided in an embodiment of the present invention, as shown below. Figure 7 As shown, the system includes:

[0131] The data acquisition and integration module 100 is used to acquire user electricity load characteristics, price sensitivity and risk preference data, and integrate them into user electricity data;

[0132] The resonance group generation module 200 is used to aggregate the user's electricity consumption data in a spatiotemporal dimension to generate resonance groups with similar electricity consumption patterns and risk tolerance.

[0133] The bidding agent module 300 is used to generate a comprehensive electricity bidding plan based on the aggregation scale of the resonance group, historical performance rate and real-time market electricity consumption data through reinforcement learning bidding agent;

[0134] The transaction matching module 400 is used to perform transaction matching based on the comprehensive electricity bidding plan, prioritize matching the power generation resources with the closest electrical distance, introduce a green energy premium coefficient, and output the transaction results;

[0135] The portfolio optimization analysis module 500 is used to perform portfolio optimization analysis based on the transaction results and generate customized electricity packages that include traditional energy, new energy and energy storage by combining risk preference data.

[0136] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0137] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0138] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A user-demand-oriented electricity trading matching and recommendation method, characterized in that, The method includes: Acquire user electricity load characteristics, price sensitivity, and risk preference data, and integrate them into user electricity data; The user electricity consumption data is aggregated in a spatiotemporal dimension to generate a resonance group with similar electricity consumption patterns and risk tolerance. By using reinforcement learning bidding agents, a comprehensive electricity bidding plan is generated based on the aggregation scale of the resonance group, historical fulfillment rate, and real-time market electricity consumption data. Based on the aforementioned comprehensive electricity consumption bidding plan, transaction matching is performed, prioritizing the matching of power generation resources with the closest electrical distance, and a green energy premium coefficient is introduced to output the transaction results; Based on the transaction results, portfolio optimization analysis is performed, and customized electricity packages including traditional energy, new energy and energy storage are generated by combining risk preference data. The acquisition of user electricity load characteristics, price sensitivity, and risk preference data includes: The peak-valley distribution, standard deviation of load fluctuation rate, and transferable load identifiers of the load curve for the next 24 hours are collected by smart meters. Based on the price elasticity matrix, the load response rate of users to the ±20% fluctuation of time-of-use electricity prices is calculated, and a price sensitivity coefficient vector is generated. Determine users' tolerance for electricity price fluctuations of ±25% and power outages of more than 1 hour by using risk preference data, and output the risk preference level. Extract historical performance rate curves of users from the power trading database and calculate the average performance rate; The process of aggregating the user electricity consumption data in a spatiotemporal dimension to generate a resonance group with similar electricity consumption patterns and risk tolerance includes: The overlap of electricity consumption periods among users is calculated based on the peak-valley time distribution of the load curve. When the overlap of time periods is >80%, it is marked as a spatiotemporally related user group. For spatiotemporal correlation groups, the shape distance of the load curves of users within the group and the standard deviation of the risk preference level of users within the group are calculated. When the shape distance is <15%, it is determined that they have similar electricity consumption patterns, and when the standard deviation is <0.5, it is determined that they have similar risk tolerance. Clustering optimization is performed on user groups that simultaneously meet the criteria of similar electricity consumption patterns and similar risk tolerance to generate resonance groups with similar electricity consumption patterns and risk tolerance, as well as resonance group feature packages containing the mean load volatility, risk preference level, and average performance rate. The generated comprehensive electricity consumption bidding plan includes: The resonance group feature package and real-time market electricity consumption data are input into the state encoder to generate a six-dimensional state vector. The DuelingDDQN network is used to process the six-dimensional state vector, and the output actions are obtained based on the price sensitivity coefficient, the proportion of transferable load, and the time period overlap, including time-sharing pricing strategy, standby capacity parameters, and ramp-up rate compensation scheme. Risk hedging parameters are dynamically set based on historical performance rate curves, and a contract for difference (CFD) hedging mechanism is embedded. Integrate output actions and risk hedging parameters to generate a comprehensive bidding plan.

2. The method according to claim 1, characterized in that, The transaction matching process prioritizes matching power generation resources with the closest electrical distance and incorporates a green energy premium coefficient. The output transaction results include: Analyze the time-sharing pricing strategy in the tender plan and initiate the auction process. When the bid reaches the reserve price of the pricing strategy, submit a technical bid application that meets the standby capacity parameters and ramp-up rate compensation plan. Calculate the electrical distance between the power generation resources and the resonance group, prioritize resources with a topological distance of <20km and that meet the ramp rate compensation scheme, and generate matching result data; The green premium is calculated using a price sensitivity coefficient, and the transaction result is output, which includes matching result data, green premium, and performance terms.

3. The method according to claim 2, characterized in that, The generated customized electricity packages, which include traditional energy, new energy, and energy storage, include: Using the matching results data as input, we construct an investment portfolio model, combine the user's risk preference level to solve for the optimal investment weight, simulate electricity price fluctuation scenarios, verify the stability of returns, and output the investment portfolio optimization parameters. Map the energy type percentage data in the matching results to asset categories in the Black-Litterman model; Based on the risk preference level data, set constraint data, and use the load volatility data to generate scenario optimization data. Combine the standby capacity parameters in the bidding plan to generate standby capacity service data. Based on the transferable load identifier, the standard deviation of load volatility, and portfolio optimization parameters, the energy storage configuration capacity and time-slot scheduling strategy are determined, and energy storage configuration data is generated. Based on energy storage configuration data, the system outputs a power package that includes energy allocation data, time-of-use pricing strategy data, standby capacity service data, and energy storage configuration data.

4. The method according to claim 1, characterized in that, The system for implementing the user-demand-oriented electricity trading matching and recommendation method includes: The data acquisition and integration module is used to acquire user electricity load characteristics, price sensitivity and risk preference data, and integrate them into user electricity data; The resonance group generation module is used to aggregate the user's electricity consumption data in a spatiotemporal dimension to generate resonance groups with similar electricity consumption patterns and risk tolerance. The bidding agent module is used to generate a comprehensive electricity bidding plan based on the aggregation scale of the resonance group, historical performance rate and real-time market electricity consumption data through reinforcement learning bidding agent. The transaction matching module is used to perform transaction matching based on the comprehensive electricity bidding plan, prioritize matching the power generation resources with the closest electrical distance, introduce a green energy premium coefficient, and output the transaction results; The portfolio optimization analysis module is used to perform portfolio optimization analysis based on the transaction results and generate customized electricity packages that include traditional energy, new energy and energy storage by combining risk preference data.

Images