Electricity selling company bidding robust optimization method and system oriented to wind power uncertainty

Abstract

The invention provides an electricity selling company bidding robust optimization method and system for wind power uncertainty, relates to the technical field of electricity market transaction and information processing, and is used for improving the problem that an electricity selling company bidding strategy cannot be effectively quantified and internal structural risks cannot be avoided in related technologies. And therefore, the technical problem of insufficient robustness under the extreme fluctuation of the market can be solved. Comprising the following steps: constructing a set comprising a plurality of discrete quotation units; generating a plurality of uncertain scene impacts; for a to-be-evaluated bidding strategy combination composed of a plurality of discrete quotation units, calculating a pore risk potential of the bidding strategy combination under the impact of the uncertain scene; executing an iterative optimization process, iteratively disturbing the bidding strategy combination and accepting or abandoning the bidding strategy combination according to a preset acceptance criterion by taking the pore risk potential and the expected income as optimization targets until a convergence condition is met, so as to obtain a final robust bidding strategy combination; and outputting the robust bidding strategy combination.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of power market transaction and information processing, and particularly relate to a robust optimization method and system for bidding of a power selling company in view of wind power uncertainty. BACKGROUND

[0002] The power market environment is becoming increasingly complex, especially under the background of large-scale grid connection of intermittent energy sources represented by wind power and photovoltaic. The randomness, volatility and low predictability of wind power output bring great challenges to the stable operation of the power system and the economic decision-making of market participants. As an important part of the power market, one of the core businesses of the power selling company is to bid for power purchase in the day-ahead power market to meet the electricity demand of its proxy users. The pros and cons of the bidding strategy directly determine the profit level and operating risk of the power selling company.

[0003] In existing technical practices, the bidding decision method of the power selling company mainly relies on the deterministic prediction of future market prices or the stochastic optimization based on probability distribution. Some methods use stochastic programming models to maximize expected revenue under multiple possible future scenarios. Other methods use robust optimization techniques to ensure a certain level of revenue in the worst-case scenario.

[0004] However, these traditional optimization methods often have some inherent limitations. First, they usually treat bidding decisions in different time periods as relatively independent variables or use only simple linear correlation coefficients to describe their relationship. This simplified treatment ignores the complex, nonlinear coupling risks that may exist between different bids in the case of market extreme fluctuations. Second, existing methods mainly focus on optimizing the final financial indicators (such as revenue or risk value), but lack internal assessment of the "health" or "structural stability" of the bidding strategy combination itself.

[0005] This leads to a serious technical problem: a bidding strategy that appears excellent under the evaluation of traditional models (for example, with a very high expected revenue) may behave abnormally fragile when encountering unexpected market shocks (such as price surges due to sudden wind power output reduction). Multiple bids within the strategy may simultaneously and chain-like incur huge losses, leading to an instant collapse of the profitability of the entire strategy combination, a phenomenon similar to "structural failure". This potential, systemic collapse risk due to unreasonable internal structure has not been fully identified, quantified and effectively avoided in existing technology. Therefore, there is an urgent need for a new method and system that can deeply analyze the internal structure of the bidding strategy, actively manage structural risks, and thus improve the robustness of the strategy in a highly uncertain environment. SUMMARY

[0006] This application provides a robust optimization method and system for power sales company bidding in response to wind power uncertainties. It aims to improve the technical problem in which power sales company bidding strategies fail to effectively quantify and avoid internal structural risks, resulting in insufficient robustness under extreme market fluctuations.

[0007] To achieve the above objectives, the embodiments of this application adopt the following technical solutions:

[0008] This application aims to address the technical problem that existing power sales companies' bidding strategies fail to effectively quantify and mitigate internal structural risks, resulting in insufficient robustness under extreme market volatility.

[0009] Firstly, this application provides a robust optimization method for electricity sales company bidding in the face of wind power uncertainty. This method quantifies the internal structural collapse risk of a bidding strategy portfolio by constructing a porosity risk potential and systematically generates bidding strategies with both high returns and high robustness through an iterative optimization process. The method includes: discretizing a continuous bidding cycle and price space to construct a set containing multiple discrete price units; generating multiple uncertainty scenario shocks, each corresponding to a set of possible wind power output and market prices; calculating the porosity risk potential under the uncertainty scenario shocks for a bidding strategy portfolio to be evaluated, which consists of multiple discrete price units, whereby the porosity risk potential is used to quantify the structural collapse risk within the bidding strategy portfolio caused by the interaction between discrete price units; executing an iterative optimization process, using the porosity risk potential and expected returns as optimization objectives, iteratively perturbing the bidding strategy portfolio and accepting or rejecting bids according to a preset acceptance criterion until convergence is met, thereby obtaining the final robust bidding strategy portfolio; and outputting the robust bidding strategy portfolio. In this way, this application no longer treats the bidding strategy as a collection of isolated bids, but optimizes it as a whole structure, actively identifying and suppressing internal risk transmission paths that may lead to the collapse of the strategy, thereby fundamentally improving the survivability and profitability stability of the bidding strategy under the market shocks brought about by the uncertainty of wind power.

[0010] In one possible implementation of the first aspect, the step of calculating the porosity risk potential under the impact of the uncertainty scenario includes: identifying the proximity relationship between any two discrete bidding units in the bidding strategy combination; obtaining the individual response state of the two discrete bidding units under the impact of the uncertainty scenario; determining the risk interaction strength between the two discrete bidding units based on the proximity relationship and the individual response state; and summing the risk interaction strengths of all discrete bidding unit pairs in the bidding strategy combination to obtain the porosity risk potential of the bidding strategy combination.

[0011] In one possible implementation of the first aspect, the proximity relationship is defined based on the distance between the two discrete pricing units in the time and price dimensions.

[0012] In one possible implementation of the first aspect, the individual response state includes a profitable state or a loss state; the intensity of the risk interaction is amplified when both discrete bidding units are in a loss state.

[0013] In one possible implementation of the first aspect, the iterative optimization process is a structural annealing optimization process, which includes: setting an initial annealing temperature and a cooling scheduling strategy; in each iteration, applying a small perturbation to the current bidding strategy combination to generate a new candidate bidding strategy combination; calculating the comprehensive evaluation value of the candidate bidding strategy combination, the comprehensive evaluation value being a function of the expected return and the porosity risk potential; generating an acceptance probability based on the comprehensive evaluation value of the current bidding strategy combination, the comprehensive evaluation value of the candidate bidding strategy combination, and the current annealing temperature; determining whether to replace the current bidding strategy combination with the candidate bidding strategy combination based on the acceptance probability, and updating the annealing temperature according to the cooling scheduling strategy.

[0014] In one possible implementation of the first aspect, the step of applying a small perturbation to the current bidding strategy combination includes randomly adding, deleting, or replacing one or more discrete bidding units from the set of discrete bidding units.

[0015] In one possible implementation of the first aspect, after generating multiple uncertain scenario shocks, the method further includes: obtaining wind power forecast information for future periods; and executing a feedforward perturbation weighting mechanism to adjust the weights of the multiple uncertain scenario shocks based on the wind power forecast information, so that the iterative optimization process pays more attention to scenario shocks that are consistent with the forecast trend.

[0016] In one possible implementation of the first aspect, after outputting the robust bidding strategy combination, the method further includes: after the bidding period ends, obtaining the actual return of the robust bidding strategy combination; executing a risk potential model adaptive correction mechanism to compare the actual return with the expected return under the corresponding scenario impact, and adjusting the preset parameters in the pore risk potential calculation model based on the comparison result to improve the accuracy of subsequent optimization.

[0017] Secondly, this application provides a robust optimization system for electricity sales company bidding in response to wind power uncertainty, including a discrete unit construction module, a scenario generation module, a risk potential calculation module, an optimization module, and a strategy output module. The modules work together to implement the aforementioned method. Attached Figure Description

[0018] Figure 1 A schematic diagram of the structure of a robust optimization system for electricity sales company bidding in response to wind power uncertainty is provided for some embodiments of this application;

[0019] Figure 2 A flowchart illustrating a robust optimization method for electricity sales company bidding in response to wind power uncertainty, provided for some embodiments of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0021] In the following description, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0022] Furthermore, in this application, directional terms such as "upper," "lower," "left," and "right" may be defined relative to the orientation of the components shown in the accompanying drawings. It should be understood that these directional terms can be relative concepts, used for relative description and clarification, and may change accordingly depending on the orientation of the components in the accompanying drawings.

[0023] In this application, unless otherwise expressly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral part; it can be a direct connection or an indirect connection through an intermediate medium. Furthermore, the term "electrical connection" can refer to the manner in which an electrical connection is used to achieve signal transmission.

[0024] As used herein, “about,” “approximately,” or “approximately” includes the stated value and a reference value within an acceptable range of deviation from the given value, characterized in that the acceptable range of deviation is determined by a person skilled in the art taking into account the measurement under discussion and the error associated with the measurement of the given quantity (i.e., the limitations of the measurement method).

[0025] This application provides a robust optimization method and system for power sales company bidding in the face of wind power uncertainty. Instead of treating bidding strategies as a simple set of isolated bids, it abstracts them into a "bid combination structure" with an internal structure. This application constructs a novel evaluation and optimization framework to identify and quantify the risk of strategy combination liquefaction under drastic market fluctuations, and systematically eliminates this risk through an optimization process, thereby generating bidding decisions that are both highly profitable and robust.

[0026] like Figure 1 As shown, one embodiment of this application provides a robust optimization system for electricity sales company bidding in response to wind power uncertainty. This system can be deployed on a single server or as a distributed system, and can be integrated into the electricity sales company's transaction decision support platform. The system may include a discrete unit construction module, a scenario generation module, a risk potential calculation module, an optimization module, and a strategy output module.

[0027] The discrete unit construction module transforms continuous bidding decision variables into discretized, structured data units that can be processed by a computer. Electricity sales companies' bidding decisions typically involve submitting the amount of electricity to purchase and the price they are willing to pay at each moment (e.g., every 15 minutes or 1 hour) over a long future period (e.g., 24 hours). These time and price variables are theoretically continuous. To facilitate subsequent combinatorial optimization, this module discretizes these continuous variables. Specifically, it discretizes the entire bidding period (e.g., from...) arrive The 96 15-minute time intervals are divided into discrete time nodes, and the possible price range (e.g., from 0 yuan / MWh to 1000 yuan / MWh) is divided into multiple price tiers. By simultaneously discretizing time and price, a finite but rich set of discrete pricing units can be constructed. A discrete pricing unit... It is an atomic pricing action, which can be defined as a triple. ,in It's the time for quoting prices. This refers to the quoted electricity volume for that specific time period. This is a quote. For example, a discrete quote unit could be... This indicates that in the 10th time period, a bid is submitted to purchase 50MWh of electricity at a price of 280 yuan / MWh. The output of this module is a pool containing all possible discrete bidding units, providing the basic building blocks for subsequently constructing bidding strategy combinations.

[0028] The scenario generation module simulates various possibilities for the future market environment arising from the uncertainties of wind power. The randomness and volatility of wind power output are major risks faced by electricity retailers, directly impacting the day-ahead market clearing price. This module does not rely on a single deterministic forecast, but rather generates a scenario containing multiple (e.g., A collection of uncertain scenarios. Every uncertain scenario impacts These all represent a path to realizing wind power output and market electricity prices over the next 24 hours. For example, in one scenario... It can be a vector sequence ,in It is a time period The realized value of wind power output, while This corresponds to the market clearing price. These scenarios can be generated through historical data analysis, Monte Carlo simulations, or stochastic process models combined with weather forecasts. The generation of these scenarios aims to cover a variety of typical and extreme market conditions, ranging from high wind power / low electricity prices to low wind power / high electricity prices, providing a stress testing environment for subsequent robustness assessments.

[0029] The scenario generation module can be implemented by those skilled in the art based on existing knowledge. For example, a typical approach is to combine the Monte Carlo method with a market simulation model. First, based on historical wind speed data and weather forecasts, scenarios are generated for each future time period. A probability distribution model is established for the wind power output, such as the commonly used Weibull distribution. Then, the Monte Carlo method is used to repeatedly sample randomly from these probability distributions, with each complete sample (covering all...) A possible wind power output curve is generated for each time period. Repeat the process. You can get it in one go. A wind power output curve.

[0030] For example, for time period The wind power output, whose probability distribution model parameters are known, is obtained by Monte Carlo sampling. Substituting the random number into the inverse function of the distribution yields a specific output value. MW. Combining the sampled values ​​from all time periods creates a complete power output curve. Then, each generated wind power output curve is... As input, a market clearing simulation model is fed in. This simulation model is a computer program that internally embeds the trading rules of the electricity market and inputs forecasts of the total market load and estimates of the bidding strategies of other major power generators (such as thermal and hydropower). The model calculates the wind power output at a given level by simulating the market clearing process. Under other boundary conditions, the market-clearing price at each time period that enables supply and demand to reach equilibrium will thus shape the wind power output curve. Transform into a market price curve In this way, the system is able to generate a series of uncertain scenario impacts covering a wide range of possibilities.

[0031] The risk potential calculation module is used to calculate a given combination of bidding strategies. The "pore risk potential" is defined in this application. Traditional risk assessments typically focus only on the expected profit or loss or variance of a single quote, neglecting the interaction between different quotes. This is specifically designed to quantify this structural, hidden coupling risk. When the market is impacted by an uncertain scenario... At that time, a combination of bidding strategies Multiple discrete bidding units in a strategy portfolio may simultaneously incur losses; this collective failure can lead to significantly larger-than-expected losses, a phenomenon known as "strategy portfolio liquefaction." The risk potential calculation module is designed to identify and quantify this risk in advance. Its input is a bidding strategy portfolio to be evaluated. (i.e., a set of discrete bidding units) and a specific uncertainty scenario impact. The module performs a series of complex calculations (see subsequent method implementation examples) to ultimately output a scalar value. The higher the value, the better the strategy combination. In this scenario The greater the risk of structural collapse, the higher the risk.

[0032] The optimization module receives multiple scenario impacts generated by the scenario generation module and uses an embedded risk potential calculation module as an evaluation tool to iteratively optimize the bidding strategy combination. The module's goal is not simply to maximize the average expected return across all scenarios, but rather to find a strategy combination that achieves the optimal balance between expected return and porosity risk potential. In one embodiment, the module employs a heuristic search algorithm called "structural annealing." This algorithm starts with a random or pre-set initial bidding strategy combination and generates new candidate combinations by continuously applying small, random "perturbations" to the current combination (e.g., adding, deleting, or replacing a discrete bid unit). For each candidate combination, the module calculates its comprehensive evaluation value across all scenarios (this value combines expected return and porosity risk potential).

[0033] Unlike traditional optimization algorithms, the optimization module doesn't just accept changes that improve the evaluation value; it also accepts changes that worsen the evaluation value with a certain probability. This acceptance probability is controlled by a parameter called the "annealing temperature." In the early stages of optimization, the "temperature" is higher, and the probability of accepting worse solutions is also higher, allowing the algorithm to "escape" the constraints of local optima and explore a wider space. As iterations proceed, the "temperature" gradually decreases according to a preset scheduling strategy (cooling table), and the algorithm becomes increasingly "greedy," eventually converging stably to a globally optimal or near-optimal solution. This final solution, the robust bidding strategy combination, is the product of systematic "stress testing" and "structural restructuring."

[0034] The strategy output module is responsible for organizing and outputting the final robust bidding strategy combination found by the optimization module. The output can be in the form of a standardized message conforming to the interface requirements of the electricity market trading system, such as a clear table listing the specific electricity volume and price to be bid for each bidding period in the next 24 hours. This module ensures that the optimization results can be seamlessly applied to actual bidding operations.

[0035] In some embodiments, the system may further include a feedforward control module and a model calibration module to achieve more advanced dynamic adaptability. The feedforward control module can dynamically adjust the weights of different scenario impacts using the latest wind power forecast information before optimization begins, allowing the optimization process to more proactively focus on the most likely future scenarios. The model calibration module forms the system's feedback loop. After each bidding cycle, it compares the actual market results (such as real returns) with the model's predictions and uses this discrepancy to automatically and iteratively correct key parameters in the risk potential calculation model.

[0036] The following details the specific steps of the robust optimization method for electricity sales company bidding in response to wind power uncertainties provided in this application. Figure 2 As shown, the method includes:

[0037] S100: Discretize the continuous bidding cycle and price space to construct a set containing multiple discrete price units.

[0038] This step aims to map the continuous decision space into discrete, combinable basic elements. Electricity retailers need to [do something] during the bidding period. (For example, within 24 hours) for a series of trading sessions Provide a quote. At the same time, the quoted price... and the quantity quoted There is also a continuous range of values. In order to construct a tractable optimization problem, this step first discretizes these continuous variables.

[0039] Specifically, the bidding period Divided into Divide 24 hours into 96 discrete time periods of equal length, for example, dividing 24 hours into 96 15-minute intervals. .

[0040] The legal price range Divided into Price tiers, for example, The price range of yuan / MWh is divided into 201 tiers in increments of 5 yuan / MWh, that is... .

[0041] Possible quoted electricity range Divided into Each battery level, for example, will The MWh capacity range is divided into 21 levels in 10MWh increments, i.e. .

[0042] Through the discretization described above, a "discrete pricing unit" can be defined. For triples It represents an atomic bidding decision. All possible discrete bidding units constitute a large basic set. A specific combination of bidding strategies It is from this basic set A subset selected from the data. .

[0043] For example, suppose time is divided into 4 periods. The price is divided into 3 tiers. Yuan / MWh, with two energy tiers. MWh. So, discrete pricing unit. This is a valid unit, representing a purchase order for 100MWh of electricity at a price of 300 yuan / MWh during the second time period. (Basic Set) Will contain There are several such units. A combination of bidding strategies to be evaluated could be... .

[0044] S200 generates multiple uncertainty scenario impacts, each of which corresponds to a set of possible realizations of wind power output and market price.

[0045] The uncertainty of wind power output is the root cause of market price fluctuations. This step aims to generate a... A collection of uncertain scenarios .

[0046] Impact of each scene All of them are for the future A complete forecast of market prices for a given period. ,in In the scene Down Market clearing prices for a given period. These price vectors can be generated based on complex models that take the predicted wind power output distribution as input. For example, Monte Carlo methods can be used to generate them first. Each wind power output curve is converted into a market price curve using a market clearing simulation model that takes into account the pricing strategies of other power generators and overall load forecasts.

[0047] This generation process can be implemented by those skilled in the art based on existing technology and the teachings of this application. A typical approach is to employ a two-stage model.

[0048] The first phase involves establishing a probability distribution model for wind power output in each future time period based on historical data and weather forecasts. For example, a widely accepted and physically meaningful model for wind power output forecasting is the Weibull distribution because it fits the statistical characteristics of wind speed well, and wind power is directly related to wind speed. For specific future time periods... The probability density function of its wind power output It can be defined as

[0049]

[0050] in, It is the wind power output value. and These are the time periods The shape and scale parameters. These two key parameters can be estimated by analyzing the power output data of this period in historical years and combining it with numerical weather prediction to forecast future wind conditions. For example, for a certain period at 2 PM, the parameters of its Weiber distribution might be obtained through data analysis. , MW. By providing each time period By establishing such a specific probability distribution model, a mathematical description of future uncertainty is completed. Then, random sampling methods such as Monte Carlo are used to repeatedly sample from these distributions. Next, to generate A statistically possible wind power output curve .

[0051] The second stage will process each wind power output curve generated in the first stage. As input, it is transformed into a corresponding market price curve through a market price generation model. To enable those skilled in the art to implement this, a simplified, exemplary market price generation function model is provided herein: First, a market price is established based on historical data regression analysis. Total market load Total wind power output The relationship between them, for example, a linear model: ,in These are the regression coefficients. When generating the scene, for each generated wind power output curve... Combined with a defined total market load forecast vector The corresponding market price curve can then be calculated using the linear model described above. Those skilled in the art will understand that more complex nonlinear models or simulation models based on the intersection of supply and demand curves can also be used, the core of which lies in establishing a deterministic mapping relationship between the uncertain input of wind power output and the output of market price.

[0052] For example, assuming there are 4 time periods, this step might generate Impact of a single scene:

[0053] (High wind power, low electricity price scenario): Yuan / MWh

[0054] (Average scenario): Yuan / MWh

[0055] (Scenario of low wind power and high electricity price impact): Yuan / MWh

[0056] These scenarios provide insights for evaluating any combination of bidding strategies. This provides a basis for performance in different market environments.

[0057] S300. For a bidding strategy combination to be evaluated, consisting of multiple discrete bidding units, calculate its porosity risk potential under the impact of the uncertainty scenario. The porosity risk potential is used to quantify the structural collapse risk within the bidding strategy combination caused by the interaction between discrete bidding units.

[0058] This is the core step of the methodology proposed in this application. Its purpose is not to assess the direct profit or loss of a single bid, but rather to quantitatively evaluate the "structural health" of the entire bidding strategy portfolio. This step constructs a mathematical model to explore whether there are harmful linkage effects between bid units that are spatially and temporally adjacent in price measurement when encountering market shocks. This linkage effect, namely the porosity risk potential, is overlooked by traditional risk assessment methods.

[0059] For a given combination of bidding strategies and a specific uncertainty scenario impact Its porosity risk potential The calculation is defined as the sum of all discrete bid units in the combination. The sum of the risk interaction strengths between them. Its calculation process includes the following sub-steps:

[0060] S310. Define proximity relationships and distance functions.

[0061] It is necessary to define any two discrete pricing units. and The "distance" between them This distance is not a physical distance, but a measure within an abstract "spatiotemporal-price" coordinate system, reflecting the degree of correlation between two pricing units in terms of business logic. A reasonable distance function can be constructed as a weighted combination of time and price distances:

[0062]

[0063] in, and These are weighting coefficients used to adjust the relative importance of time proximity and price proximity. For example, they can be set... , . It is a price normalization factor, which can be, for example, the width of the price range.

[0064] For example, consider two pricing units: and Assuming , , .

[0065] The distance between them is:

[0066] .

[0067] A small distance value indicates that the two quote units are very close in both time and price.

[0068] S320. Determine the individual response status.

[0069] Next, regarding the scene It is necessary to determine each discrete pricing unit. The individual response status. This response status is not the direct accounting profit or loss of a single quote, but rather an assessment of the contribution of that quote to the company's overall risk exposure in a specific market scenario. For electricity retailers (purchasers), the core objective of a quote is to lock in future electricity demand at an acceptable cost.

[0070] Therefore, the individual response status primarily assesses whether this objective has been achieved. When the market clearing price... Higher than the electricity purchase price At that time, the offer failed to be executed. This did not appear to result in a direct loss, but in reality, it led to a loss for the company during that period. have The company's electricity demand could not be met through the day-ahead market. This unhedged electricity gap must be filled in the higher-priced, more volatile real-time balancing market, thus exposing the company to significant price risk. This risk exposure is defined in this application as a negative individual response state, i.e., a "loss state." Conversely, if the quote is successfully executed ( If the quote is positive, it is considered that the price has successfully locked in the cost and is in a "profitable state" (or a non-lossable state).

[0071] Based on this, a binary loss state function can be defined. :

[0072] (Loss-making status), if (The quote failed, creating a risk exposure).

[0073] (Profitability), if (Quote successful, cost locked in).

[0074] For example, continue using the scenario in S200. (Scenario of low wind power and high electricity price impact): Yuan / MWh.

[0075] Consider a combination of bidding strategies ,in , .

[0076] In the scene Down:

[0077] For the quotation unit Market price Yuan / MWh. Because This bid failed to be executed. This means that the company incurred a 50 MWh power shortage in the first period, which it had to purchase in the real-time market at a price as high as 450 yuan / MWh (or higher), a significant deviation from the expected cost of 300 yuan / MWh. Therefore, the individual response status of this unit was determined to be in a loss-making state. .

[0078] For the quotation unit Market price Yuan / MWh. Similarly, because... This offer also failed to be executed, resulting in an 80 MWh risk exposure during the second session. Therefore, its response status was also a loss. .

[0079] In this way, the individual response status clearly reflects the actual impact of each bidding action on the company's overall risk profile under a specific market shock.

[0080] S330. Determine the intensity of risk interaction and calculate the porosity risk potential.

[0081] Risk interaction intensity This quantifies the coupling risk arising when two quote units simultaneously exhibit negative responses in a specific scenario. The core idea is that if two adjacent quote units incur losses simultaneously, the overall damage caused is greater than the sum of their individual losses. This non-linear amplification effect is the root of "strategy portfolio liquefaction."

[0082] The intensity of risk interaction can be defined by a function that depends on distance and the individual's response state:

[0083]

[0084] in, It is the distance defined in S310. It is a decay coefficient that controls the rate at which risk interaction decays with distance. A larger one... This means that strong interaction only occurs between very adjacent units. For example, Can be set to .

[0085] It is a risk coupling function that assigns a basic interaction value based on the individual response states of the two units. The risk coupling function... This is used to characterize a virtual risk cost or reward for synergy that the system is willing to pay for different risk interaction scenarios, and its output value is in yuan. It is defined as follows:

[0086] If both units are in a loss state (LOSS, LOSS). It is a large positive number, such as 10.0. This represents the risk cost that the system accrues for the combined risk exposure when two adjacent units fail to quote simultaneously.

[0087] If one unit incurs a loss, the other incurs a profit (LOSS, WIN). It is a small positive number or zero, such as 1.0.

[0088] If both units are profitable (WIN, WIN). It can be set to a negative number to indicate a synergistic profit effect, such as -0.5.

[0089] Finally, the entire bidding strategy combination In the scene Total porosity risk potential That is, for all different units Sum the risk interaction strengths:

[0090]

[0091] Continuing with the example of S320, , Both are operating at a loss in this scenario.

[0092] Assumption , .

[0093] distance .

[0094] Attenuation coefficient .

[0095] Risk coupling function, because both are at a loss. .

[0096] The risk interaction strength between the two is:

[0097] .

[0098] Since this combination has only two units, its total porosity risk potential is .

[0099] If there is a third unit in the combination Then it is necessary to calculate and And accumulate. This The value itself has no absolute unit; it is a relative risk measure used to compare different strategy combinations.

[0100] S400. Perform an iterative optimization process, taking the porosity risk potential and expected return as optimization objectives, by iteratively perturbing the bidding strategy combination and accepting or rejecting it according to a preset acceptance criterion, until the convergence condition is met, thereby obtaining the final robust bidding strategy combination.

[0101] This step systematically seeks a bidding strategy combination that balances expected returns and structural robustness (i.e., low porosity risk potential) through an iterative optimization process. Those skilled in the art will understand that various global optimization algorithms, such as genetic algorithms and particle swarm optimization, can be applied to the framework of this invention. The following will describe in detail a preferred embodiment using structural annealing optimization.

[0102] The process first requires defining a comprehensive evaluation value function. Used to evaluate any combination of strategies The "energy" (in the field of optimization, the objective function is often referred to as energy). This function needs to consider both benefits and risks:

[0103]

[0104] in: It is a strategy In all Average expected return in each scenario. yes In the scene The total revenue.

[0105] It is a strategy Average porosity risk potential across all scenarios.

[0106] It is a risk aversion coefficient, which is a positive number and is set by the decision-maker. The larger the potential porosity, the more risk-averse the decision-maker is, and the more likely the optimization process will choose strategies with lower porosity risk, even if this sacrifices some expected returns. For example, Can be set to .

[0107] The goal of optimization is to find... Minimize the combination of strategies .

[0108] The specific process for structural annealing optimization is as follows:

[0109] initialization:

[0110] Randomly selected from the discrete quotation unit base set Generate an initial bidding strategy combination. .

[0111] Set a high initial annealing temperature (For example, ) and a very low termination temperature (For example, ).

[0112] Set a cooling rate (A number close to 1, for example) ), used to control the cooling rate. .

[0113] Set the current temperature .

[0114] Iterative loop: when Repeat the following steps:

[0115] a. Perturbation: To the current policy Perform a small perturbation to generate a candidate policy. The perturbation operation can be:

[0116] Add: From Randomly select a unit to add to middle.

[0117] Delete: From Randomly select one unit to delete.

[0118] Replace: Performs one deletion and one addition.

[0119] b. Evaluation: Calculate the energy of the current policy. Energy of candidate strategies .

[0120] c. Accept the decision: Calculate the energy difference .

[0121] if If the candidate strategy is superior, then the perturbation should be accepted unconditionally. .

[0122] if If the candidate strategy is worse or the same, then according to the Metropolis criterion, it is selected with a certain probability. Accept the disturbance:

[0123]

[0124] Generate a Random numbers between .if Then accept the "bad" move and let Otherwise, reject the move and remain. constant.

[0125] d. Cooling: After completing a certain number of internal cycles (e.g., 100 perturbations), update the temperature once: .

[0126] Termination: When temperature Reduce to The loop terminates at this point. This refers to the final robust bidding strategy combination found. .

[0127] For example, the following traces the process of an iterative decision:

[0128] Suppose that in a certain iteration, the current temperature .

[0129] Current Strategy energy (The negative sign indicates that the expected return is 12,000 and the risk is 0, for the sake of simplifying the example).

[0130] Candidate strategies were generated by randomly replacing a unit. .

[0131] Calculate its energy Assuming The expected return decreased, but its porosity risk potential also decreased significantly, and the final calculated energy was... .

[0132] Energy difference .

[0133] because This is a "worse" solution. Calculate the acceptance probability:

[0134] .

[0135] This is a very small probability. The algorithm will generate a random number. .if Then accept this seemingly worse move, and Updated to In most cases, this move will be refused.

[0136] Now suppose there is another candidate strategy Its energy is .

[0137] Energy difference .

[0138] because This is a "better" solution, so the algorithm will accept it unconditionally. .

[0139] The core of this process lies in the fact that, even at high temperatures, It's very big. The value of may also be non-zero, enabling the algorithm to escape local optima. With As the denominator decreases, the negative value of the exponent term becomes larger. The solution rapidly approaches 0, causing the algorithm to eventually accept only "better" solutions, thus achieving convergence.

[0140] S500, Output the robust bidding strategy combination

[0141] After extensive iteration and optimization of the S400, the final strategy combination was obtained. This is a solution that performs robustly under various challenging scenarios. This step transforms it into an executable bidding plan. The strategy output module will collect... All discrete quotation units By time period Organize and generate a clear report or a document conforming to the trading system format for traders to review or submit directly to the power trading center.

[0142] To further improve the adaptability and accuracy of the method, this application also proposes a feedforward perturbation weighting mechanism. This mechanism is introduced after the scenario impact is generated in S200 and before the optimization is performed in S400. Its purpose is to dynamically adjust the static, equally weighted scenario set using the latest short-term prediction information, making the optimization process more forward-looking.

[0143] When conducting day-ahead bidding, a relatively reliable 24-hour wind power output forecast is usually available. This forecast information is typically provided by professional third-party meteorological and power forecasting agencies, or generated by the electricity sales company's in-house forecasting models based on numerical weather prediction (NWP) and machine learning algorithms (such as Long Short-Term Memory networks, LSTM). While this forecast information, such as the expected values ​​and confidence intervals of wind power output for different time periods, cannot completely eliminate uncertainty, it provides valuable clues for determining which scenarios are more likely to occur. This mechanism utilizes this forecast information to assess each pre-generated scenario impact. Assign a weight .

[0144] For example, obtaining the future Wind power prediction vector for each time period .

[0145] Impact on each scene (It implies the wind power output vector) ), calculate its relationship with the prediction vector The similarity or distance. For example, the reciprocal of the Euclidean distance can be used as the similarity: .

[0146] Weights are assigned to each scene based on similarity. And by normalization, we ensure that the sum of all weights is 1.

[0147]

[0148] In S400, the original comprehensive evaluation value function Revised to weighted form:

[0149]

[0150] For example, suppose there are 3 scenarios The implicit wind power output vectors are respectively The latest wind power prediction vector obtained is The distance is:

[0151] , , .

[0152] Similarity is: , , .

[0153] The normalized weights are:

[0154] .

[0155] .

[0156] .

[0157] .

[0158] Obtain the scene weight vector This means that in the subsequent annealing optimization, the scenario closest to the prediction will be selected. This will have a decisive impact on the final result (weighted at 0.66), while the one with the largest deviation from the prediction... The impact is significantly weakened. This mechanism can be interpreted as "directional prestressing reinforcement" of the "quotation combination structure" to cope with the most likely impact.

[0159] In some embodiments, this application also proposes an adaptive correction mechanism for the risk potential model.

[0160] This mechanism is activated after the S500 outputs its strategy and completes an actual bidding cycle, forming a feedback learning loop. Its purpose is to utilize real-world feedback to calibrate and optimize the internal risk model.

[0161] Pore ​​risk potential model Parameters in the data, such as distance weights attenuation coefficient and the risk coupling function The initial values ​​may be based on experience or general data. However, risk transmission mechanisms may differ across markets and time periods. This mechanism adjusts these core parameters in reverse by comparing the model's predictions (expected returns) with market realities (actual returns), enabling continuous model optimization.

[0162] For example, after the end of a bidding day, obtain the actual market price vector for that day. And based on the strategy adopted Actual total revenue obtained .

[0163] Find the real-world scenario closest pre-generated scene Compare the expected returns of the models in this scenario. Compared with actual benefits Calculate the profit error. .

[0164] Establish an update rule that correlates model parameters with payoff error. For example, gradient descent can be used to fine-tune the parameters. Using the risk aversion coefficient... For example, if the actual return is found to be far lower than expected, and the porosity risk on that day is high... This is very high, which may mean the model underestimates the importance of risk. The update rule could be:

[0165]

[0166] in It is a small learning rate (e.g.) If the expected return is higher than the actual return (positive in parentheses), and the risk is very high, then... This will be adjusted upwards, making the model more "conservative" in the future. Similarly, it can be used to... Other parameters in the formula (such as) Establish similar update rules.

[0167] A quantitative example is as follows:

[0168] Input: The strategy used in the previous cycle The model's expected return (in the closest scenario) Down): Yuan. The porosity risk potential of this strategy in this scenario: Actual gains: Yuan. Current risk aversion coefficient. Learning rate .

[0169] Process: The profit margin was very large, and the expectations were too optimistic.

[0170] renew :

[0171]

[0172] (The learning rate is used here for illustrative purposes; in reality, the changes are smoother.)

[0173] Updated risk aversion coefficient In the next optimization cycle, the system will penalize porosity risk potential with a higher weight, thereby generating a more conservative and robust strategy. This process can be interpreted as a continuous calibration of the internal "risk exploration model," making it increasingly reflective of real market risk.

[0174] In a preferred embodiment of this method, the key parameters of the pore risk potential model (such as...) The series of values ​​can be calibrated before deployment by back-testing historical market data. Specifically, a set of candidate parameters can be set, the method can be run using historical data, and its historical performance can be evaluated. Finally, a set of parameter combinations that achieves the best risk-return balance on historical data can be selected as the initial values. In addition, these parameters can be continuously adaptively corrected during actual operation. This parameter calibration and adaptive correction mechanism ensures that those skilled in the art can stably achieve the technical effects of the present invention without creative effort.

[0175] Those skilled in the art will understand that, in addition to the exponential decay model based on pairwise interactions described above, other mathematical forms can be used to quantify the risk of structural collapse. For example, the linkage effect between three or more units (higher-order model) can be considered, or a non-exponential distance decay function can be used. As long as the model can capture the coupling amplification effect of the proximity of the pricing units in the spatiotemporal-price dimension and their simultaneous entry into a negative response state, it should fall within the scope of protection of this application. The specific formulas provided in the specification are preferred methods for implementing this concept, and not limitations on the scope of protection.

[0176] Those skilled in the art will understand that the above embodiments are illustrative and not restrictive. Various modifications of form and detail can be made without departing from the spirit and scope of this application, and all such modifications should fall within the protection scope of this application. For example, the mathematical form of the porosity risk potential can have many variations, and the iterative optimization process can also be a genetic algorithm, particle swarm optimization algorithm, etc., as long as its core idea revolves around the "structural robustness" of evaluating and optimizing bidding strategies.

[0177] Those skilled in the art will understand that the above embodiments are illustrative and not restrictive. Various modifications of form and detail can be made without departing from the spirit and scope of this application, and all such modifications should fall within the protection scope of this application. For example, the mathematical form of the porosity risk potential can have many variations, and structural annealing optimization can be replaced by other advanced global optimization algorithms (such as genetic algorithms and particle swarm optimization), as long as their core idea revolves around the "structural robustness" of evaluating and optimizing bidding strategies.

[0178] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0179] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another apparatus, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0180] The units described as separate components may or may not be physically separate. A component shown as a unit can be one physical unit or multiple physical units; that is, it can be located in one place or distributed in multiple different places. Depending on actual needs, some or all of the units can be selected to achieve the purpose of this embodiment.

[0181] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware.

[0182] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A robust optimization method for electricity sales company bidding in the face of wind power uncertainty, characterized in that, include: The continuous bidding cycle and price space are discretized to construct a set containing multiple discrete price units; Multiple uncertainty scenarios are generated, each of which corresponds to a set of possible wind power output and market prices. For a bid strategy combination to be evaluated, consisting of multiple discrete bid units, calculate its porosity risk potential under the impact of the uncertainty scenario. The porosity risk potential is used to quantify the structural collapse risk within the bid strategy combination caused by the interaction between discrete bid units. An iterative optimization process is performed, with the porosity risk potential and expected return as the optimization objectives. The bidding strategy combination is iteratively perturbed and accepted or rejected according to a preset acceptance criterion until the convergence condition is met, thereby obtaining the final robust bidding strategy combination. Output the robust bidding strategy combination.

2. The method according to claim 1, characterized in that, The step of calculating the porosity risk potential under the impact of the uncertainty scenario includes: Identify the proximity relationship between any two discrete bid units in the bidding strategy combination; Obtain the individual response states of the two discrete bidding units under the impact of the uncertainty scenario; Based on the proximity relationship and the individual response state, the risk interaction strength between the two discrete bidding units is determined; The porosity risk potential of the bidding strategy combination is obtained by summing the risk interaction intensity of all discrete bidding unit pairs in the bidding strategy combination.

3. The method according to claim 2, characterized in that, The proximity relationship is defined based on the distance between the two discrete pricing units in the time and price dimensions.

4. The method according to claim 2, characterized in that, The individual response status includes a profitable state or a loss state; the risk interaction intensity is amplified when both discrete bidding units are in a loss state.

5. The method according to claim 1, characterized in that, The iterative optimization process is a structural annealing optimization process, which includes: Set an initial annealing temperature and a cooling scheduling strategy; In each iteration, a small perturbation is applied to the current bid strategy combination to generate a new candidate bid strategy combination; Calculate the comprehensive evaluation value of the candidate bidding strategy combination, where the comprehensive evaluation value is a function of the expected return and the porosity risk potential; Based on the comprehensive evaluation value of the current bidding strategy combination, the comprehensive evaluation value of the candidate bidding strategy combination, and the current annealing temperature, an acceptance probability is generated. Based on the acceptance probability, determine whether to replace the current bidding strategy combination with the candidate bidding strategy combination, and update the annealing temperature based on the cooling scheduling strategy.

6. The method according to claim 5, characterized in that, The step of applying a small perturbation to the current bidding strategy combination includes randomly adding, deleting, or replacing one or more discrete bidding units from the set of discrete bidding units.

7. The method according to claim 1, characterized in that, After generating multiple uncertainty scenario shocks, the process also includes: Obtain wind power forecast information for future periods; A feedforward perturbation weighting mechanism is implemented to adjust the weights of the various uncertain scenario impacts based on the wind power prediction information, so that the iterative optimization process pays more attention to scenario impacts that are consistent with the prediction trend.

8. The method according to claim 1, characterized in that, Following the output of the robust bidding strategy combination, the following is also included: After the bidding period ends, obtain the actual return of the robust bidding strategy combination; An adaptive correction mechanism for the risk potential model is implemented, which compares the actual returns with the expected returns under the corresponding scenario impact, and adjusts the preset parameters in the pore risk potential calculation model based on the comparison results to improve the accuracy of subsequent optimization.

9. A robust optimization system for electricity sales company bidding in the face of wind power uncertainty, characterized in that, include: The discrete unit construction module is configured to discretize the continuous bidding cycle and price space to construct a set containing multiple discrete price units; The scenario generation module is configured to generate multiple uncertain scenario impacts, each of which corresponds to a set of possible realizations of wind power output and market price. The risk potential calculation module is configured to calculate the porosity risk potential of a bid strategy combination consisting of multiple discrete bid units to be evaluated under the impact of the uncertainty scenario. The porosity risk potential is used to quantify the structural collapse risk within the bid strategy combination caused by the interaction between discrete bid units. The optimization module is configured to perform a structural annealing optimization process, with the porosity risk potential and expected return as optimization objectives. It iteratively perturbs the bidding strategy combination and accepts or rejects it according to a preset acceptance criterion until the convergence condition is met, thereby obtaining the final robust bidding strategy combination. The strategy output module is configured to output the robust bidding strategy combination.

10. The system according to claim 9, characterized in that, Also includes: The feedforward control module is configured to acquire wind power forecast information for future periods and execute a feedforward disturbance weighting mechanism to adjust the weights of the various uncertain scenario impacts based on the wind power forecast information. The model calibration module is configured to obtain the actual return of the robust bidding strategy combination after the bidding period ends, and execute the risk potential model adaptive calibration mechanism to compare the actual return with the expected return under the corresponding scenario impact, and adjust the preset parameters in the risk potential calculation module based on the comparison result.

Images