Power system spot market electricity price prediction method and device

Abstract

The present application relates to the technical field of electricity market price prediction, and particularly relates to a power system day-ahead spot market price prediction method and device. The method comprises the following steps: obtaining market clearing data of a historical period of a power system and system operation condition data of a prediction period; based on Tikhonov regularization, taking the market clearing data as observation data and the bidding curve of a generator set in the power system as a parameter to be inverted, a variational optimization model is constructed and solved to obtain the inverted bidding curve of each generator set in the historical period; common bidding strategies are extracted from the inverted bidding curve; the common bidding strategies and the system operation condition data of the prediction period are input into a pre-constructed power system day-ahead spot market clearing simulation model for simulation to obtain a node price prediction result in the prediction period. The present application can solve the problems of insufficient spot market price prediction accuracy, lack of physical consistency and behavior explainability of the prediction result in the prior art.

Description

Technical Field

[0001] This invention relates to the field of electricity market price forecasting technology, and in particular to a method and apparatus for forecasting day-ahead spot market electricity prices in an electricity system. Background Technology

[0002] With the deepening of global electricity market reforms, the spot market, as a core component of instantaneous pricing and optimal allocation of electricity resources, is becoming increasingly complex in its operating mechanism. Spot electricity prices directly reflect the rapid changes in supply and demand, and their accurate forecasting is a key basis for power generation companies' bidding decisions, electricity sales companies' risk management, and grid dispatch safety optimization, thus becoming an important research direction in the field of power system operation.

[0003] Currently, electricity spot market price forecasting mainly follows a "data-driven" approach, resulting in a rich methodology system ranging from traditional time series models to cutting-edge artificial intelligence models. Existing methods are primarily divided into two categories: First, classic econometric models, such as the Autoregressive Integrated Moving Average (ARIMA) and Generalized Autoregressive Moving Average (GARMA), predict prices by mining the time-series dependencies of electricity price series. Second, machine learning and deep learning methods, represented by Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), TimesNet (a two-dimensional time series variation model for general time series analysis), and various decomposition-ensemble models, construct complex nonlinear mapping models to learn the correlation between input features and predicted electricity prices from historical electricity price and load data, significantly improving forecast accuracy.

[0004] However, existing data-driven methods are essentially "black box" models. Their prediction process lacks a direct connection to the physical constraints of actual power system operation and fails to reflect the deep-seated impact of market participants' strategic bidding behavior on electricity price formation. When the market environment, trading rules, or the behavioral patterns of market participants undergo structural changes, the generalization ability and robustness of these models face severe challenges. Furthermore, existing methods often focus on mining statistical patterns from historical data, failing to construct a causal chain for electricity price formation from the perspective of market participant behavior, resulting in significant deficiencies in the physical consistency of prediction results. Summary of the Invention

[0005] This invention provides a method and apparatus for predicting day-ahead spot market electricity prices in power systems, in order to solve the problems of insufficient accuracy, lack of physical consistency and behavioral interpretability in spot price predictions in the prior art.

[0006] In a first aspect, embodiments of the present invention provide a method for forecasting day-ahead spot market electricity prices in a power system, comprising: Acquire market clearing data for historical periods of the power system and system operating condition data for predicted periods; Based on Tikhonov regularization, using the market clearing data as observation data and the reporting and pricing curves of generator units in the power system as parameters to be inverted, a variational optimization model is constructed and solved to obtain the inverted pricing curves of each generator unit in the historical period. Extract common pricing strategies from the inverted pricing curves; The common pricing strategy and the system operating condition data for the forecast period are input into a pre-built day-ahead spot market clearing simulation model of the power system for simulation, so as to obtain the nodal electricity price forecast result of the power system during the forecast period.

[0007] Secondly, embodiments of the present invention provide a power system day-ahead spot market electricity price forecasting device, comprising: The acquisition module is used to acquire market clearing data for historical periods of the power system and system operating condition data for predicted periods. The inversion module is used to construct and solve a variational optimization model based on Tikhonov regularization, using the market clearing data as observation data and the reporting and pricing curves of generator units in the power system as parameters to be inverted, to obtain the inversion pricing curves of each generator unit in the historical period. A common pricing extraction module is used to extract common pricing strategies from the inverted pricing curve; The simulation module is used to input the common pricing strategy and the system operating condition data for the forecast period into a pre-built day-ahead spot market clearing simulation model of the power system for simulation, so as to obtain the nodal electricity price forecast result of the power system during the forecast period.

[0008] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.

[0009] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect or any possible implementation thereof.

[0010] Fifthly, embodiments of the present invention provide a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect or any possible implementation thereof.

[0011] This invention provides a method and apparatus for predicting day-ahead spot market electricity prices in a power system. By acquiring market clearing data for historical periods of the power system and system operating condition data for the prediction period, and based on Tikhonov regularization, using the market clearing data as observation data and the bidding curves of generator units in the power system as parameters to be inverted, a variational optimization model is constructed and solved to obtain the inversion bidding curves of each generator unit in the historical period. Thus, Tikhonov regularization is used to transform the ill-posed inverse problem into a stable variational optimization problem, realizing the reverse deduction from the market clearing result to the bidding behavior of generators, making the prediction result fully interpretable at the behavioral logic and market mechanism levels. The common pricing strategies extracted from the inverted pricing curves solve the problem of inconsistent quality of inverted samples and provide stable and reliable behavioral input for subsequent simulations; The common pricing strategy and the system operating conditions data for the forecast period are input into a pre-constructed day-ahead spot market clearing simulation model of the power system for simulation, resulting in the nodal electricity price forecast for the power system during the forecast period. This combines the common strategy with the physical simulation model, generating electricity price forecast results under strict grid physical constraints, ensuring the physical consistency and behavioral interpretability of the forecast. Furthermore, this embodiment of the invention introduces Tikhonov regularized inversion, establishing the forecast on a causal chain of "behavioral inversion → common pricing strategy extraction → physical simulation," rather than simple data statistical fitting, fundamentally improving the accuracy of electricity price forecasting. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating the implementation of the day-ahead spot market electricity price forecasting method for the power system provided in this embodiment of the invention. Figure 2 This is a schematic diagram of a typical segmented structure of the day-ahead reporting curve for conventional generating units provided in an embodiment of the present invention; Figure 3This is a schematic diagram of the IEEE 24-node test system architecture provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the annual time series curve of the historical market bidding space before the present invention. Figure 5 In figure a, it is a schematic diagram of the historical temperature annual time series curve provided in an embodiment of the present invention. Figure 5 In diagram b, the historical humidity annual time-series curve provided in an embodiment of the present invention is shown. Figure 5 In the diagram, c is a schematic diagram of the historical wind speed per unit value annual time series curve provided in an embodiment of the present invention. Figure 5 d is a schematic diagram of the annual time series curve of historical per-unit light intensity provided in the embodiments of the present invention; Figure 6 In the diagram, 'a' is a schematic diagram of the load curves for the study day and similar days provided in an embodiment of the present invention. Figure 6 In Figure b, temperature curves for the study day and similar days provided in an embodiment of the present invention are shown. Figure 6 In the diagram, c is a schematic diagram of the humidity curves for the study day and similar days provided in an embodiment of the present invention. Figure 6 In the diagram, d is a schematic diagram of the wind speed curves for the study day and similar days provided in an embodiment of the present invention. Figure 6 In the figure, 'e' is a schematic diagram of the research day and similar solar illumination curves provided in the embodiments of the present invention; Figure 7 In diagram a, it is a schematic diagram of the inversion result of the 200MW unit quotation and pricing provided in an embodiment of the present invention. Figure 7 In diagram b, the inversion result of the 300MW unit quotation quotation is provided in an embodiment of the present invention. Figure 7 In the diagram, c is a schematic diagram of the inversion result of the 400MW unit quotation and pricing provided in an embodiment of the present invention. Figure 7 In the figure d, it is a schematic diagram of the inversion result of the quotation quotation of 500MW and above units provided in the embodiment of the present invention; Figure 8 In the diagram 'a', the actual node electricity price disclosed in the day-ahead market clearing is provided in an embodiment of the present invention. Figure 8 In Figure b, there is a schematic diagram of the nodal electricity price fitting error distribution for each time period provided in the embodiment of the present invention; Figure 9 The numbers in the diagram are schematic diagrams of the power output curve fitting of units 1-10 provided in the embodiments of the present invention. Figure 10 This is a schematic diagram of the common learning day-to-date price quotation curve results provided in an embodiment of the present invention; Figure 11 In the diagram 'a', the prediction error distribution of all nodes in each time period is shown in the embodiment of the present invention. Figure 11 In the diagram below, b is a time-series comparison of the daytime electricity price forecast and actual value for a typical node provided in this embodiment of the invention throughout the day. Figure 12 This is a schematic diagram of the unit output curve and prediction error provided in an embodiment of the present invention; Figure 13 This is a schematic diagram comparing the time series errors of the spot electricity price prediction method provided in this embodiment of the invention; Figure 14 This is a schematic diagram of the structure of the day-ahead spot market electricity price forecasting device for the power system provided in an embodiment of the present invention; Figure 15 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0014] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0015] See Figure 1 The document illustrates a flowchart of an implementation method for predicting day-ahead spot market electricity prices in a power system, as provided in an embodiment of the present invention, and details are as follows: Step 101: Obtain market clearing data for historical periods of the power system and system operating condition data for the predicted period.

[0016] In this step, the market clearing data of the power system in historical periods is obtained to provide a data foundation for subsequent inversion modeling, and the system operating condition data for the forecast period is obtained to provide a data foundation for subsequent electricity price forecasting.

[0017] Historical periods refer to past dates relative to the forecast date, where market transactions have been completed and clearing results have been publicly disclosed. Forecast periods refer to the target date for which electricity price forecasts need to be made, i.e., the various trading sessions on the forecast date, typically a 24-hour period on the forecast date.

[0018] Market clearing data refers to the official clearing results publicly released by the electricity market operator after the day-ahead market transactions conclude. For regions conducting day-ahead markets, the official day-ahead clearing results should at least include: unit configuration and unit output curves, and time-of-use pricing. Based on this, in this embodiment, market clearing data includes at least the following three core types of information: 1) Nodal marginal electricity price sequence. The nodal marginal electricity price is the Lagrange multiplier corresponding to the active power balance constraint at each node in the electricity market clearing model. It reflects the marginal value of electricity at different temporal and spatial locations and is the most core price signal in the market. In this embodiment, the nodal marginal electricity price data is represented in vector form, denoted as... ,in This indicates the number of trading sessions in the market each day, meaning each trading day contains 24 trading sessions. This indicates the number of nodes in the power system topology.

[0019] (2) Planned output curves of each generator unit. Unit output refers to the active power won by each conventional generator unit in the day-ahead market clearing process. It is the direct physical result of market scheduling and reflects the actual power generation plan of each power generation resource at different times. In this embodiment, the unit output data is represented in vector form, denoted as... .

[0020] (3) Start-up and shutdown status of each generator unit. The start-up and shutdown status of the units is a binary decision result directly output by the market clearing model. For example, 1 represents start-up and 0 represents shutdown, reflecting whether each unit is scheduled to start operation at each time period of the day. This data itself does not contain random noise from continuous observation and is an accurate discrete state information, which plays a key role in ensuring that the bidding behavior obtained from subsequent inversion conforms to the actual physical operation constraints of the power grid. In this embodiment, the start-up and shutdown status data of the units is denoted as , indicating the first Taiwanese unit in The actual start / stop status during each time period.

[0021] The three types of market clearing data described above depict the actual clearing results of the day-ahead market from three dimensions: price signals, physical scheduling results, and key operational status. Together, they constitute the observation data in the subsequent behavior inversion step.

[0022] System operating condition data for the forecast period refers to various external conditions affecting the power system's operating status and market supply and demand on the forecast date. This data serves as a crucial boundary input for subsequent day-ahead market clearing simulations, determining the system's operating conditions on the forecast date. Specifically, system operating condition data includes at least the following two categories: (1) Load data for the forecast period. Load is the most critical external driving factor that determines the operating status of the power system. In this embodiment, the system net load curve for each period of the forecast day is used. This curve reflects the electricity demand that needs to be met through day-ahead spot market bidding transactions after deducting the physical delivery portion such as the electricity volume of each market-based unit's own medium and long-term contract. The load data is represented in the form of a 24-hour time series curve.

[0023] (2) Meteorological data for the forecast period. Meteorological conditions are an important factor affecting load levels and the potential for new energy power generation. In this embodiment, meteorological data includes time-series curves of temperature, humidity, wind speed, and light intensity. These meteorological characteristics, together with the load curves, constitute the common time-series features in the subsequent similar day screening process, which are used to measure the degree of similarity of external conditions between different dates.

[0024] The aforementioned system operating condition data can be obtained before the forecast date (such as load forecast data, weather forecast data, etc.) and are routine input data in power system operation and dispatch.

[0025] Step 102: Based on Tikhonov regularization, using market clearing data as observation data and the quotation curves of generator units in the power system as parameters to be inverted, a variational optimization model is constructed and solved to obtain the inverted quotation curves of each generator unit in the historical period.

[0026] In this step, based on the historical market clearing data obtained in step 101, the generator unit quotation curve is used as the parameter to be inverted. The Tikhonov regularization theory is introduced to construct and solve a variational optimization model, thereby transforming the ill-posed behavior inversion problem into a stable optimization problem, and obtaining the generator unit inversion quotation curve for each historical period.

[0027] The parameters to be inverted are the generator set quotation curves, which are mathematically represented as parameter vectors containing the power output and price steps for each generator set. Let the number of conventional generator sets be... The number of tiered pricing segments for each generating unit is: The parameter vector to be inverted is denoted as It includes the following two types of parameters: (1) Power tier parameters of generator sets This indicates the electricity volume declared by each generating unit in each bidding segment; (2) Price ladder parameters This indicates the electricity value declared by each generating unit in each bidding segment.

[0028] Right now Its total dimension is In a typical application scenario, if the system includes 10 conventional generating units, and each unit submits 10 segments of tiered price quotation curves, then the total dimension of the parameters to be inverted is 200 dimensions.

[0029] The observation data refers to the market clearing data obtained in step 101, including the nodal marginal electricity price sequence, the planned output curves of each generating unit, and the start-up and shutdown status of each generating unit. The observation data is denoted as... superscript This indicates that the observation data contains noise or uncertainty.

[0030] Mathematically, the day-ahead market clearing process can be defined as a generalized nonlinear operator equation. This operator characterizes a day-ahead market safety-constrained economic dispatch model based on optimal power flow, i.e., given the quotation and bidding curves of each generator unit. As input, the clearing simulation model The calculation outputs the corresponding market clearing result. Its formal expression is: (1) In the formula: It is a nonlinear operator, representing the day-ahead market security-constrained economic scheduling model based on optimal power flow; For the parameter space to be inverted, For observation data space.

[0031] The equations above describe the forward process of the clearing simulation. However, this step needs to address the inverse problem: given the market clearing results (observational data)... ), and reverse-engineer the quotation curve (parameters) of the generator sets declared to achieve this clearing result. ).

[0032] Since the inverse mapping of market clearing models typically lacks continuous dependence—meaning that small changes in observed data can cause large fluctuations in the inverse solution, and the existence and uniqueness of the solution are difficult to guarantee—this inverse problem is essentially an ill-posed problem. To overcome this difficulty, this embodiment introduces Tikhonov regularization theory to construct a variational optimization model.

[0033] In one embodiment, the goal of the variational optimization model is to make the simulation clearing result obtained from the power system day-ahead spot market clearing simulation of the parameters to be inverted approximate the market clearing data. Variational optimization models include fidelity terms and stable functionals; The fidelity term is used to constrain the deviation between the simulation clearing results obtained from the simulation of the parameters to be inverted and the market clearing data; the stable functional is used to embed the structural prior constraints of the parameters to be inverted.

[0034] Optionally, this embodiment introduces Tikhonov regularization theory to transform the original ill-posed problem into a minimization problem of the following variational functional, i.e., the variational optimization model is: (2) in, For variational functionals, The fidelity term is used to quantify the fitting error between the simulated clearing result obtained from the clearing simulation of the parameters to be inverted and the observed data. The smaller the value, the better the inversion solution can reproduce the actual market clearing result. To stabilize the functional, structural prior constraints are embedded in the quantity and price quotation curves of generator sets, guiding the inversion solution to be mathematically feasible while satisfying the rationality of physical and market rules; This is a regularization parameter used to balance the relative importance of the fidelity term and the stable functional in the optimization objective. The regularization parameter can be determined automatically using methods such as Generalized Cross-Validation (GCV). In this embodiment, after preliminary testing, the selected parameter is... As a benchmark value.

[0035] This variational optimization model can obtain stable approximate solutions even when the observed data contains noise. The solutions obtained through inversion are not unique exact values, but rather a set of solutions that fall within an acceptable deviation range in both reported quantities and quoted prices. This error range can be set by the power system dispatch center according to actual needs.

[0036] In one embodiment, the fidelity term includes a weighted sum of at least one of the following error terms: nodal price fitting error, unit output fitting error, and unit start-up / shutdown state fitting error; the nodal price fitting error is the deviation between the simulated nodal price and the actual nodal price in the market-cleared data; the unit output fitting error is the deviation between the simulated unit output and the actual unit output in the market-cleared data; and the unit start-up / shutdown state fitting error is the deviation between the simulated unit start-up / shutdown state and the actual unit start-up / shutdown state in the market-cleared data.

[0037] The fidelity term is used to ensure that the market clearing simulation results corresponding to the inversion solution closely match the publicly disclosed actual observation data. Its design directly determines the degree of fit between the inversion results and the real world. The current official clearing results include unit combinations and unit output curves, time-of-use prices, and considering that nodal marginal electricity prices, unit output, and unit start-up / shutdown status are the most core data publicly disclosed in the market, the fidelity term in this embodiment is: (3) in, , , These are the weighting coefficients for each error term, used to balance the importance of different objectives in the optimization. For example, they can be weighted as 0.5, 0.3, and 0.2 respectively.

[0038] To account for the fitting error of nodal electricity prices, the simulated nodal electricity prices corresponding to the inversion results should be as close as possible to the nodal marginal electricity price series disclosed by the market. Defined as the squared L2 norm of the difference between simulated node electricity price and actual node electricity price: (4) For the purpose of inversion pricing The nodal price prediction vector obtained through clearing simulation; The actual disclosed node electricity price observation vector obtained in step 101; This represents the L2 norm of a vector.

[0039] To account for the unit output fitting error, the simulated unit output corresponding to the inversion result should be as close as possible to the planned unit output curve disclosed in the market. Defined as the square of the L2 norm of the difference between the simulated unit output and the actual unit output: (5) In the formula: For the purpose of inversion pricing The output vectors of each unit obtained from the simulation; This is the observed vector of the actual disclosed planned unit output obtained in step 101. Unit output is a direct physical result of market scheduling arrangements, and introducing this fitting error ensures that the inverted price curve remains consistent with the actual result at the physical scheduling level.

[0040] To account for the fitting error of the unit start-up and shutdown states, the inversion solution must accurately reproduce the actual unit start-up and shutdown states. The unit start-up and shutdown states are binary decision results (0 or 1) directly output by the market clearing model, and do not contain random noise from continuous observations. Ensuring that the inversion solution can accurately reproduce this discrete state is crucial to ensuring that the inversion bidding behavior conforms to the actual physical operation constraints of the power grid. Since the start-up and shutdown states are binary variables, mean square error is not applicable. This embodiment uses binary cross-entropy loss as the error metric, defined as: (6) In the formula: Indicates the unit During the period The actual start / stop status disclosed (1 for power on, 0 for power off). It is the solution from the inversion The predicted probability of the unit being in an operating state during this period, obtained from simulation, is [value missing]. . This refers to the number of conventional generator sets; Let be the number of trading sessions. The binary cross-entropy loss strongly penalizes state misjudgments, thereby driving the inversion solution to precisely match discrete start-stop plans. Ideally, this loss should approach zero when states are perfectly matched.

[0041] It is worth noting that if the fitting error of the unit's start-up and shutdown states approaches zero, the fidelity term can be simplified to a weighted sum of the fitting error of the nodal price and the fitting error of the unit output: ;(7)

[0042] In summary, through the multi-objective fidelity term composed of the above three objectives, the model can simultaneously utilize three types of core public information: market price signals (node ​​prices), physical dispatch results (unit output), and key operating status (start-stop plans) to jointly constrain the inversion process, ensuring that the inverted price curve is not only reasonable in an economic sense, but also consistent with the actual market results at the physical operation level.

[0043] Stable functionals are a core component of the Tikhonov regularization framework. Their role is to effectively solve the ill-posedness of the inversion problem and enhance the stability and interpretability of the solution by introducing structural prior knowledge and physical constraints about the parameters to be inverted from a large number of mathematically feasible solutions.

[0044] For the tiered bidding curve of generator sets in the day-ahead spot market, based on the generator cost function and the winning bid target, the bidding curve is usually divided into three typical segments, such as... Figure 2 The diagram shows a typical segmented structure of the current day application curve for conventional generating units. The horizontal axis represents power, and the vertical axis represents price. The three typical segments are: the guaranteed power segment, which uses the minimum technical output in conjunction with the lowest bid, i.e., the "floor price," to ensure basic bidding success; the winning bid segment, which, based on the increasing marginal cost characteristics, sees the bid price rise gradually with the increase in electricity volume to achieve economical bidding success; and the peak segment, which sets a higher price at the end of the bidding process to obtain excess revenue when the system load is at its peak and capacity is tight.

[0045] Based on the aforementioned market rules and the technical and economic characteristics of generating units, in one embodiment, the stability functional includes at least one of the following constraints: a monotonicity constraint, used to constrain the bidding ladder in the parameters to be inverted to exhibit a non-decreasing trend; a capacity integrity constraint, used to constrain the sum of each bidding ladder in the parameters to be inverted to equal the installed capacity of the corresponding generating unit; a smoothness constraint, used to constrain the penalty value for changes between adjacent ladders in the parameters to be inverted; and a bidding boundary constraint, used to constrain the bidding ladder in the parameters to be inverted to be within the bidding range allowed by the electricity market.

[0046] Alternatively, the stable functional is: (8) in, This is a monotonicity constraint term. For capacity integrity constraints, For smoothness constraint terms, This refers to the pricing boundary constraints.

[0047] (1) Monotonicity constraint.

[0048] According to the general rules of the electricity spot market, the bid curve submitted by generating units must be non-decreasing, meaning that as the submitted electricity volume increases, the corresponding bid price cannot decrease. This constraint ensures that the inverted bid curve meets the basic formal requirements for market access. The monotonicity constraint is defined as: (9) In the formula: Indicates the first The generator set was at the The electricity price declared for each bidding segment; Indicates the first The generator set was at the The electricity price declared for each bidding segment; The penalty function applies a squared penalty only when the current electricity price is higher than the next price, ensuring a non-decreasing price trend. Ideally, this constraint should be strictly equal to 0. It is understood that in this embodiment, for . The number of adjacent pairs required for monotonicity verification This represents the total number of segments in the tiered pricing curve for each generating unit.

[0049] (2) Capacity integrity constraints.

[0050] This constraint reflects the techno-economic logic of generator unit bidding: the sum of all declared power volumes for a generator unit should equal its available installed capacity, and the first declared power volume usually corresponds to its minimum technical output. This ensures that the inverse bidding curve can fully cover the entire generating capacity of the unit. The capacity integrity constraint is defined as follows: (10) In the formula: Indicates the first Taiwanese unit in The electricity volume declared in each bidding segment Indicates the first The electricity volume declared by the unit in the first bidding segment; and These are the installed capacity and minimum technical output of the unit, respectively. The first penalty penalizes the deviation between the sum of the declared electricity outputs for each segment and the installed capacity, while the second penalty penalizes the deviation between the declared electricity output for the first segment and the minimum technical output. This constraint should be as close to 0 as possible.

[0051] (3) Smoothness constraint term In actual bidding, power generation companies tend to submit curves with gradual changes to avoid abrupt jumps in electricity volume or price. This smoothness constraint aims to penalize excessive changes between adjacent price segments, making the inversion curve smoother and more reasonable, and reducing the risk of sudden changes due to data noise. The smoothness constraint is defined as follows: (11) In the formula, , These represent the smoothness penalty weights for price and electricity, respectively, both set to 0.5 in this embodiment. It should be noted that since the first-stage reported output of a conventional unit is typically the minimum technical output (generally above 30%), no constraint is imposed on the first stage. The smoothness constraint term uses the relative rate of change between adjacent steps as a metric, and is expressed as... Normalization is performed to ensure that the magnitude of this constraint does not change with the number of random groups. , They represent the first Taiwanese unit in The declared electricity volume and declared electricity price for each bidding segment.

[0052] (4) Price boundary constraints.

[0053] This ensures that the bidding parameters are within the reasonable physical bidding range allowed by the electricity market. The bidding boundary constraint is defined as follows: (12) In the formula, This represents the permitted price range in the electricity market; the first penalty applies to deviations where the declared price exceeds the upper limit, and the second penalty applies to deviations where the declared price falls below the lower limit. Theoretically, this constraint should be zero. , for or .

[0054] In actual day-ahead market trading, monotonicity constraints, capacity integrity constraints, and bid-ask boundaries are generally considered "hard constraints" that must be strictly satisfied. During optimization, these three terms can be directly treated as equality or inequality constraints, rather than being approximated by penalty terms in the objective function. In this case, the stable functional can be simplified to retaining only the smoothness constraint: ;(13)

[0055] Through the synergistic effect of the aforementioned fidelity term and stable functional, the original ill-posed inverse problem is transformed into a stable variational optimization problem seeking the optimal solution within a physically feasible and commercially reasonable subspace. During the solution process, appropriate constraint handling techniques are employed to directly embed hard constraints into the optimization problem's constraints. The Grey Wolf Optimizer (GWO) algorithm is used to solve the variational optimization model. By simulating the social hierarchy and hunting behavior of a wolf pack, the optimal solution that minimizes the variational functional is searched in the parameter space to be inverted. This solution process is consistent with the inversion operation for similar days in step 104; the specific solution procedure will be described in detail in subsequent steps. After the solution is completed, the inverted price curves for each generator unit within the historical time period can be obtained.

[0056] The behavioral inversion framework based on Tikhonov regularization described above enables the reverse deduction from the market clearing results to the strategic bidding behavior of generator sets, providing accurate and reliable behavioral samples for the subsequent extraction of common bidding strategies.

[0057] Step 103: Extract common pricing strategies from the inverted pricing curves.

[0058] In this step, based on the inversion price curves of each generator unit within the historical period obtained in step 102, a common pricing strategy that can represent the general market trend is extracted. A common pricing strategy refers to a representative and robust bidding behavior pattern typically adopted by market participants under similar external conditions. Extracting this common pricing strategy is a crucial bridge connecting historical behavior inversion with future electricity price prediction. Its core lies in filtering out noise and anomalies from heterogeneous historical inversion samples to extract a robust and generalizable pricing behavior pattern.

[0059] In one embodiment, extracting common pricing strategies from the inverted pricing curve includes: Select multiple similar days from historical time periods that meet the preset similarity criteria for the predicted time period; The inverted price curves corresponding to each similar day are obtained respectively, forming a set of inverted price curves; By performing weighted center estimation on the inverted pricing curve set, a common pricing strategy is obtained.

[0060] Optionally, similar day screening refers to selecting a small number of samples from a large number of historical days that have similar external conditions to the predicted day, thereby providing an effective data foundation for subsequent behavior inversion and commonality extraction. Since the factors affecting the supply and demand relationship and the bidding behavior of the main players in the electricity market can be divided into two categories: non-time-series features and time-series features, this embodiment adopts different screening strategies for the two types of features.

[0061] In one embodiment, selecting multiple similar days from historical time periods that meet preset similarity criteria with the predicted time period includes: Non-time series features and time series features for historical periods and predicted periods are obtained respectively. Non-time series features include date attributes, and time series features include power system load curves and meteorological factor curves. The historical periods with the same non-temporal characteristics as the predicted period are used as the initial sample; Calculate the relative root mean square error between the temporal characteristics of each preliminary sample and the temporal characteristics of the predicted period, and determine the preliminary samples with a relative root mean square error less than a preset threshold as similar days.

[0062] Here, non-time-series characteristics refer to date attribute characteristics that do not change with the time series. In this embodiment, non-time-series characteristics include at least date attributes, specifically seasonal attributes and weekday / holiday attributes. Seasonal attributes reflect the systematic impact of different seasons on electricity load and market supply and demand, while weekday / holiday attributes reflect the periodic impact of social production activity patterns on electricity consumption behavior.

[0063] For non-time-series features, the filtering rule is exact matching, meaning the non-time-series features of the historical date and the predicted date must be completely identical. Mathematically, this is represented as: (14) In the formula: Indicates the first Non-temporal feature vectors of historical days, This represents the non-time-series feature vector of the predicted date. In principle, the historical date and the predicted date need to simultaneously satisfy two conditions: the same season and the same weekday / holiday type. In other words, they need to be completely consistent.

[0064] Time-series characteristics refer to the curve-like features that change over time (24 hours). In this embodiment, time-series characteristics include the power system load curve (i.e., the day-ahead market bidding space curve) and meteorological factor curves, which include temperature, humidity, wind speed, and solar intensity curves. These time-series characteristics reflect the external operating environment conditions on the forecast date and are key driving factors affecting market supply and demand balance and power generator bidding behavior.

[0065] Since the time series curves are aligned to a 24-hour timeframe, this embodiment uses the root mean square error (RMSRE) to measure the similarity between historical and predicted dates on a certain time series feature. The root mean square error is defined as follows: (15) In the formula, Representing historical days With the predicted date In the The relative root mean square error on the time series feature; the smaller the value, the more similar the two dates are in terms of the time series feature. and Representing historical days and predicted date In the The time period, the first Numerical values ​​in time-series characteristics; This represents the total number of trading sessions; a typical value is 24. For the first A preset error threshold is set for each time series feature. A historical day is selected as a similar day only if the relative root mean square error of all time series features is less than the corresponding preset threshold.

[0066] It should be noted that the preset error thresholds for each time series feature are... The preset error thresholds can be pre-set by the dispatch center or power market operator according to the actual application scenario and forecast accuracy requirements. This embodiment provides an example of a set of feasible preset error threshold settings: the allowable error range for the daily load curve is 12.5%, the allowable error range for the temperature curve is 10%, the allowable error range for the humidity curve is 15%, the allowable error range for the wind speed curve is 15%, and the allowable error range for the solar intensity curve is 15%. These preset error thresholds are for illustrative purposes only and can be adjusted according to the specific system's data characteristics and forecasting needs in actual applications.

[0067] By combining precise matching of non-temporal features with threshold filtering of temporal features, multiple similar days that meet the preset similarity criteria with the predicted time period can be selected from historical time periods. The total number of similar days is denoted as . .

[0068] After completing the screening of similar days, the behavior inversion operation based on Tikhonov regularization in step 102 is performed on each similar day to obtain the inversion price curve corresponding to each similar day.

[0069] Specifically, for each selected similar day Using the market clearing data (including nodal marginal electricity price, power output of each unit, and start-up and shutdown status of each unit) of the similar day as the observation data, and the reporting and pricing curves of each generator unit on the similar day as the parameters to be inverted, the Tikhonov regularized variational optimization model as in step 102 is constructed and solved to obtain the inverted pricing curves of each generator unit on the similar day.

[0070] The inversion price curves for all similar dates are compiled into an inversion price curve set. Each inversion price curve in this set includes all conventional generator sets for that similar date. The report volume parameters and pricing parameters of the segment ladder. These inverted pricing curve samples form the data foundation for subsequent extraction of common strategies.

[0071] Although all samples in the inverted pricing curve set are derived from historical days that meet similarity criteria, they may still contain anomalous samples that deviate from the norm due to factors such as data noise, abnormal market events, or special strategies on specific days. Directly aggregating these samples using a standard weighted average would lead to excessive influence and distortion of the common strategy by these anomalous samples. Therefore, this embodiment proposes a weighted robust center estimation method. By comprehensively considering the similarity weights between each similar day and the predicted day, and introducing the Huber loss function as a robust loss metric, it effectively suppresses the excessive influence of anomalous samples on the center estimate, thereby extracting a more representative and stable common pricing strategy.

[0072] In one embodiment, a weighted center estimation is performed on the inverted pricing curve set to obtain a common pricing strategy, including: The weight of each similar day is determined based on the similarity between the temporal characteristics of each similar day and the temporal characteristics of the predicted period. Based on the weights of each similar day and the set of inverted price curves, the Huber loss function is used to perform robust center estimation. The center price curve that minimizes the weighted robust loss is then used as a common price strategy.

[0073] Optionally, since non-time-series features have already been precisely matched as hard constraints during the screening process, the weight allocation for similar days is based solely on time-series common features. Considering the impact on the economic behavior of market-oriented generating units, daily load curves, temperature curves, humidity curves, wind speed curves, and solar intensity curves are selected as key time-series common features.

[0074] First, calculate the relative root mean square error between each similar day and the predicted day on each time-series feature. Then, the relative root mean square errors of each time series feature are weighted and summed to obtain the comprehensive distance metric: (16) In the formula, For similar days The overall distance metric between the predicted date and the predicted date is a very small indicator, meaning that the smaller the value, the more similar the overall external conditions of the two dates are. In this embodiment, the total number of types of common temporal features is [not specified]. These correspond to five characteristics: daily load, temperature, humidity, wind speed, and light intensity, respectively. For the first The weighting coefficients of time-series features in the comprehensive distance metric are used to reflect the differences in the degree of influence of different features on electricity market behavior.

[0075] In this embodiment, the weighting coefficients of each time-series characteristic were determined by experts and are set as follows: daily load curve weight is 0.4, temperature curve weight is 0.2, humidity curve weight is 0.1, wind speed curve weight is 0.15, and solar intensity curve weight is 0.15. These weighting settings reflect the priority of daily load as the most direct factor affecting the supply and demand relationship in the electricity market, while also taking into account the significant impact of meteorological conditions on load and renewable energy output.

[0076] Weights for each similar day based on a comprehensive distance metric. Determined in the following ways: (17) The weight of similar days is inversely proportional to their comprehensive distance metric—the more similar a similar day is to the predicted day in terms of comprehensive external conditions, the greater its weight is assigned in subsequent center estimation. The sum of the weights of all similar days is normalized to 1.

[0077] Optionally, after obtaining the weights of each similar day... Then, the common pricing strategy is obtained by solving the following weighted robust center estimation problem: (18) In the formula: Let be the common pricing strategy vector to be solved; To optimize the variables, the central price curve represents the candidate; For the first Inversion price curves for similar days; This is the robust loss function.

[0078] In this specific implementation, the Huber loss function is used as the robust loss function. The Huber loss function combines the advantages of mean squared error (penalizing samples near the center with squared error) and mean absolute error (penalizing outliers far from the center with linear error), thus fully utilizing the information from normal samples while effectively suppressing the negative impact of outliers. The specific form of the Huber loss function is: (19) In the formula: The distance between the candidate center and the sample is the Euclidean distance. The threshold parameter is used to distinguish between "normal samples" and "abnormal samples." Based on robust statistical standards, the Huber loss threshold... It is usually taken as a constant multiple of the Median Absolute Deviation (MAD) value to achieve a balance between efficiency and robustness.

[0079] The mechanism of the Huber loss function can be understood as follows: when a sample is close to the candidate center ( When a sample is far from the candidate center, it is considered a normal sample, and the loss function uses a quadratic penalty to ensure it has a normal impact on the center estimation; when a sample is far from the candidate center ( When the sample is considered a potential outlier, the loss function is reduced from a squared penalty to a linear penalty, thus significantly weakening its ability to excessively influence the center estimation direction.

[0080] Threshold parameter The choice of directly affects the robustness of the Huber loss function. If If the value is too small, a large number of normal samples will be treated as anomalies, resulting in the loss of too much effective information and reduced estimation efficiency; if... If the threshold is too large, outliers cannot be effectively suppressed, resulting in insufficient robustness. To achieve a reasonable balance between efficiency and robustness, this specific implementation method uses MAD to adaptively determine the threshold parameter.

[0081] In one embodiment, when using the Huber loss function for robust center estimation, the threshold parameter of the Huber loss function is determined in the following manner: Calculate the absolute deviation of the median of the inverted price curve set; scale the absolute deviation of the median using a preset scaling factor to obtain the threshold parameter.

[0082] MAD is a more robust measure of data dispersion than standard deviation, and it is insensitive to outliers. It is defined as follows: (20) In the formula, This indicates that the reported quantity / quotation combination curves of each conventional unit under all similar days obtained by inversion are processed separately in the actual calculation of this embodiment. The MAD values ​​corresponding to the reported quantity parameter vector sequence and the quotation parameter vector sequence are calculated separately. Then it means the first The volume / price combination curve for similar days.

[0083] After obtaining the MAD value, the threshold parameter Determined in the following ways: ; This is a preset scaling factor. According to standard practices in robust statistics, the scaling factor... The typical value range is 1.0 to 2.0, and in this embodiment, we take... This setting provides good robustness against outliers while ensuring estimation efficiency.

[0084] After solving the above weighted robust center estimation optimization problem, the center price curve that minimizes the weighted robust loss is obtained. The central pricing curve represents the common pricing strategy extracted in this step.

[0085] In a mathematical sense, common pricing strategy It can be considered a robust "central estimator" of the distribution of historical pricing behavior. Functionally, the common pricing strategy effectively filters out anomalous inversion samples caused by data noise or occasional market events, thus more reliably capturing and reflecting the general behavioral patterns of the market under normal conditions. This common pricing strategy includes all conventional generator sets... The common quantity and price parameters of the segment tier will serve as inputs for the subsequent day-ahead spot market clearing simulation model, providing stable and interpretable strategy inputs for electricity price forecasting.

[0086] Through the three sub-steps of similar day screening, inversion pricing curve set construction, and weighted robust center estimation, this embodiment realizes the complete extraction from historical behavior samples to common pricing strategies, forming a complete technical path of behavior decoding and strategy extraction, laying a solid foundation for the physical simulation and electricity price prediction in the subsequent step 104.

[0087] Step 104: Input the common pricing strategy and system operating condition data for the forecast period into the pre-built day-ahead spot market clearing simulation model of the power system for simulation, and obtain the nodal electricity price forecast results of the power system during the forecast period.

[0088] In this step, the common pricing strategy obtained in step 103 is used as the input of the pricing curve of each generator unit during the prediction period. The system operating condition data of the prediction period obtained in step 101 is used as the boundary condition and input into the pre-built day-ahead spot market clearing simulation model of the power system for simulation. By solving the security-constrained economic dispatch problem, the nodal price prediction results of each node during the prediction period are obtained.

[0089] The current spot market clearing simulation model is the physical core of power system operation simulation. Its core function is to integrate the learned common bidding strategies of generating units with the real-time system conditions on the forecast date. Under the premise of satisfying the physical operation of the power grid and market rule constraints, it minimizes the total system power purchase cost by optimizing decision variables (mainly the winning bid volume of each generating unit) and directly generates the final predicted value of the nodal price. This optimization problem consists of an objective function and a series of equality and inequality constraints.

[0090] In one embodiment, the day-ahead spot market clearing simulation model is constructed based on a DC power flow model. The objective function is to minimize the total power purchase cost of the power system, and the constraints are power balance constraints, generator operation constraints, and line power flow constraints. The generator operation constraints include upper and lower output limits, ramp rate constraints, and bid segment accumulation constraints. The nodal price prediction results are obtained by solving the Lagrange multipliers of the objective function under the active power balance constraints.

[0091] DC power flow models are a simplified form of power system power flow calculation. Their core assumptions include: transmission line resistance is much smaller than reactance, thus ignoring active power losses; voltage amplitudes at each node are approximately 1 per unit; and the voltage phase angle difference between the two ends of the line is small. Under these assumptions, the active power flow distribution depends only on the node voltage phase angle, thereby simplifying the complex AC power flow problem into a linear one. This significantly reduces computational complexity while maintaining sufficient accuracy, and is therefore widely used in power market clearing simulations.

[0092] The current spot market clearing aims to maximize social welfare. In market operation, social welfare equals the electricity purchase revenue of users minus the generation cost of generators. Since this example assumes that users participate by reporting quantity without bidding, their electricity demand is considered rigid, meaning it does not adjust with price changes. Therefore, maximizing social welfare is equivalent to minimizing the total electricity purchase cost of generators.

[0093] After winning the bid, each generating unit receives its corresponding power generation revenue according to the price of its bid for each segment. The total system power purchase cost is the sum of the bid costs of all winning units. Therefore, the objective function is defined as minimizing the total system bid cost for each trading period within the forecast period: ;(twenty one) In the formula, To use common pricing strategies The extracted first Taiwan unit The price range (RMB / MWh) is derived from the common pricing strategy extracted in step 103 and is a known constant. For the first Taiwanese crew during the period , No. The winning bid volume (MWh) in the bidding segment is the decision variable to be optimized.

[0094] The objective function above indicates that the clearing of each trading period only considers the bid price and the winning bid volume in that period, and the optimization between each period is independent of each other, which is consistent with the actual operating mechanism of time-of-use clearing in the electricity spot market.

[0095] The current market clearing simulation must meet the following physical and security constraints, which must be met at any trading session.

[0096] (1) Power balance constraint.

[0097] Based on the DC power flow model, the system only considers active power, and the net injected power of each node in each time period must be zero, that is, the active power injected into each node and the active power flowing out of each node are kept in balance. ;(twenty two) In the formula: For nodes During the period The active power injection for power generation is the sum of the output of all winning units connected to this node; For nodes During the period The active load is provided by the system operating condition data for the forecast period; This is the set of system nodes.

[0098] Power balance constraints are the most fundamental physical constraints for the operation of power systems, and their corresponding Lagrange multipliers are the electrical energy components of nodal prices.

[0099] (2) Generator set operation constraints.

[0100] Within the framework of optimal DC power flow, generator sets must simultaneously meet both their physical characteristics and market pricing rules, mainly including the following constraints: 2a) Output upper and lower limit constraints.

[0101] The total output of each unit must be between its minimum and maximum technical output. ;(twenty three) In the formula: and The first Minimum technical output and installed capacity of the unit; For the unit In the The winning bid volume for the quoted price range, For the unit In the The total output of the bid segment. This constraint ensures that the unit's operating point is always within its physically feasible range.

[0102] 2b) Climbing rate constraint.

[0103] The rate of change in power output between adjacent time periods of each unit must not exceed its physical ramping capacity limit: ;(twenty four) In the formula: and The first The maximum downhill and uphill rates of the unit; For the unit In the current time period The winning bid volume for the quoted price range, For the unit In the previous period The winning bid volume for the quoted segment. This constraint reflects the physical characteristics of conventional generating units such as thermal power plants, which cannot significantly adjust their output in a short period of time due to thermodynamic and mechanical inertia limitations.

[0104] 2c) Cumulative constraint on price segments.

[0105] The winning bid volume for each unit in each bidding segment must not exceed its declared capacity, and the volume of each segment must be allocated sequentially according to the order of declaration. (25) In the formula: To use common pricing strategies The extracted first Taiwan unit The declared electricity volume (MW) for each bidding segment. The first formula ensures that the winning bid volume for each segment is non-negative and does not exceed the corresponding declared electricity volume; the second formula applies a cumulative constraint, meaning that any previous... The sum of the winning bids for each bidding segment shall not exceed the previous one. The sum of the declared electricity amounts for each quotation segment is used to ensure that the calling order of each segment conforms to the actual calling logic of the quotation segment from low to high.

[0106] (3) Line power flow constraints.

[0107] To ensure the safe and stable operation of the power system under predicted daily operating conditions, the simulation model must strictly meet network physical security constraints. In DC power flow calculations, the influence of node voltage amplitudes is not considered; therefore, only line active power flow constraints need to be considered. That is, the active power flow of each transmission line must always be controlled within its thermal stability limits. (26) In the formula: For the line During the period The active power flow is determined by the node voltage phase angle difference and the line reactance. For the line The thermal stability limit, i.e., the maximum allowable transmission capacity; This is the set of system lines. This constraint is presented in absolute value form, ensuring that the power transmission, whether forward or reverse, does not exceed the line's safety limit.

[0108] After solving the above optimization problem, the Lagrange multipliers (i.e., shadow prices) corresponding to the active power balance constraints at each node in each time period can be obtained. During the period marginal electricity price This is given by the Lagrange multiplier. In AC power flow models, nodal charges typically consist of three parts: energy component, congestion component, and loss component. Since this specific implementation uses a DC power flow model, it assumes that the transmission line resistance is 0 and there is no active power loss in the system; therefore, the loss component is 0. The formula for calculating nodal charges is: (27) In the formula: The unified electrical energy component of the system corresponds to the Lagrange multiplier of the system's active power balance constraint, reflecting the marginal value of electrical energy at the system level, and has the same value for all nodes within the same trading period. For nodes During the period The congestion component corresponds to a linear combination of the Lagrange multipliers of the line power flow constraints, reflecting the differentiated impact of transmission line congestion on the electricity price at each node. For the loss component, in the DC power flow model Therefore, the nodal electricity price can be simplified to the sum of the system-wide uniform electrical energy component and the node-specific congestion component.

[0109] By using the aforementioned day-ahead spot market clearing simulation, we can obtain the nodal electricity price forecasts for all nodes and all trading periods within the forecast period. ( , The forecast results include both the spatiotemporal value signal of electricity and the comprehensive impact of the strategic bidding behavior of market participants, while strictly meeting the physical operation constraints of the power system, thus possessing both physical consistency and behavioral interpretability.

[0110] In this embodiment, the aforementioned day-ahead spot market clearing optimization model is solved using the Matpower toolkit. Matpower is an open-source power system simulation toolkit that runs on the MATLAB platform and provides various calculation functions such as DC power flow, AC power flow, and optimal power flow, enabling efficient solutions to power system optimization operation problems. During the solution process, the common pricing strategy obtained in step 103 is used as the pricing curve for each generating unit. The system operating condition data for the forecast period (including load forecasts for each node, network topology parameters, line parameters, and generating unit technical parameters) are used as input boundary conditions. The DC optimal power flow solution function of Matpower is called for simulation calculation to obtain the optimal output plan for each generating unit and the marginal electricity price for each node and each time period, forming the final day-ahead node price forecast result.

[0111] The following is a case study analysis of the method for forecasting day-ahead spot market electricity prices in the power system.

[0112] In this embodiment, the IEEE 24-node test system is selected as the research case, such as... Figure 3 The diagram shown illustrates the architecture of the IEEE 24-node (BUS1-BUS24) test system, which simulates and generates several days of day-ahead market trading data based on its network topology and unit parameters.

[0113] To ensure comparability with existing electricity price forecasting studies, this application follows common practices in the field and uses the mean absolute percentage error (MAPE) in both the performance comparison and results analysis sections. However, this index may be unstable during the optimization process. Therefore, in the parameter inversion and training phases of the model, we adopt a loss function that is more suitable for optimization, namely Equation (2).

[0114] 1. Input basic parameters and filter similar days.

[0115] This example uses the IEEE 24-node test system as the simulation basis. Its network topology and the baseline load (peak value) of each node are as follows: Figure 3 As shown. Here, the load fluctuations of each node in the topology are considered to be consistent in order to construct the load curve of the entire system. This application assumes that the load of each node has a similar daily variation pattern. Specifically, the following text Figure 4 The curve showing the overall network bidding space is the benchmark, but the distribution of new energy units is unknown.

[0116] On the supply side, conventional generating units participating in the day-ahead spot market bidding are the core entities for meeting system load demands. Key parameters such as the connection node, installed capacity, minimum technical output, and ramp rate of each unit are detailed in Appendix 1. It should be noted that the unit at node 14 is a synchronous condenser, whose main function is to provide reactive power support to maintain voltage stability; it does not have active power output capability and is therefore not included in the day-ahead electricity market trading scope. Furthermore, in this embodiment, the price range for all conventional generating units is uniformly set at [0, 1000] yuan / MWh, and this price upper and lower limits remain unchanged throughout all simulated trading days.

[0117] Table 1 Basic Information of Conventional Units On the demand side, the core input to this example is the day-ahead forecast net load curve, which represents the electricity demand that needs to be met through competitive bidding in the day-ahead spot market. For example... Figure 4 The historical date-ahead market bidding space annual time series curve shown is illustrated. This curve represents the Class B bidding space, and its value has deducted the physical delivery portion, such as the electricity volume allocated under medium- and long-term contracts for each market-based generating unit. This forecast value is mainly affected by meteorological conditions and date type (such as weekday / holiday) and includes inherent forecast error disturbances.

[0118] In actual market operations, the supply and demand relationship represented by this bidding space is a key basis for conventional generating units to formulate their quotation strategies. However, due to the constant presence of forecast disturbances and uncertainties, directly relying on single-day load curves for behavioral extrapolation may lead to unstable results. Therefore, to enhance the robustness and generalization ability of the model, in addition to the day-ahead market bidding space feature, this study also extracts common features related to market behavior from it, which serve as stable inputs for subsequent screening of similar days and learning of common strategies.

[0119] In selecting common features, this application mainly relies on the source driving factors that affect electricity load and market behavior. Specifically, two types of features are selected: one is time attributes (such as seasons, weekdays / holidays), used to characterize the periodic patterns of social activities; the other is meteorological attributes (such as temperature, humidity, wind speed, and light intensity), used to reflect the direct impact of weather conditions on load and the potential of new energy power generation.

[0120] On the other hand, in this example, the renewable energy units do not participate in market transactions; their actual output will be uniformly released as the hourly output of wind and solar power across the entire grid after the transaction concludes. This means that although transaction users can only rely on day-ahead wind and solar forecast data with inherent errors when making decisions, they can retrospectively verify the renewable energy power generation situation of that day through publicly available actual output information. Therefore, when extracting features, it is not necessary to introduce the outcome indicator of "grid-wide renewable energy output"; natural resource forecast data such as wind speed and solar irradiance can be directly used as candidate common features, such as... Figure 5 The diagrams shown in the image are: annual time series curves of historical temperature, annual time series curves of historical humidity, annual time series curves of historical wind speed per unit value, and annual time series curves of historical light intensity per unit value.

[0121] The common characteristics of the non-time series of the study are shown in Table 2. The simulation scenario in this example is set in a coastal province in southern my country, so the seasons are divided into spring (February–April), summer (May–July), autumn (August–October) and winter (November–January).

[0122] Table 2 After screening for common features of non-time series data, 27 summer holidays were extracted. Further, common time series features such as bidding space for the study days were obtained. The allowable error ranges for each time series common feature curve are shown in Table 3.

[0123] Table 3 Therefore, according to equation (15), three representative similar days were extracted, and the common characteristics of the time series of the study day and all similar days were as follows: Figure 6As shown in Figure ae, the dotted line graph at the center represents the common time series characteristic curves of the study days, and the common time series characteristics of similar days are presented in the form of error bands.

[0124] Therefore, the weight distribution of each similar day can be calculated based on the comprehensive similarity according to equations (16) and (17), as shown in Table 4.

[0125] Table 4 2. Inversion of economic behavior of market entities.

[0126] In this simulation embodiment, since the inversion of market entity economic behavior is a high-dimensional, ill-posed inverse problem, the specific parameter settings for the GWO algorithm in this high-dimensional space are shown in Table 5 to ensure its global exploration capability and convergence accuracy. For the similar days extracted above, this embodiment specifically describes the inversion process of market entity economic behavior for each day. The inversion result is an acceptable range in terms of both quantity and price deviations; based on the market entity economic behavior inversion method, the acceptable error for the dispatch center is set at 5%, while ensuring that the unit's reported quantity and price still meet all constraints.

[0127] Table 5 Here, we present the inversion results for the most similar day 2 to the study day from Table 4. The tiered pricing curve for conventional generator sets submitted before the date of study is as follows: Figure 7 As shown, where, Figure 7 In the diagram, 'a' represents the inversion result of the quotation and pricing for a 200MW unit. Figure 7 Figure b is a schematic diagram of the inversion results of the quotation and pricing for a 300MW unit. Figure 7 In the middle, c is a schematic diagram of the inversion results of the quotation and pricing for a 400MW unit. Figure 7 The diagram in 'd' shows the inversion results of the quantity and price quotation for units of 500MW and above. The inversion results are all 10-segment stepped quantity and price quotation curves. Each segment has a reasonable error (5%) in both quantity and price. Only the curves of each unit at the error center are shown here.

[0128] According to the inversion Figure 7 The tiered bidding curve diagram of the generating units shown in the diagram (AD) reveals a market behavior pattern with a clear economic explanation. First, the initial bid for all units is the market floor price (0 yuan / MWh), reflecting the rational strategy of conventional generating units to bid low to secure the bid in order to ensure minimum technical output operation and avoid shutdown and restart costs.

[0129] Secondly, the middle section of the price quotation curve shows a gradual upward trend, which is consistent with the traditional pricing logic based on increasing marginal costs. However, at the end of the curve, there is a significant increase in prices, which can be explained as a strategic pricing behavior targeting peak loads in the system. During periods of scarce power generation capacity, the market power of generating units is enhanced, and they can obtain excess profits by raising prices.

[0130] Furthermore, horizontal comparisons show that units with larger installed capacity generally have lower overall bid levels, stemming from the economies of scale in power generation. The lower unit generation cost of large-capacity units allows them to adopt more competitive low-price strategies, securing a share of baseload power and aligning with their market positioning as a baseload power source.

[0131] The reliability of the inversion results for similar day 2 behavior will be discussed next.

[0132] (1) Nodal price fitting error.

[0133] On a similar day 2, the actual node electricity price diagram disclosed by the market clearing is as follows: Figure 8 As shown in Figure a, the distribution of nodal electricity price fitting errors for each time period is illustrated in the diagram below. Figure 8 As shown in b. Combined with Figure 8 In section b, the fitting errors for all nodes and time periods (24 hours × 24 nodes = 576 samples) were statistically analyzed, and their distribution is summarized in Table 6. Table 6 shows that as many as 90.1% of the samples had fitting errors controlled within 10%, indicating that the Tikhonov regularized inversion framework proposed in this application can accurately reproduce market electricity prices in most cases. Furthermore, samples with errors exceeding 15% accounted for 7.12%. These outliers are usually related to extreme congestion at specific nodes or small changes in unit output at critical states, and their impact is limited to a local spatiotemporal range.

[0134] Table 6 (2) The error in the fitting of the unit start-up and shutdown and output curves.

[0135] In Similar Day 2, the comparison between the simulated output of each unit based on the inverted price curve and the actual market clearing results is as follows: Figure 9 As shown in aj, where, Figure 9 In the diagram, aj represents the output curve fitting diagrams for units 1-10. Analysis shows that the inversion results of each unit's output are highly consistent with the actual results in terms of trend, and the start-up and shutdown states of all units are completely consistent with the actual information disclosed in the market. Therefore, the fitting error of the unit start-up and shutdown states calculated according to equation (6) is zero. The average fitting error of the unit output curves is calculated to be 5.81%, indicating a good fitting effect.

[0136] (3) Stable functional.

[0137] according to Figure 7 In the first case, the price tiers of each conventional generator set show a non-decreasing trend, and the price range is all within the given [0, 1000] yuan / MWh. Therefore, the monotonicity constraint term and the price boundary constraint term are both 0. On the other hand, the sum of the quoted quantities of each conventional generator set is equal to its installed capacity, so the capacity integrity constraint term is also 0. Therefore, equation (8) can be simplified to equation (13), and only the price smoothness constraint term needs to be studied.

[0138] For the inversion results of all similar days, the smoothness constraint term is calculated. The GWO optimization results show that the calculation result of equation (11) is 0.0779, of which the smoothness of reporting volume and the smoothness of price quotation are 0.0809 and 0.0749, respectively; this shows that the average change of the reporting volume step of each unit and each section is 8.09%, and the average change of the price quotation step is 7.49%, which is at a relatively ideal level.

[0139] 3. Learning outcomes of common pricing strategies.

[0140] According to equation (20), the residual MAD values ​​of all similar daily report volume and price inversion results in this example are 0.065 and 0.024, respectively. Therefore, the Huber loss thresholds for report volume and price are 0.0975 and 0.036, respectively.

[0141] Furthermore, by extracting the quotation curves of conventional generating units for similar days, the quotation curves of each conventional generating unit for the study day can be finally obtained, such as... Figure 10 The diagram showing the common learning day's reported quantity and price curve results also only presents the reported quantity and price curves of each unit at the error center (the error range acceptable to the dispatch center is 5%). The first segment of reported quantity for each unit is the minimum technical output, and the price is the "floor price".

[0142] 4. Market clearing simulation prediction results.

[0143] Based on Figure 10 The resulting final quotation curve will be used as the reporting strategy for each unit on the research day and imported into [the system / platform]. Figure 3 In the IEEE 24-node system topology shown, and in conjunction with the predicted net load curve for the study day, a day-ahead market clearing simulation is performed to obtain the electricity price prediction results for all nodes and all time periods.

[0144] The simulated electricity price predictions for each node are as follows: Figure 11 As shown in ab. To evaluate prediction accuracy, Figure 11 Figure 'a' shows a schematic diagram illustrating the distribution of prediction errors for all nodes across different time periods. Figure 11The diagram in section b further illustrates the time-series comparison of the daytime electricity price forecast and actual value for typical nodes throughout the day.

[0145] Based on the comprehensive forecast results across all time periods and all nodes, the overall error distribution of nodal electricity prices is shown in Table 7. Analysis indicates that the forecast accuracy remains high under most operating conditions: up to 87.15% of the samples (i.e., the total percentage of errors ≤10%) have forecast errors within 10%, with approximately 59.72% of these samples having errors controlled within 5%, resulting in a comprehensive nodal price error (MAPE) of 4.79%. This demonstrates the effectiveness of the proposed method in day-ahead electricity price forecasting.

[0146] Table 7 Further analysis of the error structure reveals a significant temporal distribution. Samples with errors exceeding 15% account for 7.12%, and these high-error points are not randomly distributed but primarily concentrated between 9:00 AM and 10:00 AM. This period typically coincides with a rapid load ramp-up, and the combined uncertainty of renewable energy output and load forecasting can amplify subtle changes in the system equilibrium point and congestion patterns through market clearing models, leading to sensitive fluctuations in nodal electricity prices. A very small percentage of nodes exhibited forecast errors exceeding 20% ​​during specific periods. This is mainly attributed to random disturbances artificially introduced into the simulation data, which correspond to unpredictable extreme events or noise in the real market. In contrast, forecast accuracy is generally higher during relatively stable peak and off-peak periods.

[0147] Figure 12 This is a schematic diagram of the unit's output curve and prediction error, from... Figure 12 It can be seen that the clearing simulation based on the reported quantity and price curve obtained from common learning is completely consistent with the actual start-up and shutdown status of the units in each time period. At the same time, the predicted output values ​​of each unit are highly consistent with the actual values, and the prediction error is controlled within 10% for most time periods, with only a few units showing larger deviations in a few time periods.

[0148] 5. Comparison of prediction methods.

[0149] To verify the effectiveness of the proposed method, two mainstream data-driven spot electricity price forecasting methods, such as MHG-LSTM and ICEEMDAN-ARIMA-TimesNet, were selected based on the IEEE 24-bus system and compared with the proposed day-ahead spot market electricity price forecasting method (TBR-CL method). The day-ahead spot electricity price forecasting results are shown in Table 8. Figure 13 As shown, Figure 13 The diagram shown is a time series comparison of errors in spot electricity price forecasting methods.

[0150] Table 8 As shown in Table 8, the day-ahead spot market electricity price forecasting method (TBR-CL method) proposed in this application, although relatively less computationally efficient than the current mainstream data-driven forecasting methods, demonstrates a certain advantage in the accuracy of electricity price forecasting, with its average error reduced by approximately 25% to 30% compared to mainstream data-driven forecasting methods. More importantly, the day-ahead spot market electricity price forecasting method (TBR-CL method) proposed in this application is superior in terms of physical interpretability, providing economic behavioral interpretability support for unit output while obtaining electricity price forecast results.

[0151] Furthermore, the time-series errors in electricity price forecasting using various methods are considered. For example... Figure 13 As shown, the time-series errors of the MHG-LSTM and ICEEMDAN-ARIMA-TimesNet methods are relatively stable overall, with error peaks mainly occurring between 8–10 pm (i.e., the rapid load increase phase) and 4–20 pm (i.e., the peak load period). In contrast, the day-ahead spot market electricity price forecasting method for the power system proposed in this application exhibits higher forecast accuracy across all time periods, especially during peak load periods; however, its forecast accuracy is slightly inferior to mainstream data-driven methods during the rapid load increase phase.

[0152] Therefore, the day-ahead spot market electricity price forecasting method proposed in this application is generally superior to mainstream data-driven methods in terms of forecasting accuracy. It can also overcome the shortcomings of poor physical interpretability of mainstream data-driven methods and avoid the limitation of oversimplification of behavioral assumptions by a single physical model. It provides an effective solution for high-precision, interpretable, and robust short-term electricity price forecasting in the electricity market environment.

[0153] In current day-ahead spot electricity price forecasting research, to overcome the limitations of traditional data-driven models with weak physical foundations and poor interpretability, as well as the oversimplification of market participant behavior assumptions in existing physical models, this application proposes a method for forecasting day-ahead spot market electricity prices in the power system. Through theoretical modeling and simulation analysis, the following main conclusions are drawn: (1) The constructed Tikhonov regularized behavior inversion framework can robustly invert the unit price curve that conforms to market rules from publicly available market data, realizing accurate decoding of price signals into behavior signals; (2) The proposed weighted robust commonality learning method can extract pricing strategies that reflect the general laws of the market from historical behavior samples, providing stable and reliable behavioral input for prediction; (3) The integrated prediction paradigm of “behavior inversion-strategy learning-physical simulation” has significantly improved prediction accuracy compared with mainstream data-driven prediction methods, while ensuring that the prediction results are highly consistent with the physical constraints of the power grid.

[0154] With the deepening of electricity market reforms and the increasing complexity of decision-making by stakeholders, higher demands are placed on the accuracy, interpretability, and decision support capabilities of electricity price forecasts. The method proposed in this application provides market participants and operators with an analytical tool that combines theoretical rigor with practical feasibility. Future research will focus on more complex market scenarios, such as considering the inversion and prediction of strategic behaviors by renewable energy power plants and electricity users actively participating in bidding.

[0155] This invention provides a method for predicting day-ahead spot market electricity prices in a power system. The method involves acquiring historical market clearing data and system operating condition data for the prediction period. Based on Tikhonov regularization, a variational optimization model is constructed and solved using market clearing data as observation data and the bidding curves of generating units in the power system as parameters to be inverted. This yields the inverted bidding curves of each generating unit within the historical period. Common bidding strategies are extracted from the inverted bidding curves. These common bidding strategies, along with the system operating condition data for the prediction period, are input into a pre-constructed day-ahead spot market clearing simulation model for simulation, resulting in the predicted nodal electricity prices for the power system during the prediction period.

[0156] Traditional data-driven methods rely directly on the statistical correlation between historical electricity prices and load data for prediction. When market supply and demand or external conditions change, this statistical correlation may drift, leading to increased prediction bias. In contrast, this invention introduces Tikhonov regularized inversion, establishing predictions on a causal chain of "behavioral inversion → common pricing strategy extraction → physical simulation," rather than simple data statistical fitting, fundamentally improving the reliability of predictions.

[0157] Traditional data-driven methods rely solely on historical statistical patterns for extrapolation, which may lead to predictions that deviate from the actual operational constraints of the power system, resulting in physically infeasible predictions. In contrast, the predictions provided by this invention are generated under strict adherence to the aforementioned physical and safety constraints. The simulation errors of unit output and start-up / shutdown states remain within acceptable limits, ensuring a high degree of consistency between the predictions and actual grid operational constraints. This physical consistency allows the predictions to not only be used as a reference but also to directly serve grid dispatching decisions.

[0158] Traditional data-driven methods cannot explain why electricity prices are predicted to a certain value, and their decision-making process lacks transparency. However, the nodal price predicted by this invention can be traced back to the common bidding strategies of generator units under specific load and weather conditions. Furthermore, the composition of each component of the nodal price (electrical energy component and congestion component) can be explained through the constraints in the clearing model and the Lagrange multiplier. The prediction results have complete interpretability at the levels of behavioral logic and market mechanism.

[0159] Traditional data-driven methods suffer from a sharp decline in generalization ability when structural changes occur in the market environment, trading rules, or the behavior patterns of market participants, due to the inconsistency between the distribution of training data and the prediction scenario. In contrast, this invention relies solely on recent historical data, continuously updating the behavioral sample library through similar day filtering, and filtering out noise and anomalies through weighted robust commonality extraction. This allows it to reflect the dynamic evolution of market behavior patterns in real time and maintain good predictive performance even during market-oriented reform transitions or rule adjustments.

[0160] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0161] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0162] Figure 14 The diagram shows a schematic representation of a day-ahead spot market electricity price forecasting device for a power system according to an embodiment of the present invention. For ease of explanation, only the parts relevant to the embodiment of the present invention are shown, and are described in detail below: like Figure 14 As shown, the day-ahead spot market electricity price forecasting device for the power system includes: an acquisition module 141, an inversion module 142, a common price extraction module 143, and a simulation module 144.

[0163] The acquisition module 141 is used to acquire market clearing data for historical periods of the power system and system operating condition data for predicted periods. Inversion module 142 is used to construct and solve a variational optimization model based on Tikhonov regularization, using market clearing data as observation data and the reporting and pricing curves of generator units in the power system as parameters to be inverted, to obtain the inversion pricing curves of each generator unit in the historical period. Common pricing extraction module 143 is used to extract common pricing strategies from the inverted pricing curve; The simulation module 144 is used to input common pricing strategies and system operating condition data for the forecast period into a pre-built day-ahead spot market clearing simulation model of the power system for simulation, and to obtain the nodal electricity price forecast results of the power system during the forecast period.

[0164] In one possible implementation, the goal of the variational optimization model is to make the simulation clearing result obtained from the power system day-ahead spot market clearing simulation of the parameters to be inverted approximate the market clearing data. Variational optimization models include fidelity terms and stable functionals; The fidelity term is used to constrain the deviation between the simulation clearing results obtained from the simulation of the parameters to be inverted and the market clearing data; the stable functional is used to embed the structural prior constraints of the parameters to be inverted.

[0165] In one possible implementation, the fidelity term includes a weighted sum of at least one of the following error terms: nodal price fitting error, unit output fitting error, and unit start-up / shutdown state fitting error; the nodal price fitting error is the deviation between the simulated nodal price and the actual nodal price in the market-cleared data; the unit output fitting error is the deviation between the simulated unit output and the actual unit output in the market-cleared data; and the unit start-up / shutdown state fitting error is the deviation between the simulated unit start-up / shutdown state and the actual unit start-up / shutdown state in the market-cleared data.

[0166] In one possible implementation, the stable functional includes at least one of the following constraints: a monotonicity constraint to constrain the bidding ladder in the parameters to be inverted to have a non-decreasing trend; a capacity integrity constraint to constrain the sum of the bidding ladders in the parameters to be inverted to be equal to the installed capacity of the corresponding generating unit; a smoothness constraint to constrain the penalty value of the change between adjacent ladders in the parameters to be inverted; and a bidding boundary constraint to constrain the bidding ladder in the parameters to be inverted to be within the bidding range allowed by the electricity market.

[0167] In one possible implementation, when the common pricing extraction module 143 extracts the common pricing strategy from the inverted pricing curve, it is used for: Select multiple similar days from historical time periods that meet the preset similarity criteria for the predicted time period; The inverted price curves corresponding to each similar day are obtained respectively, forming a set of inverted price curves; By performing weighted center estimation on the inverted pricing curve set, a common pricing strategy is obtained.

[0168] In one possible implementation, the common quotation extraction module 143 filters out multiple similar days and times from historical time periods that meet preset similarity conditions with the predicted time period, for the purpose of: Non-time series features and time series features for historical periods and predicted periods are obtained respectively. Non-time series features include date attributes, and time series features include power system load curves and meteorological factor curves. The historical periods with the same non-temporal characteristics as the predicted period are used as the initial sample; Calculate the relative root mean square error between the temporal characteristics of each preliminary sample and the temporal characteristics of the predicted period, and determine the preliminary samples with a relative root mean square error less than a preset threshold as similar days.

[0169] In one possible implementation, the common pricing extraction module 143 performs weighted center estimation on the inverted pricing curve set to obtain the common pricing strategy, which is then used for: The weight of each similar day is determined based on the similarity between the temporal characteristics of each similar day and the temporal characteristics of the predicted period. Based on the weights of each similar day and the set of inverted price curves, the Huber loss function is used to perform robust center estimation. The center price curve that minimizes the weighted robust loss is then used as a common price strategy.

[0170] In one possible implementation, when the common quote extraction module 143 uses the Huber loss function for robust center estimation, the threshold parameter of the Huber loss function is determined in the following way: Calculate the absolute deviation of the median of the inverted price curve set; scale the absolute deviation of the median using a preset scaling factor to obtain the threshold parameter.

[0171] In one possible implementation, the day-ahead spot market clearing simulation model is built on a DC power flow model. The objective function is to minimize the total power purchase cost of the power system, and the constraints are power balance constraints, generator operation constraints, and line power flow constraints. The generator operation constraints include upper and lower limits of output, ramp rate constraints, and bid segment accumulation constraints. The nodal price prediction results are obtained by solving the Lagrange multipliers of the objective function under the active power balance constraints.

[0172] The above embodiments provide a power system day-ahead spot market price forecasting device. The device acquires historical market clearing data and system operating condition data for the forecast period through an acquisition module. An inversion module, based on Tikhonov regularization, uses market clearing data as observation data and the power system generator unit quotation curves as inversion parameters to construct and solve a variational optimization model, obtaining the inversion quotation curves for each generator unit within the historical period. A common quotation extraction module extracts common quotation strategies from the inversion quotation curves. A simulation module inputs the common quotation strategies and the system operating condition data for the forecast period into a pre-constructed power system day-ahead spot market clearing simulation model for simulation, obtaining the nodal price forecast results for the power system during the forecast period.

[0173] Traditional data-driven methods rely directly on the statistical correlation between historical electricity prices and load data for prediction. When market supply and demand or external conditions change, this statistical correlation may drift, leading to increased prediction bias. In contrast, this invention introduces Tikhonov regularized inversion, establishing predictions on a causal chain of "behavioral inversion → common pricing strategy extraction → physical simulation," rather than simple data statistical fitting, fundamentally improving the reliability of predictions.

[0174] Traditional data-driven methods rely solely on historical statistical patterns for extrapolation, which may lead to predictions that deviate from the actual operational constraints of the power system, resulting in physically infeasible predictions. In contrast, the predictions provided by this invention are generated under strict adherence to the aforementioned physical and safety constraints. The simulation errors of unit output and start-up / shutdown states remain within acceptable limits, ensuring a high degree of consistency between the predictions and actual grid operational constraints. This physical consistency allows the predictions to not only be used as a reference but also to directly serve grid dispatching decisions.

[0175] Traditional data-driven methods cannot explain why electricity prices are predicted to a certain value, and their decision-making process lacks transparency. However, the nodal price predicted by this invention can be traced back to the common bidding strategies of generator units under specific load and weather conditions. Furthermore, the composition of each component of the nodal price (electrical energy component and congestion component) can be explained through the constraints in the clearing model and the Lagrange multiplier. The prediction results have complete interpretability at the levels of behavioral logic and market mechanism.

[0176] Traditional data-driven methods suffer from a sharp decline in generalization ability when structural changes occur in the market environment, trading rules, or the behavior patterns of market participants, due to the inconsistency between the distribution of training data and the prediction scenario. In contrast, this invention relies solely on recent historical data, continuously updating the behavioral sample library through similar day filtering, and filtering out noise and anomalies through weighted robust commonality extraction. This allows it to reflect the dynamic evolution of market behavior patterns in real time and maintain good predictive performance even during market-oriented reform transitions or rule adjustments.

[0177] Figure 15 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 15 As shown, the electronic device 15 of this embodiment includes a processor 150 and a memory 151. The memory 151 stores a computer program 152. When the processor 150 executes the computer program 152, it implements the steps in the various method embodiments described above. Alternatively, when the processor 150 executes the computer program 152, it implements the functions of each module / unit in the various device embodiments described above.

[0178] For example, computer program 152 may be divided into one or more modules / units, which are stored in memory 151 and executed by processor 150 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 152 in electronic device 15.

[0179] Electronic device 15 may include, but is not limited to, processor 150 and memory 151. Those skilled in the art will understand that... Figure 15 This is merely an example of electronic device 15 and does not constitute a limitation on electronic device 15. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 15 may also include input / output devices, network access devices, buses, etc.

[0180] The processor 150 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0181] The memory 151 can be an internal storage unit of the electronic device 15, such as a hard disk or RAM of the electronic device 15. The memory 151 can also be an external storage device of the electronic device 15, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the electronic device 15. Furthermore, the memory 151 can include both internal and external storage units of the electronic device 15. The memory 151 is used to store the computer program 152 and other programs and data required by the electronic device 15. The memory 151 can also be used to temporarily store data that has been output or will be output.

[0182] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0183] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0184] This invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0185] Computer programs include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0186] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0187] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for predicting day-ahead spot market electricity prices in a power system, characterized in that, include: Acquire market clearing data for historical periods of the power system and system operating condition data for predicted periods; Based on Tikhonov regularization, using the market clearing data as observation data and the reporting and pricing curves of generator units in the power system as parameters to be inverted, a variational optimization model is constructed and solved to obtain the inverted pricing curves of each generator unit in the historical period. Extract common pricing strategies from the inverted pricing curves; The common pricing strategy and the system operating condition data for the forecast period are input into a pre-built day-ahead spot market clearing simulation model of the power system for simulation, so as to obtain the nodal electricity price forecast result of the power system during the forecast period.

2. The method for forecasting day-ahead spot market electricity prices in the power system according to claim 1, characterized in that, The objective of the variational optimization model is to make the simulation clearing result obtained from the power system day-ahead spot market clearing simulation of the parameters to be inverted approximate the market clearing data. The variational optimization model includes a fidelity term and a stable functional; The fidelity term is used to constrain the deviation between the simulation clearing result obtained from the simulation of the parameters to be inverted and the market clearing data; The stable functional is used to embed structural prior constraints on the parameters to be inverted.

3. The method for forecasting day-ahead spot market electricity prices in the power system according to claim 2, characterized in that, The fidelity factor includes a weighted sum of at least one of the following error factors: nodal price fitting error, unit output fitting error, and unit start-up / shutdown state fitting error; the nodal price fitting error is the deviation between the simulated nodal price and the actual nodal price in the market clearing data; the unit output fitting error is the deviation between the simulated unit output and the actual unit output in the market clearing data; and the unit start-up / shutdown state fitting error is the deviation between the simulated unit start-up / shutdown state and the actual unit start-up / shutdown state in the market clearing data.

4. The method for forecasting day-ahead spot market electricity prices in the power system according to claim 2, characterized in that, The stable functional includes at least one of the following constraints: a monotonicity constraint, used to constrain the bidding steps in the parameters to be inverted to have a non-decreasing trend; a capacity integrity constraint, used to constrain the sum of the bidding steps in the parameters to be inverted to be equal to the installed capacity of the corresponding generating unit; a smoothness constraint, used to constrain the penalty value of the change between adjacent steps in the parameters to be inverted; and a bidding boundary constraint, used to constrain the bidding steps in the parameters to be inverted to be within the bidding range allowed by the electricity market.

5. The method for forecasting day-ahead spot market electricity prices in the power system according to any one of claims 1-4, characterized in that, The extraction of common pricing strategies from the inverted pricing curve includes: Select multiple similar days from the historical time period that meet the preset similarity criteria for the predicted time period; The inversion price curves corresponding to each of the aforementioned similar days are obtained respectively, forming a set of inversion price curves; The common pricing strategy is obtained by performing a weighted center estimation on the set of inverted pricing curves.

6. The method for forecasting day-ahead spot market electricity prices in the power system according to claim 5, characterized in that, The step of selecting multiple similar days from the historical time period that meet the preset similarity criteria with the predicted time period includes: The non-time-series features and time-series features of the historical period and the predicted period are obtained respectively. The non-time-series features include date attributes, and the time-series features include power system load curves and meteorological factor curves. The historical periods that have the same non-temporal characteristics as the predicted period are selected as the initial samples; Calculate the relative root mean square error between the temporal characteristics of each of the initial selected samples and the temporal characteristics of the predicted period, and determine the initial selected samples whose relative root mean square error is less than a preset threshold as the similar days.

7. The method for forecasting day-ahead spot market electricity prices in the power system according to claim 6, characterized in that, The step of performing weighted center estimation on the inverted pricing curve set to obtain the common pricing strategy includes: The weight of each similar day is determined based on the similarity between the temporal characteristics of each similar day and the temporal characteristics of the predicted time period. Based on the weights of each similar day and the set of inverted pricing curves, the Huber loss function is used to perform robust center estimation, and the center pricing curve that minimizes the weighted robust loss is solved as the common pricing strategy.

8. The method for forecasting day-ahead spot market electricity prices in the power system according to claim 7, characterized in that, When using the Huber loss function for robust center estimation, the threshold parameter of the Huber loss function is determined in the following way: Calculate the median absolute deviation of the inverted price curve set; scale the median absolute deviation using a preset scaling factor to obtain a threshold parameter.

9. The method for forecasting day-ahead spot market electricity prices in the power system according to claim 1, characterized in that, The day-ahead spot market clearing simulation model is built on a DC power flow model. The objective function is to minimize the total power purchase cost of the power system. The constraints are power balance constraints, generator operation constraints, and line power flow constraints. The generator operation constraints include upper and lower output limits, ramp rate constraints, and bid segment accumulation constraints. The nodal price prediction results are obtained by solving the Lagrange multipliers of the objective function under the active power balance constraints.

10. A device for predicting day-ahead spot market electricity prices in a power system, characterized in that, include: The acquisition module is used to acquire market clearing data for historical periods of the power system and system operating condition data for predicted periods. The inversion module is used to construct and solve a variational optimization model based on Tikhonov regularization, using the market clearing data as the observation data and the reporting and pricing curves of generator units in the power system as the parameters to be inverted, to obtain the inversion pricing curves of each generator unit in the historical period. A common pricing extraction module is used to extract common pricing strategies from the inverted pricing curve; The simulation module is used to input the common pricing strategy and the system operating condition data for the forecast period into a pre-built day-ahead spot market clearing simulation model of the power system for simulation, so as to obtain the nodal electricity price forecast result of the power system during the forecast period.

Images