Spot market bidding parameter prediction method based on improved bounded rationality model
By improving the bounded rationality model for predicting bidding parameters in the spot market, and by constructing behavioral profiles and clustering to screen representative competitors, combined with belief update probability and psychological confidence variables, the problems of insufficient sample representativeness and insufficient multi-scale feature capture in existing technologies are solved, and more accurate bidding parameter prediction is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUAZHONG UNIV OF SCI & TECH
- Filing Date
- 2026-04-29
- Publication Date
- 2026-07-24
AI Technical Summary
Existing computer prediction schemes suffer from problems such as insufficient sample representativeness, difficulty in modeling bounded rationality behavior, insufficient multi-scale feature capture, and difficulty in fusing multi-source heterogeneous inputs in predicting spot market bidding parameters, resulting in inadequate prediction performance.
By constructing an improved bounded rationality model, statistical and temporal morphological features are extracted based on competitors' historical bidding parameters. A behavioral map is constructed and clustered to select representative competitors. Combining belief update probability and psychological confidence variables, a predictive model framework is built, incorporating shock response terms and boundary constraints to achieve personalized prediction.
It significantly improves the representativeness and accuracy of bidding parameter prediction, can finely characterize bidding behavior features, and improves the generalization ability and performance of the prediction model.
Smart Images
Figure CN122453447A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of computer information processing and artificial intelligence technology, and more specifically, relates to a method for predicting spot market bidding parameters based on an improved bounded rationality model. Background Technology
[0002] As the informatization of the electricity spot market continues to improve, the market clearing system has accumulated massive amounts of historical bidding and clearing data. Market participants typically rely on computer forecasting models to analyze competitors' bidding parameters to support data preparation for power generation planning and bidding decisions. However, existing computer forecasting solutions still have several technical shortcomings in data processing and model building.
[0003] First, in the competitor sample selection stage, existing methods mostly rely on manual rules or subjective experience, lacking an objective data mining mechanism that can automatically identify representative competitors from massive historical data. This results in insufficient representativeness of the samples input into the prediction model, leading to increased data bias in subsequent prediction results. Second, existing prediction models are generally based on the assumption of perfect rationality or simple statistical fitting, and their model structures are unable to accurately characterize the bounded rationality behaviors such as conservative bias and trend extrapolation bias in actual bidding behavior. The model's insufficient ability to learn and express actual bidding behavior directly leads to high prediction errors. Third, competitor bidding behavior data has complex multi-scale characteristics, including time-series dependence, local volatility, intraday periodicity, and inter-subject heterogeneity. Existing technologies, if only using single-scale feature extraction or modeling, cannot fully capture these multi-scale characteristics, resulting in weak generalization ability of the prediction model on unseen data. In addition, bidding parameters are affected by multiple input sources, including long-term reference levels, short-term market signals, and sudden shocks, and different competitors exhibit significant heterogeneity in their response strength to similar signals. Existing methods struggle to adaptively fuse such multi-source heterogeneous inputs and estimate differentiated response parameters, resulting in insufficient feature utilization and model underfitting. In summary, existing computer prediction schemes generally suffer from insufficient predictive performance. Summary of the Invention
[0004] In view of the shortcomings of the prior art, the purpose of this application is to provide a spot market bidding parameter prediction method based on an improved bounded rationality model, which aims to solve the technical problem of insufficient performance of existing computer prediction schemes for bidding parameters.
[0005] The first aspect of this application relates to a method for predicting spot market bidding parameters based on an improved bounded rationality model, comprising: The historical bidding parameters of competitors before the prediction period are input into the corresponding bidding parameter prediction model to obtain the bidding parameters for the prediction period; the bidding parameter prediction model is obtained through the following steps: The comprehensive similarity between competitors is measured based on the statistical characteristics and temporal patterns of their historical bidding parameters. Then, a behavioral graph containing all competitors is constructed based on the comprehensive similarity. After clustering the behavioral graph, a representative competitor is selected from each cluster. A predictive model framework is constructed based on an improved bounded rationality model. In this framework, the subjective probability that a representative competitor believes they are in a trend extrapolation state is measured by the belief update probability; the strength of the representative competitor's confirmation of the decision is measured by the psychological confidence variable; and the bidding parameters of the representative competitor are predicted by linking the belief update probability and the psychological confidence variable. By fitting the historical bidding parameters of each representative competitor to the prediction model framework, the prediction model parameters of each representative competitor are obtained. Substituting the prediction model parameters of each representative competitor into the prediction model framework, the bidding parameter prediction model corresponding to each representative competitor is obtained.
[0006] Preferably, the representative competitors are selected through the following steps: Multiple statistical features are extracted from competitors' historical bidding parameters and concatenated to obtain behavioral feature vectors. The similarity of statistical features between different competitors is measured based on the distance between these behavioral feature vectors. Standardized bidding trajectories are obtained by standardizing the historical bidding parameters of competitors. The similarity of temporal morphological features between different competitors is measured based on the distance between the standardized bidding trajectories. By integrating the statistical feature similarity and temporal morphological feature similarity, a comprehensive similarity among different competitors is obtained; based on the comprehensive similarity, a behavioral map containing all competitors is constructed. After performing spectral clustering on the behavioral graph, multiple clusters are obtained. Based on the centroid selection mechanism, a representative competitor is selected from each cluster.
[0007] Preferably, the temporal morphological similarity between different competitors is obtained through the following steps: Standardization eliminates the dimensional differences of bidding parameters at different scales, resulting in a standardized bidding trajectory. The distance between standardized bidding trajectories is calculated using a dynamic time warping method to represent the temporal morphological similarity between different competitors.
[0008] Preferably, the behavioral map is obtained through the following steps: Gaussian kernel mapping is applied to the statistical feature similarity and the temporal morphological feature similarity respectively to obtain the Gaussian kernel similarity of the statistical features and the Gaussian kernel similarity of the temporal morphological features; We use a weighted fusion of Gaussian kernel similarity based on statistical features and temporal morphological features to obtain the comprehensive similarity between different competitors. Construct a behavioral graph containing all competitors based on the comprehensive similarity between each pair of all competitors. ,in A matrix composed of all competitors. A matrix consisting of the comprehensive similarity between all pairs of competitors.
[0009] Preferably, the representative competitors are selected through the following steps: A degree matrix is constructed based on the comprehensive similarity matrix between all pairs of competitors in the aforementioned behavior graph; Based on the degree matrix, a normalized graph Laplacian matrix is constructed. Then, eigenvalue decomposition is performed on the normalized graph Laplacian matrix, and the smallest eigenvalue is selected. The eigenvectors corresponding to each eigenvalue constitute a spectral embedding matrix. Based on the spectral embedding results, competitor classification is performed, resulting in... One cluster; The distance between competitors in a cluster is measured by a weighted sum of statistical similarity and temporal morphological similarity among competitors. Within a cluster, the competitor with the smallest sum of distances to other competitors is defined as the representative competitor.
[0010] Preferably, the prediction model framework is constructed through the following steps: The change in bidding parameters is obtained based on the historical bidding parameters of representative competitors; The Bayesian rule is used to capture the immediate posterior judgment of being in a trend extrapolation state from the changes in bidding parameters; the conservative bias is introduced as the weight of the immediate posterior judgment to construct the belief update probability to measure the subjective probability of representative competitors believing that they are in a trend extrapolation state. The directional consistency and volatility of the changes in bidding parameters over a period of time are obtained. Based on the directional consistency and volatility, a psychological confidence variable is constructed using a logistic function to measure the strength of confirmation of the decision by representative competitors. By linking the belief update probability and the psychological confidence variable, the representativeness bias is obtained. Based on the representativeness bias, the proportion of representative competitors in a conservative state and a trend extrapolation state is adjusted to achieve the prediction of the bidding parameters of representative competitors.
[0011] Preferably, the belief update probability is specifically: ; in, Representative competitors exist The probability of belief updates at any given moment. Representative competitors Conservative bias parameter, For immediate posterior judgment, the following condition must be met: ; in, Representative competitors exist The conditional probability distribution of the changes in bidding parameters when the bidder considers itself to be in a trend extrapolation state at any given moment; Representative competitors exist The conditional probability distribution of the change in bidding parameters when the bidder considers itself to be in a conservative state at any given moment.
[0012] The specific psychological confidence variables are: ; in, Representative competitors exist The psychological confidence variable at any given moment. The baseline psychological confidence level parameter, To characterize the reinforcing effect of continuous and consistent signals on the degree of psychological confirmation, To characterize the weakening effect of local fluctuations on the degree of psychological confirmation, Represents an exponential function. Representative competitors exist Consistency of the direction of bidding parameters at all times Representative competitors exist The volatility of the bidding parameters at any given time satisfies: ; ; in, For symbolic functions, As competitors exist The change in bidding parameters at any given time. for The average change in bidding parameters within a time unit; For indicator functions, The value is 1 if the condition in parentheses is true, and 0 otherwise.
[0013] Preferably, the bidding parameter prediction specifically includes: ; in, Representative competitors exist Predicted values of bidding parameters at any given time. Representative competitors exist Actual values of bidding parameters at any given time. Representative competitors exist Actual values of bidding parameters at any given time; This indicates the base adjustment speed parameter for bidding parameters. Representative competitors Long-term reference level parameters, Indicates the shock response adjustment parameters. For the impact response term, the following condition must be met: ; in, Representative competitors Impact recognition threshold parameters, Representative competitors exist Changes in bidding parameters at any given time; To find the function with the maximum value, It is a symbolic function; Representative competitors exist The representativeness deviation at any given time satisfies: ; in, Representative competitors The parameter representing the degree to which psychological belief amplifies the tendency to extrapolate trends. Representative competitors exist The probability of belief updates at any given moment. Representative competitors exist The psychological confidence variable at any given moment. To find the function with the maximum value, To find the minimum value of the function.
[0014] Preferably, the predicted bidding parameters also need to be pruned based on boundary constraints: ; in, Representative competitors exist Predicted values of bidding parameters at any given time. Let be the projection function, representing the projection function. Projected onto representative competitors feasible range Among them Indicates the lower limit of the interval. This indicates the upper limit of the interval.
[0015] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art: (1) This application extracts statistical features from historical bidding parameters to mine long-term behavioral differences among competitors, and extracts temporal morphological features from historical bidding parameters to mine differences in bidding trajectories among competitors. Then, it uses Gaussian kernel weighted fusion to construct a comprehensive similarity among competitors, and constructs a behavioral map of all competitors based on this. Subsequently, it divides competitors into several clusters through spectral clustering, and selects representative competitors from each cluster using the intra-cluster centroid mechanism. This technical solution significantly improves the objectivity and representativeness of representative competitor identification, ensuring that the subsequent computer prediction model is built on representative competitors with typicality and coverage, and laying a solid data foundation for improving the performance of the computer prediction model.
[0016] (2) Based on representative competitors, this application simulates the subjective hesitation process of competitors between "conservative regression state" and "trend extrapolation state" by designing belief update probability. The design of psychological confidence variables further characterizes the amplification or inhibition effect of recent signal consistency and volatility on the degree of confidence of competitors' decisions. At the same time, it also incorporates conservative bias, representativeness bias, psychological confidence dynamics, shock response mechanism and parameter boundary constraints into the prediction model framework. This makes the prediction model framework no longer limited to the trend fitting of data for the dynamic prediction of bidding parameters, but has the ability to characterize the bounded rationality behavior characteristics such as the conservative inertia, bandwagon reinforcement and sudden shock response of representative competitors, thereby further improving the prediction performance of the computer prediction model.
[0017] (3) This application innovatively introduces psychological confidence variables on the basis of the traditional bounded rationality model framework, and constructs a dual cognitive prediction model framework that includes the coupling of belief update probability and psychological confidence dynamics, thereby realizing a refined characterization of the intensity of competitors' behavioral responses. The traditional bounded rationality model only relies on belief update probability to describe cognitive state, which is difficult to explain the difference in adjustment magnitude shown by the behavioral subjects under similar subjective judgments due to different degrees of certainty. The solution of this application constructs a Logistic form psychological confidence variable through directional consistency and local volatility within a rolling window, and connects the psychological confidence variable with the belief update probability extrapolated by the trend to obtain the representativeness bias intensity. This enables the model framework to distinguish between various behavioral patterns such as the tentative adjustment of "weak belief and weak confidence", the hesitant follow-up of "strong belief and weak confidence" and the resolute extrapolation of "strong belief and strong confidence", thereby further improving the predictive performance of the computer prediction model.
[0018] (4) This application simultaneously embeds an impact response term and hard boundary constraints during the bidding parameter prediction process, realizing unified modeling of normal evolution and abnormal jumps, free adjustment and rule restrictions. The impact response term identifies significant information impacts by setting an impact response threshold. When the impact response threshold is exceeded, additional adjustments are made according to the impact direction and intensity, effectively capturing the overreaction behavior of competitors to sudden market signals. The parameter boundary constraints ensure that the model output meets the physical rules and trading rules of the electricity spot market by projecting the updated results to a preset feasible range. This technology enables the prediction model to maintain behavioral flexibility while taking into account the feasibility of the output results, avoiding the parameter out-of-bounds or outlier problems that may occur in pure statistical models, thereby further improving the prediction performance of the computer prediction model.
[0019] (5) This application implements independent prediction model parameter fitting for each representative competitor, realizing differentiated quantification and personalized prediction of bounded rationality behavior characteristics. By estimating an independent set of parameters for each type of representative competitor, the prediction model can capture the differences in the strength of conservative bias, trend extrapolation tendency, psychological confidence sensitivity and shock response sensitivity of different behavioral groups, thereby further improving the prediction performance of the computer prediction model. Attached Figure Description
[0020] Figure 1 This is a schematic diagram of the computer prediction model construction process and parameter prediction process of the spot market bidding parameter prediction method based on the improved bounded rationality model provided in the embodiments of this application.
[0021] Figure 2 This is a schematic diagram of the process for selecting representative competitors from all competitors, provided in an embodiment of this application.
[0022] Figure 3 This is a flowchart illustrating the framework for constructing a prediction model based on an improved bounded rationality model, as provided in an embodiment of this application.
[0023] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0025] In this application, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A existing alone, A and B existing simultaneously, and B existing alone. In this application, the symbol " / " indicates that the related objects are in an "or" relationship, for example, A / B means A or B.
[0026] In this application, the terms "first" and "second," etc., are used to distinguish different objects, not to describe a specific order of objects. For example, "first statistical characteristic" and "second statistical characteristic," etc., are used to distinguish different statistical characteristics, not to describe a specific order of statistical characteristics.
[0027] In this application, the term "electrical connection" can refer to a direct circuit connection or a signal transmission via a communication protocol.
[0028] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.
[0029] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more. For example, multiple statistical features means two or more statistical features, and multiple bidding parameters means two or more bidding parameters.
[0030] First, the technical terms involved in the embodiments of this application will be introduced.
[0031] Bidding parameters are obtained by fitting the bidding curves of competitors; the coefficients of the fitted function are the bidding parameters. These parameters allow us to understand the bidding patterns of competitors.
[0032] The embodiments of this application are described below with reference to the accompanying drawings. Figure 1 As shown in the figure, this application discloses a method for predicting spot market bidding parameters based on an improved bounded rationality model, which specifically includes the following steps: S1. Collect historical bidding parameters of all competitors from the current spot market.
[0033] In this embodiment, the day-ahead spot market refers to the time-of-use electricity auction market organized the day before the market trading day.
[0034] The bidding parameters are as follows: Parameters and The parameters here are preset to a quadratic function curve representing the competitor's pricing curve. Parameters and The parameters are the coefficients of the linear and quadratic terms of the quadratic function.
[0035] In some embodiments, the spot market may be other auction markets, and the auction parameters may be other auction parameters.
[0036] To ensure comparability among different competitors, the historical bidding parameters of each competitor are aligned based on a unified time granularity, and the complete time-series trajectory is used as the basic input for subsequent bidding parameter prediction. Competitors At any moment The bidding parameters are expressed as follows: ; in, For the collection of all competitors, For the duration of history.
[0037] S2. Select representative competitors based on the historical bidding parameters of all competitors. The comprehensive similarity between competitors is measured based on the statistical characteristics and temporal patterns of their historical bidding parameters. Then, a behavioral graph containing all competitors is constructed based on this comprehensive similarity. After clustering the behavioral graph, a representative competitor is selected from each cluster. For example... Figure 2 The steps shown are as follows: S21. Extract various statistical features from the historical bidding parameters of competitors and concatenate them to obtain a behavioral feature vector. Measure the similarity of statistical features between different competitors based on the distance between the behavioral feature vectors. To reduce the dimensionality of the original time-series data and enhance the interpretability of the clustering results, statistical features that reflect the long-term behavioral style of competitors are first extracted from the sequence of historical bidding parameters, and then constructed as a behavioral feature vector of competitors.
[0038] In this embodiment, specifically for Parameter sequence and The following statistical features can be extracted from the parameter sequence: S211, Horizontal and Discrete Features ; ; in, Let be the mean of the parameter sequence. Let the standard deviation of the parameter sequence be . For parameter sequences.
[0039] S212, Trend Characteristics Linear trend is estimated using the least squares method: ; in, The average over time. This represents the average of the bidding parameters.
[0040] S213, Inertial Characteristics The first-order autocorrelation coefficient is used to describe the persistence and path dependence of bidding adjustments: ; in, The first-order autocorrelation function, specifically the Pearson linear correlation coefficient, satisfies: ; Let covariance function be used. Let X and Y be the standard deviations.
[0041] S214, Intraday Cycle Characteristics Aggregate and statistically analyze bidding parameters by hour, setting the hour as... The average value is: ; For a given time period. Further define the intraday amplitude indicator: .
[0042] S215, Impact Response Characteristics Based on first-order difference sequences: .
[0043] The average absolute difference of its higher quantile portion is used to characterize the degree of response to local anomalous fluctuations: ; Represents all historical samples The 90th percentile of a sequence defines normal and abnormal fluctuations; This represents an indicator function; it takes the value 1 if the condition within the parentheses is true, and 0 otherwise. This indicates the average magnitude of violent fluctuations.
[0044] S216, Trend Continuation Characteristics The correlation between adjacent difference terms can be used to reflect whether competitors are engaging in short-term trend-following behavior. .
[0045] S217, respectively targeting Parameter sequence and After extracting the above statistical features from the parameter sequence, they are concatenated to form the competitor's behavioral feature vector: ; in, As competitors behavioral feature vectors, It is a behavioral feature vector The length.
[0046] In this embodiment, Parameters and The parameters all selected six statistical features. After concatenation, the feature vector... length .
[0047] In other embodiments, other bidding parameters and other statistical characteristics may be selected, and the technical solution of this application is not specifically limited.
[0048] S218. Standardize the behavioral feature vectors and use Euclidean distance to measure the statistical feature similarity between different competitors in the feature space: ; That is, competitors after standardization behavioral feature vectors, As a competitor after standardization behavioral feature vectors, To find the Euclidean distance. This refers to the similarity of statistical characteristics between competitors.
[0049] S22. Standardize the historical bidding parameters of competitors to obtain standardized bidding trajectories, and measure the temporal morphological similarity between different competitors based on the distance between the standardized bidding trajectories. While statistical features can effectively summarize the long-term bidding style of competitors, they are essentially a compressed representation of sequences and cannot fully preserve the dynamic evolutionary information in the original trajectory. Therefore, we further model competitors from the perspective of their original time-series morphology.
[0050] This embodiment has two bidding parameters: Parameters and Therefore, the parameters need to be standardized first to eliminate the dimensional differences between different bidding parameters, so that the bidding parameters at different scales are comparable and clusterable.
[0051] S221. The dimensional differences of bidding parameters at different scales are eliminated through standardization to obtain a standardized bidding trajectory. ; in, Indicate competitors Bidding parameters sequence standard deviation As competitors Bidding parameters sequence standard deviation Indicate competitors Bidding parameters Sequence mean competitors Bidding parameters Sequence mean As competitors exist Bidding parameters at any time , As competitors exist Bidding parameters at any time , As competitors The standardized bidding trajectory.
[0052] S222. The distance between standardized bidding trajectories is calculated using the dynamic time warping method as a measure of the temporal morphological similarity between different competitors: For trajectory similarity measurement, the Dynamic Time Warping (DTW) method is used to calculate the temporal morphological distance between any two competitors' standardized bidding trajectories. Its local matching cost is: ; As competitors exist Standardized bidding trajectory points at any given moment and competitors exist Standardized bidding trajectory points at any given moment The cost of local matching.
[0053] The advancement of DTW distance is manifested in the following ways: ; Ultimately gained the support of competitors and Similarity in temporal morphological features between them: ; S23. Integrate the statistical feature similarity and temporal morphological feature similarity to obtain the comprehensive similarity between different competitors; construct a behavioral map containing all competitors based on the comprehensive similarity; S231. Apply Gaussian kernel mapping to the statistical feature similarity and temporal morphological feature similarity respectively to obtain the Gaussian kernel similarity of the statistical features. Gaussian kernel similarity to temporal morphological features : ; ; in, The Gaussian kernel bandwidth parameter is a statistical characteristic. The Gaussian kernel bandwidth parameter represents the temporal morphological characteristics.
[0054] S232. Weighted fusion of Gaussian kernel similarity of statistical features and temporal morphological features yields the comprehensive similarity between different competitors: ; in, For Gaussian kernel fusion weights, For all A set; For all gather. It is the set of comprehensive similarities between all pairs of competitors.
[0055] S233. Construct a behavioral graph containing all competitors based on the comprehensive similarity between all pairs of competitors. ,in A matrix composed of all competitors. A matrix consisting of the comprehensive similarity between all pairs of competitors.
[0056] S24. After performing spectral clustering on the behavior graph, multiple clusters are obtained. Based on the centroid selection mechanism, a representative competitor is selected from each cluster.
[0057] S241. Construct a degree matrix based on the comprehensive similarity between all pairs of competitors in the aforementioned behavioral graph: ; in, It represents the total number of competitors in the spot market. As competitors The sum of overall similarities with other competitors, by Composition degree matrix .
[0058] S242. Construct a normalized graph Laplacian matrix based on the degree matrix. : ; in, It is an identity matrix.
[0059] Perform eigenvalue decomposition on the normalized graph Laplacian matrix and select the smallest eigenvalue. The eigenvectors corresponding to the eigenvalues constitute the spectral embedding matrix: ; Finally, based on the spectral embedding results, the competitors were categorized, resulting in... There are several clusters.
[0060] S243, Utilizing the similarity of statistical characteristics among competitors Similarity with temporal morphological features The weighted sum measures the distance between competitors in a cluster. : ; in, Assign weights to similarity.
[0061] S244, the first Clusters Within a given area, the competitor whose sum of distances to all other competitors is the representative competitor. : ; ; in, for The first cluster in the cluster A cluster, As competitors The sum of distances from other competitors To minimize the objective function.
[0062] Therefore, a central point sample is selected from each cluster as a representative competitor.
[0063] S3. Constructing a predictive model framework based on an improved bounded rationality model. In this embodiment, the prediction model framework uses belief update probability to measure the subjective probability that a representative competitor believes they are in a trend extrapolation state; it uses psychological confidence variables to measure the strength of the representative competitor's confirmation of the decision; and it combines the belief update probability and psychological confidence variables to predict the bidding parameters of the representative competitor; as follows: Figure 3 As shown, the specific steps include the following: S31. Obtain the change in bidding parameters based on the historical bidding parameters of representative competitors. In the At any given time, its bidding parameters and The change is: ; ; Because this embodiment has two bidding parameters and Therefore, the bidding parameters need to be set first. and A comprehensive analysis was conducted to identify representative competitors. In the Comprehensive feedback signal at any moment : ; in, Representative competitors Bidding parameters standard deviation Representative competitors Bidding parameters standard deviation and Parameters and The comprehensive weight parameters satisfy: ; The meaning of comprehensive feedback signals lies in using a unified scale to comprehensively depict the direction and intensity of competitors' bidding adjustments. When When the value is positive and the absolute value is large, it indicates that the competitor... In the The bidding behavior at any given moment tends to be aggressively adjusted; when When it is negative, it indicates that the competitor In the Bidding behavior is becoming more contractionary or conservative.
[0064] In other embodiments, if only one bidding parameter is predicted, the step of combining bidding parameters is not required.
[0065] S32. Capture the immediate posterior judgment of being in a trend extrapolation state from the changes in bidding parameters using Bayesian rules; introduce conservative bias as the weight of the immediate posterior judgment to construct the belief update probability to measure the subjective probability of representative competitors believing they are in a trend extrapolation state. Drawing on the basic ideas of traditional bounded rationality models, we assume representative competitors. At any given moment Two potential cognitive states are faced: ; in, Indicates a conservative state. This indicates the trend extrapolation state.
[0066] Set comprehensive feedback signals for the two cognitive states respectively. Conditional probability distribution: ; ; Representative competitors In a conservative state The distribution variance parameter; Representative competitors When trend extrapolation is in progress The distribution variance parameter; Representative competitors The baseline trend strength parameter under trend extrapolation; It is a symbolic function; for In length The mean over a given time period satisfies: ; Based on the newly arrived integrated feedback signal The competitor first forms an immediate posterior judgment about the trend extrapolation state according to Bayesian rules: .
[0067] Introducing Conservatism Bias As the immediate a posteriori judgment The weights are used to construct the belief update probability. Measuring the subjective probability that representative competitors believe they are in a state of trend extrapolation: ; in, Representative competitors exist The probability of belief updates at any given moment. Representative competitors Conservative bias parameter.
[0068] S33. Obtain the directional consistency and volatility of the changes in bidding parameters over a period of time, and construct a psychological confidence variable using a logistic function based on the directional consistency and volatility to measure the strength of representative competitors' confirmation of the decision. Update probability based solely on belief This is insufficient to fully describe the intensity of competitors' behavioral responses. In reality, even if actors form similar subjective judgments, they may exhibit different degrees of bidding adjustments due to varying degrees of conviction about the current trend. Therefore, in addition to the belief updating mechanism, it is necessary to further introduce psychological confidence variables. This is used to characterize the strength of competitors' subjective confirmation of recent pattern recognition results.
[0069] set up Indicates recent Number of observations within a given period that align with the direction of the current rolling average signal: ; set up This indicates the local fluctuation level of the integrated feedback signal within the same window: ; in, It is a symbolic function; For indicator functions, The value is 1 if the condition in parentheses is true, and 0 otherwise.
[0070] Construct psychological confidence variables using the Logistic formula: ; in, Representative competitors exist The psychological confidence variable at any given moment. The baseline psychological confidence level parameter, To characterize the reinforcing effect of continuous and consistent signals on the degree of psychological confirmation, To characterize the weakening effect of local fluctuations on the degree of psychological confirmation, This represents an exponential function.
[0071] S34. Connecting the belief update probability and the psychological confidence variable yields the representativeness bias: ; in, Representative competitors The parameter representing the degree to which psychological confidence amplifies the tendency to extrapolate trends. The larger the size, the stronger the competitor. exist The more a timeframe tends to extrapolate recent patterns as future trends, the lower the weighting, indicating stronger competition. exist Adjustments should be made conservatively based on long-term reference levels.
[0072] Based on the aforementioned representativeness bias, the weighting of representative competitors in conservative and trend extrapolation states is adjusted to predict the bidding parameters of representative competitors: Representative competitors exist Time-based bidding parameters The prediction equation is: ; Representative competitors exist Time-based bidding parameters The prediction equation is: ; in, Indicates bidding parameters Adjust the basic speed parameters. Indicates bidding parameters Adjust the basic speed parameters.
[0073] Representative competitors Bidding parameters Long-term reference level parameters; Representative competitors Bidding parameters Long-term reference level parameters.
[0074] Representative competitors Bidding parameters Shock response adjustment parameters Representative competitors Bidding parameters The impact response adjustment parameters.
[0075] For the impact response term, the following condition must be met: ; in, Representative competitors Impact recognition threshold parameters, To find the function with the maximum value, It is a symbolic function.
[0076] When the comprehensive feedback signal The absolute value did not exceed the threshold. At that time, it is believed that market information is still within the normal fluctuation range, and the shock response term is zero; when it exceeds the threshold... At that time, the predictive model assumes that competitors perceive a significant impact and makes additional adjustments to the two bidding parameters based on the direction and intensity of the impact. .
[0077] The predicted bidding parameters still need to be trimmed based on boundary constraints: To ensure the model output meets the realistic feasibility of the bidding parameters, the following measures are taken: Parameters and The parameter update results are subject to projection constraints. Representative competitors are defined respectively. of Feasible range of parameters Feasible range of parameters: ; After completing the dynamic update, representative competitors will be analyzed. The predicted value of the bidding parameters at each time point is projected onto the interval: ; ; in, This indicates projecting the variable onto the interval. Functions on.
[0078] S4. Fit the historical bidding parameters of each representative competitor to the aforementioned prediction model framework to obtain the prediction model parameters for each representative competitor. For each type of representative competitor Each parameter is estimated as a set of independent parameters to characterize its unique bounded rational behavior. This set of parameters is denoted as the prediction model parameter set. : ; Given the historical bidding parameters of each representative competitor, the parameters are estimated by minimizing the weighted squared error between the predicted and actual values of the prediction model, and a regularization term is added to the objective function: ; in, and Representative competitors exist The actual value of the bidding parameters at any given time. and Indicates the set of parameters in the prediction model Under these conditions, representative competitors exist Predicted values of bidding parameters at any given time. and The fitting error weights for the two bidding parameters are... This is the regularization coefficient. This represents finding the set of parameters for the prediction model under the minimization objective. .
[0079] By estimating the parameters of representative competitors, we can obtain the differences in parameters of different types of competitors in terms of conservatism bias, trend extrapolation tendency, changes in psychological confidence, and shock response sensitivity, thus providing a quantitative basis for the interpretation of behavioral characteristics.
[0080] S5. Substitute the prediction model parameters of each representative competitor into the prediction model framework to obtain the bidding parameter prediction model corresponding to each representative competitor.
[0081] S6. When predicting bidding parameters: The most recent period before the prediction period By inputting historical bidding parameters from various periods into the bidding parameter prediction model, the bidding parameters of each representative competitor within the prediction period can be obtained. Parameter predictions and The parameter prediction values are used as the prediction results of the bidding parameters of competitors in the spot market in the previous day.
[0082] In summary, this application proposes a method for predicting competitor bidding parameters in the day-ahead spot market. It constructs a multi-perspective behavioral map combining behavioral statistical features and time-series dynamic patterns, and uses spectral clustering and centroid mechanisms to identify representative competitors. Furthermore, it constructs an improved BSV (Bounded Rationality) bidding parameter evolution model, incorporating conservatism bias, representativeness bias, psychological confidence dynamics, shock response terms, and parameter boundary constraints into a unified framework. Finally, it obtains the predicted parameters for various representative competitors through parameter estimation, achieving effective prediction of competitor bidding parameters in the day-ahead spot market.
[0083] Based on the methods in the above embodiments, this application provides an electronic device, such as... Figure 4 As shown, the electronic device may include a processor, a communications interface, a memory, and a communication bus, wherein the processor, communications interface, and memory communicate with each other via the communication bus. The processor can invoke logical instructions stored in the memory to execute the methods described in the above embodiments.
[0084] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.
[0085] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.
[0086] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.
[0087] It is understood that the processor in the embodiments of this application 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, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.
[0088] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.
[0089] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).
[0090] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.
[0091] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.
Claims
1. A method for predicting spot market bidding parameters based on an improved bounded rationality model, characterized in that, include: The historical bidding parameters of competitors before the prediction period are input into the corresponding bidding parameter prediction model to obtain the bidding parameters for the prediction period; the bidding parameter prediction model is obtained through the following steps: The comprehensive similarity between competitors is measured based on the statistical characteristics and temporal patterns of their historical bidding parameters. Then, a behavioral graph containing all competitors is constructed based on the comprehensive similarity. After clustering the behavioral graph, a representative competitor is selected from each cluster. A predictive model framework is constructed based on an improved bounded rationality model. In this framework, the subjective probability that a representative competitor believes they are in a trend extrapolation state is measured by the belief update probability; the strength of the representative competitor's confirmation of the decision is measured by the psychological confidence variable; and the bidding parameters of the representative competitor are predicted by linking the belief update probability and the psychological confidence variable. By fitting the historical bidding parameters of each representative competitor to the prediction model framework, the prediction model parameters of each representative competitor are obtained. Substituting the prediction model parameters of each representative competitor into the prediction model framework, the bidding parameter prediction model corresponding to each representative competitor is obtained.
2. The spot market bidding parameter prediction method according to claim 1, characterized in that, The representative competitors were selected through the following steps: Multiple statistical features are extracted from competitors' historical bidding parameters and concatenated to obtain behavioral feature vectors. The similarity of statistical features between different competitors is measured based on the distance between these behavioral feature vectors. Standardized bidding trajectories are obtained by standardizing the historical bidding parameters of competitors. The similarity of temporal morphological features between different competitors is measured based on the distance between the standardized bidding trajectories. By integrating the statistical similarity and temporal morphological similarity, a comprehensive similarity among different competitors is obtained; Based on the comprehensive similarity, a behavioral graph containing all competitors is constructed; After performing spectral clustering on the behavioral graph, multiple clusters are obtained. Based on the centroid selection mechanism, a representative competitor is selected from each cluster.
3. The spot market bidding parameter prediction method according to claim 2, characterized in that, The temporal morphological similarity between different competitors is obtained through the following steps: Standardization eliminates the dimensional differences of bidding parameters at different scales, resulting in a standardized bidding trajectory. The distance between standardized bidding trajectories is calculated using a dynamic time warping method to represent the temporal morphological similarity between different competitors.
4. The spot market bidding parameter prediction method according to claim 1 or 2, characterized in that, The behavioral map is obtained through the following steps: Gaussian kernel mapping is applied to the statistical feature similarity and the temporal morphological feature similarity respectively to obtain the Gaussian kernel similarity of the statistical features and the Gaussian kernel similarity of the temporal morphological features; We use a weighted fusion of Gaussian kernel similarity based on statistical features and temporal morphological features to obtain the comprehensive similarity between different competitors. Construct a behavioral graph containing all competitors based on the comprehensive similarity between each pair of all competitors. ,in A matrix composed of all competitors. A matrix consisting of the comprehensive similarity between all pairs of competitors.
5. The spot market bidding parameter prediction method according to claim 1 or 2, characterized in that, The representative competitors were selected through the following steps: A degree matrix is constructed based on the comprehensive similarity matrix between all pairs of competitors in the aforementioned behavior graph; Based on the degree matrix, a normalized graph Laplacian matrix is constructed. Then, eigenvalue decomposition is performed on the normalized graph Laplacian matrix, and the smallest eigenvalue is selected. The eigenvectors corresponding to each eigenvalue constitute a spectral embedding matrix. Based on the spectral embedding results, competitor classification is performed, resulting in... One cluster; The distance between competitors in a cluster is measured by a weighted sum of statistical similarity and temporal morphological similarity among competitors. Within a cluster, the competitor with the smallest sum of distances to other competitors is defined as the representative competitor.
6. The spot market bidding parameter prediction method according to claim 1, characterized in that, The prediction model framework is constructed through the following steps: The change in bidding parameters is obtained based on the historical bidding parameters of representative competitors; The Bayesian rule is used to capture the immediate posterior judgment of being in a trend extrapolation state from the changes in bidding parameters; the conservative bias is introduced as the weight of the immediate posterior judgment to construct the belief update probability to measure the subjective probability of representative competitors believing that they are in a trend extrapolation state. The directional consistency and volatility of the changes in bidding parameters over a period of time are obtained. Based on the directional consistency and volatility, a psychological confidence variable is constructed using a logistic function to measure the strength of confirmation of the decision by representative competitors. By linking the belief update probability and the psychological confidence variable, the representativeness bias is obtained. Based on the representativeness bias, the proportion of representative competitors in a conservative state and a trend extrapolation state is adjusted to achieve the prediction of the bidding parameters of representative competitors.
7. The spot market bidding parameter prediction method according to claim 1 or 6, characterized in that, The specific probability of belief update is as follows: ； in, Representative competitors exist The probability of belief updates at any given moment. Representative competitors Conservative bias parameter, For immediate posterior judgment, the following condition must be met: ； in, Representative competitors exist The conditional probability distribution of the changes in bidding parameters when the bidder considers itself to be in a trend extrapolation state at any given moment; Representative competitors exist The conditional probability distribution of the change in bidding parameters when the bidder considers itself to be in a conservative state at any given moment.
8. The spot market bidding parameter prediction method according to claim 1 or 6, characterized in that, The specific psychological confidence variables are: ； in, Representative competitors exist The psychological confidence variable at any given moment. As the baseline psychological confidence level parameter, To characterize the reinforcing effect of continuous and consistent signals on the degree of psychological confirmation, To characterize the weakening effect of local fluctuations on the degree of psychological confirmation, Represents an exponential function. Representative competitors exist Consistency of the direction of bidding parameters at all times. Representative competitors exist The volatility of the bidding parameters at any given time satisfies: ； ； in, For symbolic functions, As competitors exist The change in bidding parameters at any given time. for The average change in bidding parameters within a time unit; For indicator functions, The value is 1 if the condition in parentheses is true, and 0 otherwise.
9. The spot market bidding parameter prediction method according to claim 1 or 6, characterized in that, The bidding parameter prediction is specifically as follows: ； in, Representative competitors exist Predicted values of bidding parameters at any given time. Representative competitors exist Actual values of bidding parameters at any given time. Representative competitors exist Actual values of bidding parameters at any given time; This indicates the base adjustment speed parameter for bidding parameters. Representative competitors Long-term reference level parameters, Indicates the shock response adjustment parameters. For the impact response term, the following condition must be met: ； in, Representative competitors Impact recognition threshold parameters, Representative competitors exist Changes in bidding parameters at any given time; To find the function with the maximum value, It is a symbolic function; Representative competitors exist The representativeness deviation at any given time satisfies: ； in, Representative competitors The parameter representing the degree to which psychological belief amplifies the tendency to extrapolate trends. Representative competitors exist The probability of belief updates at any given moment. Representative competitors exist The psychological confidence variable at any given moment. To find the function with the maximum value, To find the minimum value of the function.
10. The spot market bidding parameter prediction method according to claim 1 or 6, characterized in that, The predicted bidding parameters still need to be trimmed based on boundary constraints: ； in, Representative competitors exist Predicted values of bidding parameters at any given time. Let be the projection function, representing the projection function. Projected onto representative competitors feasible range Among them Indicates the lower limit of the interval. This indicates the upper limit of the interval.