XGBoost-based power generation quotation rationality evaluation method, system and equipment and medium

Abstract

The invention discloses an XGBoost-based power generation quotation rationality evaluation method, system and equipment and a medium, and relates to the technical field of power market operation and supervision, and the method comprises the steps: collecting historical operation and market environment data, obtaining five evaluation indexes through a prediction model, elastic analysis, matching degree calculation and clustering analysis, fusing the indexes to construct a feature vector, and obtaining a power generation quotation rationality evaluation result. And training a classification model to evaluate the rationality of quotation, and generating a monitoring early warning signal when the matching degree is too low, the price response is too strong, the quotation deviation is too large and the behavior mode is abnormal by continuously tracking the change trend of key indexes. According to the invention, by fusing the multi-dimensional dynamic features and adopting the elastic time sequence matching and interpretable machine learning model, a power generation quotation behavior intelligent supervision closed loop integrating comprehensive evaluation, accurate quantification and active early warning is constructed, and the comprehensiveness, accuracy, objectivity and timeliness of market supervision are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of power market operation and supervision technology, specifically to a method, system, equipment, and medium for evaluating the reasonableness of power generation bids based on XGBoost. Background Technology

[0002] With the deepening of power market reform, the bidding behavior of market participants has become increasingly complex and strategic. The rationality assessment and supervision of power generation companies' bidding behavior has become a key technical issue for maintaining the fair and efficient operation of the market. In recent years, some studies have begun to introduce machine learning methods to identify non-competitive behavior by directly classifying or detecting anomalies in bidding sequences. In addition, time series forecasting techniques, such as ARIMA and LSTM, have also been used to establish benchmark models for bidding. However, existing technologies mostly approach the issue from a single perspective (such as price elasticity, market power, or time series forecasting bias) or treat bidding behavior as a static feature for analysis. There is a lack of a systematic assessment framework that can integrate multi-dimensional dynamic indicators and quantify the comprehensive impact.

[0003] The shortcomings of existing technologies are mainly reflected in three aspects: First, in terms of the comprehensiveness of evaluation dimensions, most methods fail to organically combine the bidding behavior of power generation entities with their physical operating characteristics and inherent behavioral patterns, resulting in one-sided evaluation results that are difficult to fully reflect strategic aspects. Second, in terms of the adaptability of data processing and analysis methods, electricity market bidding data exhibits high-dimensionality, segmentation, and time-series misalignment characteristics. Existing methods are insufficient in dealing with data feature extraction and similarity measurement, especially in the quantitative comparison of curve shapes. Conventional measurement methods such as Euclidean distance cannot effectively handle the phase difference and nonlinear deformation of time series. Finally, in terms of the real-time performance and early warning capabilities of evaluation, existing research mainly focuses on post-event retrospective analysis and lacks an early warning mechanism based on continuous monitoring and dynamic thresholds. This makes it difficult to identify and warn of potential strategic bidding and market power abuse risks in a timely manner, leading to regulatory lag. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention provides a method, system, device and medium for evaluating the rationality of power generation pricing based on XGBoost, in order to solve the problems in the prior art that it is difficult to fully reflect the strategic nature, that there are deficiencies in the feature extraction and similarity measurement of data, and that there is a lack of early warning mechanism based on continuous monitoring and dynamic thresholds.

[0005] To address the aforementioned technical issues, a method for evaluating the reasonableness of power generation pricing based on XGBoost is proposed, including: Historical operating data of power generation entities and electricity market environment data are collected. Based on the historical operating data, the first prediction model is used to obtain the bidding prediction data of the power generation entities, and the first evaluation index is calculated. Based on the historical operating data and electricity market environment data, the first elasticity analysis is used to obtain the responsiveness of the power generation entities to changes in market prices, which serves as the second evaluation index. The deviation between the bidding price of the power generation entities and the market price is calculated as the third evaluation index. Based on the historical operating data and electricity market environment data, the first matching degree index is used to calculate the degree of fit between the output of the power generation entities and the system demand, which serves as the fourth evaluation index. The first clustering analysis method is used to identify the bidding behavior patterns of the power generation entities, which are encoded as the fifth evaluation index. The first to fifth evaluation indices are standardized and feature fused to construct a multi-dimensional feature vector. This vector is then used with a preset rationality label and input into the first classification model for training to obtain a bidding rationality evaluation model. The feature vector obtained by processing the real-time data of the power generation entities to be evaluated is input into the bidding rationality evaluation model, and the classification results of the bidding behavior rationality are output. The bidding behavior of the power generation entities is continuously evaluated, and the changing trends of the key evaluation indices of the power generation entities are tracked. When the evaluation indices deviate from the threshold, a monitoring and early warning signal is generated.

[0006] As a preferred embodiment of the XGBoost-based power generation pricing rationality assessment method described in this invention, the calculation of the first assessment index includes: preprocessing historical pricing data, integrating the original pricing data of different time granularities and different number segments into a single time series data of a unified format by taking the average value, and filling in the missing values ​​in the series. The integrated time series data is modeled using the first prediction model. Based on the first prediction model that has completed parameter learning and fitting, the price values ​​for future time windows are rolled out to generate a sequence of price prediction values ​​containing the predicted values ​​and confidence intervals. Align the actual bid sequence and the bid prediction sequence of the power generation entities within the same time window, and calculate the average of the absolute differences between the two.

[0007] As a preferred embodiment of the XGBoost-based power generation pricing rationality assessment method described in this invention, the second assessment index includes: obtaining the power generation entity's pricing volume sequence and the market day-ahead clearing price sequence from the preprocessed data, aligned by time points. Logarithmic transformations are performed on the quotation volume series and the market price series to convert the exponential growth relationship into a linear relationship, satisfying the assumptions of the first elasticity analysis method regarding data stationarity and model form. The two logarithmically transformed sequence data are input into the first elasticity analysis model, and a linear equation is fitted by minimizing the error term. The slope coefficient, i.e., price elasticity, is extracted from the fitted linear equation, and the slope coefficient value is used as a second evaluation index to characterize the sensitivity.

[0008] As a preferred embodiment of the XGBoost-based power generation pricing rationality assessment method described in this invention, the fourth assessment index includes obtaining the actual power output curve and system load prediction curve of the power generation entity within the assessment period with the same time granularity and the same time series length. Using the first matching degree index calculation method, the two curves are non-linearly aligned and compared globally. That is, by constructing a cumulative cost matrix, the optimal bending path in the matching of all point pairs between the two curves is found, so as to minimize the cumulative alignment cost. Based on the optimal curvature path, the final cumulative distance value is calculated and used as the fourth evaluation metric.

[0009] As a preferred embodiment of the XGBoost-based power generation pricing rationality assessment method described in this invention, the first clustering analysis method includes: collecting historical segmented pricing curve data of the target power generation entity and similar power generation entities to form an original pricing curve set; and converting all pricing curves into a daily pricing time series of uniform length by taking the daily average value. The DTW algorithm is used to calculate the similarity distance between two curves in the original set of price quote curves, and a cumulative cost matrix is ​​constructed. The final cumulative cost value is the DTW distance between the two sequences. The DTW distance matrix is ​​obtained by calculating the similarity distance for all curve pairs, as expressed by the formula: in, Store the distances between all pairs of points on the two curves in a local distance matrix. This is the index of the time point on the actual output curve Q. This refers to the index of time points on the system load forecast curve L. Let Q be the actual output value of curve Q at time point i. Let L be the predicted system load value at time point j. For point With point Distance function between, Store the minimum cumulative distance from the starting point to each pair of points in the cumulative cost matrix. Let (i, j) be the path from the point pair (i-1, j) to (i, j). Let (i, j-1) be the path from point pair (i, j) to (i, j). Let (i, j) be the path from point pair (i-1, j-1) to (i, j). The dynamic time-warped distance between curve Q and curve L. The lower right element of the cumulative cost matrix S. The length of curve Q is the total number of time points. Let L be the length of curve L; Based on prior business knowledge or the elbow rule, the number of clusters is preset, and an optimization method based on the distance matrix is ​​used to initialize curves from the original set of curves as initial cluster centers. For each historical price quote curve in the set, calculate the DTW distance between it and the cluster center curve, and assign it to the cluster represented by the nearest cluster center according to the principle of minimum distance; For each newly formed cluster, the dynamic centroid averaging algorithm is used to average all curves within the cluster. Using the current cluster center as the initial template, the DTW alignment is used to align all curves within the cluster with the template, and the average value of all aligned points is calculated as the new template. This process is iterated until the template is stable, generating a new cluster center curve for the current cluster. Repeat the iteration until the change in all cluster center curves compared to the previous iteration is less than a preset threshold, indicating that the algorithm has converged and obtained stable clustering results. Each cluster represents a pricing behavior paradigm. The historical bidding curves of the power generation entities to be evaluated are calculated, and their DTW distances to the cluster centers are determined. The entities are then classified into the clusters with the smallest distances. The cluster labels represent the bidding behavior paradigm types. The current type is then converted into a binary vector using one-hot encoding, and the vector is used as the fifth evaluation indicator.

[0010] As a preferred embodiment of the XGBoost-based power generation pricing rationality assessment method described in this invention, the data standardization and feature fusion include processing the continuous numerical features represented by the first assessment indicator (price prediction deviation), the second assessment indicator (price elasticity coefficient), the third assessment indicator (price deviation degree), and the fourth assessment indicator (DTW matching distance) using a minimum-maximum normalization method. For each feature, the minimum and maximum values ​​in the current batch of data are calculated, and all feature values ​​are linearly scaled to the [0, 1] interval. The minimum-maximum normalization method formula is expressed as: in, These are the original eigenvalues. The standardized value. The minimum value of the original feature. The maximum value of the original feature; For the categorical features represented by the fifth evaluation index, namely the obtained seven-dimensional one-hot encoded vector, it is directly used as the discrete feature part. The normalized continuous feature vector and the one-hot encoded discrete feature vector are concatenated in the order of feature dimensions to form an 11-dimensional unified multi-dimensional feature vector.

[0011] As a preferred embodiment of the XGBoost-based power generation pricing rationality assessment method described in this invention, the training of the first classification model includes dividing the dataset containing 11-dimensional multi-dimensional feature vectors of all samples and reasonable or unreasonable manually labeled data into a training set and a test set. Configure the hyperparameters of the XGBoost classifier, set the objective function, set the maximum tree depth, learning rate, number of base learners, fix the random seed, and set the L1 and L2 regularization parameters; The features and labels of the training set are input into the model for training. The gradient boosting framework is used to iteratively build multiple CART decision trees. In each iteration, a new tree is built to fit the residual between the current model prediction result and the true label. The structure of the tree and the weight of the leaf nodes are learned by minimizing the loss function containing first-order and second-order derivative information and regularization terms. The performance of the trained model is evaluated using the test set. The precision, recall, F1 score, and accuracy of the model on the test set are calculated. The preset performance thresholds for the model to be effective are: precision and recall of reasonable classes are not less than 80%, precision and recall of unreasonable classes are not less than 85%, overall F1 score is not less than 0.8, and accuracy is not less than 80%. When the test results reach or exceed the thresholds, the model training is confirmed to be effective, and the final model structure and parameters are saved. The classification results of the reasonableness of the output pricing behavior include the generation logic based on the feature importance weights and decision rules learned internally by the model; The feature importance weights are automatically obtained during model training. According to the gain importance ranking provided by the XGBoost model, the fourth evaluation indicator, namely the output curve matching degree, has the highest importance. The third evaluation indicator, namely the degree of price deviation, and the second evaluation indicator, namely the price elasticity, are next in importance. The first evaluation indicator, namely the price prediction deviation, is next in importance. The fifth evaluation indicator, namely the feature importance of each dimension of the price behavior type, has the lowest importance. The decision rules include the model's comprehensive nonlinear judgment of the input feature values, the model learning the critical threshold of each feature during training, and the model's collective voting of all decision trees to obtain the final classification result.

[0012] The beneficial effects of this preferred technical solution are as follows: Standardized processing ensures fair weighting of each indicator in model training. The XGBoost model not only provides high-precision classification results, but its built-in feature importance analysis function automatically reveals the contribution of evaluation dimensions to the rationality of the judgment, providing an explanatory basis for regulatory decisions, increasing the persuasiveness and credibility of the evaluation conclusions, and avoiding the "black box" problem of machine learning models.

[0013] As a preferred embodiment of the XGBoost-based power generation pricing rationality assessment system of the present invention, it is characterized by including a data acquisition and preprocessing module, a multi-dimensional feature extraction module, a feature engineering and fusion module, an XGBoost model training and optimization module, a rationality assessment and decision interpretation module, and a continuous monitoring and dynamic early warning module.

[0014] The data acquisition and preprocessing module is used to collect historical segmented price quotes and actual output curves of power generation entities, as well as market environment data such as system load forecasts and market clearing prices. By performing mean aggregation on data of different formats and linear interpolation on missing values, all data are unified to the same time granularity and time length.

[0015] The multi-dimensional feature extraction module is used to generate price predictions and calculate price prediction deviations using the Prophet time series prediction model, fit the relationship between price volume and market price using a log-log linear regression model to calculate price elasticity, calculate the degree of deviation between price and market price using the arithmetic mean, calculate the matching degree between the force curve and the load curve using the dynamic time warping algorithm, and identify and encode the price behavior paradigm type based on the DTW distance time series clustering method.

[0016] The feature engineering and fusion module is used to perform Min-Max standardization on the bid prediction deviation, bid elasticity, deviation degree and matching degree, and scale it to the [0,1] interval. It performs one-hot encoding on the discrete bid behavior paradigm type indicators, converts them into binary vectors, and concatenates the standardized continuous feature vector with the encoded discrete feature vector to obtain a comprehensive multidimensional feature vector.

[0017] The XGBoost model training and optimization module is used to divide the labeled multidimensional feature vector dataset into training and test sets, use the XGBoost gradient boosting decision tree framework for model training, configure parameters, and iteratively build decision trees to minimize the loss function during training to obtain a strong classifier. After the model training is completed, the effectiveness of the model is judged by verifying whether the precision, recall and F1 score metrics reach the preset thresholds on the test set, and the optimal model is saved.

[0018] The rationality assessment and decision interpretation module is used to input the feature vector obtained by processing the data to be evaluated into the trained XGBoost model and directly output the classification result.

[0019] The continuous monitoring and dynamic early warning module is used to generate a time series trajectory of the rationality of the behavior of each power generation entity by periodically and automatically executing the entire process from data collection to rationality assessment, and preset monitoring thresholds for key indicators. It makes a comprehensive judgment based on the persistence of behavioral paradigm types. When it is detected that an entity continuously triggers multiple early warning conditions, the system automatically generates an early warning signal and marks it as a high-risk object.

[0020] A computer device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a method for assessing the reasonableness of power generation pricing based on XGBoost.

[0021] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for assessing the reasonableness of electricity generation pricing based on XGBoost.

[0022] The beneficial effects of this invention are as follows: By integrating five dimensions of indicators, this invention overcomes the one-sidedness of single-perspective evaluation, making it difficult for power generation entities to circumvent regulation through local strategy adjustments, thus significantly improving the systematic nature and anti-circumvention capability of the evaluation; by employing a dynamic time warping algorithm to elastically quantify the morphological similarity between the output curve and the load curve, it effectively eliminates misjudgments caused by time delays or phase differences, achieving accurate measurement of the physical operation response capability; by using time series clustering based on DTW distance to summarize high-dimensional pricing curves into standardized behavioral paradigm labels, it provides key qualitative features for machine learning and regulatory profiling, achieving efficient dimensionality reduction and representation of complex behavioral patterns; by using the XGBoost model to fuse heterogeneous features and train it, it not only ensures high classification accuracy through preset performance thresholds, but also provides transparent decision-making basis through feature importance analysis, enhancing the interpretability and credibility of the results; and by tracking the continuous changing trends of key indicators and correlating them with the long-term stability of behavioral patterns, a dynamic early warning mechanism is established, transforming regulation from static ex-post judgment to proactive and forward-looking risk prevention and control. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0024] Figure 1The overall flowchart of the XGBoost-based power generation pricing rationality assessment method provided in one embodiment of the present invention is shown below.

[0025] Figure 2 The flowchart of a system scheme for a power generation pricing rationality assessment system based on XGBoost, provided as an embodiment of the present invention. Detailed Implementation

[0026] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0027] Example 1, referring to Figure 1 As one embodiment of the present invention, a method for evaluating the reasonableness of power generation pricing based on XGBoost is provided, comprising: S100: Collect historical operating data of power generation entities and electricity market environment data. Based on the historical operating data, obtain the bidding prediction data of power generation entities through the first prediction model, and calculate the first evaluation index.

[0028] S200: Based on historical operating data and electricity market environment data, the first elasticity analysis is used to obtain the degree of response of power generation entities to changes in market prices, which serves as the second evaluation indicator. The deviation between the power generation entity's bid and the market price is calculated as the third evaluation indicator.

[0029] S300: Based on historical operating data and electricity market environment data, the degree of matching between the power output of the power generation entity and the system demand is calculated through the first matching degree index, which serves as the fourth evaluation index. The bidding behavior pattern of the power generation entity is identified through the first cluster analysis method and coded as the fifth evaluation index.

[0030] S400: Standardize and fuse the data of the first to fifth evaluation indicators to construct a multi-dimensional feature vector, and input it with the preset rationality label into the first classification model for training to obtain the pricing rationality evaluation model. Input the feature vector obtained by processing the real-time data of the power generation entity to be evaluated into the pricing rationality evaluation model, and output the classification result of the pricing behavior rationality.

[0031] S500: Continuously evaluates the bidding behavior of power generation entities, tracks the changing trends of key evaluation indicators of power generation entities, and generates monitoring and early warning signals when evaluation indicators deviate from the threshold.

[0032] It should be noted that this invention achieves a leap from single evaluation to three-dimensional cognition by constructing multi-dimensional evaluation indicators. Through continuous evaluation and a multi-indicator combination early warning mechanism, it upgrades from post-event analysis to in-event intervention and pre-event prevention. It also uses the XGBoost model for efficient fusion and intelligent decision-making, and uses feature importance analysis to reveal the decision-making basis, ensuring the high accuracy and interpretability of the evaluation results.

[0033] Example 2, refer to Figure 1 This is a second embodiment of the present invention, which provides a method for evaluating the reasonableness of power generation pricing based on XGBoost, including: In step S100, data collection includes steps S101 to S103: S101: Collect historical operating data of the target power generation entity and corresponding electricity market environment data for the corresponding period.

[0034] The historical operating data includes multi-dimensional historical price data and actual output data, while the electricity market environment data covers system load forecasts and day-ahead market clearing prices.

[0035] S102: Convert the original 60-day × 96-point / day × 7-segment price data into a single-day price sequence in 60×1 format by taking the daily average.

[0036] S103: Preprocess all time series data, including the converted price series, actual output in 60×96 format, system load forecast, and market clearing price. Fill in missing values ​​using linear interpolation and ensure that all data are strictly aligned on the time axis, thus building a complete, well-organized, and time-synchronized data foundation for subsequent analysis.

[0037] Furthermore, in this embodiment of the application, in step S100, calculating the first evaluation index includes using the Prophet time series forecasting model as the first forecasting model to calculate the deviation between the actual bid price of the power generation entity and the predicted bid price of the model, specifically including steps S111~S113: S111: Organize the preprocessed daily price history series into the format required by the model, initialize the parameters according to the characteristics of the electricity market, set the long-term trend as linear, simulate medium-term fluctuations, and enable annual and weekly seasonal effects.

[0038] S112: The model is trained using the data from the previous 48 days as the training set. The model decomposes the time series into trend, period, and error terms. After training, the model predicts future time windows and generates a price quote series containing predicted values ​​and confidence intervals. The formula for decomposing a time series into a trend term, a period term, and an error term is as follows: in, The quote for time t, To analyze periodic changes in time series, The trend term reflects non-periodic changes in the time series. This is the error term.

[0039] S113: Within the test set or evaluation period, the first evaluation index is obtained by comparing the actual bids with the model-predicted bids and calculating the average of the absolute differences between the two. The smaller the value of the first evaluation index, the more the bidding behavior of the power generation entity conforms to its own historical patterns and market expectations. The first evaluation indicator was obtained: in, The actual quote for time t. For the predicted quote at time t, To assess the length of the time series.

[0040] In an optional implementation, in step S100, the calculation of the first evaluation index further includes: using a seasonal autoregressive integral moving average model to perform a stationarity test on the preprocessed price time series; eliminating trends through differencing operations; identifying the seasonal and non-seasonal autoregressive order and moving average order of the model by analyzing the autocorrelation plot and partial autocorrelation plot of the series; determining the model parameters through the maximum likelihood estimation method; and using the fitted model to perform rolling forecasts to obtain the predicted values ​​and confidence intervals of future prices; and calculating the average absolute deviation between the actual value and the predicted value as the evaluation index.

[0041] In another optional implementation, in step S100, the calculation of the first evaluation index may further include constructing the price quotation time series into a supervised learning sample format, that is, using past sequences as input features to predict the price quotation for the next day, building an LSTM neural network, using the internal gating mechanism to learn long-term dependencies, training the network with historical data, optimizing the network weights through backpropagation and gradient descent algorithms to minimize the prediction error, and after the model training is completed, inputting the latest historical sequence can obtain the price quotation prediction.

[0042] In this embodiment of the application, obtaining the second evaluation index in step S200 includes using a log-log linear regression model as the first elasticity analysis method to quantify the sensitivity of the quoted price to the market price, specifically including steps S201 to S203: S201: Prepare the historical bid volume sequence of the power generation entity and the corresponding market day-ahead clearing price sequence, and ensure that the two are precisely aligned in time.

[0043] S202: Establish a log-log linear regression model and fit it using the least squares method to analyze the nonlinear relationship between quoted volume and market price. The formula is expressed as: in, The price quoted by the power generation entity o at time t. The market clearing price for time t. This is a constant term, representing the base price level. The regression coefficient is the price elasticity. This is the random error term.

[0044] S203: The regression coefficient β obtained by fitting the model using the least squares method is the second evaluation index. Its economic meaning is: the average percentage change in the bidding volume of power generation entities for every 1% change in market price. When β>0, it means that the bidding volume increases with the rise in market price. When β<0, it means that a price reduction strategy is adopted during periods of high electricity price in order to win the bid. The larger the |β|, the more sensitive it is to price.

[0045] In an optional implementation, in step S200, obtaining the second evaluation index further includes using a panel data fixed effects model to collect panel data from multiple power generation entities and multiple time periods. When constructing the model, the explained variable is set as the bid quantity, the core explanatory variable is the market price, and individual fixed effects dummy variables representing different power generation entities are introduced to control the inherent characteristics of each entity that do not change over time. The coefficient of the market price variable is estimated through panel regression. The coefficient reflects the average impact of market price changes on the bid quantity after controlling for individual heterogeneity, and can be used as an adjusted elasticity measure.

[0046] In another optional implementation, in step S200, obtaining the second evaluation index may further include using a quantile regression model. Unlike ordinary linear regression, which focuses on the conditional mean, quantile regression analyzes the impact of market price changes on the bid volume at different conditional quantiles (low price bid quantile, high price bid quantile), and fits the regression model at different quantiles to obtain the elasticity coefficient.

[0047] Furthermore, in step S200, calculating the third evaluation index includes steps S211-S212: S211: When calculating the third assessment indicator, namely the deviation between the quoted price and the market price, the aim is to directly measure the absolute difference between the power generation entity's quoted price and the overall market clearing price. The formula is expressed as: in, The price quoted by the power generation entity at time t. As the third evaluation indicator, The market clearing price for time t. To assess the length of the time series, For time indexing.

[0048] S212: This indicator is obtained by calculating the average absolute difference between the power generation entity's bid and the corresponding market clearing price at each point in time during the assessment period. It reflects the deviation of the power generation entity's overall bid level from the market benchmark. The higher the value of the third assessment indicator, the greater the deviation between the bid and the market price, and the more likely there is abnormal bidding behavior. This provides a direct quantitative basis for subsequent comprehensive assessment.

[0049] In this embodiment of the application, in step S300, the calculation of the first matching degree index includes using the Dynamic Time Warping (DTW) algorithm as the first matching degree index to evaluate the degree to which the power generation output tracks the load of the system, specifically including steps S301 to S303: S301: Obtain the actual power output curve of the power generation entity and the system load forecast curve for the same period within the assessment period.

[0050] S302: The DTW algorithm calculates the distance between all pairs of points on two curves by constructing a local distance matrix, and uses dynamic programming to find the optimal curved path from the starting point to the ending point, which minimizes the cumulative cost when the two curves are aligned. The DTW algorithm formula is expressed as: in, Store the distances between all pairs of points on the two curves in a local distance matrix. This is the index of the time point on the actual output curve Q. This refers to the index of time points on the system load forecast curve L. Let Q be the actual output value of curve Q at time point i. Let L be the predicted system load value at time point j. For point With point Distance function between, Store the minimum cumulative distance from the starting point to each pair of points in the cumulative cost matrix. Let (i, j) be the path from the point pair (i-1, j) to (i, j). Let (i, j-1) be the path from point pair (i, j) to (i, j). Let (i, j) be the path from point pair (i-1, j-1) to (i, j). The dynamic time-warped distance between curve Q and curve L. The lower right element of the cumulative cost matrix S. The length of curve Q is the total number of time points. Let L be the length of curve L.

[0051] S303: The cumulative total distance of the current optimal path is defined as the fourth evaluation index. The smaller the DTW distance value, the more similar the actual output curve of the power generation entity is to the system load curve in terms of shape and trend. The higher the degree of matching between the output arrangement and system demand, the stronger the ability to respond to market signals.

[0052] In an optional implementation, in step S300, the calculation of the first matching index further includes using the Pearson correlation coefficient to calculate the Pearson correlation coefficient of the numerical sequences of the two curves at the same time point, which is used to measure the degree of linear correlation between the trend and direction of the two curves. The correlation coefficient is between -1 and 1, and the closer the value is to 1, the more consistent the trend of the two curves rising and falling synchronously.

[0053] In another optional implementation, in step S300, the calculation of the first matching index may further include using a similarity measure based on morphological feature extraction to extract key morphological features from the two curves, such as the number and position of extreme points (peaks and troughs), the slope of the rising / falling segments, and the curvature of the curves, and constructing feature vectors from the extracted features, and calculating the distance (Euclidean distance or cosine similarity) between the two feature vectors as a measure of the overall morphological similarity of the two curves.

[0054] Furthermore, in this embodiment of the application, in step S300, the first clustering analysis method includes using a time series K-Means clustering method based on DTW distance as the first clustering analysis method to identify and encode bidding behavior patterns, specifically including steps S311~S313: S311: Collect a large number of historical price curves and use the DTW algorithm to calculate the morphological similarity between any two curves to construct a distance matrix.

[0055] S312: Initialize the data, with a preset number of clusters K=7. Randomly select K curves as initial cluster centers, calculate the DTW distance between each price quote curve and all cluster centers, and assign it to the nearest cluster. For each cluster, update the cluster centers using the Dynamic Centroid Average (DBA) algorithm. Repeat the steps of assigning clusters and updating centers until the change in cluster centers is less than a preset threshold. The algorithm converges, resulting in 7 typical price quote behavior paradigm clusters.

[0056] DBA generates a new representative center curve by iteratively aligning all curves within the cluster and averaging the points.

[0057] S313: Several typical bidding behavior paradigms are formed, the historical bidding curves of the power generation entities to be evaluated are classified into a specific paradigm, and converted into a multi-dimensional binary vector through unique thermal coding, which serves as the fifth evaluation index.

[0058] In an optional implementation, in step S300, the first clustering analysis method further includes using K-means clustering based on Euclidean distance combined with feature dimensionality reduction to extract a set of statistical features (mean, variance, skewness, kurtosis) and time series features (autocorrelation coefficient) from each price quote curve, using principal component analysis to reduce the dimensionality of the high-dimensional features, and using Euclidean distance to perform standard K-means clustering in the dimensionality-reduced feature space to classify the price quote behavior.

[0059] In another optional implementation, in step S300, the first clustering analysis method may further include: using shape feature-based and hierarchical clustering to extract key morphological features of the quotation curve to form feature vectors, and using a hierarchical clustering algorithm to calculate the distance between feature vectors, merging or splitting clusters from bottom to top or from top to bottom to form a tree-like clustering structure, and cutting at a specified level to obtain behavioral classification.

[0060] Furthermore, pricing behavior paradigms include cost-oriented, market-responsive, strategy-adjusted, dynamically adjusted, gradual, highly elastic, and stable pricing behaviors.

[0061] It should be noted that there are several types of pricing strategies: Cost-oriented: pricing based on marginal cost; Market-responsive: pricing closely follows market price changes; Strategy-adjusted: pricing has multiple price plateau periods, corresponding to peak load periods; Dynamically adjusted: pricing changes frequently without obvious long-term plateau periods; Gradual pricing: pricing changes in a smooth gradient over time; Highly elastic pricing: pricing fluctuates significantly and is very sensitive to market price changes; Stable pricing: pricing curve changes gently, stably, and with small fluctuations.

[0062] In step S400, the data standardization and feature fusion include steps S401 to S403: S401: Perform Min-Max standardization on the continuous numerical features of the first to fourth evaluation indicators, linearly scaling them to the range of [0,1] to eliminate the dimensional differences between different indicators; The Min-Max normalization formula is expressed as: in, These are the original eigenvalues. The standardized value. The minimum value of the original feature. This represents the maximum value of the original feature.

[0063] S402: The fifth evaluation index, as a 7-dimensional one-hot encoded vector, is already a standard binary discrete feature and does not require further conversion.

[0064] S403: The four standardized continuous features and the seven discrete feature vectors are concatenated in dimensional order to form a unified multidimensional feature vector containing 11 dimensions, which serves as the final input for the subsequent classification model.

[0065] Furthermore, in step S400, obtaining the quotation reasonableness assessment model includes steps S411~S414: S411: The complete dataset containing all samples' 11-dimensional multi-dimensional feature vectors and corresponding "reasonable" or "unreasonable" manually labeled data is randomly divided into a training set and a test set in an 8:2 ratio. The training set is used for model parameter learning, and the test set is used to verify the model's generalization ability.

[0066] S412: Configure the hyperparameters of the XGBoost classifier, including setting the objective function to multi:softmax for binary classification, setting the maximum tree depth to 5, the learning rate to 0.1, the number of base learners (trees) to 100, and fixing the random seed to 42 to ensure reproducibility of results. At the same time, set the L1 and L2 regularization parameters to prevent overfitting.

[0067] S413: Input the features and labels of the training set into the model for training. The training process adopts a gradient boosting framework to iteratively build multiple CART decision trees. In each iteration, a new tree is built to fit the residual (negative gradient) between the current model prediction result and the true label. The structure of the tree and the weights of the leaf nodes are learned by minimizing the loss function containing first-order and second-order derivative information and regularization terms.

[0068] S414: Use the test set to evaluate the performance of the trained model, and calculate the model's precision, recall, F1 score, and accuracy on the test set. The preset performance thresholds for a valid model include: precision and recall of reasonable categories should be no less than 80%, precision and recall of unreasonable categories should be no less than 85%, overall F1 score should be no less than 0.8, and accuracy should be no less than 80%. When the test results reach or exceed the thresholds, the model training is confirmed to be effective, the final model structure and parameters are saved, and the final pricing rationality evaluation model is saved.

[0069] Furthermore, in step S400, the classification result of the reasonableness of the output quotation behavior includes steps S421~S423: S421: The logic for generating classification results is based on the feature importance weights and decision rules learned internally by the model.

[0070] S422: Feature importance weights are automatically obtained during model training. According to the gain importance ranking provided by the XGBoost model, the fourth evaluation indicator (output curve matching degree) is the most important, followed by the third evaluation indicator (price deviation degree) and the second evaluation indicator (price elasticity), then the first evaluation indicator (price prediction deviation), and the fifth evaluation indicator (price behavior type) has the lowest feature importance across all dimensions.

[0071] S423: The decision rules include the model's comprehensive nonlinear judgment of the input feature values, the model learning the critical threshold of each feature during training, and when the normalized matching degree (fourth indicator) is lower than 0.6, the price deviation (third indicator) is higher than 0.5, the absolute value of price elasticity (second indicator) is greater than 0.8, or the price prediction deviation (first indicator) is greater than 0.15 in the feature vector to be evaluated, the internal decision path of the model will tend to output the "unreasonable" classification label. The final classification result is the result of collective voting or probability averaging of all decision trees in the model.

[0072] In step S500, generating the monitoring and early warning signal includes steps S501 to S503: S501: Regularly (e.g., daily or weekly) collect the latest market data and power generation entity data, repeat steps S100 to S400, realize continuous rolling evaluation and dynamic classification of power generation entity bidding behavior, and form a historical trajectory of the rationality of behavior over time.

[0073] S502: For each power generation entity being evaluated, establish time-series tracking records for key evaluation indicators (second, third, fourth, and fifth evaluation indicators), and set monitoring and warning thresholds for each indicator, including: matching degree (fourth indicator) consistently below 0.6, absolute value of price elasticity (second indicator) consistently above 0.8, price deviation (third indicator) consistently above 0.5, and the pricing behavior paradigm (fifth indicator) being judged as "strategy adjustment type" or "high elasticity pricing behavior" for a long period (such as multiple consecutive evaluation cycles).

[0074] S503: When the monitoring system detects that a power generation entity simultaneously meets or repeatedly meets multiple pre-set alarm conditions, it automatically triggers the system's early warning mechanism, generates a report containing the current entity's identifier, alarm indicators, and trend analysis, and marks it as a high-risk entity that requires key attention and further verification in market operations.

[0075] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

[0076] Example 3, referring to Figure 2 This is the third embodiment of the present invention, which provides a power generation pricing rationality assessment system based on XGBoost, including a data acquisition and preprocessing module, a multi-dimensional feature extraction module, a feature engineering and fusion module, an XGBoost model training and optimization module, a rationality assessment and decision interpretation module, and a continuous monitoring and dynamic early warning module.

[0077] The data acquisition and preprocessing module is used to collect historical segmented price quotes and actual output curves of power generation entities, as well as market environment data such as system load forecasts and market clearing prices. By performing mean aggregation on data of different formats and linear interpolation on missing values, all data are unified to the same time granularity and time length.

[0078] The multi-dimensional feature extraction module is used to generate price predictions and calculate price prediction deviations using the Prophet time series prediction model, fit the relationship between price volume and market price using a log-log linear regression model to calculate price elasticity, calculate the degree of deviation between price and market price using the arithmetic mean, calculate the matching degree between the force curve and the load curve using the dynamic time warping algorithm, and identify and encode the price behavior paradigm type based on the DTW distance time series clustering method.

[0079] The feature engineering and fusion module is used to perform Min-Max standardization on the bid prediction deviation, bid elasticity, deviation degree and matching degree, and scale it to the [0,1] interval. It performs one-hot encoding on the discrete bid behavior paradigm type indicators, converts them into binary vectors, and concatenates the standardized continuous feature vector with the encoded discrete feature vector to obtain a comprehensive multidimensional feature vector.

[0080] The XGBoost model training and optimization module is used to divide the labeled multidimensional feature vector dataset into training and test sets, use the XGBoost gradient boosting decision tree framework for model training, configure parameters, and iteratively build decision trees to minimize the loss function during training to obtain a strong classifier. After the model training is completed, the effectiveness of the model is judged by verifying whether the precision, recall and F1 score metrics reach the preset thresholds on the test set, and the optimal model is saved.

[0081] The rationality assessment and decision interpretation module is used to input the feature vector obtained by processing the data to be evaluated into the trained XGBoost model and directly output the classification result.

[0082] The continuous monitoring and dynamic early warning module is used to generate a time series trajectory of the rationality of the behavior of each power generation entity by periodically and automatically executing the entire process from data collection to rationality assessment, and preset monitoring thresholds for key indicators. It makes a comprehensive judgment based on the persistence of behavioral paradigm types. When it is detected that an entity continuously triggers multiple early warning conditions, the system automatically generates an early warning signal and marks it as a high-risk object.

[0083] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

[0084] Example 4, the fourth embodiment of the present invention, differs from the previous three embodiments in that: If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, 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 invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0085] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0086] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0087] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

Claims

1. A method for evaluating the reasonableness of power generation pricing based on XGBoost, characterized by: include, Collect historical operating data of power generation entities and electricity market environment data. Based on the historical operating data, obtain the bidding prediction data of power generation entities through the first prediction model, and calculate the first evaluation index. Based on historical operating data and electricity market environment data, the first elasticity analysis is used to obtain the degree of response of power generation entities to changes in market prices, which serves as the second evaluation indicator. The deviation between the power generation entity's bid and the market price is calculated as the third evaluation indicator. Based on historical operating data and electricity market environment data, the degree of matching between the power output of the power generation entity and the system demand is calculated through the first matching degree index, which serves as the fourth evaluation index. The bidding behavior pattern of the power generation entity is identified through the first cluster analysis method and coded as the fifth evaluation index. The first to fifth evaluation indicators are standardized and feature fused to construct a multi-dimensional feature vector. This vector is then used in conjunction with a pre-set rationality label to train the first classification model, thereby obtaining a pricing rationality evaluation model. The feature vector obtained by processing the real-time data of the power generation entity to be evaluated is then input into the pricing rationality evaluation model, and the classification result of the pricing behavior rationality is output. The bidding behavior of power generation entities is continuously evaluated, and the changing trends of key evaluation indicators of power generation entities are tracked. When the evaluation indicators deviate from the threshold, a monitoring and early warning signal is generated.

2. The XGBoost-based power generation pricing rationality assessment method as described in claim 1, characterized in that: The calculation of the first evaluation index includes preprocessing historical quotation data, integrating the original quotation data of different time granularities and different number segments into a single time series data of uniform format by taking the average, and filling in the missing values ​​in the series. The integrated time series data is modeled using the first prediction model. Based on the first prediction model that has completed parameter learning and fitting, the price values ​​for future time windows are rolled out to generate a sequence of price prediction values ​​containing the predicted values ​​and confidence intervals. Align the actual bid sequence and the bid prediction sequence of the power generation entities within the same time window, and calculate the average of the absolute differences between the two.

3. The XGBoost-based power generation pricing rationality assessment method as described in claim 2, characterized in that: The second evaluation metric includes obtaining the bidding volume sequence of power generation entities and the market day-ahead clearing price sequence from the preprocessed data, aligned by time points. Logarithmic transformations are performed on the quotation volume series and the market price series to convert the exponential growth relationship into a linear relationship, satisfying the assumptions of the first elasticity analysis method regarding data stationarity and model form. The two logarithmically transformed sequence data are input into the first elasticity analysis model, and a linear equation is fitted by minimizing the error term. The slope coefficient, i.e., price elasticity, is extracted from the fitted linear equation, and the slope coefficient value is used as a second evaluation index to characterize the sensitivity.

4. The XGBoost-based power generation pricing rationality assessment method as described in claim 3, characterized in that: The fourth evaluation indicator includes obtaining the actual power output curve and system load prediction curve of the power generation entity within the evaluation period, using the same time granularity and the same time series length. Using the first matching degree index calculation method, the two curves are non-linearly aligned and compared globally. That is, by constructing a cumulative cost matrix, the optimal bending path in the matching of all point pairs between the two curves is found, so as to minimize the cumulative alignment cost. Based on the optimal curvature path, the final cumulative distance value is calculated and used as the fourth evaluation metric.

5. The XGBoost-based power generation pricing rationality assessment method as described in claim 4, characterized in that: The first clustering analysis method includes collecting historical segmented price curve data of the target power generation entity and similar power generation entities to form an original price curve set, and converting all price curves into a daily price time series of uniform length by taking the daily average value. The DTW algorithm is used to calculate the similarity distance between two curves in the original set of price quote curves, and a cumulative cost matrix is ​​constructed. The final cumulative cost value is the DTW distance between the two sequences. The DTW distance matrix is ​​obtained by calculating the similarity distance for all curve pairs, as expressed by the formula: in, Store the distances between all pairs of points on the two curves in a local distance matrix. This is the index of the time point on the actual output curve Q. This refers to the index of time points on the system load forecast curve L. Let Q be the actual output value of curve Q at time point i. Let L be the predicted system load value at time point j. For point With point Distance function between, Store the minimum cumulative distance from the starting point to each pair of points in the cumulative cost matrix. Let (i, j) be the path from the point pair (i-1, j) to (i, j). Let (i, j-1) be the path from point pair (i, j) to (i, j). Let (i, j) be the path from point pair (i-1, j-1) to (i, j). The dynamic time-warped distance between curve Q and curve L. The lower right element of the cumulative cost matrix S. The length of curve Q is the total number of time points. Let L be the length of curve L; Based on prior business knowledge or the elbow rule, the number of clusters is preset, and an optimization method based on the distance matrix is ​​used to initialize curves from the original set of curves as initial cluster centers. For each historical price quote curve in the set, calculate the DTW distance between it and the cluster center curve, and assign it to the cluster represented by the nearest cluster center according to the principle of minimum distance; For each newly formed cluster, the dynamic centroid averaging algorithm is used to average all curves within the cluster. Using the current cluster center as the initial template, the DTW alignment is used to align all curves within the cluster with the template, and the average value of all aligned points is calculated as the new template. This process is iterated until the template is stable, generating a new cluster center curve for the current cluster. Repeat the iteration until the change in all cluster center curves compared to the previous iteration is less than a preset threshold, indicating that the algorithm has converged and obtained stable clustering results. Each cluster represents a pricing behavior paradigm. The historical bidding curves of the power generation entities to be evaluated are calculated, and their DTW distances to the cluster centers are determined. The entities are then classified into the clusters with the smallest distances. The cluster labels represent the bidding behavior paradigm types. The current type is then converted into a binary vector using one-hot encoding, and the vector is used as the fifth evaluation indicator.

6. The XGBoost-based method for evaluating the reasonableness of power generation pricing as described in claim 5, characterized in that: The data standardization and feature fusion process includes processing the continuous numerical features represented by the first evaluation indicator (price prediction deviation), the second evaluation indicator (price elasticity coefficient), the third evaluation indicator (price deviation degree), and the fourth evaluation indicator (DTW matching distance) using the minimum-maximum normalization method. For each feature, the minimum and maximum values ​​in the current batch of data are calculated, and all feature values ​​are linearly scaled to the range [0, 1]. The minimum-maximum normalization method formula is expressed as: in, These are the original eigenvalues. The standardized value. The minimum value of the original feature. The maximum value of the original feature; For the categorical features represented by the fifth evaluation index, namely the obtained seven-dimensional one-hot encoded vector, it is directly used as the discrete feature part. The normalized continuous feature vector and the one-hot encoded discrete feature vector are concatenated in the order of feature dimensions to form an 11-dimensional unified multi-dimensional feature vector.

7. The XGBoost-based power generation pricing rationality assessment method as described in claim 6, characterized in that: The training of the first classification model includes dividing the dataset containing 11-dimensional multi-dimensional feature vectors of all samples and reasonable or unreasonable manually labeled data into a training set and a test set. Configure the hyperparameters of the XGBoost classifier, set the objective function, set the maximum tree depth, learning rate, number of base learners, fix the random seed, and set the L1 and L2 regularization parameters; The features and labels of the training set are input into the model for training. The gradient boosting framework is used to iteratively build multiple CART decision trees. In each iteration, a new tree is built to fit the residual between the current model prediction result and the true label. The structure of the tree and the weight of the leaf nodes are learned by minimizing the loss function containing first-order and second-order derivative information and regularization terms. The performance of the trained model is evaluated using the test set. The precision, recall, F1 score, and accuracy of the model on the test set are calculated. The preset performance thresholds for the model to be effective are: precision and recall of reasonable classes are not less than 80%, precision and recall of unreasonable classes are not less than 85%, overall F1 score is not less than 0.8, and accuracy is not less than 80%. When the test results reach or exceed the thresholds, the model training is confirmed to be effective, and the final model structure and parameters are saved. The classification results of the reasonableness of the output pricing behavior include the generation logic based on the feature importance weights and decision rules learned internally by the model; The feature importance weights are automatically obtained during model training. According to the gain importance ranking provided by the XGBoost model, the fourth evaluation indicator, namely the output curve matching degree, has the highest importance. The third evaluation indicator, namely the degree of price deviation, and the second evaluation indicator, namely the price elasticity, are next in importance. The first evaluation indicator, namely the price prediction deviation, is next in importance. The fifth evaluation indicator, namely the feature importance of each dimension of the price behavior type, has the lowest importance. The decision rules include the model's comprehensive nonlinear judgment of the input feature values, the model learning the critical thresholds of each feature during training, and the model's collective voting of all decision trees to obtain the final classification result.

8. A power generation pricing rationality assessment system based on XGBoost, employing the power generation pricing rationality assessment method based on XGBoost as described in any one of claims 1 to 7, characterized in that, It includes modules for data acquisition and preprocessing, multi-dimensional feature extraction, feature engineering and fusion, XGBoost model training and optimization, rationality assessment and decision interpretation, and continuous monitoring and dynamic early warning. The data acquisition and preprocessing module is used to collect historical segmented price quotes and actual output curves of power generation entities, as well as market environment data such as system load forecasts and market clearing prices. It also unifies all data to the same time granularity and time length by performing mean aggregation on data of different formats and linear interpolation on missing values. The multi-dimensional feature extraction module is used to generate price predictions and calculate price prediction deviations using the Prophet time series prediction model, fit the relationship between price volume and market price using a log-log linear regression model to calculate price elasticity, calculate the degree of deviation between price and market price using the arithmetic mean, calculate the matching degree between the force curve and the load curve using the dynamic time warping algorithm, and identify and encode the price behavior paradigm type based on the DTW distance time series clustering method. The feature engineering and fusion module is used to perform Min-Max standardization on the bid prediction deviation, bid elasticity, deviation degree and matching degree, and scale it to the [0,1] interval. It performs one-hot encoding on the discrete bid behavior paradigm type indicators, converts them into binary vectors, and concatenates the standardized continuous feature vector with the encoded discrete feature vector to obtain a comprehensive multidimensional feature vector. The XGBoost model training and optimization module is used to divide the labeled multidimensional feature vector dataset into training and test sets, use the XGBoost gradient boosting decision tree framework for model training, configure parameters, and iteratively build decision trees to minimize the loss function during training to obtain a strong classifier. After the model training is completed, the effectiveness of the model is judged by verifying whether the precision, recall and F1 score metrics reach the preset thresholds on the test set, and the optimal model is saved. The rationality assessment and decision interpretation module is used to input the feature vector obtained by processing the data to be evaluated into the trained XGBoost model and directly output the classification result. The continuous monitoring and dynamic early warning module is used to generate a time series trajectory of the rationality of the behavior of each power generation entity by periodically and automatically executing the entire process from data collection to rationality assessment, and preset monitoring thresholds for key indicators. It makes a comprehensive judgment based on the persistence of behavioral paradigm types. When it is detected that an entity continuously triggers multiple early warning conditions, the system automatically generates an early warning signal and marks it as a high-risk object.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the XGBoost-based power generation pricing rationality assessment method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the XGBoost-based power generation pricing rationality assessment method as described in any one of claims 1 to 7.

Images