Data analysis-based power spot transaction data model construction method

Abstract

The present application relates to the technical field of data model construction, and especially relates to a power spot transaction data model construction method based on data analysis, which comprises the following steps: collecting key data in real time; generating a training set through data preprocessing; constructing an incremental model; determining the stability of training data; identifying abnormal conduction types; generating a bias vector adjustment parameter; and updating the model through incremental training. In the clearing process of the high-volatility limited power spot market, the present application simultaneously introduces multi-dimensional parameters such as the change rate of node marginal electricity price, the climbing gradient of main power output units, the real-time order book fractal dimension, and the satellite remote sensing cloud movement speed for data analysis, effectively solving the problems of poor model adaptability, insufficient prediction accuracy, and further causing risk warning delay and poor market regulation effect due to the lack of timely identification and response to market supply and demand fluctuations and abnormal behaviors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data model construction, in particular to a power spot transaction data model construction method based on data analysis. BACKGROUND

[0002] With the large-scale access of renewable energy, the influence of meteorological factors on the power system is increasingly prominent. Weather phenomena such as cloud movement and wind speed change directly affect the output fluctuation of new energy power generation, and further cause changes in supply and demand and price fluctuations in the power market. How to effectively integrate real-time meteorological data and market transaction data to accurately reflect the influence of meteorological changes on market behavior has become an important direction to improve the risk monitoring and regulation ability of the power spot market. In the face of the spatio-temporal heterogeneity and dynamic complexity of multi-source data, it is still a major technical challenge to realize the deep integration and accurate modeling of meteorological information and market data.

[0003] Chinese patent application publication No. CN116433261A discloses a power spot transaction data model modeling method, which comprises the following steps: S110, constructing the basic architecture of the power spot transaction data model, referring to the CIM modeling standard, and setting the specific level content; S120, collecting power spot transaction data, including transaction center data, dispatching system data and other data for auxiliary transaction decision analysis; S130, constructing data tables and index attribute assignments, specifically, according to the collected and sorted power spot transaction data, following the constructed data model architecture, performing data table construction and index attribute assignment on the power spot transaction data; S140, constructing a full-process data business domain model, specifically, according to the basic transaction rules of the power spot market, the transaction full-cycle business process and the market member categories, sorting out the full-cycle business process of the power spot transaction data, and constructing a full-process data business domain model; S150, forming a complete data model of the power spot transaction, specifically, integrating the data physical model content of steps S110 to S140, establishing the inter-table association relationship data logic model, and forming a complete data model of the power spot transaction.

[0004] It can be seen that the power spot transaction data model modeling method has the following problems: the method focuses on static data model construction and business process sorting, lacks modeling and analysis of real-time volatility, supply and demand response dynamic changes and price fluctuations in the power spot market, and is difficult to meet the real-time regulation needs of high-volatility markets; the method is mainly based on transaction center, dispatching system and auxiliary decision data, and does not involve the integration of multi-dimensional, multi-source heterogeneous data such as real-time meteorological information, remote sensing data and market behavior fractal characteristics, limiting the sensitivity and identification ability of the model to external environment and market abnormal behavior. SUMMARY

[0005] To this end, the application provides a power spot transaction data model construction method based on data analysis, which is used to overcome the problems of poor model adaptability, insufficient prediction accuracy, and thus delayed risk warning and poor market regulation effect caused by the lack of timely identification and response to market supply and demand fluctuations and abnormal behavior in the prior art by fusing multi-source dynamic data and a real-time abnormality determination mechanism.

[0006] To achieve the above-mentioned purpose, the application provides a power spot transaction data model construction method based on data analysis, comprising:

[0007] Real-time collection of the node marginal price change rate in the clearing process of the high fluctuation limited type power spot market, the ramping gradient of the main output generator set, the fractal dimension of the real-time order book, and the cloud layer movement speed in the satellite remote sensing image to obtain a collection result, and preprocessing the collection results within a preset training period to obtain an initial training set;

[0008] Construction of an incremental model according to the initial training set;

[0009] Based on the incremental model, steady state determination is performed according to the node marginal price change rate, the ramping gradient, and a preset steady state threshold to obtain a first determination result or a second determination result;

[0010] Based on the first determination result, abnormal conduction path types are determined according to the cloud layer movement speed and the fractal dimension to obtain a supply and demand disturbance type or a price manipulation type;

[0011] Based on the supply and demand disturbance type, a structural bias vector is generated according to the node marginal price change rate and the ramping gradient, and based on the price manipulation type, a behavior bias vector is generated according to the fractal dimension and the cloud layer movement speed;

[0012] The preset steady state threshold and the preset training period are adjusted according to the structural bias vector and the behavior bias vector;

[0013] Based on the second determination result, the incremental model is trained according to the collection results within the next preset training period to obtain a target model.

[0014] Further, the process of steady state determination according to the node marginal price change rate, the ramping gradient, and a preset steady state threshold to obtain a first determination result and a second determination result comprises:

[0015] A plurality of price fluctuation values are determined according to the marginal price change rate from the initial time to each time within a preset steady state determination period;

[0016] A plurality of supply and demand response accelerations are determined according to the ramping gradient within the preset steady state determination period;

[0017] determining a steady state index at an end time of a preset steady state determination period according to all the price fluctuation values and all the supply-demand response accelerations;

[0018] performing steady state determination based on a comparison result of the steady state index and the preset steady state threshold to obtain the first determination result or the second determination result.

[0019] Further, the process of determining a steady state index at an end time of a preset steady state determination period according to all the price fluctuation values and all the supply-demand response accelerations comprises:

[0020] calculating the steady state index according to all the price fluctuation values, a preset price weight, all the supply-demand response accelerations, and a preset response weight.

[0021] Further, based on the first determination result, determining an abnormal conduction path type according to the cloud layer movement speed and the fractal dimension to obtain a supply-demand disturbance type or a price manipulation type comprises:

[0022] determining a direction consistency coefficient of a wind cloud vector field in the satellite remote sensing image according to the cloud layer movement speed;

[0023] determining a periodicity index of the order book structure according to the fractal dimension;

[0024] determining a weather market influence index according to the direction consistency coefficient and the periodicity index;

[0025] determining the abnormal conduction path type to be the supply-demand disturbance type or the price manipulation type according to the weather market influence index.

[0026] Further, the process of determining a weather market influence index according to the direction consistency coefficient and the periodicity index comprises:

[0027] determining the weather market influence index according to the direction consistency coefficient and the periodicity index, a preset consistency weight, and a preset periodicity weight.

[0028] Further, the process of determining an abnormal conduction path type to be the supply-demand disturbance type or the price manipulation type according to the weather market influence index comprises:

[0029] determining the abnormal conduction path type to be the supply-demand disturbance type or the price manipulation type according to a comparison result of the weather market influence index and a preset influence index threshold.

[0030] Further, based on the supply-demand disturbance type, the process of generating a structural bias vector according to the node marginal price change rate and the climbing gradient comprises:

[0031] determining a price fluctuation amplitude according to the node marginal price change rate within the preset time window, and determining a gradient acceleration according to the ramping gradient within the preset time window;

[0032] synchronizing the price fluctuation amplitude and the ramping gradient acceleration according to the preset time window to form a two-dimensional deviation feature sequence;

[0033] calculating a mean deviation and an extreme deviation within each preset time window according to the two-dimensional deviation feature sequence, and combining the mean deviation and the extreme deviation into a plurality of structural deviation components;

[0034] connecting all the structural deviation components in chronological order to obtain the structural deviation vector.

[0035] Further, based on the price manipulation type, a process of generating a behavior deviation vector according to the fractal dimension and the cloud layer movement speed includes:

[0036] determining a fractal dimension change rate according to the fractal dimension within the preset time window, and calculating a direction change rate of the cloud layer movement speed within the preset time window;

[0037] synchronizing and pairing the fractal dimension change rate and the cloud layer movement speed direction change rate according to the preset time window to obtain a two-dimensional behavior feature sequence;

[0038] determining a plurality of behavior deviation components according to a relative change deviation and a correlation coefficient deviation within each preset time window according to the two-dimensional behavior feature sequence;

[0039] connecting all the behavior deviation components in chronological order to obtain the behavior deviation vector.

[0040] Further, a process of adjusting the preset steady state threshold and the preset training period according to the structural deviation vector and the behavior deviation vector includes:

[0041] determining a deviation correlation degree according to all the structural deviation vectors and all the behavior deviation vectors from the time when the first determination result is obtained to the next steady state determination;

[0042] when the absolute value of the deviation correlation degree is greater than a preset deviation correlation threshold, adjusting the preset steady state threshold and the preset training period according to the deviation correlation degree and the preset deviation correlation threshold.

[0043] Further, a process of adjusting the preset steady state threshold and the preset training period according to the deviation correlation degree and the preset deviation correlation threshold includes:

[0044] According to the absolute value of the deviation correlation degree and the relative deviation of the preset deviation correlation threshold, the preset first adjustment coefficient increases the preset steady state threshold value;

[0045] According to the absolute value of the deviation correlation degree and the relative deviation of the preset deviation correlation threshold, the preset second adjustment coefficient and the threshold period lower limit decrease the preset training period.

[0046] Compared with the prior art, the beneficial effects of the present application are that by simultaneously introducing node marginal price change rate, main output unit climbing gradient, real-time order book fractal dimension and satellite remote sensing cloud movement speed and other multi-dimensional parameters in the clearing process of the high fluctuation limited type electricity spot market, the price fluctuation characteristics, supply and demand dynamic characteristics, market microstructure characteristics and external weather disturbance characteristics can be cooperatively reflected in the unified model; in the steady state determination, abnormal conduction path identification and deviation vector generation process, a step-by-step progressive mapping relationship from price change to supply and demand response, to market behavior and external disturbance is established between the parameters, thereby realizing accurate characterization of the market state and adaptive adjustment of the dynamic threshold and training period, ensuring that the model still has high generalization ability and prediction accuracy when facing different disturbance types and change scenarios, and can continuously optimize the decision effect in subsequent incremental training, effectively solving the problems of poor model adaptability, insufficient prediction accuracy, and further causing risk warning delay and poor market regulation effect due to the lack of timely identification and response to market supply and demand fluctuations and abnormal behavior.

[0047] Further, by quantifying the standard deviation of the node marginal price change rate and the first-order difference of the unit climbing gradient as the price fluctuation value and the supply and demand response acceleration respectively, and determining the steady state index, accurate characterization of the market operation state is realized. Price fluctuation reflects the sensitivity of the market to supply and demand relationship, transaction activity and external disturbance, and supply and demand response acceleration reflects the dynamic matching level of unit regulation capacity and load change, and the combination of the two can comprehensively reveal the dynamic coupling relationship between market fluctuation amplitude and regulation rate. By comparing the steady state index with the preset threshold value, it can be distinguished whether the data is in stable operation or in severe fluctuation state, providing reliable basis for training set screening, abnormal path determination and parameter dynamic adjustment in subsequent model incremental training, thereby improving the adaptability and accuracy of the model in response to different market scenarios.

[0048] Further, by normalizing the price fluctuation value and the supply-demand response acceleration respectively, market fluctuation characteristics of different dimensions and value ranges can be quantitatively compared under the same evaluation system, and by combining the weighting calculation of the preset price weight and the preset response weight, a steady-state index that can comprehensively reflect the market price change amplitude and the unit regulation capacity is obtained, so as to quantitatively evaluate the market operation state at the end of the preset steady-state determination period, which helps to accurately identify the superimposed effect of short-term market fluctuation and supply-demand imbalance, and improve the sensitivity and accuracy of steady-state determination.

[0049] Further, by using the cloud movement speed to reflect the dynamic influence of weather systems on regional meteorological conditions, and by using the optical flow algorithm to extract the wind cloud motion vector field and calculate the direction consistency coefficient to measure the stability and concentration of meteorological disturbances in space; at the same time, by analyzing the complexity of the market real-time order book structure through fractal dimension, and by extracting the main periodic component through Fourier transform to calculate the periodicity index, the regularity of market transaction behavior in the time dimension is described; the weather market influence index is formed by combining the meteorological disturbance characteristics and the market structure characteristics, so that the conduction effect of meteorological changes on power supply and demand fluctuations and price anomalies can be quantified and distinguished, thereby effectively identifying the supply-demand disturbance type and the price manipulation type two types of abnormal paths, providing accurate basis for risk warning and transaction strategy optimization of the electricity spot market.

[0050] Further, by normalizing the direction consistency coefficient and the periodicity index within the preset type determination period, the comparability of parameters of different dimensions and different numerical scales is realized, ensuring that meteorological disturbance characteristics and market order structure characteristics play a role in a unified evaluation framework; the direction normalization coefficient reflects the stability of the spatial distribution of meteorological disturbances, and the periodicity index describes the time periodicity of the market order structure, both of which form the weather market influence index under the action of weight parameters, so that the conduction effect of meteorological changes on market price and supply-demand relationship can be quantified, and the accuracy and sensitivity of abnormal conduction path identification can be improved without distorting the original information, providing more targeted decision-making basis for the warning and strategy adjustment of the electricity spot market.

[0051] Further, by comparing the weather market influence index with the preset influence index threshold, the quantitative demarcation of meteorological disturbance characteristics and market structure characteristics is realized: when the index value is high, it indicates that the change of meteorological conditions has a significant impact on the supply and demand side, thereby triggering the determination of the supply-demand disturbance type; when the index value is low, it indicates that market price fluctuations are more due to abnormal transaction structure or behavior pattern, pointing to the price manipulation type. This determination mechanism combines meteorological dynamics, market order characteristics and price change trends, so that different abnormal causes can be quickly distinguished.

[0052] Further, by combining the node marginal price change rate with the dynamic information of the generator group climbing gradient, the comprehensive characteristics reflecting the changes in supply and demand structure are extracted. First, the price fluctuation amplitude is calculated by the price change rate to depict the fluctuation intensity at the market price level. Then, the gradient acceleration is calculated by the climbing gradient to reflect the speed and amplitude of the change in power generation output. Subsequently, the two types of parameters are time-synchronized within a unified time window to ensure that the changes in price and supply and demand response are comparable. Based on the synchronized two-dimensional feature sequence, the mean deviation and extreme deviation of each window are calculated to capture the dual information of stable deviation and sudden deviation. Finally, the deviation components are connected in time sequence to form a structural deviation vector, realizing the continuous quantitative description of the dynamic deviation pattern of market supply and demand structure, and providing accurate basis for identifying structural abnormalities.

[0053] Further, based on the synchronous construction of the fractal dimension change rate and the cloud layer movement speed direction change rate, the behavior feature sequence is effectively revealed, which effectively reveals the interaction between the complex dynamics of the market order structure and the changes in the external environment. By calculating the relative change deviation and the correlation coefficient deviation, the time sequence characteristics of abnormal behavior can be accurately depicted, and then the behavior deviation vector with time sequence coherence is formed. This method fully reflects the internal relationship between market internal structure fluctuation and environmental disturbance, improves the identification ability and response sensitivity of the model to price manipulation behavior, and thus provides scientific and dynamic support for risk monitoring and regulation of the electricity spot market.

[0054] Further, by calculating the correlation between the structural deviation vector and the behavior deviation vector, the internal relationship between market supply and demand dynamics and abnormal behavior is effectively captured. When the deviation correlation exceeds the preset threshold, the threshold and training period of steady state determination can be dynamically adjusted to improve the sensitivity and adaptability of the model to market fluctuations, so as to more accurately reflect the real-time state and potential risks of the electricity spot market. This dynamic adjustment mechanism reasonably utilizes the synergistic change relationship between multi-dimensional parameters, realizes fine description and effective response to complex market environment, and enhances the accuracy and reliability of risk monitoring and early warning.

[0055] Further, by comparing the deviation correlation with the preset deviation correlation threshold, and adjusting the steady state threshold and training period based on the relative deviation, the model can automatically increase the steady state determination threshold and shorten the model update period when the structural deviation and behavior deviation are highly correlated, thereby enhancing the sensitivity and response speed to sudden market disturbances. When the correlation is low, a longer training period and a lower steady state threshold are maintained to maintain the stability and generalization ability of the model. This mechanism utilizes the dynamic coupling relationship between parameters to make the threshold adjustment amplitude proportional to the correlation strength, while introducing a period lower limit to prevent excessive adjustment, balancing real-time and robustness, and effectively improving the accuracy and reliability of risk identification and regulation of the electricity spot market. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 A flow chart of the data analysis-based power spot transaction data model construction method of the present embodiment is shown in FIG. 1, which provides a data analysis-based power spot transaction data model construction method, comprising:

[0057] Figure 2 A determination logic diagram for steady state determination of the present embodiment is shown in FIG. 2;

[0058] Figure 3 A determination logic diagram for determining the type of abnormal conduction path of the present embodiment is shown in FIG. 3;

[0059] Figure 4 A determination logic diagram for adjusting the preset steady state threshold and the preset training period of the present embodiment is shown in FIG. 4. DETAILED DESCRIPTION

[0060] In order to make the objects and advantages of the present application clearer, the present application will be further described below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0061] The preferred embodiments of the present application will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present application and do not limit the protection scope of the present application.

[0062] Please refer to Figure 1 FIG. 1, which is a flow chart of the data analysis-based power spot transaction data model construction method of the present embodiment, provides a data analysis-based power spot transaction data model construction method, comprising:

[0063] The node marginal price change rate in the clearing process of the high fluctuation limited power spot market, the climbing gradient of the main output generator, the fractal dimension of the real-time order book and the cloud moving speed in the satellite remote sensing image are collected in real time to obtain the collection results, and the collection results in the preset training period are preprocessed to obtain an initial training set;

[0064] An incremental model is constructed according to the initial training set;

[0065] Based on the incremental model, steady state determination is performed according to the node marginal price change rate, the climbing gradient and a preset steady state threshold to obtain a first determination result or a second determination result;

[0066] Based on the first determination result, the type of abnormal conduction path is determined according to the cloud moving speed and the fractal dimension to obtain a supply-demand disturbance type or a price manipulation type;

[0067] generate a structural deviation vector according to the node marginal price change rate and the ramping gradient based on the supply-demand disturbance type, and generate a behavioral deviation vector according to the fractal dimension and the cloud layer movement speed based on the price manipulation type;

[0068] adjust the preset steady-state threshold and the preset training period according to the structural deviation vector and the behavioral deviation vector;

[0069] train the to-be-incremental model according to the collection result in the next preset training period based on the second determination result, to obtain a target model.

[0070] In this embodiment, the high-volatility-constrained power spot market refers to a power spot trading market in which the node marginal price has large fluctuations in a short time, but is limited in a certain range due to operating constraints such as unit ramping capacity, cross-region transmission channel capacity, frequency regulation reserve, and safety margin. This market is commonly found in regions with a high proportion of new energy and rapid load changes. The clearing process refers to the entire process of generating a price-quantity allocation scheme by a market operator, which includes calculating the marginal price (LMP) and node clearing quantity of each trading node based on the bid information submitted by the generation side and the consumption side, combining power flow calculation, operating constraints, transmission capacity limitations, and dispatching strategies, and using a clearing algorithm. In the above clearing process, in order to simultaneously depict the comprehensive influence of the internal supply-demand state and external meteorological disturbance of the market, four types of parameters are collected in real time: the node marginal price change rate (the percentage change of LMP between two adjacent clearing periods, obtained through the market operator price data interface and time series difference calculation), the ramping gradient of the main power generator (unit time active power change rate, obtained and calculated through the dispatching center SCADA or EMS system), the fractal dimension of the real-time order book (describing the complexity of the order distribution, calculated through the order book snapshot data of the trading platform and using the multi-scale box counting method), and the cloud layer movement speed introduced according to the significant influence of new energy output on cloud coverage (obtained by receiving continuous cloud image data from geostationary or low-orbit meteorological satellites and using the optical flow method or feature point matching algorithm to extract cloud displacement vector).

[0071] In this embodiment, the to-be-incremental model constructed based on the initial training set is trained by the node marginal price change rate, the ramping gradient, the real-time order book fractal dimension, and the cloud layer movement speed. The model can capture the operating rules and potential abnormal patterns of the power spot market under different weather and load conditions.

[0072] The preset training period is a time window for the model to count and analyze data before incremental training. It depends on the frequency of power spot market price fluctuations, unit response characteristics, and external disturbance change period, and is usually set between 1 hour and 24 hours. In this embodiment, it is set to 6 hours, which can capture market short-period fluctuation characteristics and reduce invalid data interference while considering data representativeness and calculation timeliness.

[0073] The preset steady-state threshold is a steady-state index critical value for determining whether the market operating state is stable. It depends on the statistical distribution characteristics of historical operation data, the setting of different parameter weights, and the tolerance of abnormal fluctuations, and is usually set between 0.3 and 0.7. In this embodiment, it is set to 0.5, which can effectively distinguish whether the market is in a stable adjustment stage or a disturbance fluctuation stage, and provide a clear standard for model training data screening.

[0074] By introducing node marginal price change rate, main output unit climbing gradient, real-time order book fractal dimension, and satellite remote sensing cloud movement speed in the clearing process of high fluctuation limited power spot market, the price fluctuation characteristics, supply and demand dynamic characteristics, market microstructure characteristics, and external meteorological disturbance characteristics can be reflected in a unified model. In the steady-state determination, abnormal conduction path identification, and bias vector generation process, a progressive mapping relationship is established between the parameters from price change to supply and demand response, and then to market behavior and external disturbance, so as to realize accurate description of market state and adaptive adjustment of dynamic threshold and training period. This ensures that the model still has high generalization ability and prediction accuracy when facing different disturbance types and change scenarios, and can continuously optimize the decision effect in subsequent incremental training, effectively solving the problems of poor adaptability, insufficient prediction accuracy, risk warning delay, and poor market regulation effect caused by lack of timely identification and response to market supply and demand fluctuations and abnormal behavior.

[0075] Please refer to Figure 2As shown, it is the determination logic diagram for steady state determination of the embodiment. In the embodiment, the process of determining the steady state according to the node marginal electricity price change rate, the ramping gradient and the preset steady state threshold to obtain the first determination result and the second determination result includes: calculating the standard deviation of the marginal electricity price change rate from the initial time to each time in the preset steady state determination period to determine a number of price fluctuation values; calculating the first order difference of the ramping gradient of each arbitrary adjacent two time in the preset steady state determination period to determine a number of supply and demand response accelerations; determining the steady state index of the end time of the preset steady state determination period according to all the price fluctuation values and all the supply and demand response accelerations; determining that the incremental training data is unstable when the steady state index is greater than the preset steady state threshold to obtain the first determination result; determining that the incremental training data is stable when the steady state index is less than or equal to the preset steady state threshold to obtain the second determination result.

[0076] The preset steady state determination period is a continuous time window selected when calculating the steady state index, which depends on the dynamic change rate of electricity price fluctuation and unit response and the market clearing frequency, and is usually set between 5 minutes and 60 minutes. In the embodiment, it is set to 30 minutes, which can balance the short time sensitivity and fluctuation smoothness while accurately reflecting the market transient characteristics.

[0077] By quantifying the standard deviation of the node marginal electricity price change rate and the first order difference of the unit ramping gradient into price fluctuation values and supply and demand response accelerations respectively, and determining the steady state index, the market operation state is accurately described. The price fluctuation reflects the sensitivity of the market to the supply and demand relationship, transaction activity and external disturbance, and the supply and demand response acceleration reflects the dynamic matching level of the unit regulation capacity and load change. The combination of the two can comprehensively reveal the dynamic coupling relationship between the market fluctuation amplitude and the regulation rate. By comparing the steady state index with the preset threshold, it can be distinguished whether the data is in stable operation or in severe fluctuation state, providing a reliable basis for subsequent training set screening, abnormal path determination and parameter dynamic adjustment of model incremental training, thereby improving the adaptability and accuracy of the model in dealing with different market scenarios.

[0078] Specifically, the process of determining the steady state index of the end time of the preset steady state determination period according to all the price fluctuation values and all the supply and demand response accelerations includes: normalizing the price fluctuation values to obtain price normalized values, and normalizing the supply and demand response accelerations to obtain response normalized values; calculating the steady state index of the end time of the preset steady state determination period according to the price normalized values, the response normalized values, the preset price weight and the preset response weight, Q = a × A + b × B, wherein Q is the steady state index, a is the preset price weight, A is the price normalized value, b is the preset response weight, and B is the response normalized value.

[0079] The preset price weight is a proportion of influence of a price normalized value in the calculation of the steady state index, and is determined by the importance of price fluctuation in market stability evaluation and historical fluctuation rules, and is usually set between 0.4 and 0.7, and is set to 0.6 in the embodiment, so as to highlight the contribution of price change to the steady state determination.

[0080] The preset response weight is a proportion of influence of a supply-demand response normalized value in the calculation of the steady state index, and is determined by the importance of unit climbing characteristics and supply-demand adjustment speed, and is usually set between 0.3 and 0.6, and is set to 0.4 in the embodiment, so as to take into account the auxiliary judgment effect of supply-demand response on market stability.

[0081] By normalizing the price fluctuation value and the supply-demand response acceleration, market fluctuation characteristics of different dimensions and value ranges can be quantitatively compared in the same evaluation system, and the steady state index that can comprehensively reflect the market price change amplitude and the unit adjustment capacity is obtained by combining the preset price weight and the preset response weight, so as to quantitatively evaluate the market operation state at the end of the preset steady state determination period, which helps to accurately identify the superposition effect of short-term market fluctuation and supply-demand imbalance, and improves the sensitivity and accuracy of the steady state determination.

[0082] Specifically, based on the first determination result, the process of determining the abnormal conduction path type according to the cloud layer movement speed and the fractal dimension includes:

[0083] The optical flow algorithm is used to calculate the displacement vector of the cloud pixel block in the two continuous satellite remote sensing images, so as to obtain the wind-cloud movement vector field of each pixel point, wherein the modulus value of the vector represents the cloud movement speed, and the direction represents the cloud movement direction. The direction angle of all effective vectors in the vector field is converted into a unit direction vector, and the modulus length of the vector sum is normalized to obtain a direction consistency coefficient. The Fourier transform is performed on the fractal dimension sequence to extract the main periodic component, and a periodicity index is calculated. The weather market influence index is determined according to the direction consistency coefficient and the periodicity index. The abnormal conduction path type is determined to be the supply-demand disturbance type or the price manipulation type according to the weather market influence index.

[0084] By using the cloud movement speed to reflect the dynamic influence of weather system on regional meteorological conditions, and by using the optical flow algorithm to extract the wind cloud movement vector field and calculate the direction consistency coefficient to measure the stability and concentration of meteorological disturbance in space, and by using the fractal dimension to analyze the complexity of market real-time order book structure and by using the Fourier transform to extract the main periodic component to calculate the periodicity index to describe the regularity of market transaction behavior in time dimension, the weather market influence index is formed by combining the meteorological disturbance characteristics and the market structure characteristics, so that the conduction effect of meteorological change on power supply and demand fluctuation and price anomaly can be quantified and distinguished, thereby effectively identifying the two types of abnormal paths of supply and demand disturbance type and price manipulation type, and providing accurate basis for risk warning and transaction strategy optimization of power spot market.

[0085] Specifically, the process of determining the weather market influence index according to the direction consistency coefficient and the periodicity index includes:

[0086] According to the direction consistency coefficient of the end time of the preset type determination period, the direction normalization coefficient is obtained by normalizing the direction consistency coefficient of the preset type determination period, and according to the periodicity index of the end time of the preset type determination period, the periodic normalization index is obtained by normalizing the periodicity index of the preset type determination period, and according to the preset periodicity weight, the preset consistency weight, the direction normalization coefficient and the periodic normalization index, the weather market influence index is determined, U = i x I + j x J, wherein U is the weather market influence index, i is the preset consistency weight, I is the direction normalization coefficient, j is the preset periodicity weight, and J is the periodic normalization index.

[0087] The preset type determination period is a time window used for statistical and normalization of direction consistency coefficient and periodicity index in abnormal path type judgment, which depends on the response delay characteristics of target power spot market price and supply and demand fluctuation, and is usually set between 5 minutes and 60 minutes, and is set to 30 minutes in this embodiment, which can balance the propagation timeliness of meteorological disturbance and the stability of market structure feature change.

[0088] The preset consistency weight is the weight coefficient of the direction consistency coefficient in the fusion calculation of the weather market influence index, which depends on the sensitivity of meteorological disturbance to price fluctuation, and is usually set between 0.3 and 0.7, and is set to 0.5 in this embodiment, which can balance the contribution of meteorological factors and market order structure factors in index calculation.

[0089] By normalizing the direction consistency coefficient and the periodicity index within a preset type determination period, the comparability of parameters with different dimensions and different numerical scales is achieved, and the meteorological disturbance characteristics and the market order structure characteristics are ensured to play a role in a unified evaluation framework; the direction normalization coefficient reflects the stability of the spatial distribution of meteorological disturbances, and the periodicity normalization index describes the time periodicity variation characteristics of the market order structure, and under the action of the weight parameter, the weather market influence index is formed, so that the conduction effect of meteorological changes on the market price and the supply and demand relationship can be quantified, the multi-source heterogeneous parameters can be synergistically fused without distorting the original information, and the accuracy and sensitivity of the abnormal conduction path identification are improved, thereby providing more targeted decision basis for the early warning and strategy adjustment of the power spot market.

[0090] Please refer to Figure 3 As shown in FIG. 6, which is a determination logic diagram for determining the type of abnormal conduction path in the embodiment, the process of determining the type of abnormal conduction path as the supply and demand disturbance type or the price manipulation type according to the weather market influence index in the embodiment includes: determining the type of abnormal conduction path as the supply and demand disturbance type when the weather market influence index is greater than or equal to a preset influence index threshold value; and determining the type of abnormal conduction path as the price manipulation type when the weather market influence index is less than the preset influence index threshold value.

[0091] The preset influence index threshold value is a limit value for distinguishing the supply and demand disturbance type from the price manipulation type, and depends on the distribution characteristics of the weather market influence index in the historical data under the two types of abnormal situations. The preset influence index threshold value is usually set to be between 0.4 and 0.6, and is set to be 0.5 in the embodiment, so that the type of abnormal path can be accurately determined.

[0092] By comparing the weather market influence index with the preset influence index threshold value, the quantitative demarcation of the meteorological disturbance characteristics and the market structure characteristics is achieved: when the index value is high, it indicates that the change of meteorological conditions has a significant impact on the supply and demand side, thereby triggering the determination of the supply and demand disturbance type; when the index value is low, it indicates that the market price fluctuation is more caused by the abnormality of the transaction structure or the behavior mode, and points to the price manipulation type. The determination mechanism combines the meteorological dynamics, the market order characteristics and the price change trend, so that different abnormal causes can be quickly distinguished.

[0093] Specifically, based on the supply-demand disturbance type, the process of generating a structural deviation vector according to the node marginal price change rate and the ramping gradient includes: determining a price fluctuation amplitude according to the node marginal price change rate within a preset time window, and determining a gradient acceleration according to the ramping gradient within the preset time window; time-synchronizing the price fluctuation amplitude and the ramping gradient acceleration according to the preset time window to form a two-dimensional deviation feature sequence; calculating the mean deviation and the extreme deviation within each preset time window according to the two-dimensional deviation feature sequence, and combining the mean deviation and the extreme deviation into a plurality of structural deviation components; connecting all the structural deviation components in time sequence to obtain the structural deviation vector.

[0094] The preset time window is a time interval for dividing data collection and analysis, and is usually set between 5 minutes and 6 hours according to the fluctuation frequency of the electricity spot market and the monitoring accuracy requirement. In this embodiment, the preset time window is set to 15 minutes, which can effectively capture the short-term change characteristics of the price and the ramping gradient, and ensure the timeliness and accuracy of the structural deviation vector.

[0095] By combining the node marginal price change rate with the dynamic information of the generator set ramping gradient, the comprehensive characteristics reflecting the structural change of supply and demand are extracted. First, the price fluctuation amplitude is calculated based on the price change rate to describe the fluctuation intensity of the market price. Then, the gradient acceleration is calculated based on the ramping gradient to reflect the speed and amplitude of the change of power output. Subsequently, the two types of parameters are time-synchronized within a unified time window to ensure the comparability of the changes of price and supply-demand response. Based on the synchronized two-dimensional feature sequence, the mean deviation and the extreme deviation of each window are calculated to capture the dual information of stable deviation and sudden deviation. Finally, these deviation components are connected in time sequence to form a structural deviation vector, realizing continuous quantitative description of the dynamic deviation pattern of market supply-demand structure, and providing accurate basis for identifying structural abnormalities.

[0096] Specifically, based on the price manipulation type, the process of generating a behavior deviation vector according to the fractal dimension and the cloud layer moving speed includes: determining a fractal dimension change rate according to the fractal dimension within a preset time window, and calculating a direction change rate of the cloud layer moving speed within the preset time window; synchronously pairing the fractal dimension change rate and the cloud layer moving speed direction change rate according to the preset time window to obtain a two-dimensional behavior feature sequence; determining a plurality of behavior deviation components according to the relative change deviation and the correlation coefficient deviation within each preset time window according to the two-dimensional behavior feature sequence; connecting all the behavior deviation components in time sequence to obtain the behavior deviation vector.

[0097] The behavior characteristic sequence is constructed based on the fractal dimension change rate and the cloud layer moving speed direction change rate, which effectively reveals the interaction between the complex dynamics of the market order structure and the external environment changes. By calculating the relative change deviation and the correlation coefficient deviation, the time sequence characteristics of abnormal behavior can be accurately described, and then the behavior deviation vector with time sequence continuity is formed. This method fully reflects the internal relationship between market internal structure fluctuations and environmental disturbances, improves the identification ability of price manipulation behavior and the response sensitivity of the model, thereby providing scientific and dynamic support for risk monitoring and regulation of the electricity spot market.

[0098] Referring to Figure 4 The process of adjusting the preset steady state threshold and the preset training period according to the structural deviation vector and the behavior deviation vector in the embodiment includes: calculating the Pearson correlation coefficient of all the structural deviation vectors and behavior deviation vectors from the time when the first determination result is obtained to the next steady state determination, to obtain a deviation correlation degree; when the absolute value of the deviation correlation degree is greater than a preset deviation correlation threshold, adjusting the preset steady state threshold and the preset training period according to the deviation correlation degree and the preset deviation correlation threshold.

[0099] The preset deviation correlation threshold is a threshold for judging the correlation degree between the structural deviation vector and the behavior deviation vector, which depends on the statistical characteristics of historical market data and the demand for abnormal sensitivity of the model, and is usually set between 0.6 and 0.9. In the embodiment, it is set to 0.75, which can effectively distinguish significant deviation correlation from normal fluctuations, thereby supporting dynamic adjustment of the steady state threshold and the training period, and improving the flexibility and accuracy of the model.

[0100] By calculating the correlation degree between the structural deviation vector and the behavior deviation vector, the internal relationship between market supply and demand dynamics and abnormal behavior can be effectively captured. When the deviation correlation degree exceeds the preset threshold, the threshold and the training period of the steady state determination can be dynamically adjusted, the sensitivity and adaptability of the model to market fluctuations can be improved, and the real-time state and potential risks of the electricity spot market can be more accurately reflected. This dynamic adjustment mechanism reasonably utilizes the synergistic change relationship between multi-dimensional parameters, realizes fine description and effective response to complex market environment, and enhances the accuracy and reliability of risk monitoring and early warning.

[0101] Specifically, the process of adjusting the preset steady-state threshold and the preset training period according to the deviation correlation degree and the preset deviation correlation threshold comprises: increasing the preset steady-state threshold according to the absolute value of the deviation correlation degree and the relative deviation of the preset deviation correlation threshold, a preset first adjustment coefficient, H' = H x [1 + k1 x (R - R0) / R0], wherein H' is the increased preset steady-state threshold, H is the preset steady-state threshold before increasing, k1 is the preset first adjustment coefficient, R is the absolute value of the deviation correlation degree, and R0 is the preset deviation correlation threshold; and decreasing the preset training period according to the absolute value of the deviation correlation degree and the relative deviation of the preset deviation correlation threshold, a preset second adjustment coefficient, and a threshold period lower limit, T' = T x max [1 - k2 x (R - R0) / R0], P], wherein T' is the decreased preset training period, T is the preset training period before decreasing, k2 is the preset second adjustment coefficient, and P is the threshold period lower limit.

[0102] The preset deviation correlation threshold is a reference value for determining whether the correlation degree of the structural deviation vector and the behavioral deviation vector reaches a significant level, which depends on the correlation distribution characteristics of the two types of deviation vectors in historical market data and the sensitivity requirements of risk identification, and is usually set to be between 0.3 and 0.7, and is set to 0.5 in the embodiment. In the embodiment, the market disturbance situation with a higher deviation correlation degree can be accurately identified while considering the sensitivity and false alarm rate of anomaly detection.

[0103] The preset first adjustment coefficient is a proportional factor for dynamically increasing the steady-state threshold according to the relative deviation amplitude of the deviation correlation degree, which depends on the sensitivity requirements and fault tolerance range of steady-state determination under different market volatility levels, and is usually set to be between 0.1 and 0.5, and is set to 0.2 in the embodiment. The steady-state threshold can be moderately increased when the correlation degree is significantly increased, so as to avoid excessive triggering of model training caused by short-term abnormal correlation.

[0104] The preset second adjustment coefficient is a proportional factor for dynamically shortening the training period according to the relative deviation amplitude of the deviation correlation degree, which depends on the balance requirement between the model update speed and the stability of the training data, and is usually set to be between 0.1 and 0.4, and is set to 0.15 in the embodiment. The model update rhythm can be accelerated in the high correlation degree situation to introduce the latest market features in time to improve the risk response speed, and the cycle lower limit is combined to prevent excessive frequent adjustment.

[0105] By comparing the deviation correlation degree with a preset deviation correlation threshold and respectively adjusting the steady state threshold and the training period based on the relative deviation, the model can automatically raise the steady state determination threshold and shorten the model updating period when the structural deviation and the behavioral deviation are highly correlated, thereby enhancing the sensitivity and response speed to sudden market disturbances; when the correlation degree is low, a longer training period and a lower steady state threshold are maintained to maintain the stability and generalization ability of the model. The mechanism uses the dynamic coupling relationship between parameters to make the threshold adjustment amplitude change in proportion to the strength of the correlation, while introducing a lower limit to the period to prevent excessive adjustment, balancing real-time performance and robustness, and effectively improving the accuracy and reliability of risk identification and regulation in the power spot market.

[0106] The above only describes the preferred embodiments of the present application and is not intended to limit the present application; for those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A data analysis-based power spot transaction data model construction method, characterized by, The application comprises the following steps: Collecting the node marginal price change rate in the clearing process of the high volatility limited electricity spot market, the climbing gradient of the main output generator unit, the fractal dimension of the real-time order book, and the cloud movement speed in the satellite remote sensing image to obtain the collection results, and preprocessing the collection results in the preset training period to obtain an initial training set; Building an incremental model according to the initial training set; Based on the incremental model, performing steady state judgment according to the node marginal price change rate, the climbing gradient and a preset steady state threshold to obtain a first judgment result or a second judgment result; Based on the first judgment result, determining the abnormal conduction path type according to the cloud movement speed and the fractal dimension to obtain a supply and demand disturbance type or a price manipulation type; Based on the supply and demand disturbance type, generating a structural bias vector according to the node marginal price change rate and the climbing gradient, and based on the price manipulation type, generating a behavior bias vector according to the fractal dimension and the cloud movement speed; Adjusting the preset steady state threshold and the preset training period according to the structural bias vector and the behavior bias vector; Based on the second judgment result, training the incremental model according to the collection results in the next preset training period to obtain a target model; The process of performing steady state judgment according to the node marginal price change rate, the climbing gradient and a preset steady state threshold to obtain a first judgment result or a second judgment result comprises the following steps: calculating the standard deviation of the marginal price change rate from the initial time to each time in the preset steady state judgment period to determine a plurality of price fluctuation values; calculating the first-order difference of the climbing gradient of any two adjacent times in the preset steady state judgment period to determine a plurality of supply and demand response accelerations; determining a steady state index at the end time of the preset steady state judgment period according to all the price fluctuation values and all the supply and demand response accelerations; when the steady state index is greater than the preset steady state threshold, determining that the incremental training data is unstable to obtain the first judgment result; when the steady state index is less than or equal to the preset steady state threshold, determining that the incremental training data is stable to obtain the second judgment result; The process of determining the abnormal conduction path type according to the cloud movement speed and the fractal dimension to obtain a supply and demand disturbance type or a price manipulation type based on the first judgment result comprises the following steps: Using an optical flow algorithm to calculate the displacement vector of the cloud pixel block in two consecutive satellite remote sensing images to obtain the wind cloud movement vector field of each pixel point, wherein the modulus of the vector represents the cloud movement speed and the direction represents the cloud movement direction, converting the direction angle of all effective vectors in the vector field into unit direction vectors and performing modulus length normalization processing on the vector sum to obtain a direction consistency coefficient; performing Fourier transform on the fractal dimension sequence to extract the main periodic component and calculating a periodicity index; determining a weather market influence index according to the direction consistency coefficient and the periodicity index; determining the abnormal conduction path type to be the supply and demand disturbance type or the price manipulation type according to the weather market influence index; The process of generating a structural deviation vector according to the node marginal price change rate and the ramping gradient based on the supply-demand disturbance type includes: determining a price fluctuation amplitude according to the node marginal price change rate within a preset time window, and determining a gradient acceleration according to the ramping gradient within the preset time window; time-synchronizing the price fluctuation amplitude and the ramping gradient acceleration according to the preset time window to form a two-dimensional deviation feature sequence; calculating the mean deviation and the extreme deviation within each preset time window according to the two-dimensional deviation feature sequence, and combining the mean deviation and the extreme deviation into a plurality of structural deviation components; connecting all the structural deviation components in chronological order to obtain the structural deviation vector; The process of generating a behavior deviation vector according to the fractal dimension and the cloud layer moving speed based on the price manipulation type includes: determining a fractal dimension change rate according to the fractal dimension within a preset time window, and calculating a direction change rate of the cloud layer moving speed within the preset time window; synchronously pairing the fractal dimension change rate and the cloud layer moving speed direction change rate according to the preset time window to obtain a two-dimensional behavior feature sequence; determining a plurality of behavior deviation components according to the relative change deviation and the correlation coefficient deviation within each preset time window according to the two-dimensional behavior feature sequence; connecting all the behavior deviation components in chronological order to obtain the behavior deviation vector.

2. The data analysis based power spot trading data model building method according to claim 1, characterized in that, The process of determining a steady-state index at the end of a preset steady-state determination period according to all the price fluctuation values and all the supply-demand response accelerations includes: The process of calculating the steady-state index according to all the price fluctuation values, a preset price weight, all the supply-demand response accelerations, and a preset response weight includes:

3. The data analysis based power spot trading data model building method according to claim 2, characterized in that, The process of determining a weather market influence index according to the direction consistency coefficient and the periodicity index includes: The process of determining the weather market influence index according to the direction consistency coefficient and the periodicity index within a preset type, a preset consistency weight, and a preset periodicity weight includes:

4. The data analysis based power spot trading data model building method according to claim 3, characterized in that, The process of determining that the abnormal conduction path type is the supply-demand disturbance type or the price manipulation type according to the weather market influence index includes: The process of determining that the abnormal conduction path type is the supply-demand disturbance type or the price manipulation type according to a comparison result of the weather market influence index and a preset influence index threshold value includes:

5. The data analysis based power spot trading data model building method according to claim 4, characterized in that, The process of adjusting the preset steady-state threshold and the preset training period according to the structural deviation vector and the behavior deviation vector includes: determining a deviation correlation degree according to all the structural deviation vectors and all the behavior deviation vectors from the time when the first determination result is obtained to the next steady-state determination; when the absolute value of the deviation correlation degree is greater than a preset deviation correlation threshold value, adjusting the preset steady-state threshold and the preset training period according to the deviation correlation degree and the preset deviation correlation threshold value.

6. The data analysis based power spot trading data model building method according to claim 5, characterized in that, The process of adjusting the preset steady-state threshold and the preset training period according to the deviation correlation degree and the preset deviation correlation threshold value includes: increasing the preset steady-state threshold by a preset first adjustment coefficient according to the relative deviation of the absolute value of the deviation correlation degree and the preset deviation correlation threshold value; According to the absolute value of the deviation correlation degree and the relative deviation of the preset deviation correlation threshold, the preset second adjustment coefficient and the threshold period lower limit, the preset training period is reduced.

Images