Market subject quotation preference quantification extraction method based on public disclosure data

Abstract

The application relates to the technical field of power spot quantitative transaction, and discloses a market subject quotation preference quantitative extraction method based on open disclosure data, which comprises the following steps: obtaining text data packets, meteorological telemetry signals and power consumption load pulse sequences from a transaction server network card interface, a meteorological sensing collection node and a power grid dispatching gateway, extracting initial quotation preference feature label strings, calculating a new energy physical output estimation value, and differentiating the whole power consumption physical demand quantity to obtain a marginal unit physical bidding space quantity, generating a real-time preference correction coefficient and correcting the same to obtain dynamic quotation quantitative parameters, and then generating an automatic order control instruction; the application fuses open disclosure text features and real-time meteorological power physical boundaries, overcomes the problem that a static preference deviates from the actual situation, naturally avoids misleading disclosure content, and makes the quotation strategy always follow objective physical laws.

Description

Technical Field

[0001] This invention relates to the field of quantitative trading technology for electricity spot transactions, and in particular to a method for quantitatively extracting market participants' pricing preferences based on publicly disclosed data. Background Technology

[0002] In the field of electricity spot market analysis, extracting market participants' pricing preferences from massive amounts of publicly disclosed documents using natural language processing (NLP) technology is a common method to assist in quantitative trading decisions. This technology uses semantic analysis models to parse unstructured texts such as corporate annual reports and business announcements, and combines this with clustering algorithms to map text features into fixed quantitative parameters reflecting pricing tendencies. For example, statements about declining coal costs are transformed into labels indicating a preference for low-price competition. This type of method directly solidifies discrete textual semantics into static pricing features for subsequent short-term electricity price forecasting models.

[0003] However, in actual transaction execution, the pricing preference parameters extracted from historical texts often misalign with the real-time operating conditions of thermal power units within a specific time period. This causes numerical conflicts between the quantitative pricing parameters output by the system and the actual boundary conditions of the electricity spot market during that period. For example, the system might parse information about lower coal procurement costs from a recent report disclosed by a power plant and assign it a low-price pricing preference feature. However, during a trading session on a particular day, a sudden severe convective weather event causes a sharp drop in renewable energy output, forcing the thermal power unit to operate at full capacity in an emergency peak state, resulting in a significant increase in actual marginal costs. The system fails to detect this real-time change and mechanically substitutes the static low-price preference parameters into the bidding model, outputting a clearing prediction value far lower than the actual bid price. Traders see this erroneous low-price prediction on their terminal interface, formulate and execute trading strategies accordingly, leading to a significant deviation of the actual transaction price from expectations and resulting in losses. Therefore, reducing the misalignment between unstructured features and real-time operating conditions in the process of using natural language processing technology to assist in electricity spot market forecasting is a problem that urgently needs to be solved in this field. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a method for quantitatively extracting the pricing preferences of market participants based on publicly disclosed data. By comprehensively processing publicly available text data, meteorological telemetry data, and power grid load data, the method dynamically quantifies and extracts the pricing preferences of market participants, thereby improving the accuracy of price prediction and the efficiency of automated transactions.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a method for quantitatively extracting market participants' pricing preferences based on publicly disclosed data, comprising:

[0007] The transaction server network card interface receives text capture network data packets, simultaneously acquires meteorological sensing telemetry signals through meteorological sensor acquisition nodes, and reads the entire network power load pulse sequence from the power grid dispatch gateway.

[0008] Natural language segmentation and clustering operations are performed on the text-grabbing network data packets to extract objective state descriptive words related to cost changes and fuel inventory and convert them into initial price preference feature tag strings;

[0009] Based on the meteorological sensing and telemetry signals, the estimated value of the physical output of new energy is calculated. The overall physical demand of electricity is obtained by integrating and accumulating the pulse sequence of the total grid electricity load. The estimated value of the physical output of new energy is then subtracted from the overall physical demand of electricity to generate the physical bidding space of marginal units.

[0010] Based on the physical bidding space of the marginal unit, a real-time preference correction coefficient is generated, and the initial bid preference feature tag string is adjusted to obtain dynamic bid quantification parameters, which are finally converted into automatic order placement control instructions.

[0011] Preferably, receiving text capture network data packets through the transaction server's network card interface includes:

[0012] Within a specified time window, the network interface card of the control transaction server establishes a Transmission Control Protocol (TCP) connection with the Internet Protocol address of the target competitor's web server, requests the target competitor to disclose the web page in accordance with Hypertext Transfer Protocol or Hypertext Transfer Security Protocol, and captures the response message byte stream returned by the target competitor's web page server frame by frame at the data link layer.

[0013] If the captured response message byte stream is an encrypted transmission message, it is decrypted in real time to obtain a plaintext byte stream;

[0014] All response message byte streams captured within a specified time window are concatenated in order of arrival time and encapsulated into a text-based network data packet capture.

[0015] Preferably, the synchronous acquisition of meteorological telemetry signals through meteorological sensor acquisition nodes includes:

[0016] Each meteorological sensor acquisition node uses a unified sampling clock signal as a reference to acquire light intensity and wind speed pulse values ​​and record the sampling timestamp in each sampling sub-cycle of the current clearing period.

[0017] The arithmetic mean of the light intensity values ​​of each sampling sub-cycle within the current clearing period is used to obtain the representative light intensity value of the geographical region. The arithmetic mean of the wind speed pulse values ​​of each sampling sub-cycle within the current clearing period is used to obtain the representative wind speed pulse value of the geographical region.

[0018] The light intensity value and wind speed pulse value of the geographic region, which are represented by the sampling timestamp, are encapsulated into a meteorological sensing telemetry signal.

[0019] Preferably, reading the entire network power load pulse sequence from the power grid dispatch gateway includes:

[0020] According to the unified sampling clock signal, in each sampling sub-cycle of the current clearing period, the newly added pulse count value in the current sampling sub-cycle is read from the hardware data interface of the power grid dispatch gateway, and the reading timestamp is recorded.

[0021] The pulse count values ​​read from each sampling sub-cycle during the current clearing period are arranged in order of reading timestamp to generate a pulse sequence of the entire network's power load.

[0022] Preferably, the step of performing natural language segmentation and clustering operations on the text-grabbed network data packets to extract objective state descriptive words related to cost changes and fuel inventory and converting them into initial price preference feature tag strings includes:

[0023] The text-captured network data packets are parsed to remove the message control header and retain valid character entities;

[0024] A pre-defined text clustering algorithm is used to perform feature word frequency statistics on the effective character entities, and objective state description words related to cost changes and fuel inventory are extracted.

[0025] The objective state descriptors are converted into corresponding numerical labels according to a preset quantization mapping rule, and then sequentially concatenated to generate an initial price preference feature tag string.

[0026] Preferably, the step of using a pre-set text clustering algorithm to perform feature word frequency statistics on the effective character entities and extracting objective state descriptive words related to cost changes and fuel inventory includes:

[0027] Perform Chinese word segmentation on the effective character entities, count the feature word frequency of each segment, and generate a feature word frequency vector;

[0028] A pre-defined text clustering algorithm is used to cluster the feature word frequency vectors to generate a set of text clusters.

[0029] From the text cluster set, match the cost fuel topic cluster and extract objective state descriptive words related to cost changes and fuel inventory.

[0030] Preferably, the step of calculating the estimated physical output value of new energy sources based on the meteorological sensing telemetry signal includes:

[0031] The solar irradiance values ​​of each geographical region in the meteorological remote sensing signal are substituted into the photovoltaic power generation conversion formula to calculate the photovoltaic physical output component, and the wind speed pulse values ​​of each geographical region in the meteorological remote sensing signal are substituted into the wind power generation conversion formula to calculate the wind power physical output component.

[0032] The summation of the photovoltaic physical output component and the wind power physical output component yields the estimated value of the new energy physical output.

[0033] Preferably, the amount of physical bidding space for generating marginal units includes:

[0034] The overall physical demand for electricity in the corresponding time period is obtained by integrating and accumulating the pulse sequence of the total power load of the entire network.

[0035] The remaining supply-demand gap is obtained by subtracting the estimated physical output of new energy sources from the total physical demand for electricity. When the remaining supply-demand gap is greater than zero, it is used as the marginal unit physical bidding space to be undertaken by thermal power. When the remaining supply-demand gap is not greater than zero, the marginal unit physical bidding space is determined to be zero.

[0036] Preferably, the generation of real-time preference correction coefficients based on the physical bidding space of the marginal units includes:

[0037] Extract the physical bidding space of the marginal units, divide the physical bidding space of the marginal units by the physical upper limit of the thermal power installed capacity stored in the form of a preset fixed constant parameter, and calculate the supply and demand tension approach rate.

[0038] The supply-demand tension approach rate is nonlinearly mapped to generate a real-time preference correction coefficient. When the supply-demand tension approach rate is greater than a set threshold, the real-time preference correction coefficient is output in an exponentially amplified manner.

[0039] Preferably, the step of generating real-time preference correction coefficients by performing nonlinear mapping calculations on the supply-demand tension approximation rate includes:

[0040] In the sub-threshold range where the supply-demand tension approach rate is no greater than the set threshold, the real-time preference correction coefficient is made to increase linearly with the supply-demand tension approach rate in a linear form.

[0041] In the threshold range where the supply-demand tension approach rate is greater than a set threshold, an exponential term is superimposed on the linear value at the set threshold. The independent variable of the exponential term is the portion of the supply-demand tension approach rate that exceeds the set threshold, so that the real-time preference correction coefficient in the threshold range increases monotonically on the basis of the sub-threshold range.

[0042] Preferably, the step of adjusting the initial quote preference feature marker string to obtain dynamic quote quantification parameters, and finally converting them into automatic order placement control instructions, includes:

[0043] The initial pricing preference feature string is multiplied by the real-time preference correction coefficient for numerical correction, generating dynamic pricing quantification parameters that eliminate static text bias;

[0044] Based on the dynamic price quantification parameters, the buy and sell clearing price values ​​are directly calculated, encapsulated into an automatic order placement control instruction in a standard trading format, and sent to the trading gateway for execution.

[0045] Secondly, this invention provides a system for quantitatively extracting market participant pricing preferences based on publicly disclosed data, comprising:

[0046] The receiving and acquisition module is used to receive text capture network data packets through the network card interface of the transaction server, simultaneously acquire meteorological telemetry signals through the meteorological sensor acquisition node, and read the pulse sequence of the power load of the entire network from the power grid dispatch gateway.

[0047] The text segmentation and clustering module is used to perform natural language segmentation and clustering operations on the text-grabbing network data packets, extract objective state descriptive words related to cost changes and fuel inventory, and convert them into initial price preference feature tag strings;

[0048] The supply and demand difference calculation module is used to calculate the estimated value of new energy physical output based on the meteorological sensing telemetry signal, perform integral accumulation calculation on the pulse sequence of the whole network power load to obtain the overall physical demand of electricity, and subtract the estimated value of new energy physical output from the overall physical demand of electricity to generate the physical bidding space of marginal units.

[0049] The dynamic correction and order placement module is used to generate a real-time preference correction coefficient based on the physical bidding space of the marginal unit, adjust the initial bid preference feature tag string to obtain dynamic bid quantification parameters, and finally convert them into automatic order placement control instructions.

[0050] Compared to existing technologies, the advantages of this invention are as follows: This invention effectively solves the problem of conflict between price preference parameters and actual boundary conditions caused by the misalignment between unstructured features and real-time operating conditions by dynamically multiplying and correcting the initial price preference feature tag string extracted from static text with the marginal unit physical bidding space quantity calculated based on real-time physical quantity differential calculation. Raw data is directly collected from three types of hardware interfaces: the transaction server network card interface, meteorological sensor acquisition nodes, and the power grid dispatch gateway, and synchronized with a unified sampling clock signal. This ensures strict alignment of physical quantities on both the supply and demand sides on the time axis, eliminating the release delay error of forwarded data collection under sudden weather conditions, and guaranteeing that the marginal unit physical bidding space quantity corresponds synchronously with the physical processes of electricity consumption and new energy. A piecewise nonlinear mapping is used to generate real-time preference correction coefficients, which amplify exponentially when the supply-demand tension approach rate exceeds a set threshold, accurately depicting the physical law of the marginal cost of thermal power first leveling off and then sharply increasing. This forces the product correction to raise the static low-price preference to the high-price range, achieving automatic coverage of static text features by real-time physical constraints. This mechanism, without requiring a separate authenticity verification module, inherently possesses the ability to circumvent misleading disclosures from competitors, avoiding strategic losses caused by false low-price text. Ultimately, the corrected dynamic pricing quantitative parameters are converted into automatic order placement control instructions, ensuring that trading decisions are always based on rapidly changing objective physical laws, thus improving the accuracy and automation level of multi-source data fusion quantitative analysis in the electricity spot trading scenario. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart of the method for quantitatively extracting market participants' pricing preferences based on publicly disclosed data in this invention;

[0053] Figure 2 This is a schematic diagram of the deployment structure of the meteorological sensor acquisition node in an embodiment of the present invention;

[0054] Figure 3 This is a schematic diagram illustrating the alignment of the unified sampling clock signal in an embodiment of the present invention;

[0055] Figure 4 This is a schematic diagram of text capture network data packet deconstruction in an embodiment of the present invention;

[0056] Figure 5 This is a schematic diagram of text cluster feature space partitioning in an embodiment of the present invention;

[0057] Figure 6 This is a schematic diagram of the transformation of the preset quantization mapping rule structure in an embodiment of the present invention;

[0058] Figure 7 This is a schematic diagram illustrating the differential generation of physical bidding space quantity for marginal units in an embodiment of the present invention;

[0059] Figure 8 This is a schematic diagram of the product correction of dynamic quotation quantification parameters in an embodiment of the present invention;

[0060] Figure 9 This is a schematic diagram of the automatic order placement control command message structure in an embodiment of the present invention;

[0061] Figure 10 This is a functional module diagram of the market participant pricing preference quantitative extraction system based on publicly disclosed data in this invention. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] Example 1:

[0064] Please see Figure 1 As shown, this embodiment provides a method for quantitatively extracting market participants' pricing preferences based on publicly disclosed data, including:

[0065] This embodiment provides a method for quantitatively extracting market participant bidding preferences based on publicly disclosed data, applied to the scenario of quantitative trading and management in the electricity spot market. In the electricity spot market, total electricity consumption constitutes the demand side, and renewable energy generation constitutes the supply side. Both the demand and supply sides are highly influenced by meteorological conditions. The remaining portion after subtracting the total renewable energy supply from the total demand is borne by thermal power units, which play a marginal role in the power system. The bids of thermal power units determine the clearing price of the entire market. Since the actual generation costs of each thermal power plant are kept confidential, this embodiment obtains first-hand data through hardware interfaces such as the trading server network card interface, meteorological sensor acquisition nodes, power grid dispatch gateway, and trading gateway. It performs reverse analysis on publicly disclosed documents and dynamically corrects the text extraction results by combining underlying meteorological and power grid load physical quantities, ultimately generating automatic order placement control commands for execution. The method includes steps S1 to S4, which are described in detail below in conjunction with the hardware deployment and data processing flow.

[0066] S1: Receives text capture network data packets through the network card interface of the transaction server, simultaneously acquires meteorological telemetry signals through the meteorological sensor acquisition node, and reads the pulse sequence of the entire network's power load from the power grid dispatch gateway.

[0067] The clearing price in the electricity spot market is determined by the real-time bids of marginal thermal power units. The bid range for thermal power depends on the rapidly changing physical quantities of meteorological supply and electricity demand. Traditional financial data terminals acquire market and load data through multi-level server forwarding. The secondary data delay introduced by the forwarding links causes the collected data to deviate from the actual time of occurrence in the physical world. Under sudden weather conditions, this deviation will distort supply and demand calculations. To ensure that data acquisition and physical occurrence are strictly synchronized, step S1 directly acquires raw text, meteorological, and load data from three types of hardware interfaces: the trading server network card interface, the meteorological sensor acquisition node, and the power grid dispatch gateway. The meteorological and load data are synchronously collected based on a unified sampling clock signal based on satellite time synchronization to ensure that the physical quantities on the supply side and the physical quantities on the demand side are strictly aligned on the time axis.

[0068] Further, step S1 includes:

[0069] S11: Control the network card interface of the transaction server to capture the underlying protocol messages of the competitor's disclosed web pages within a specified time window, and obtain text capture network data packets.

[0070] Competitors' operational status information is disclosed in unstructured text format across web page files. If obtained through a third-party data service provider, the text content will lose its original timestamps and complete fields due to secondary editing. Directly capturing the underlying protocol messages of the disclosed web pages within a specified time window can preserve the original timing and complete payload of the messages, providing an undisturbed data source for subsequent text parsing.

[0071] Specifically, the network interface card (NIC) of the trading server is an Ethernet interface controller deployed on the physical host of the trading server. It operates at the data link layer of the Transmission Control Protocol (TCP) and Internet Protocol (IP) stack, and supports the capture of raw bytes of incoming and outgoing packets.

[0072] The method for determining the designated time window is as follows: taking the bidding and declaration cycle of the electricity spot market as the benchmark, the start time to the end time of a single bidding and declaration cycle is taken as the designated time window. For example, the designated time window can be set as the daily declaration period from 8:00 to 9:00.

[0073] Within a specified time window, the network interface card of the control transaction server establishes a Transmission Control Protocol (TCP) connection with the Internet Protocol address of the target competitor's web server, requests the target competitor to disclose the web page according to Hypertext Transfer Protocol (HTTP) or Hypertext Transfer Security Protocol (HTTP), and captures the response message byte stream returned by the target competitor's web page server frame by frame at the data link layer. If the captured response message byte stream is an encrypted transmission message, it is decrypted in real time through a front-end security gateway or decryption module to obtain a plaintext byte stream.

[0074] All response message byte streams captured within a specified time window (after decryption) are concatenated in order of arrival time and encapsulated into a text-captured network data packet. The data structure of the text-captured network data packet is a sequence of byte streams with capture timestamps, and each byte stream unit contains a message header field and a message payload field.

[0075] S12: Real-time collection of light intensity and wind speed pulse values ​​by meteorological sensor acquisition nodes deployed in different geographical areas, which are then converted into meteorological remote sensing signals.

[0076] The subsequent step S31 requires using the light intensity values ​​and wind speed pulse values ​​covering different geographical areas as input to calculate the physical output of renewable energy. However, renewable energy power plants are distributed across different geographical zones, and the meteorological conditions in each zone vary spatially. A single measuring point cannot provide meteorological input covering the entire spatial distribution. Step S12 requires deploying meteorological sensor acquisition nodes in different geographical areas to collect the light intensity values ​​and wind speed pulse values ​​of each geographical zone, providing spatially resolved meteorological input for estimating the physical output of renewable energy.

[0077] Specifically, the meteorological sensor acquisition nodes include illuminance sensors and wind speed pulse sensors, deployed respectively at concentrated geographical coordinate points of photovoltaic power plants and wind farms within the power grid coverage area. The number of meteorological sensor acquisition nodes deployed corresponds to the number of geographical zones of the new energy power plants. Illuminance sensors are deployed in concentrated photovoltaic power plant areas because the illuminance value needs to characterize the actual solar radiation power received by the photovoltaic module's illuminated surface; wind speed pulse sensors are deployed in concentrated wind farm areas because the wind speed pulse value needs to characterize the actual incoming wind speed at the hub height of the wind turbine. See also Figure 2 This is a schematic diagram of the deployment structure of the meteorological sensor acquisition node provided in an embodiment of this application. Figure 2As shown in the figure, the left and right geographical zones are marked with dashed bounding boxes. The left geographical zone contains a solar image and a photovoltaic panel image, labeled as a concentrated area of ​​photovoltaic power plants, with a light intensity sensor module connected below it. The right geographical zone contains wind streamlines and a windmill image, labeled as a concentrated area of ​​wind farms, with a wind speed pulse sensor module connected below it. The light intensity sensor module and the wind speed pulse sensor module converge at the bottom meteorological sensor acquisition node module via an arrow. The light intensity sensor is deployed in the concentrated area of ​​photovoltaic power plants to characterize the actual solar radiation power received by the photovoltaic module's light-receiving surface; the wind speed pulse sensor is deployed in the concentrated area of ​​wind farms to characterize the actual incoming wind speed at the hub height of the wind turbine. The meteorological sensor acquisition node is deployed according to the geographical zones and aggregates the light intensity values ​​and wind speed pulse values ​​of each zone, providing spatially resolved meteorological input for estimating the physical output of new energy sources, avoiding the deficiency of a single measuring point being unable to cover the differences in spatially distributed meteorological conditions.

[0078] The light intensity sensor collects the solar radiation power received per unit area and outputs the light intensity value in watts per square meter.

[0079] The wind speed pulse sensor generates pulses by rotating a wind cup to drive a photoelectric encoder. The number of pulses per unit time is proportional to the wind speed, and the output is a wind speed pulse value. The conversion method from the wind speed pulse value to the physical quantity of wind speed is as follows: Take the pulse frequency and wind speed linearity coefficient calibrated at the factory of the wind speed pulse sensor, multiply the number of pulses per unit sampling time by the pulse frequency and the wind speed linearity coefficient to obtain the wind speed, in meters per second. Here, the pulse frequency refers to the number of pulses per unit time (1 second). If the actual number of pulses is accumulated over the entire sampling sub-cycle, the accumulated number of pulses must first be divided by the number of seconds in the sampling sub-cycle to convert it into the number of pulses per unit time, i.e., the frequency, and then multiplied by the wind speed linearity coefficient to obtain the wind speed, in order to avoid proportional deviations caused by inconsistencies in the duration.

[0080] The current clearing period refers to a single clearing and settlement period in the electricity spot market. For example, the current clearing period can be set to 15 minutes. The sampling sub-period is determined by: determining it according to the pulse refresh cycle of the metering module of the power grid dispatch gateway, and taking a number of seconds not less than the pulse refresh cycle as the sampling sub-period. For example, the sampling sub-period can be set to 10 seconds. The current clearing period duration is divided by the sampling sub-period to obtain the number of sampling points. For example, a 15-minute current clearing period corresponds to 90 sampling points.

[0081] The unified sampling clock signal is a satellite-synchronized second pulse signal, distributed to the data acquisition modules of the meteorological sensor acquisition nodes and the power grid dispatch gateway. This triggers the meteorological sensor acquisition nodes and the power grid dispatch gateway to complete sampling and mark consistent timestamps at the same time. Each meteorological sensor acquisition node, using the unified sampling clock signal as a reference, acquires light intensity and wind speed pulse values ​​in each sampling sub-cycle of the current clearing period and records the sampling timestamp. For each geographic region, the arithmetic mean of the light intensity values ​​in each sampling sub-cycle within the current clearing period is used to obtain the representative light intensity value for that geographic region. Similarly, the arithmetic mean of the wind speed pulse values ​​in each sampling sub-cycle within the current clearing period is used to obtain the representative wind speed pulse value for that geographic region. The representative light intensity value and representative wind speed pulse value for each geographic region, with their sampling timestamps, are then encapsulated into a meteorological telemetry signal. The data structure of this meteorological telemetry signal consists of multiple telemetry records indexed by geographic region number. Each telemetry record includes the geographic region number, sampling timestamp, representative light intensity value for the geographic region, and representative wind speed pulse value for the geographic region.

[0082] S13: Periodically read the pulse count value representing the real-time electricity demand of the whole society from the hardware data interface of the power grid dispatch gateway, and generate the pulse sequence of the power load of the whole network.

[0083] Real-time electricity demand across society serves as the demand-side benchmark for determining the pricing space of the power grid. This demand is generated in real-time in the form of metering pulses within the power grid dispatching system. If load statistics are obtained through forwarding from financial data terminals, these statistics, after aggregation and delayed release, lag behind the actual electricity consumption time, making it impossible to align with meteorological telemetry signals on the timeline. Furthermore, the common forwarding-based data acquisition method introduces release delay errors in the differential calculation stage under sudden operating conditions due to the time misalignment between supply and demand. Directly reading pulse count values ​​from the power grid dispatching gateway hardware data interface provides first-hand demand data generated synchronously with the physical electricity consumption process.

[0084] Specifically, the power grid dispatch gateway is a hardware data interface device that provides real-time operational data to the power grid dispatch system. Its internal metering module outputs one pulse for each fixed energy increment measured, and the number of pulses is proportional to the total cumulative electricity consumption of the entire society. The fixed energy increment is the factory-set metering constant of the power grid dispatch gateway's metering module; for example, the fixed energy increment can be set to 1 megawatt-hour of energy per pulse.

[0085] Using the same unified sampling clock signal as the meteorological sensor acquisition nodes, the newly added pulse count value in each sampling sub-cycle of the current clearing period is read from the hardware data interface of the power grid dispatch gateway, and the reading timestamp is recorded.

[0086] The pulse count values ​​read from each sampling sub-cycle within the current clearing period are arranged in order of reading timestamp to generate a network-wide power load pulse sequence. The data structure of the network-wide power load pulse sequence is a time series indexed by sampling sub-cycle number. Each sequence element contains the sampling sub-cycle number, the reading timestamp, and the pulse count value within the current sampling sub-cycle.

[0087] The entire grid's electricity load pulse sequence shares the same unified sampling clock signal with all telemetry records from meteorological sensors. This ensures that the demand-side pulse count value and the supply-side geographical region's representative light intensity value and geographical region's representative wind speed pulse value are captured at the same sampling time and marked with the same sampling timestamp. In the electricity spot market, the load statistics and meteorological statistics forwarded by the financial data terminal are published through their respective aggregation links. The publication times of the two types of data do not overlap. When the output of renewable energy generation drops sharply due to sudden weather, the delayed load statistics and meteorological statistics are misaligned on the time axis, causing the supply and demand difference calculation to be superimposed with the publication delay error. The system directly reads first-hand pulse count values ​​and meteorological physical quantities with the same unified sampling clock timestamp from the power grid dispatch gateway hardware data interface and meteorological sensor acquisition nodes. This ensures strict alignment of demand-side and supply-side physical quantities on the time axis, eliminating the need for subsequent supply-demand differential calculations in step S33 due to release delay errors. This ensures that the marginal unit physical bidding space corresponds synchronously with the electricity consumption and renewable energy processes from the source of the data link, avoiding the failure of forwarding-based data acquisition due to time misalignment between supply and demand sides under sudden weather conditions. (See also...) Figure 3 This is a schematic diagram of the unified sampling clock signal alignment provided in an embodiment of this application. For example... Figure 3 As shown in the diagram, the satellite timing icon at the top transmits a unified sampling clock signal, which is distributed via dashed lines to the meteorological sensor acquisition node box on the left and the power grid dispatch gateway box on the right. Below the meteorological sensor acquisition node box, a meteorological telemetry signal sequence block is drawn, and below the power grid dispatch gateway box, a full-network power load pulse sequence block is drawn. The alignment between the two sequence blocks is marked by a horizontal dashed line. A vertical time axis runs through both sequences and is marked with a lock symbol and sampling timestamp text. The unified sampling clock signal simultaneously triggers the meteorological sensor acquisition node and the power grid dispatch gateway to complete sampling at the same time and mark the same sampling timestamp. This ensures that the light intensity value representing the geographical region, the wind speed pulse value representing the geographical region, and the pulse count value in the full-network power load pulse sequence are captured at the same sampling time. The lock symbol indicates that the physical quantities on the demand side and the physical quantities on the supply side are forcibly aligned on the time axis, avoiding the release delay error introduced by the misalignment of time on the supply and demand sides during sudden weather events in forwarding-style data acquisition.

[0088] S2: Perform natural language segmentation and clustering operations on the text-grabbed network data packets, extract objective state descriptive words related to cost changes and fuel inventory, and convert them into initial price preference feature tag strings.

[0089] The message payload in the text-captured network data packets carries descriptions of competitors' operational status published in disclosure documents. These descriptions exist in unstructured natural language, reflecting macroscopic, long-term, and static operational status. Without converting this unstructured text into unified numerical quantification parameters, the pricing preference information cannot be incorporated into subsequent numerical correction and clearing price calculation processes. Step S2 first removes the message control header, retaining valid character entities. Then, it extracts objective status descriptive words related to cost changes and fuel inventory through natural language segmentation and text clustering. Finally, it converts these objective status descriptive words into an initial pricing preference feature tag string according to preset quantification mapping rules, providing a calculable numerical benchmark for subsequent physical quantity correction. This initial pricing preference feature tag string only carries static text information and has not yet incorporated the real-time supply and demand physical constraints of the current period.

[0090] Further, step S2 includes:

[0091] S21: Parse the text content of the captured network data packets, remove the message control header, and retain the valid character entities.

[0092] In the text-grabbing network data packets output in step S11, each byte stream unit contains both a header field and a payload field. The header field carries transmission control information but does not contain operational status semantics. If the complete byte stream is directly segmented, the control characters in the header field will be mixed into the word frequency statistics and pollute the clustering results. Step S21 requires stripping the header field, retaining only the valid character entities that carry text semantics.

[0093] Specifically, according to the Hypertext Transfer Protocol (HTTP) message format specification, each byte stream unit in the text-captured network data packet is split into fields, and the message header fields are identified and removed. The message header fields refer to the status line, response header fields, and blank line separator in the HTTP response message, which are located before the message payload.

[0094] The payload field of the message after removing the header fields is decoded according to the character encoding declared in the message, and the tag control characters are removed according to the tag syntax of the Hypertext Markup Language, retaining only the text content between the tags.

[0095] The decoded text content is then processed by character normalization, removing consecutive whitespace and invisible control characters, and retaining Chinese characters, numeric characters, and punctuation characters to obtain valid character entities. The data structure of these valid character entities is a character sequence with an identifier of the source disclosure webpage and a capture timestamp. See also... Figure 4 This is a schematic diagram of the text capture network data packet deconstruction provided in an embodiment of this application. For example... Figure 4 As shown, the outermost box labels the captured network data packets, nested within byte stream unit boxes. Each byte stream unit box contains two sub-boxes: the header field and the payload field. A cross icon indicates removal from the header field, while a funnel-shaped icon indicates filtering and parsing from the payload field, pointing to the valid character entities at the bottom. Below, three vertically arranged boxes label the source disclosure webpage identifier, the capture timestamp, and the character sequence. The header field, carrying transmission control information but lacking operational status semantics, is removed to prevent control characters from contaminating the word frequency statistics and clustering results. The payload field, after decoding and label removal, retains only the text content between the labels, ultimately yielding a character sequence with the source disclosure webpage identifier and capture timestamp—the valid character entities. This provides unedited text semantic input for subsequent Chinese word segmentation and feature word frequency statistics.

[0096] S22: Use a pre-built text clustering algorithm to perform feature word frequency statistics on effective character entities and extract objective state description words related to cost changes and fuel inventory.

[0097] The descriptive words carrying the semantics of operating status in the effective character entities are mixed with a large amount of irrelevant narrative text, and the wording of cost and fuel-related descriptions is inconsistent in different disclosure documents. Direct matching by fixed keywords will miss synonymous expressions. The text clustering algorithm can aggregate semantically similar text units into the same cluster based on the distribution of words in the feature space, so that the descriptive words related to cost changes and fuel inventory form clusters in the feature space. Step S22 first segments the effective character entities and counts the feature word frequency to generate a feature word frequency vector, then performs clustering on the feature word frequency vector to obtain a set of text clusters, and finally extracts the objective status descriptive words from the text clusters that match the cost and fuel topic.

[0098] Further, step S22 includes:

[0099] S221: Perform Chinese word segmentation on valid character entities, count the feature word frequency of each segment, and generate a feature word frequency vector.

[0100] The subsequent text clustering algorithm in step S222 requires calculating the distance between text units using numerical vectors as input. However, the effective character entities output in step S21 are continuous character sequences, which have not yet been segmented into word units with independent semantics, nor converted into numerical form. Step S221 needs to segment the effective character entities into words and count word frequencies to generate feature word frequency vectors that can be used for clustering.

[0101] Specifically, taking valid character entities as input, a dictionary-based maximum matching word segmentation algorithm is used for Chinese word segmentation: starting from the beginning of the character sequence, the longest matching word in a pre-set word segmentation dictionary is searched as a word, and the segmentation position is moved to the end of the matched word. This search is repeated until the end of the character sequence, and the word segmentation sequence is output. The pre-set word segmentation dictionary is supplemented with professional terms in the fields of electricity spot trading and thermal power fuel on the basis of a general Chinese dictionary. The professional terms include coal, gas, procurement cost, inventory, maintenance, generating unit, and output.

[0102] Stop words are removed from the segmented sequence to obtain the effective segmented sequence. The stop words are function words and punctuation marks included in a pre-set stop word list.

[0103] Using the effective word segmentation sequence as input, the feature word frequency value of each effective word segment is calculated using the term frequency and inverse document frequency method: the number of times the effective word segment appears in the effective character entity of the current traversal is counted as the term frequency; the natural logarithm of the result after dividing the total number of effective character entities by the number of effective character entities containing the effective word segment of the current traversal is taken as the inverse document frequency; the feature word frequency value of the effective word segment of the current traversal is obtained by multiplying the term frequency and the inverse document frequency.

[0104] Using the word order of a pre-set feature word list as the dimension order, the feature word frequency values ​​of each effective word segment are filled into the corresponding dimension to generate a feature word frequency vector. The dimension of the feature word frequency vector is equal to the total number of words in the pre-set feature word list; the pre-set feature word list consists of professional terms related to cost, fuel, and unit operation from a pre-set word segmentation dictionary.

[0105] S222: Use a pre-set text clustering algorithm to cluster the feature word frequency vectors and generate a set of text clusters.

[0106] If objective state descriptions are retrieved directly using single keywords without clustering based on feature word frequency vectors, synonymous expressions of the same business fact in different disclosure documents will be judged as different items due to differences in word choice, resulting in the fragmentation and omission of cost and fuel-related information. Clustering based on feature word frequency vectors allows semantically similar text units to be aggregated into the same text cluster, laying the foundation for the complete extraction of objective state descriptions from the topic clusters.

[0107] Specifically, the feature word frequency vectors corresponding to each valid character entity are used as sample points, and the K-means clustering algorithm is used for clustering.

[0108] The method for determining the number of clusters in the K-means clustering algorithm is as follows: take the cost fuel topic and the irrelevant topic as the two topics as the number of clusters, that is, set the number of clusters to 2.

[0109] The K-means clustering algorithm is executed as follows: Two feature word frequency vector sample points are randomly selected from the feature word frequency vector sample set as initial cluster centers. For each feature word frequency vector sample point, the cosine distance between the currently traversed feature word frequency vector sample point and the two cluster centers is calculated. The currently traversed feature word frequency vector sample point is assigned to the cluster to which the cluster center with the smallest cosine distance belongs. For each cluster, the mean vector of all feature word frequency vector sample points within the cluster is recalculated, and this mean vector is normalized to a unit vector and used as a new cluster center to maintain consistency with the cosine distance metric. This process of partitioning and updating is repeated until the change in the cluster center position between two adjacent iterations is less than the clustering convergence threshold. The cosine distance is 1 minus the cosine similarity of the two vectors, where the cosine similarity is the dot product of the two vectors divided by the product of their Euclidean moduli. The clustering convergence threshold can be preset based on the feature word frequency vector dimension and engineering experience. For example, the square root of the inverse of the feature word frequency vector dimension can be taken as the clustering convergence threshold, that is, the clustering is considered to have converged when the sum of the Euclidean distances of the two cluster centers in two adjacent iterations is less than the clustering convergence threshold.

[0110] The two clusters after clustering convergence and their cluster centers are encapsulated into a text cluster set. The data structure of the text cluster set is a set containing two cluster records. Each cluster record contains a cluster number, a cluster center vector, and a list of valid character entity identifiers belonging to the current cluster.

[0111] S223: Match cost fuel topic clusters from the text cluster set and extract objective state descriptive words related to cost changes and fuel inventory.

[0112] The text cluster set output in step S222 only groups valid character entities by topic, without specifying which cluster corresponds to the cost and fuel topics, nor extracting specific descriptive terms from the clusters. Step S223 needs to identify the cost and fuel topic clusters and extract noun phrases from them to obtain objective state descriptive terms that can be used for quantification and mapping.

[0113] Specifically, a pre-defined set of cost fuel seed terms is provided. This set consists of standard terms in the fields of cost changes and fuel inventory, including procurement costs, coal prices, gas prices, inventory levels, and unit maintenance.

[0114] For each cluster record in the text clustering cluster set, construct a seed word frequency vector using the feature word frequency values ​​of various sub-words in the cost fuel seed word set. Calculate the cosine similarity between the cluster center vector of the currently traversed cluster record and the seed word frequency vector, and take the cluster record with the largest cosine similarity as the cost fuel topic cluster.

[0115] For each valid character entity belonging to the cost fuel topic cluster, part-of-speech tagging is performed, and noun phrases are extracted. Noun phrases that simultaneously meet the following two conditions are retained as objective state description words: the currently traversed noun phrase contains at least one seed word from the cost fuel seed word set; the currently traversed noun phrase contains a trend word representing rise, fall, increase, decrease. The trend words are words included in a pre-set trend word table, including rise, fall, increase, decrease, sufficient, and tense.

[0116] The objective state descriptors refer to noun phrases that only state verifiable business facts such as cost increases and decreases and fuel inventory changes, without subjective evaluation. Objective state descriptors characterize the business attributes of competitors; for example, objective state descriptors include decreased coal procurement costs and sufficient fuel inventory. The extracted objective state descriptors are arranged in order of their appearance in the disclosure documents, outputting an objective state descriptor sequence.

[0117] S23: Convert objective state descriptions into corresponding numerical labels according to preset quantization mapping rules, and then concatenate them sequentially to generate an initial price preference feature tag string.

[0118] The subsequent step S43 requires product correction of the price preference parameters in numerical form with the real-time preference correction coefficient. However, the objective state description word sequence output in step S223 is a sequence of natural language phrases and cannot be directly used in numerical calculations. Step S23 needs to convert each objective state description word into a numerical label according to a preset quantization mapping rule and concatenate them sequentially to generate an initial price preference feature tag string that can be used for numerical correction.

[0119] Specifically, the preset quantitative mapping rules are stored in the form of a key-value table. The key is the standard category of the objective state descriptor, and the value is the corresponding price preference benchmark label. The price preference benchmark label ranges from 0 to 2 in a closed interval. A value of 1 indicates a neutral price level, a value less than 1 indicates a tendency to bid low relative to the benchmark electricity price, and a value greater than 1 indicates a tendency to bid high relative to the benchmark electricity price.

[0120] The initial calibration method for each bid preference benchmark label in the preset quantitative mapping rule key-value table is as follows: The objective state descriptive term category appearing in historical disclosure documents is used as the classification basis; the ratio of the actual bid price of the same market entity in the electricity spot market to the benchmark electricity price of the same period after the corresponding objective state descriptive term category appears is used as the target variable; a regression matrix is ​​constructed using the unique thermal encoding of the objective state descriptive term category as the independent variable; linear regression is performed on no less than 100 historical samples according to the minimum residual sum of squares criterion; and the ratio obtained from the regression fit is taken as the bid preference benchmark label for the corresponding objective state descriptive term category and written into the key-value table. For example, the bid preference benchmark label corresponding to the category of decreasing coal procurement costs can be calibrated to 0.7, and the bid preference benchmark label corresponding to the category of tight fuel inventory can be calibrated to 1.2.

[0121] When a new objective state descriptor category not covered in the key-value table appears, the quote preference benchmark label of the existing category that is closest to the new objective state descriptor category in the cost-fuel semantic space is used as the initial value to supplement the key-value table. The semantic distance is measured by the cosine distance of the seed word frequency vectors of the two categories.

[0122] After each electricity spot market clearing period ends, the system uses the ratio of the actual clearing price to the benchmark price for the corresponding clearing period as the observation value, and updates the bid preference benchmark label for the corresponding objective state descriptor category using an exponentially weighted moving average method. The decay factor of the exponentially weighted moving average is set to 0.1.

[0123] For each objective state descriptor in the objective state descriptor sequence, a preset quantization mapping rule key-value table is queried to obtain the corresponding pricing preference benchmark label. These benchmark labels are then concatenated sequentially according to the order of the objective state descriptor sequence to generate an initial pricing preference feature tag string. The initial pricing preference feature tag string is an ordered numerical sequence, with a sequence length equal to the number of objective state descriptors in the objective state descriptor sequence. The element is the first element in the sequence of objective state descriptive words. The pricing preference benchmark labels corresponding to each objective state descriptor. See also Figure 6 This is a schematic diagram of the transformation of the preset quantization mapping rule structure provided in the embodiments of this application. For example... Figure 6As shown, the left side uses curly braces to summarize the sequence of objective state descriptive terms, and three vertically arranged boxes below represent sequence elements. A dashed arrow points to the middle pre-defined quantization mapping rule table, which is divided into a key-value table standard category column and a pricing preference benchmark label column. A thick black arrow points to the initial pricing preference feature tag string on the right, and five squares below form an ordered numerical sequence, with one square filled with a diagonal shading to indicate a converted numerical label. Each objective state descriptive term in the sequence is queried from the pre-defined quantization mapping rule key-value table to obtain its corresponding pricing preference benchmark label. These labels are then concatenated sequentially to generate the initial pricing preference feature tag string, converting the sequence of natural language phrases into a pricing preference benchmark label sequence that can be used for numerical calculations. This provides a computable numerical benchmark for the product correction in subsequent step S43.

[0124] S3: Calculate the estimated value of new energy physical output based on meteorological sensor telemetry signals, integrate and accumulate the pulse sequence of the entire grid's power load to obtain the overall physical demand for electricity, and subtract the estimated value of new energy physical output from the overall physical demand for electricity to generate the physical bidding space of marginal units.

[0125] The initial bid preference feature marker string output in step S23 only reflects the static text operating status of competitors and does not include the real-time supply and demand physical boundary of thermal power in the current period. The clearing price of the electricity spot market is determined by the bid of marginal thermal power units, and the generation space required by thermal power is equal to the remaining part of the total social electricity demand after deducting the supply of renewable energy generation. This remaining part changes frequently with weather conditions and electricity load. If the clearing price is predicted directly using the initial bid preference feature marker string without introducing the real-time supply and demand physical boundary, it will deviate from the actual clearing price under sudden weather conditions. Step S3 calculates the estimated value of renewable energy physical output based on meteorological telemetry signals and calculates the overall physical demand of electricity based on the pulse sequence of the total grid electricity load. The difference between the two yields the physical bidding space of marginal units undertaken by thermal power, providing a real-time supply and demand boundary for subsequent physical correction of static text preferences.

[0126] Further, step S3 includes:

[0127] S31: Substitute the light intensity value and wind speed pulse value in the meteorological sensor telemetry signal into the fixed meteorological power generation conversion formula to obtain the estimated value of the physical output of new energy for the corresponding time period.

[0128] The physical output of new energy sources consists of two parts: photovoltaic power generation and wind power generation. Photovoltaic power generation is determined by sunlight intensity, while wind power generation is determined by wind speed. These two components follow different physical conversion relationships and are distributed across different geographical regions. The meteorological telemetry signal output in step S12 carries the representative sunlight intensity value and representative wind speed pulse value for each geographical region, indexed by geographical region number. These values ​​need to be substituted into the photovoltaic power generation conversion formula and the wind power generation conversion formula respectively to calculate their respective physical output components, and then summarized into an estimated value of the physical output of new energy sources covering the entire network.

[0129] Further, step S31 includes:

[0130] S311: Substitute the light intensity values ​​of each geographical region in the meteorological remote sensing signal into the photovoltaic power generation conversion formula to calculate the photovoltaic physical output component.

[0131] The subsequent step S313 requires using photovoltaic power generation as one of the supply components to participate in the aggregation of new energy physical output estimates. However, the geographic region-represented light intensity value in the meteorological sensor telemetry signal is the original meteorological physical quantity and has not yet been converted into power generation. Step S311 requires converting the geographic region-represented light intensity value into a photovoltaic physical output component using the photovoltaic power generation conversion formula.

[0132] Specifically, the solar irradiance value representing the geographical region in each telemetry record indexed by geographical region number in the meteorological sensor telemetry signal is used as input, and photovoltaic power generation conversion is performed for each geographical region.

[0133] The formula for converting photovoltaic power generation is: ,in, Geographical partition numbering The corresponding geographic region represents the light intensity value, measured in watts per square meter, and is taken from the geographic region number in the meteorological sensor telemetry signal. Telemetry records; Geographical partition numbering The total light-receiving area of ​​the internal photovoltaic modules, in square meters, is taken from the new energy power station ledger interface of the power grid dispatch system. The photovoltaic module's photoelectric conversion efficiency is a dimensionless constant, taken from the photovoltaic module's factory specifications, and is an example. 0.20 is acceptable; This is the conversion factor from watts to megawatts; Geographical partition numbering The corresponding photovoltaic physical output quantum quantity, in megawatts.

[0134] The photovoltaic physical output components are obtained by summing the photovoltaic physical output quantities of all geographical regions covered by meteorological remote sensing signals, with the unit being megawatts.

[0135] For example, when geographic zone number 1 represents a light intensity of 800 watts per square meter and the total light-receiving area of ​​the photovoltaic modules is... When the area is 1 square meter and the photoelectric conversion efficiency is 0.20, the photovoltaic physical output quantum quantity of geographical zone number 1 is: Megawatts; after summing the photovoltaic physical output quantities of each geographical region, the exemplary photovoltaic physical output component is 800 megawatts.

[0136] S312: Substitute the wind speed pulse values ​​of each geographical region in the meteorological sensor telemetry signal into the wind power generation conversion formula to calculate the physical output component of wind power.

[0137] The photovoltaic physical output component output in step S311 only represents the photovoltaic portion of the new energy supply and does not yet include the wind power portion, while the proportion of wind power output in the new energy supply cannot be ignored. Wind power generation is proportional to the cube of wind speed. It is necessary to convert the wind speed pulse value representing the geographical region in the meteorological remote sensing signal into wind speed and then input it into the wind power generation conversion formula to obtain the wind power physical output component.

[0138] Specifically, the wind speed pulse value representing the geographical region in each telemetry record of the meteorological sensor telemetry signal is used as input, and the wind power generation conversion is performed for each geographical region.

[0139] First, convert the wind speed pulse value representing the geographical region into wind speed: According to the conversion method described in step S12, the wind speed is equal to the number of wind speed pulses representing the geographical region per unit time (1 second) multiplied by the pulse frequency and wind speed linearity coefficient calibrated by the wind speed pulse sensor at the factory. If the wind speed pulse value representing the geographical region is the cumulative number of pulses in the entire sampling sub-cycle, first divide it by the number of seconds in the sampling sub-cycle to convert it into the number of pulses per unit time, then substitute it into the above conversion, and record the geographical region number. The corresponding wind speed is The unit is meters per second.

[0140] The formula for converting wind power generation is: ,in, The density is air, expressed in kilograms per cubic meter, and is taken as 1.225 under standard conditions. Geographical partition numbering The total swept area of ​​the wind turbine rotor of the internal wind turbine, in square meters, is taken from the new energy power station ledger interface of the power grid dispatch system; Let be the wind energy utilization coefficient, and be a dimensionless constant, taken as an engineering value under the Betz limit constraint, for example. 0.40 is acceptable; Geographical partition numbering Corresponding wind speed; This is the conversion factor from watts to megawatts; Geographical partition numbering The corresponding wind power output quantum quantity, in megawatts.

[0141] The wind power physical output components are obtained by summing the wind power physical output quantities of all geographical regions covered by meteorological sensing and telemetry signals, with the unit being megawatts.

[0142] For example, when the wind speed in geographic zone 1 is 10 meters per second, the air density is 1.225 kilograms per cubic meter, and the total area swept by the wind turbine is... When the area is 1 square meter and the wind energy utilization factor is 0.40, the wind power physical output quantum quantity of geographical zone number 1 is: Megawatts; after summing the wind power physical output units of each geographical region, the exemplary wind power physical output component is 1200 megawatts.

[0143] S313: Summing the physical output components of photovoltaic power and wind power to obtain the estimated value of new energy physical output.

[0144] The subsequent step S33 requires the total supply of new energy power generation to be included in the supply-demand difference calculation, while the photovoltaic physical output component and wind power physical output component output in steps S311 and S312 respectively only represent a single power generation type of new energy supply. Step S313 needs to merge the two types of physical output components into a new energy physical output estimate that covers all types of new energy power generation.

[0145] Specifically, the photovoltaic physical output component output in step S311 and the wind power physical output component output in step S312 are used as inputs, and the unit of both the photovoltaic physical output component and the wind power physical output component is megawatt.

[0146] The physical output components of photovoltaic power and wind power are added together to obtain the estimated value of new energy physical output for the corresponding time period, in megawatts. The estimated value of new energy physical output represents the power generation supply of all new energy power plants within the grid coverage area for the corresponding time period.

[0147] For example, when the physical output of photovoltaic power is 800 MW and the physical output of wind power is 1200 MW, the estimated physical output of new energy is 2000 MW.

[0148] S32: Perform integral and cumulative calculations on the pulse sequence of the power load of the entire network to obtain the overall physical demand for electricity in the corresponding time period.

[0149] The power load pulse sequence of the entire network is composed of pulse counts from each sampling sub-cycle. Each pulse count only represents the increase in electrical energy consumption within the corresponding sampling sub-cycle, and is in discrete count form, so it cannot directly represent the overall power consumption for the corresponding time period. The pulse count value has a fixed proportional relationship with the increase in electrical energy consumption. Based on this, the discrete pulse count values ​​can be integrated and accumulated to restore the overall physical demand for electricity in the corresponding time period.

[0150] Specifically, the entire network power load pulse sequence output in step S13 is used as input.

[0151] The pulse count values ​​of each sequence element in the power load pulse sequence of the entire network are summed in the order of the sampling sub-cycle number to obtain the cumulative pulse count value in the current clearing period.

[0152] Multiply the cumulative pulse count by the fixed energy increment of the power grid dispatch gateway metering module to obtain the cumulative power consumption during the current clearing period, in megawatt-hours.

[0153] Dividing the cumulative electricity consumption by the duration of the current clearing period yields the overall physical demand for electricity during the corresponding time period, expressed in megawatts. This overall physical demand represents the average power consumption of the entire society within the grid's coverage area during the corresponding time period. The duration of the current clearing period is taken as the difference between the start and end times of the current clearing period determined in step S12.

[0154] For example, when the cumulative pulse count during the current clearing period is 1500, the fixed energy increment is 1 MWh per pulse, and the duration of the current clearing period is 0.25 hours, the cumulative energy consumption is 1500 MWh, and the overall physical demand for electricity is... Megawatts.

[0155] S33: Subtract the estimated value of new energy physical output from the total physical demand for electricity to obtain the remaining supply and demand gap; when the remaining supply and demand gap is greater than zero, the remaining supply and demand gap is used as the marginal unit physical bidding space to be undertaken by thermal power; when the remaining supply and demand gap is not greater than zero, the marginal unit physical bidding space is determined to be zero.

[0156] The overall physical demand for electricity output in step S32 represents the total demand side, while the estimated physical output of new energy output in step S313 represents the total supply side. Both are one-sided physical quantities and do not represent the supply-demand gap that thermal power, as a marginal unit, actually needs to fill. Step S33 requires subtracting the total demand side from the total supply side to obtain the physical bidding space for marginal units undertaken by thermal power.

[0157] Specifically, the overall physical demand for electricity output in step S32 and the estimated output of new energy physical output in step S313 are used as inputs, with the units of the overall physical demand for electricity and the estimated output of new energy physical output being megawatts.

[0158] The remaining supply-demand gap is obtained by subtracting the estimated output of new energy sources from the overall physical demand for electricity. When the remaining supply-demand gap is greater than zero, it is used as the marginal unit bidding space; when the remaining supply-demand gap is not greater than zero, the marginal unit bidding space is set to zero. The unit is megawatt. The marginal unit bidding space represents the power gap that needs to be filled by thermal power units within a corresponding time period, i.e., the generation space of thermal power as marginal units. See also... Figure 7 This is a schematic diagram illustrating the differential generation of the physical bidding space quantity of marginal units provided in an embodiment of this application. For example... Figure 7 As shown, the top long bar indicates the overall physical demand for electricity, and the right side of it is filled with a diagonal shaded line to indicate the exposed area not covered by the new energy section below; the two short bars below indicate the physical output components of photovoltaic power and wind power respectively, and the two are joined together in curly braces to indicate the estimated physical output value of new energy; a broken line is drawn from the exposed shaded area to indicate the remaining supply and demand gap and points to the simplified diagram of thermal power units below, and the bottom is marked with curly braces to indicate the physical bidding space of marginal units. The remaining supply-demand gap is the exposed portion after subtracting the sum of the photovoltaic and wind power physical output components in the estimated value of new energy physical output from the overall physical demand for electricity. When the remaining supply-demand gap is greater than zero, it is used as the marginal unit physical bidding space to be undertaken by thermal power units; when the remaining supply-demand gap is not greater than zero, the marginal unit physical bidding space is determined to be zero. This difference result is obtained by coupling two independent physical measurement channels: the electricity pulse count of the metering module of the grid dispatch gateway and the light intensity value and wind speed pulse value of the meteorological sensor acquisition node. When the estimated value of new energy physical output drops sharply due to sudden weather, the difference result immediately rises, so that the physical condition of thermal power being forced to operate at full load peak is captured in real time.

[0159] For example, when the overall physical demand for electricity is 6,000 MW and the estimated physical output of new energy is 2,000 MW, the remaining supply and demand gap is 4,000 MW, and the physical bidding space for marginal units is 4,000 MW; when a sudden weather event causes the estimated physical output of new energy to drop sharply to 500 MW, the physical bidding space for marginal units will rise to 5,500 MW.

[0160] The marginal unit physical bidding space is obtained by the difference between the overall physical demand for electricity and the estimated value of renewable energy physical output. The two differencing components come from independent physical measurement channels: the overall physical demand for electricity originates from the electricity pulse count of the metering module of the grid dispatch gateway, reflecting the physical process of electricity consumption on the demand side; the estimated value of renewable energy physical output originates from the light intensity value and wind speed pulse value of the meteorological sensor acquisition node, converted by the meteorological power generation conversion formula, reflecting the meteorological-driven physical process on the supply side. If only the overall physical demand for electricity on the demand side is retained without the estimated value of renewable energy physical output, the supply share already filled by renewable energy cannot be deducted, the marginal unit physical bidding space will be systematically overestimated, the bidding space for thermal power will be exaggerated, and the subsequent supply-demand tension approximation rate calculated using this as the numerator will be artificially high. If only the estimated value of renewable energy physical output on the supply side is retained without the overall physical demand for electricity, the demand benchmark is lost, the level of renewable energy output cannot be correlated with the gap that thermal power needs to fill, and the estimated value of renewable energy physical output degenerates into an isolated supply reading and cannot be mapped into bidding space. The differential coupling of the two physical measurement channels enables the marginal unit physical bidding space quantity to simultaneously possess both the demand-side total benchmark and the supply-side deduction capability. When sudden extreme weather causes a sharp drop in the estimated value of new energy physical output, the differential result immediately rises, and the harsh physical condition of thermal power being forced to operate at full capacity peak is instantly quantified and captured. This quantification result is transmitted to the subsequent step S41 as the numerator for calculating the supply and demand tension approach rate, so that the blind spot of sudden physical conditions that cannot be covered by relying solely on historical price average predictions is filled by real-time supply and demand differentials. The absence of either of the two measurement channels will cause the marginal unit physical bidding space quantity to lose its physical meaning corresponding to supply and demand.

[0161] S4: Based on the physical bidding space of marginal units, a real-time preference correction coefficient is generated, and the initial bid preference feature tag string is adjusted to obtain dynamic bid quantification parameters, which are finally converted into automatic order placement control instructions.

[0162] The initial bid preference feature marker string output in step S23 reflects the macro-level, long-term, and static bid preferences presented by competitors in their disclosure documents. The marginal unit physical bidding space quantity output in step S33 reflects the high-frequency, real-time, and dynamic supply and demand physical boundary of thermal power in the current period. Existing feature extraction methods typically map unstructured text directly to fixed bid parameters, without establishing a dynamic alignment and verification mechanism between text semantic features and real-time high-frequency time series physical quantities. This leads to numerical conflicts between static preferences and real-time boundaries under sudden physical conditions. Step S4 first generates a real-time preference correction coefficient from the marginal unit physical bidding space quantity, then corrects the initial bid preference feature marker string by multiplication to obtain dynamic bid quantification parameters. Finally, it calculates the buy-sell clearing price value based on this and encapsulates it into an automatic order placement control instruction, which is then sent to the trading gateway for execution, thus achieving a forced concession of static text features to real-time physical boundaries.

[0163] Further, step S4 includes:

[0164] S41: Extract the physical bidding space of marginal units, divide the physical bidding space of marginal units by the physical upper limit of thermal power installed capacity stored in the form of preset fixed constant parameters, and calculate the supply and demand tension approach rate.

[0165] The marginal unit physical bidding space output in step S33 is an absolute power value measured in megawatts. Its value varies with the grid size and cannot directly represent the degree of tension in thermal power supply approaching its physical upper limit. Normalizing the marginal unit physical bidding space to the physical upper limit of thermal power installed capacity yields a dimensionless supply-demand tension approach rate, providing a unified comparison benchmark for the degree of supply-demand tension under different grid sizes.

[0166] Specifically, the physical bidding space of the marginal unit output in step S33 is used as input.

[0167] The physical upper limit of thermal power installed capacity is stored in the form of a preset fixed constant parameter for the current clearing cycle. It refers to the maximum power generation of all available thermal power units within the grid coverage area when running at full load, in megawatts. It is taken from the sum of the rated capacity of each thermal power unit in the thermal power unit ledger of the grid dispatch system and deducted from the planned maintenance capacity. For example, the physical upper limit of thermal power installed capacity can be set to 6000 megawatts.

[0168] Dividing the marginal unit physical bidding space by the physical upper limit of thermal power installed capacity yields the supply-demand tension approach rate. This supply-demand tension approach rate is dimensionless and ranges from 0 to 1 (closed interval). The closer the supply-demand tension approach rate is to 1, the closer the supply-demand gap that thermal power needs to fill is to the physical upper limit of full-load thermal power generation, indicating a tighter supply-demand situation. Conversely, the closer the supply-demand tension approach rate is to 0, the more ample the thermal power supply margin.

[0169] For example, when the marginal unit physical bidding space is 4000 MW and the physical upper limit of thermal power installed capacity is 6000 MW, the supply and demand tension rate is approaching. When the marginal unit physical bidding space rises to 5500 MW, the supply-demand tension rate rises to... .

[0170] S42: Perform nonlinear mapping calculation on the supply and demand tension approach rate to generate a real-time preference correction coefficient. When the supply and demand tension approach rate is greater than the set threshold, the output is an exponentially amplified real-time preference correction coefficient.

[0171] In the electricity spot market, the marginal cost of thermal power exhibits a non-linear surge as supply approaches its physical limit. Common linear mapping methods, failing to capture this sharp increase in marginal cost near the peak of full-load thermal power, systematically underestimate the magnitude of price increases, causing the output price preference parameter to deviate from the true marginal cost under emergency peak conditions. Step S42 employs a piecewise non-linear mapping, causing the real-time preference correction coefficient to amplify exponentially after the supply-demand tension rate exceeds a set threshold, thus matching the non-linear physical law of the thermal power marginal cost initially leveling off and then surging.

[0172] Specifically, the supply-demand tension approximation rate output in step S41 is used as input, and the supply-demand tension approximation rate is denoted as... .

[0173] The expression for calculating the nonlinear mapping of the real-time preference correction coefficient is as follows:

[0174]

[0175] in, This is a real-time preference correction coefficient; The supply-demand tension rate is taken from step S41; To set a threshold; This is the subthreshold linear adjustment coefficient; The magnitude coefficient of the over-threshold exponent; This is the steepness coefficient of the overthreshold exponential.

[0176] The set threshold The method for determining this is as follows: Statistically analyze the supply-demand tension approach rate corresponding to when the marginal cost of thermal power begins to rise non-linearly in the historical electricity spot market. Calculate the lower quartile of this statistical supply-demand tension approach rate as a set threshold. For example... It can be set to 0.80.

[0177] The subthreshold linear adjustment coefficient The method for determining the target variable is as follows: The incremental ratio of the actual clearing price to the benchmark electricity price within the historical sample where the supply-demand tension approach rate does not exceed a set threshold is used as the target variable. The supply-demand tension approach rate is used as the independent variable. Linear regression is performed according to the minimum residual sum of squares criterion, and the slope obtained from the fit is taken as the target value. Exemplary 0.25 is acceptable.

[0178] The overthreshold exponent amplitude coefficient With the overthreshold exponential steepness coefficient The method for determining the target variable is as follows: The target variable is the difference between the ratio of the actual clearing price to the benchmark electricity price within the historical sample where the supply-demand tension approach rate exceeds a set threshold, minus the linear value taken at the set threshold. The independent variable is the portion of the supply-demand tension approach rate exceeding the set threshold. An exponential function curve is fitted according to the minimum residual sum of squares criterion. The amplitude coefficient and steepness coefficient obtained from the fitting are taken as... and Exemplary 0.30 is acceptable. Version 8.0 is recommended.

[0179] The design principle of the nonlinear mapping calculation expression for the real-time preference correction coefficient is: subthreshold segment The linear form indicates that when the supply-demand tension approach rate does not exceed a set threshold, the thermal power supply margin is sufficient, the marginal cost increases moderately with the supply-demand gap, and the real-time preference correction coefficient increases slightly linearly with the supply-demand tension approach rate; the threshold-exceeding segment is represented by an exponential term. Superimposed on the linear values ​​at a set threshold, the independent variable of the exponential term is the portion of the supply-demand tension approach rate that exceeds the set threshold. When the supply-demand tension approach rate is exactly equal to the set threshold, the exponential term takes a value of 0, ensuring that the mapping is continuous at the set threshold without any jumps. When the supply-demand tension approach rate exceeds the set threshold, the exponential term increases sharply with the excess, indicating a nonlinear surge in marginal cost caused by unit ramp-up restrictions and start-up / shutdown constraints when thermal power is nearing full-load peak. The exponential term is in the numerator and is additively superimposed on the sub-threshold value, causing the real-time preference correction coefficient in the over-threshold segment to monotonically increase on the basis of the sub-threshold segment. The overall design goal of the expression is to make the real-time preference correction coefficient approach a neutral value when supply and demand are relaxed, and to amplify exponentially when supply and demand are near the physical upper limit of thermal power, that is, when the supply-demand tension approach rate increases and exceeds the set threshold, the output rises sharply, which is consistent with the physical law that the marginal cost of thermal power first flattens and then rises sharply.

[0180] For example, when the supply and demand tension rate is 0.67, and the threshold is set at 0.80, When the value is 0.25, because the supply and demand tension approach rate does not exceed the set threshold, the real-time preference correction coefficient is... When the supply and demand tensions approach a rate of 0.92, For 0.30, When the value is 8.0, the real-time preference correction coefficient is adjusted because the supply and demand imbalance rate exceeds the set threshold. .

[0181] S43: Multiply the initial bid preference feature string by the real-time preference correction coefficient for numerical correction, and generate dynamic bid quantification parameters that eliminate static text bias.

[0182] The subsequent step S44 requires using the real-time physically corrected price preference value as input to calculate the buy-sell clearing price value. However, the initial price preference feature marker string output in step S23 only carries static textual price tendencies, and the real-time preference correction coefficient output in step S42 only carries the supply and demand tension; the two have not yet been combined. Step S43 needs to apply the real-time preference correction coefficient to the initial price preference feature marker string to generate dynamic price quantification parameters that incorporate real-time physical constraints.

[0183] Specifically, the initial price preference feature tag string output in step S23 and the real-time preference correction coefficient output in step S42 are used as inputs.

[0184] Each element in the ordered numerical sequence of the initial bid preference feature marker string is multiplied by a real-time preference correction coefficient to obtain the corrected ordered numerical sequence, i.e., the dynamic bid quantification parameter. The data structure of the dynamic bid quantification parameter is an ordered numerical sequence of the same length as the initial bid preference feature marker string, where the element in the sequence is... The element equals the first element in the initial quote preference feature tag string. The product of each element and the real-time preference correction coefficient.

[0185] Since elements with values ​​less than 1 in the initial price preference feature tag string represent a tendency to bid low and elements with values ​​greater than 1 represent a tendency to bid high, and the real-time preference correction coefficient is greater than 1 and amplifies exponentially when supply and demand are tight, the product correction causes the elements that originally represented a tendency to bid low to be raised to the high price range greater than 1 when supply and demand are tight, thus covering the static text preference with real-time physical constraints.

[0186] For example, when an element in the initial bid preference feature marker string is 0.7 and the real-time preference correction coefficient is 1.68, the corresponding element in the dynamic bid quantification parameter is... The original value of 0.7, which represented a tendency towards low prices, was corrected to 1.18, which represents a tendency towards high prices. See also Figure 8 This is a schematic diagram of the product correction of dynamic quotation quantification parameters provided in an embodiment of this application. Figure 8As shown, the top row of small squares represents the original sequence, with one square sunk and marked as indicating a low-price bidding tendency. The large box on the left marks the initial price preference feature string, and a cross-multiplication symbol connects the middle to the larger box on the right marking the real-time preference correction coefficient. The two lines converge at the central product correction text box, and the arrow below points to the elongated and thickened rectangle marking the dynamic price quantification parameter. The right side marks the high-price bidding tendency. Each element in this ordered numerical sequence of the initial price preference feature string is multiplied by the real-time preference correction coefficient to obtain the dynamic price quantification parameter. Since elements in the initial price preference feature string with values ​​less than 1 represent a low-price bidding tendency, and the real-time preference correction coefficient is greater than 1 and amplifies exponentially during periods of tight supply and demand, the product correction raises the elements that originally represented a low-price tendency to the high-price bidding tendency range (greater than 1) during periods of tight supply and demand. This uses real-time physical constraints to cover static text preferences, eliminating static text bias.

[0187] Step S43 applies the real-time preference correction coefficient to the initial bid preference feature tag string in a product manner. Its original design goal is to eliminate numerical conflicts between static text preferences and real-time supply and demand physical boundaries due to different time scales. The real-time preference correction coefficient is derived from the marginal unit physical bidding space calculated in step S33 based on meteorological telemetry signals and the pulse sequence of the entire grid's electricity load. The value of this physical quantity is completely independent of the text content of the competitor's disclosed documents and is not affected by any textual descriptions. It is this mathematical structure, which is independent of text content, that enables step S43 to achieve the original goal of time scale alignment while naturally exhibiting the ability to structurally circumvent misleading disclosures: when competitors deliberately exaggerate the decline in coal procurement costs in their historical disclosure reports to create the illusion of low-cost pricing and suppress the initial price preference feature tag string to the low-price range, as long as the estimated value of new energy physical output drops sharply due to sudden weather in the current period and the supply and demand tension approach rate exceeds the set threshold, the real-time preference correction coefficient will be exponentially amplified. The product correction will forcefully raise the suppressed elements to the high-price range, and the misleading text tags will be automatically canceled by real-time physical constraints. The ability to circumvent misleading disclosures was not the original design goal of step S43. Step S43 was only designed for time scale alignment between static text preferences and real-time physical boundaries. However, since the value of the real-time preference correction coefficient does not depend on the text content, the system naturally acquires the implicit ability to resist misleading low-price texts and avoid significant losses under emergency peak conditions of the power grid without introducing an independent disclosure authenticity judgment module. This overcomes the hidden defect that the pricing strategy based solely on publicly disclosed static data cannot defend against competitors' deliberate disclosure of misleading information.

[0188] S44: Based on the dynamic quotation quantitative parameters, directly calculate the buy and sell clearing price values, encapsulate them into automatic order placement control instructions in a standard trading format, and send them to the trading gateway for execution.

[0189] The dynamic price quantification parameters output in step S43 are dimensionless ordered numerical sequences, representing only the price preference level after real-time physical correction. They are neither price values ​​with monetary units nor instruction formats recognizable by the trading system, and therefore cannot be directly issued for execution. Step S44 requires converting the dynamic price quantification parameters into buy / sell clearing price values ​​and encapsulating them into automatic order placement control instructions recognizable by the trading gateway, thus completing the closed loop from quantification parameters to actual trading actions.

[0190] Specifically, the dynamic price quantification parameters output in step S43 are used as input.

[0191] The arithmetic mean of all elements in the ordered numerical sequence of dynamic pricing quantification parameters is used to obtain the pricing preference comprehensive coefficient. This coefficient represents the ratio of the overall pricing level of market participants to the benchmark electricity price after real-time physical correction.

[0192] The benchmark electricity price is obtained from the day-ahead market reference price for the corresponding time period published by the power trading center's information release interface, and the unit is yuan per megawatt-hour.

[0193] Multiplying the benchmark electricity price by a comprehensive bidding preference coefficient yields the buy-sell clearing price, expressed in yuan per megawatt-hour. This buy-sell clearing price refers to the price that the market participant intends to submit to the electricity spot market after comprehensively adjusting for competitors' static operating preferences and its own current real-time supply and demand physical boundaries. It is used to generate its own order placement instructions and is not a prediction of competitors' bids or market clearing results.

[0194] For example, when the arithmetic mean of the elements of the dynamic pricing quantification parameters is 1.18 and the benchmark electricity price is 300 yuan per megawatt-hour, the pricing preference coefficient is 1.18, and the buy-sell clearing price is... Yuan per megawatt-hour.

[0195] The buy / sell clearing price, transaction direction, declared electricity quantity, and declared time period fields are filled and encapsulated according to the standard transaction message format of the electricity spot market to generate an automatic order placement control instruction. The data structure of the automatic order placement control instruction is a structured transaction message conforming to the power trading center's message specifications, including the transaction direction field, buy / sell clearing price field, declared electricity quantity field, and declared time period field. See also... Figure 9 This is a schematic diagram of the automatic order placement control instruction message structure provided in an embodiment of this application. For example... Figure 9As shown, the file label with a folded corner in the upper left corner indicates the standard transaction message format. The outer label indicates the automatic order placement control instruction and connects to the content below with a vertical line. Inside, four squares are arranged horizontally, indicating the transaction direction field, the buy / sell clearing price value field, the declared electricity quantity field, and the declared time period field, respectively. The buy / sell clearing price value, transaction direction field, declared electricity quantity field, and declared time period field are filled and encapsulated according to the standard transaction message format of the electricity spot market to generate the automatic order placement control instruction. The data structure of this instruction is a structured transaction message conforming to the message specifications of the electricity trading center, containing the above four types of fields. It is sent to the trading gateway through the network card interface of the trading server. The trading gateway submits the declaration to the electricity spot market according to the buy / sell clearing price value, completing the execution of the action from quantitative parameters to actual transaction action.

[0196] The automatic order placement control command is sent to the trading gateway through the network card interface of the trading server. The trading gateway then submits a declaration to the electricity spot market according to the buy-sell clearing price value, thus completing the automatic order placement execution.

[0197] In this embodiment, steps S1 to S4 constitute a complete closed loop from the underlying hardware physical quantity to the actual transaction action: the text capture network data packet captured by the network card interface of the transaction server is converted into an initial price preference feature tag string carrying static text information in step S2; the meteorological telemetry signal output by the meteorological sensor acquisition node and the pulse sequence of the total power load output by the power grid dispatch gateway are calculated in step S3 to obtain the marginal unit physical bidding space quantity carrying the real-time supply and demand physical boundary; step S4 uses the real-time preference correction coefficient obtained by normalizing and nonlinearly mapping the marginal unit physical bidding space quantity to perform product correction on the initial price preference feature tag string, and then converts the corrected dynamic price quantification parameter into a buy-sell clearing price value and encapsulates it into an automatic order placement control instruction and sends it to the transaction gateway for execution. Throughout the entire data processing process, every intermediate variable is acquired by the preceding hardware interface or generated by the preceding steps and received and processed by the subsequent steps. There are no variables that appear out of thin air or are not used. The input end of the method is anchored to the real physical quantities acquired by three types of hardware: the transaction server network card interface, the meteorological sensor acquisition node, and the power grid dispatch gateway. The output end is grounded in the actual declaration action of the transaction gateway. This forces the long-cycle text patterns extracted by the natural language processing module to be constrained within the objective physical boundaries constructed by real-time meteorological telemetry signals and power grid load pulses, ensuring that the automatically generated transaction instructions always follow the rapidly changing objective meteorological and electricity consumption physical laws.

[0198] Example 2:

[0199] This embodiment, based on Embodiment 1, provides a system for quantitatively extracting market participant pricing preferences based on publicly disclosed data, such as... Figure 10 As shown, it includes:

[0200] The receiving and acquisition module is used to receive text capture network data packets through the network card interface of the transaction server, simultaneously acquire meteorological telemetry signals through the meteorological sensor acquisition node, and read the power load pulse sequence of the entire network from the power grid dispatch gateway.

[0201] Text segmentation and clustering module: used to perform natural language segmentation and clustering operations on the text-grabbing network data packets, extract objective state descriptive words related to cost changes and fuel inventory, and convert them into initial price preference feature tag strings;

[0202] Supply and demand difference calculation module: used to calculate the estimated value of new energy physical output based on the meteorological sensor telemetry signal, perform integral accumulation calculation on the pulse sequence of the whole network power load to obtain the overall physical demand of electricity, and subtract the estimated value of new energy physical output from the overall physical demand of electricity to generate the physical bidding space of marginal units;

[0203] Dynamic correction and order placement module: used to generate real-time preference correction coefficients based on the physical bidding space of the marginal units, adjust the initial bid preference feature marker string to obtain dynamic bid quantification parameters, and finally convert them into automatic order placement control instructions.

Claims

1. A method for quantitatively extracting market participants' pricing preferences based on publicly disclosed data, characterized in that, include: The transaction server network card interface receives text capture network data packets, simultaneously acquires meteorological sensing telemetry signals through meteorological sensor acquisition nodes, and reads the entire network power load pulse sequence from the power grid dispatch gateway. Natural language segmentation and clustering operations are performed on the text-grabbing network data packets to extract objective state descriptive words related to cost changes and fuel inventory and convert them into initial price preference feature tag strings; Based on the meteorological sensing and telemetry signals, the estimated value of the physical output of new energy is calculated. The overall physical demand of electricity is obtained by integrating and accumulating the pulse sequence of the total grid electricity load. The estimated value of the physical output of new energy is then subtracted from the overall physical demand of electricity to generate the physical bidding space of marginal units. Based on the physical bidding space of the marginal unit, a real-time preference correction coefficient is generated, and the initial bid preference feature tag string is adjusted to obtain dynamic bid quantification parameters, which are finally converted into automatic order placement control instructions.

2. The method for quantitatively extracting market participant pricing preferences based on publicly disclosed data according to claim 1, characterized in that, The process of receiving text capture network data packets through the transaction server's network card interface includes: Within a specified time window, the network interface card of the control transaction server establishes a Transmission Control Protocol (TCP) connection with the Internet Protocol address of the target competitor's web server, requests the target competitor to disclose the web page in accordance with Hypertext Transfer Protocol or Hypertext Transfer Security Protocol, and captures the response message byte stream returned by the target competitor's web page server frame by frame at the data link layer. If the captured response message byte stream is an encrypted transmission message, it is decrypted in real time to obtain a plaintext byte stream; All response message byte streams captured within a specified time window are concatenated in order of arrival time and encapsulated into a text-based network data packet capture.

3. The method for quantitatively extracting market participant pricing preferences based on publicly disclosed data according to claim 1, characterized in that, The synchronous acquisition of meteorological telemetry signals through meteorological sensor acquisition nodes includes: Each meteorological sensor acquisition node uses a unified sampling clock signal as a reference to acquire light intensity and wind speed pulse values ​​and record the sampling timestamp in each sampling sub-cycle of the current clearing period. The arithmetic mean of the light intensity values ​​of each sampling sub-cycle within the current clearing period is used to obtain the representative light intensity value of the geographical region. The arithmetic mean of the wind speed pulse values ​​of each sampling sub-cycle within the current clearing period is used to obtain the representative wind speed pulse value of the geographical region. The light intensity value and wind speed pulse value of the geographic region, which are represented by the sampling timestamp, are encapsulated into a meteorological sensing telemetry signal.

4. The method for quantitatively extracting market participant pricing preferences based on publicly disclosed data according to claim 1, characterized in that, The step of reading the entire network power load pulse sequence from the power grid dispatch gateway includes: According to the unified sampling clock signal, in each sampling sub-cycle of the current clearing period, the newly added pulse count value in the current sampling sub-cycle is read from the hardware data interface of the power grid dispatch gateway, and the reading timestamp is recorded. The pulse count values ​​read from each sampling sub-cycle during the current clearing period are arranged in order of reading timestamp to generate a pulse sequence of the entire network's power load.

5. The method for quantitatively extracting market participant pricing preferences based on publicly disclosed data according to claim 1, characterized in that, The step of performing natural language segmentation and clustering operations on the text-grabbed network data packets, extracting objective state descriptive words related to cost changes and fuel inventory, and converting them into initial price preference feature tags includes: The text-captured network data packets are parsed to remove the message control header and retain valid character entities; A pre-defined text clustering algorithm is used to perform feature word frequency statistics on the effective character entities, and objective state description words related to cost changes and fuel inventory are extracted. The objective state descriptors are converted into corresponding numerical labels according to a preset quantization mapping rule, and then sequentially concatenated to generate an initial price preference feature tag string.

6. The method for quantitatively extracting market participant pricing preferences based on publicly disclosed data according to claim 5, characterized in that, The step of using a pre-set text clustering algorithm to perform feature word frequency statistics on the effective character entities and extracting objective state descriptive words related to cost changes and fuel inventory includes: Perform Chinese word segmentation on the effective character entities, count the feature word frequency of each segment, and generate a feature word frequency vector; A pre-defined text clustering algorithm is used to cluster the feature word frequency vectors to generate a set of text clusters. From the text cluster set, match the cost fuel topic cluster and extract objective state descriptive words related to cost changes and fuel inventory.

7. The method for quantitatively extracting market participant pricing preferences based on publicly disclosed data according to claim 1, characterized in that, The calculation of the estimated physical output of new energy sources based on the meteorological remote sensing signals includes: The solar irradiance values ​​of each geographical region in the meteorological remote sensing signal are substituted into the photovoltaic power generation conversion formula to calculate the photovoltaic physical output component, and the wind speed pulse values ​​of each geographical region in the meteorological remote sensing signal are substituted into the wind power generation conversion formula to calculate the wind power physical output component. The summation of the photovoltaic physical output component and the wind power physical output component yields the estimated value of the new energy physical output.

8. The method for quantitatively extracting market participant pricing preferences based on publicly disclosed data according to claim 1, characterized in that, The physical bidding space for generating marginal units includes: The overall physical demand for electricity in the corresponding time period is obtained by integrating and accumulating the pulse sequence of the total power load of the entire network. The remaining supply-demand gap is obtained by subtracting the estimated physical output of new energy sources from the total physical demand for electricity. When the remaining supply-demand gap is greater than zero, it is used as the marginal unit physical bidding space to be undertaken by thermal power. When the remaining supply-demand gap is not greater than zero, the marginal unit physical bidding space is determined to be zero.

9. The method for quantitatively extracting market participant pricing preferences based on publicly disclosed data according to claim 1, characterized in that, The real-time preference correction coefficient generated based on the physical bidding space of the marginal unit includes: Extract the physical bidding space of the marginal units, divide the physical bidding space of the marginal units by the physical upper limit of the thermal power installed capacity stored in the form of a preset fixed constant parameter, and calculate the supply and demand tension approach rate. The supply-demand tension approach rate is nonlinearly mapped to generate a real-time preference correction coefficient. When the supply-demand tension approach rate is greater than a set threshold, the real-time preference correction coefficient is output in an exponentially amplified manner.

10. The method for quantitatively extracting market participant pricing preferences based on publicly disclosed data according to claim 9, characterized in that, The step of generating real-time preference correction coefficients by nonlinear mapping calculation of the supply-demand tension approximation rate includes: In the sub-threshold range where the supply-demand tension approach rate is no greater than the set threshold, the real-time preference correction coefficient is made to increase linearly with the supply-demand tension approach rate in a linear form. In the threshold range where the supply-demand tension approach rate is greater than a set threshold, an exponential term is superimposed on the linear value at the set threshold. The independent variable of the exponential term is the portion of the supply-demand tension approach rate that exceeds the set threshold, so that the real-time preference correction coefficient in the threshold range increases monotonically on the basis of the sub-threshold range.

11. The method for quantitatively extracting market participant pricing preferences based on publicly disclosed data according to claim 1, characterized in that, The process of adjusting the initial price preference feature marker string to obtain dynamic price quantification parameters, which are ultimately converted into automatic order placement control instructions, includes: The initial pricing preference feature string is multiplied by the real-time preference correction coefficient for numerical correction, generating dynamic pricing quantification parameters that eliminate static text bias; Based on the dynamic price quantification parameters, the buy and sell clearing price values ​​are directly calculated, encapsulated into an automatic order placement control instruction in a standard trading format, and sent to the trading gateway for execution.

12. A system for quantitatively extracting market entity pricing preferences based on publicly disclosed data, used to implement the method for quantitatively extracting market entity pricing preferences based on publicly disclosed data as described in any one of claims 1 to 11, characterized in that, The system includes: The receiving and acquisition module is used to receive text capture network data packets through the network card interface of the transaction server, simultaneously acquire meteorological telemetry signals through the meteorological sensor acquisition node, and read the pulse sequence of the power load of the entire network from the power grid dispatch gateway. The text segmentation and clustering module is used to perform natural language segmentation and clustering operations on the text-grabbing network data packets, extract objective state descriptive words related to cost changes and fuel inventory, and convert them into initial price preference feature tag strings; The supply and demand difference calculation module is used to calculate the estimated value of new energy physical output based on the meteorological sensing telemetry signal, perform integral accumulation calculation on the pulse sequence of the whole network power load to obtain the overall physical demand of electricity, and subtract the estimated value of new energy physical output from the overall physical demand of electricity to generate the physical bidding space of marginal units. The dynamic correction and order placement module is used to generate a real-time preference correction coefficient based on the physical bidding space of the marginal unit, adjust the initial bid preference feature tag string to obtain dynamic bid quantification parameters, and finally convert them into automatic order placement control instructions.

Images