A human-machine collaborative intelligent agent decision-making method and system for daily rolling matching of electricity transactions

Abstract

This invention discloses a human-machine collaborative intelligent agent decision-making method and system for daily rolling matching transactions in the power market, belonging to the field of power market transaction auxiliary decision-making technology. The method includes the following steps: collecting power spot market transaction data, user load data, and historical operation record data; uniformly storing and processing the collected data to form a basic data set containing market status information, load change information, and operational behavior information. This invention improves transaction efficiency and accuracy and reduces human error through an integrated process and automated execution mechanism; it achieves strategy evaluation and continuous optimization based on a quantitative model and log system; it constructs an embedded dynamic risk control mechanism to ensure the safety and controllability of the transaction process; and it enhances the system's adaptability and scalability to different trading environments through a layered architecture design.

Description

Technical Field

[0001] This invention relates to the field of power market transaction auxiliary decision-making technology, specifically to a human-machine collaborative intelligent agent decision-making method and system for daily rolling matching transactions of electricity. Background Technology

[0002] With the deepening of my country's power market reform and the accelerated development of the electricity spot market, intraday rolling matching transactions, as an important component of the spot market, allow market participants to continuously buy and sell electricity within a near-real-time time window, thus providing flexible adjustment tools for renewable energy consumption and load fluctuations. Under this trading model, the market clearing price is affected by multiple uncertainties, including real-time load changes, fluctuations in renewable energy (wind and solar) output, and grid congestion, exhibiting characteristics of rapid fluctuations, complex changes, and high predictability. Against this backdrop, market participants such as electricity retailers urgently need to utilize efficient and intelligent trading support tools to improve their responsiveness to market changes and decision-making efficiency.

[0003] Existing technologies mainly include a human-driven multi-tool combination mode and a simple programmatic trading interface mode, both of which have significant shortcomings: In the human-driven multi-tool mode, processes such as load forecasting, electricity price forecasting, strategy formulation, and transaction submission are scattered across multiple independent systems. Data flow relies on manual operation, resulting in fragmented workflows, low efficiency, and susceptibility to human error. At the same time, strategies rely on experience and are difficult to quantify and backtest, and risk control mainly relies on ex-post or manual judgment, lacking real-time and systematicity. On the other hand, the programmatic trading mode suffers from strong dependence on the trading center API, tight coupling between strategy and execution, and a lack of flexible adjustment capabilities. Moreover, its risk control mechanism is mostly based on simple threshold control, which is difficult to adapt to dynamically changing market rules. It also lacks effective human-machine collaboration and emergency interruption mechanisms, posing a risk of loss of control in extreme market conditions, and is unable to meet the requirements of high real-time performance, security, and refined control in rolling matching transactions of electricity.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a human-machine collaborative intelligent agent decision-making method and system for daily rolling matching transactions of electricity, so as to solve the problems in the background art mentioned above.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a human-machine collaborative intelligent agent decision-making method for daily rolling matching transactions of electricity, comprising the following steps: Collect historical data of medium- and long-term rolling matching transactions of electricity, historical load data of agent users, historical meteorological data, historical data of new energy output and thermal power bidding space, and uniformly store and process the collected data to form a historical data set that includes market status information, load change information and user private output data. The system performs load forecasting, bidding space forecasting, and electricity price forecasting on historical data sets, and generates standardized trading strategy instructions based on the forecast results, including the trading target day, time period, trading direction, target price, and matched volume. For standardized trading strategy instructions, the system obtains real-time rolling matching trading price and transaction information, calculates the difference between the total matching target volume and the already traded volume to obtain the current remaining operating power, and generates corresponding trading declaration decisions based on changes in the trading price. During the execution of transaction declarations, the transaction declaration decisions are subject to constraints on declaration volume, price range, declaration frequency, and market validity. Based on the verification results, automatic declarations or manual confirmations are completed. At the same time, all operational information during the transaction execution process is recorded for subsequent review and analysis.

[0007] Regarding the handling of proportional relationships in the process of decomposing new energy output, stable load forecasting results are formed by analyzing and combining different types of output data. The steps are as follows: The system receives total renewable energy data (weekly) and wind power day-ahead output data. It performs timestamp matching on the two types of data, divides them into 96 continuous time series data points according to 15-minute intervals, and performs interpolation to complete the remaining time points. Then, it calculates the difference between the two types of data to form an estimated photovoltaic output data sequence. Read the day-ahead photovoltaic power output data, calculate the ratio of it to the estimated photovoltaic power output data hour by hour, generate a photovoltaic proportional coefficient sequence, and perform screening processing on abnormal proportional values, including setting upper and lower limits of the proportional and removing data points that exceed the range. The proportional coefficient series is grouped according to the date dimension, and the mean is calculated, the discrete points are smoothed and the continuity is corrected for each time segment to form stable proportional curve data. Perform trend scanning processing on the proportional curve to extract the starting point, peak point and ending point of change, construct the corresponding time index set, and output the load forecast data sequence.

[0008] By combining the time series variation characteristics, and through key node extraction and data fitting processing, a continuous wind power output variation trend curve is constructed. The steps are as follows: Perform a full traversal analysis on the scale curve, identify the start point, end point and peak point by detecting changes in the curve slope, and record the corresponding time index and scale value; Based on the identification results, read the new energy output data and wind power ratio data at the corresponding time points, and map the two types of data in a unified manner to construct a key node data set; A weighted combination calculation is performed on the set of key nodes, the wind power output value of each node is derived according to the proportional relationship, and a set of discrete wind power output data points is formed by inter-node interpolation. Curve fitting is performed on discrete data points to generate a continuous wind power output trend curve, and the complete output sequence is output in chronological order.

[0009] Based on the correlation between changes in new energy output and historical bidding space data, the future bidding space is derived and calculated. The steps are as follows: Extract the target date bidding space data and read the corresponding date new energy output data. Perform unified time granularity processing on the two types of data and construct a data alignment sequence. Differential calculations are performed on the continuous date renewable energy output data to obtain a time-by-time output change sequence, and the changes are marked as positive or negative. The output change is mapped to the bidding space data, and the original bidding space data is corrected time by time through addition and subtraction operations to form the bidding space calculation result for the target day. The calculation results are subjected to continuity verification processing, including outlier correction and sequence smoothing, and a complete bidding space prediction curve is generated.

[0010] Based on the price differences between different markets, trading signals are generated by comparing multiple price data. The steps are as follows: Read medium- and long-term price data and day-ahead clearing forecast prices, and perform time alignment and format standardization processing on the two types of data; Perform a comparison operation at each time point, extract the minimum value from the medium- and long-term prices and the day-ahead clearing forecast prices, and generate a reference price series. Acquire real-time price data from the matching market, and calculate the difference between it and the reference price every hour to form initial price difference data; Perform sign recognition and numerical output processing on the price spread data to generate a standardized price spread signal sequence.

[0011] Based on the changes in electricity volume during the transaction execution process, an operable electricity control mechanism is calculated using the difference between the target electricity volume and the transaction volume. The steps are as follows: Extract the total matching target from the strategy instructions and divide it into multiple time-segment target sequence according to the time dimension; Read the current real-time market transaction data, perform cumulative calculations on the declared and traded electricity volumes, and form a sequence of currently traded electricity volumes; The difference between the total matching target and the already traded volume is calculated at each time step to obtain the sequence of remaining eligible volumes. The remaining declared electricity is mapped to the current trading cycle to limit the range of declared electricity and update the execution status data.

[0012] Based on the dynamic changes in the remaining available electricity volume sequence, the target volume sequence for each time period is synchronously corrected, and the correction result is matched and verified with the current traded electricity volume sequence to generate an updated operable electricity volume allocation result. At the same time, the allocation result is linked to the transaction declaration process to form continuous execution control data.

[0013] Based on the price fluctuation characteristics and monitoring parameter configuration, a corresponding trading strategy is formed by determining the price range. The steps are as follows: Receive monitoring price parameters and simultaneously read real-time buyer and seller price data that meet the reporting rules to construct a set of order book price data; Compare the monitored price with the buyer's price, and output the result of the corresponding relationship judgment. The position of the monitored price within the range between the buyer's price and the seller's price is determined, forming a range judgment result; The comparison results and the interval judgment results are combined to generate the declared price data and the transaction direction data.

[0014] Based on the multi-dimensional data conditions in the transaction declaration process, consistency checks are performed on the declaration behavior using multiple verification rules. The steps are as follows: The scope of the declared electricity volume is tested to determine whether it meets the minimum and maximum declared volume requirements, and the test results are marked. Perform range matching on the declared prices to determine whether they are within the preset price range and record the verification status; Perform statistical calculations on the number of applications within a unit of time, compare the statistical results with a preset threshold, and output the frequency detection results; The order book volume is filtered to identify orders below a preset threshold, and the corresponding price signals are filtered to form the final verification result.

[0015] A human-machine collaborative intelligent agent decision-making system for daily rolling matching transactions in the power sector includes a data acquisition and processing module, a prediction and strategy generation module, an automatic monitoring and reporting module, and a data and log module. Data acquisition and processing module: used to collect historical data of long-term rolling matching market transactions in the spot electricity market, historical load data of agent users, historical meteorological data, new energy output data, historical data of thermal power bidding space and historical operation record data, and to perform unified processing on the collected data to form a basic historical data set containing market status information, load change information, new energy output information and operation behavior information. The prediction and strategy generation module is used to replace manual multi-tool analysis and provide data-driven decision-making. It performs decoupling of new energy output, calculation of photovoltaic proportional coefficient and curve fitting processing, generates estimated wind power output curves and photovoltaic output curves, calculates the target day bidding space and makes time-segmented adjustments, further performs quantitative prediction of electricity price based on bidding space and new energy output, and constructs strategy decision-making logic based on recycling rules and price difference signals, outputting standardized strategy instructions including target day, time period, trading direction, target price and matched volume. The automatic market monitoring and reporting module displays market price information, electricity price forecasts, and changes in open positions on a visual monitoring screen, and provides strategy configuration and monitoring parameter settings. It executes standardized strategy commands and includes a monitoring module, a decision-making module, and a reporting module. The monitoring module acquires market data and tracks the total matching target volume and the volume already traded to calculate the open electricity volume. The decision-making module generates transaction reporting decisions based on real-time prices using unidirectional or bidirectional monitoring logic. The reporting module completes transaction reporting and performs verification of the reported volume, price, frequency, and market volume. It also supports semi-automatic mode, fully automatic mode, and one-click pause. The data and log module is used to record strategy reception, market data, decision-making process, reporting instructions and execution results, and to perform transaction review and profit attribution analysis based on the recorded data.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention significantly improves the efficiency and accuracy of transaction execution. By constructing an integrated workflow system that connects data acquisition, strategy generation, and automated execution, it eliminates inefficient steps such as repeated data switching and manual data transfer between different systems, enabling continuous and automated operation of the transaction process. Traders no longer need to perform tedious manual data entry, allowing them to focus their main efforts on higher-value aspects such as strategy monitoring, parameter optimization, and anomaly handling. Simultaneously, the automated data entry mechanism, combined with multi-level verification logic, rigorously checks key parameters such as electricity volume, price, and frequency before submission, preventing errors caused by human intervention and significantly improving the stability and accuracy of transaction execution.

[0017] This invention realizes a scientific transformation of trading strategies from experience-driven to data-driven. By constructing a quantitative decision-making model and a full-link log recording system, the strategy formulation process becomes calculable, verifiable, and optimizable. Each strategy can be quantitatively simulated before execution and analyzed based on real data after execution, thus forming a complete evaluation loop. By continuously accumulating historical execution data and analysis results, strategy parameters can be systematically optimized, gradually transforming strategies from artistic judgments relying on personal experience to scientific decisions based on data feedback, effectively improving the stability, replicability, and long-term evolution capabilities of the strategies.

[0018] This invention constructs a proactive, dynamic, and reliable real-time risk control system, embedding risk control mechanisms into the entire transaction execution process, rather than relying on post-trade verification or static threshold limits. During execution, by continuously monitoring transaction progress and key state variables, the operable range is dynamically calculated and execution strategies are adjusted in real time, ensuring that risk control and trading behavior are synchronized. Especially in continuous matching trading scenarios, the remaining operable space can be continuously updated based on real-time transaction data, effectively avoiding the triggering of medium- to long-term position ratio constraints or buyback mechanisms due to the accumulation of multiple transactions, thereby keeping trading behavior within safe boundaries in a complex and rapidly changing market environment.

[0019] This invention enhances the system's adaptability and scalability to heterogeneous trading environments. By designing execution capabilities as independent layers, it allows for flexible adjustments to implementation methods based on the interface conditions of different trading platforms. When the trading environment provides a standard interface, direct system integration enables automated execution; when only a graphical interface is available, operations can be completed through simulated interaction, ensuring consistent execution results under various technical conditions. This layered structure makes strategy generation and execution processes independent, allowing upper-layer strategy logic to be quickly migrated to different markets or platforms, reducing system modification costs while improving overall scalability and application flexibility. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0021] Figure 1 This is a flowchart of a human-machine collaborative intelligent agent decision-making method for daily rolling matching transactions of electricity, according to the present invention.

[0022] Figure 2 This is a schematic diagram of a human-machine collaborative intelligent decision-making system for daily rolling matching transactions of electricity, according to the present invention. Detailed Implementation

[0023] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0024] This invention provides, for example Figure 1The human-machine collaborative intelligent agent decision-making method for daily rolling matching transactions of electricity is shown below, with the following specific steps: The load forecasting process based on similar days and proportional coefficients aims to address the instability in load and electricity price forecasts caused by fluctuations in renewable energy output. Since renewable energy output (especially wind and solar power) exhibits significant randomness and volatility, it directly impacts power system load balance and market price formation. Therefore, it is necessary to refine the modeling of renewable energy output through data decomposition, proportional modeling, and curve fitting to provide a reliable foundation for subsequent bidding space estimation.

[0025] New energy output decoupling and data preprocessing: First, during the decoupling process of new energy output, weekly total new energy data is obtained from public or private data sources. This data represents the total power output of wind power and other new energy sources such as solar power within the statistical period. It also includes day-ahead wind power output data. This data represents the wind power output that has been confirmed in the market recently.

[0026] By calculating the difference between the two, the estimated photovoltaic power output data for subsequent analysis is obtained. The calculation relationship is as follows: through minus get .in: : Indicates the total output of new energy sources; : Indicates wind power output; : Indicates the estimated photovoltaic output.

[0027] This process achieved the initial separation of wind power and photovoltaic output, providing a data foundation for subsequent proportional modeling and effectively eliminating coupling interference caused by mixed data sources.

[0028] Photovoltaic scaling factor modeling and robustness enhancement: In the process of modeling the photovoltaic scaling factor, the actual daily output of photovoltaic power is calculated. and The ratio between these two values ​​yields the actual photovoltaic proportional gain at each time point. Specifically: This indicates the actual photovoltaic output in the market at the current time. : Indicates the estimated photovoltaic output.

[0029] The ratio of these two factors reflects the proportion of photovoltaic power output in the total output of new energy sources. Subsequently, the historical average of this ratio is calculated by day and by time period (e.g., a 96-point discrete time series) to obtain stable statistical results, which are then adopted as the photovoltaic ratio. .

[0030] This coefficient This method is used to characterize the typical proportion of photovoltaic power output in total renewable energy output, effectively reducing the impact of daily fluctuations and making the model more robust and generalizable. In engineering applications, this step can be further improved by introducing sliding window averaging or weighted exponential smoothing algorithms to address the impact of seasonal variations on the proportionality coefficient.

[0031] Curve fitting and output reconstruction mechanism: During the curve fitting and output reconstruction stage, in the proportional coefficient curve Identify key feature points, including start point, end point, and peak point. , , By extracting the historical renewable energy output and wind power output data corresponding to these feature points, a weighted average method is used to fit and calculate the key points.

[0032] The key calculation relationship is as follows: In this formula: point, point, Point: Represents the key moment point selected in the proportional coefficient curve; Point B and Point C: Represent the total renewable energy output at the corresponding moment; Point A, Point B, and Point C: Represent the proportion of wind power output in renewable energy output at the corresponding moment.

[0033] Using the above calculation method, The wind power output of the point is estimated, and combined with... Point and Using known data points, a quadratic function curve opening upwards is obtained through fitting. This curve describes the changing trend of wind power output over a periodic scale, accurately capturing the nonlinear characteristics of wind power output and providing mathematical support for high-resolution time series forecasting.

[0034] Generation and detailed representation of 96-point output curve: During the generation of the 96-point power output curve, the fitted quadratic function curve is used to decompose the total renewable energy output data for the forecast day and the day before into two 96-point predicted wind power output curves. and .in: This represents the predicted daily wind power output curve; This represents the wind power output curve for the previous day.

[0035] Further based on total power output data of new energy sources and Calculate the corresponding photovoltaic power output curve: in: and These represent the total output of new energy sources on the forecast date and the previous day, respectively. and These represent the corresponding photovoltaic power output curves.

[0036] This process enables a more detailed representation of new energy output over time, transforming macroscopic total data into microscopic time-specific data, providing high-precision foundational data support for subsequent bidding space calculations. In patented implementation, this step can be combined with meteorological forecast data (such as irradiance and wind speed) to... and Dynamic corrections are made to further improve the accuracy of short-term forecasts.

[0037] Bidding Space Calculation and Cross-Day Linkage Model: During the bidding space calculation phase, read Daily 96-point bidding space data This data reflects The available electricity volume for market trading at different times of the day. This is combined with the aforementioned calculations. Estimated output of new energy sources and Estimated output of new energy sources Calculated using the following formula Daily bidding range: in: express Daily bidding space; express Forecasted daily renewable energy output; express Forecasted daily renewable energy output; Indicate target Daily bidding space.

[0038] The physical meaning of this formula is that changes in the bidding space are primarily driven by changes in renewable energy output. When renewable energy output increases, conventional generating units need to relinquish more market share, leading to a decrease in the market's tradable space; conversely, it increases. Thus, by using the difference relationship, the inter-day bidding space can be calculated, constructing a dynamic mapping mechanism between renewable energy output and market capacity.

[0039] Human experience fusion and interactive correction: During the integration of human experience, an interactive interface is provided to receive traders' expected volatility for specific time periods, such as adjustments needed for morning peak, mid-load, and evening peak periods. Based on this input, the system adjusts... The curve is adjusted in different time periods.

[0040] This approach ensures that the prediction results incorporate both objective calculations from the data model and practical trading experience, thereby generating predictions that better reflect actual market conditions. Bidding space prediction curve. This process effectively compensates for the shortcomings of pure data models in abnormal situations such as extreme weather, equipment maintenance, or sudden policy changes, improving the practicality and reliability of the prediction results. This human-machine collaborative model is an important part of engineering implementation, ensuring the adaptability of the algorithm model in real-world complex environments.

[0041] Through the coordinated processing of the above-mentioned multiple stages, a complete derivation process from raw renewable energy output data to bidding space prediction results was achieved. This process not only preserves the physical and statistical characteristics of renewable energy output, but also improves prediction accuracy through proportional coefficients and curve fitting methods, and enhances the model's adaptability through a human experience fusion mechanism.

[0042] Ultimately, the system provides a stable and reliable data foundation for subsequent electricity price forecasting and trading strategy formulation, significantly reducing market risks caused by fluctuations in new energy sources, and has extremely high value for promotion and application and potential for patent protection.

[0043] In-depth analysis of the quantitative forecasting process for D+2 electricity prices based on bidding space and renewable energy output. This technical process revolves around the impact of changes in renewable energy output on market bidding space and electricity prices. It aims to achieve a quantitative analysis of the basis for electricity price formation on the target day (D+2) through mathematical methods such as proportional coefficient modeling, key point identification, curve fitting, and time-period correction. At the data level, the process uses weekly renewable energy output as a benchmark and, through refined processing at each time period (96 points), transforms macro-level output data into high-precision input that can be directly used for electricity market decision-making.

[0044] The entire derivation path strictly follows the engineering logic of data cleaning, feature extraction, model fitting, spatial calculation, and empirical correction, ensuring that the prediction results have both the objectivity of data-driven approaches and the practicality of market experience.

[0045] Construction and correction of the photovoltaic scaling factor: First, differential operations are performed on the basic fields in the power output data source to obtain the intermediate variable used to calculate the photovoltaic proportional coefficient—the estimated photovoltaic power output. The calculation formula is as follows: It represents the total output of new energy sources in the statistical week containing the target date D+2, and its physical meaning includes the sum of wind power, photovoltaic power and other new energy sources.

[0046] This indicates the wind power output value that has been locked in and confirmed in the market as of D+1 day prior.

[0047] Its physical meaning is that, under the constraint of known wind power output, the estimated scale of photovoltaic output is derived by working backward from the total amount, and it serves as the benchmark denominator for subsequent proportional calculations.

[0048] Subsequently, the estimated values ​​were calibrated using historical real data, and the actual photovoltaic proportional coefficients for all times were calculated: Represents the corresponding moment in the market today. The actual output of photovoltaic power reflects the true historical operating level.

[0049] Correction mechanism: By comparing actual values ​​with estimated values The ratio of the actual photovoltaic output to the theoretical estimate quantifies the model error. This coefficient directly reflects the deviation between the actual photovoltaic output and the theoretical estimate, and is used to correct systematic errors in the model in subsequent steps, making the calculation closer to the actual operating conditions of the power grid.

[0050] To eliminate the interference of random fluctuations, historical data from two consecutive weeks were selected as the sample interval. Statistical averaging was performed on a daily basis and over 96 discrete time periods (with a time resolution of 15 minutes) to obtain a stable photovoltaic proportion coefficient. : This coefficient It depicts the typical proportion pattern of photovoltaic power output in the total output of new energy sources, effectively smooths out the noise caused by daily meteorological fluctuations, and significantly improves the robustness of the prediction model.

[0051] Key feature point identification and wind power output curve fitting: In time series In the process, the boundaries of the active photovoltaic power generation range are identified: Non-zero intervals: The consecutive periods when the median value is not 0 correspond to the active periods of photovoltaic power output during the day.

[0052] Boundary point definition: Find the first endpoints with a value of 0 on both sides of the non-zero interval, and denote them as point B (sunrise boundary) and point C (sunset boundary) respectively.

[0053] Physical significance: Points B and C mark the moments when photovoltaic power output starts from zero and returns to zero, which are key nodes for dividing the day and night modes.

[0054] Read 96 data points from the wind power ratio coefficient sequence, locate the minimum point, and denote it as point A.

[0055] Physical significance: Point A usually corresponds to the moment when wind power output accounts for the lowest proportion in the new energy structure. It is a key reference anchor point for describing the intraday fluctuation characteristics of wind power and also the core control point for subsequent quadratic curve fitting.

[0056] For a given target date (e.g., 20260316), based on known data from points B and C, the weekly wind power output at point A is derived using weighted interpolation. The calculation formula is as follows: in: The wind power output at point A (to be determined). The weekly total energy output corresponding to points B and C are respectively; The wind power ratio coefficients (i.e., the proportion of wind power output to new energy output) are for points A, B, and C, respectively.

[0057] Derivation logic: This formula uses the relationship between the total amount of new energy and the proportion of wind power at points B and C, and through proportional scaling and weighted combination, derives the wind power output benchmark at point A, thereby constructing a complete wind power output feature set for points A, B, and C.

[0058] Given that the weekly wind power output at points B and C equals their weekly renewable energy output (since photovoltaic power is zero at this time), and combining this with the derived data for point A, we can utilize these three points... Fit a quadratic function curve that opens upwards: This curve depicts the trend of wind power output over time. Based on this curve and weekly renewable energy output data for a specified date... (Target date) and (The previous day) Two predicted wind power output curves with 96 points each were generated. and : Periods B to C: Estimated wind power output is calculated based on the fitted curve.

[0059] For other periods: the estimated wind power output is directly equal to the weekly renewable energy output for the corresponding date.

[0060] Furthermore, the estimated photovoltaic power output curve is derived by applying the principle of total conservation. and : in, and The photovoltaic output curves representing the target day and the previous day, respectively, achieve complete decoupling of the new energy structure.

[0061] Introducing a proportionality coefficient After correcting the photovoltaic component, the final estimated renewable energy output curve is synthesized: In this expression, For the fitted wind power output, This represents the proportionally adjusted photovoltaic output. The two figures are superimposed to form a complete estimated renewable energy output curve containing data from two days, preserving the fitting trend of wind power while correcting for prediction biases in photovoltaics.

[0062] D+2 Bidding Space Calculation and Time-Based Adjustment: During the bidding space calculation phase, read the 96-point data column D+1 bidding space (denoted as...). ), combined with estimated output (D+1) and (D+2), the bidding space of D+2 is derived through the following equilibrium equation. : Parameter definition: It refers to the market bidding space on day D+1. This is the estimated output of new energy sources on D+1 day. The estimated output of new energy sources is for D+2 day. It is the target bidding space for D+2 days.

[0063] Market Mechanism: This formula profoundly reflects the negative correlation between new energy output and market potential. When the output of new energy in D+2 increases ( When this happens, it encroaches on the space of conventional generating units, resulting in a reduction in bidding space. );vice versa.

[0064] To integrate expert experience with actual market operation characteristics, a manually set fluctuation value is introduced. The estimated data is adjusted according to the peak, off-peak, and valley periods: Through the aforementioned segmented adjustment mechanism, the model output can flexibly respond to market fluctuation expectations under different load characteristics, ultimately generating a high-precision D+2 bidding space curve.

[0065] Through the above process, a complete derivation path was realized from new energy output data to bidding space and then to basic data for electricity price prediction, so that the prediction results have both data-driven and experience-corrected characteristics, improving the overall prediction accuracy and practicality.

[0066] The core of the quantitative process for matching strategies based on buyback rules and price spread signals lies in transforming trading behavior, which originally relied on experience-based judgment, into a quantitative decision-making mechanism with clear input-output relationships. This enables the trading strategy to maintain consistency and executability under different market conditions. With profit maximization as its objective, and considering market rule constraints, the process achieves a systematic expression of trading behavior through profit model construction, buyback boundary definition, price spread signal calculation, and the establishment of a strategy logic decision matrix.

[0067] In the process of constructing the profit model, the actual revenue model of the electricity sales side in the rolling matching transaction is first defined as follows: In this expression, This represents the actual profit outcome under a given transaction state; This represents the amount of electricity locked in through medium- to long-term contracts; this variable remains relatively stable in the short term. This represents the volume of electricity that has been traded in the rolling matching market, and this variable changes in real time as the trading process progresses; This indicates the electricity price level corresponding to medium- and long-term contracts; This indicates the current transaction price in the matching market; This indicates the market clearing price at the recent close; This represents a predicted value for users' future electricity demand. The model reflects the relationship between revenue and multiple factors, among which... It is a static variable, while As a dynamic variable, it is this dynamic variable that determines the need for real-time adjustments and refined decision-making in trading strategies.

[0068] In determining the recovery boundaries, the upper and lower limits of recovery for each period are calculated based on the operating rules of the spot market and load forecasts. The upper limit of recovery is determined as follows: The lower limit of recycling is determined as follows: Here, load forecast represents the predicted electricity load for various future time periods, with 115% and 55% being the proportional thresholds set in market rules. Based on the above calculations, the current position is divided into three intervals: above the upper limit of the recovery limit, below the lower limit of the recovery limit, and between the upper and lower limits. This division method is used to determine whether the current position has recovery risk or arbitrage potential, thus providing constraints for subsequent strategy selection.

[0069] During the calculation of the price spread signal, the price spread signal is calculated in real time, and its expression is as follows: in, This indicates a price spread signal, used to measure the degree of deviation of the current matching market price from the reference price; Indicates the price of medium- to long-term contracts; This indicates a forecast of the day-to-day clearing price; This means selecting the lower of the two values ​​as the benchmark price, thereby ensuring that the strategy still has room for profit under conservative conditions; This represents the current transaction price in the matching market. This calculation method yields the difference between the matching price and the reference price, thus determining whether the current market price has trading value.

[0070] In the process of constructing the strategy logic decision matrix, the recovery boundary and price spread signals are combined and analyzed to form decision logic under different market conditions. Specifically, this includes the following scenarios: Scenario 1: If current holdings > maximum buyback limit and In this scenario, current open interest has exceeded the redemption limit, and the matching market price is lower than the reference price, indicating a low-price buying opportunity exists in the market. The optimal strategy here is to sell medium- to long-term contracts and buy in the matching market to reduce overall electricity purchase costs. In this scenario, the profit... The conditions are: (Buy) in, This represents the trading volume in the matching market; a positive value indicates that a buy operation has been executed.

[0071] Scenario 2: If the current holdings are less than the lower limit of the buyback and In this scenario, the current open interest is below the lower limit of the buyback, while the matched market price is higher than the reference price, indicating a high-price selling opportunity exists in the market. The optimal strategy here is to buy medium- to long-term contracts and sell them in the matched market to profit from the price difference. In this scenario, the profit... The conditions are: (Sold) in, A negative value indicates that a sell operation has been performed.

[0072] Scenario 3: If the position is between the upper and lower limits of the buyback, then it is necessary to... The sign and value of the sign are comprehensively judged. In this case, calculation is performed to make... Required The process involves determining the quantity and direction of the transaction, while simultaneously assessing the potential recovery risks associated with the operation, to decide whether to execute the transaction and the scale of electricity to be traded. This process reflects the trade-off between risk and reward in the strategy.

[0073] During the strategy output standardization process, based on the aforementioned decision-making logic, the calculation results are transformed into a structured strategy instruction data template. This template is in the form of a standardized data table, including at least the following fields: target date, time period, trading direction (buy or sell), target price, and total matching volume. Specifically, the target date represents the date corresponding to the transaction, the time period represents the specific time interval, the trading direction indicates whether to execute a buy or sell operation, the target price represents the expected transaction price, and the total matching volume represents the planned trading volume to be executed in the matching market. This standardized output method allows the strategy to be directly invoked in subsequent execution processes, thus achieving a seamless transition from decision-making to execution.

[0074] Through the above process, a complete quantitative path from profit modeling to strategy output is realized, enabling trading behavior to shift from experience-driven to data-driven, while ensuring the executability and risk controllability of the strategy under the constraints of market rules.

[0075] The intelligent agent module, as the core execution part of the entire system, is responsible for the entire execution process from strategy instructions to actual trading operations. This part operates as an independent software entity, and its operational logic strictly adheres to the principle of rule supremacy; that is, all operations must meet preset rule constraints to ensure the standardization, security, and consistency of trading behavior. During overall operation, it continuously senses changes in the market environment and internal states to achieve dynamic response and refined control.

[0076] Market Monitoring Module: This module is responsible for continuously sensing the market environment and internal status, and is the basic data source for realizing dynamic decision-making.

[0077] The latest publicly available market data can be obtained in real time from the power trading center system through simulation or API calls. This data includes, but is not limited to: the best bid and ask prices, the latest transaction prices, weighted average prices, and order book volume for each underlying asset on a daily and hourly basis. These data together constitute a real-time snapshot of the market condition.

[0078] We continuously monitor and update several key status variables, including the total matching target volume, completed volume, pending matching volume, and unmatched volume for each target day and time period. The continuous updating of these variables allows the system to monitor the current strategy execution progress in real time.

[0079] The core variable, exposed electricity, is calculated in real time, and its calculation relationship is as follows: In this formula: the open volume represents the trading volume that still needs to be completed at the current moment, which is a key indicator for measuring the remaining task volume of the strategy execution; the total matching target volume represents the total target trading volume initially set by the strategy; and the completed volume represents the trading volume that has been completed up to the current moment.

[0080] By calculating this difference, the remaining amount of electricity to be executed can be accurately determined, thus avoiding problems such as over-trading or under-execution. This variable is dynamically updated throughout the execution process and is the core foundation for achieving refined control.

[0081] The decision-making module, as the core decision-making unit of the intelligent agent, is responsible for transforming policy instructions into specific reporting behaviors.

[0082] It receives standardized strategy instructions generated from the forecasting and strategy section, and parses the total matching target for each target day and each time period to clarify the execution target for each time slice.

[0083] Decision-making is based on the interface configuration parameters (including unidirectional or bidirectional, price fluctuation range, etc.) and real-time market data.

[0084] One-way market monitoring: Taking buying as an example, the bid price is continuously monitored. When the bid price is lower than the target monitoring price, an order is triggered. At this time, the order price is: > bid price + price fluctuation range. The price fluctuation range is a configurable, small positive number that slightly increases the quoted price based on the current market price to improve the probability of a transaction.

[0085] Monitoring status of both parties: Taking buying as an example again, the decision-making logic can be divided into three cases: When the monitored price is lower than the best bid price: it indicates that the market price is higher than expected, the condition is not met, and therefore the order will not be executed; When the bid price is less than or equal to the monitored price and the monitored price is less than or equal to the ask price: it indicates that the market price is within an acceptable range. In this case, submit a bid at the bid price plus the floating value to increase the probability of a transaction. When the monitored price is higher than the best ask price, it indicates that the market price is better than expected. In order to complete the transaction as soon as possible, submit an order at the best ask price.

[0086] This logic can automatically adjust the pricing strategy according to market changes, thereby avoiding frequent manual price modifications and improving execution efficiency.

[0087] The application module is responsible for translating decision results into actual transaction operations and embedding multi-level risk control mechanisms during the execution process.

[0088] Includes two modes: semi-automatic and fully automatic. Semi-automatic mode: The system automatically fills in the transaction form, including the target day, time period, electricity volume and electricity price information, and highlights it. Then the operation is paused and the system waits for manual confirmation. Fully automatic mode: Under the condition that all risk control conditions are met, the system automatically completes the form filling and submission operations, thereby achieving fully automated execution.

[0089] Several mandatory checks must be conducted before the application is submitted for implementation: It provides a global one-click pause function. When this function is triggered, the system immediately interrupts all automatic execution processes and enters a safe state, waiting for manual takeover, thereby providing security in extreme market conditions.

[0090] The above structure enables a complete closed loop from market perception, status tracking, decision generation to execution control, making trading activities real-time, controllable, and highly reliable.

[0091] Human-computer interaction and monitoring enable human-in-the-loop collaboration through a visually appealing large-screen monitor. The core objective is to introduce a human intervention mechanism into the automated trading execution system, allowing traders to maintain continuous awareness and intervention capabilities regarding market conditions, position status, and strategy execution during system operation, thereby establishing an effective balance between automation and human control. Through this module, traders can not only obtain real-time market information but also adjust system behavior at key points, thus improving the overall security and controllability of the trading system.

[0092] Regarding the monitoring dashboard functionality, the interface clearly displays daily and time-period market information for each asset (best bid / best ask / latest price), day-ahead electricity price forecasts, and confidence intervals. Specifically, the interface presents market data in a structured format, where the best bid price represents the highest current buy price, the best ask price represents the lowest current sell price, and the latest price represents the most recent transaction price. This price information reflects the real-time state of market supply and demand. Meanwhile, the day-ahead electricity price forecast demonstrates the judgment on future electricity price trends, while the confidence interval expresses the range of uncertainty in the forecast, thus helping traders understand the reliability of the forecast results.

[0093] The position monitoring section displays in real-time the user load forecast curve (actual fitted curve), the medium- and long-term contract curve, and the currently dynamically changing position ratio curve. The user load forecast curve reflects the changing trend of future electricity demand, and its actual fitted curve represents the forecast result obtained by fitting historical data; the medium- and long-term contract curve represents the distribution of electricity locked through medium- and long-term agreements; and the position ratio curve reflects the ratio of currently held electricity to load demand. By comparing and displaying these three types of curves, traders can intuitively determine whether the current position status deviates from a reasonable range, thereby identifying potential risks.

[0094] This highly integrated visualization method presents market information, forecasts, and position information on a single interface, enabling traders to make overall judgments without switching between multiple systems. Through clear and concise graphical representation, it significantly reduces information acquisition costs and improves decision-making efficiency.

[0095] In addition, the batch configuration and operation area provides a user-friendly interface for uploading standardized strategy templates and configuring monitoring parameters (direction, floating price, order limits, etc.). This area primarily handles strategy configuration and execution parameter settings. Standardized strategy templates exist in structured data format, containing key fields such as target day, time period, trading direction, target price, and trading volume. Traders can upload these templates through the interface, allowing the system to directly load and execute the strategy. Monitoring parameter configuration includes trading direction (buy or sell), price fluctuation range, and order limit conditions. These parameters are used to adjust the system's behavior during execution.

[0096] In terms of configuration, it supports both daily and full configuration to adapt to different team collaboration modes. Daily configuration is suitable for scenarios requiring fine-tuning for a specific date, such as temporary strategy optimization for special market conditions; full configuration is suitable for batch processing of multiple trading days, improving operational efficiency. In team collaboration mode, different roles can configure and adjust strategies according to their permissions, thereby achieving division of labor and collaboration.

[0097] During the overall operation, it also serves the functions of information aggregation and status feedback. By updating interface data in real time, traders can continuously obtain the system's execution status, including strategy execution progress, completed transactions, and the amount of remaining tasks to be executed, thereby achieving dynamic monitoring of the system's operation.

[0098] At the human-machine collaboration level, the large-screen monitoring platform serves as the core platform, effectively connecting automated execution with human decision-making. The automated system handles high-frequency, repetitive operations, while humans use this interface to set key parameters and intervene when necessary. This collaborative approach ensures efficiency while avoiding the risks associated with full automation.

[0099] From an engineering implementation perspective, it needs high-concurrency data processing capabilities to support real-time market data updates, as well as excellent interface responsiveness to ensure smooth user operation. Furthermore, the interface display needs to support synchronous updates of multi-dimensional data to ensure consistency between different pieces of information.

[0100] Through the above structural and functional design, unified monitoring and management of market conditions, strategy execution status, and user position status are achieved, enabling traders to maintain a clear understanding in complex market environments, thereby improving the reliability and practicality of the overall trading system.

[0101] The data and log module is responsible for the unified management, recording, and analysis of data throughout the entire intelligent agent's operation. It is a crucial component in ensuring the system's traceability, interpretability, and optimizability. By structurally recording and subsequently analyzing the data across the entire process, transactions not only possess execution capabilities but also the ability to backtrack and continuously optimize, thus forming a closed-loop operating mechanism.

[0102] Regarding the end-to-end operation logs, all events throughout the agent's lifecycle are recorded in structured formats such as JSON. The logs cover the entire process from policy reception to final execution, including: when the policy was received, market data snapshots, decision-making basis, generated reporting instructions, risk control check results, submission time, and return status (success / failure / error code). All of this information is stored in structured fields, ensuring good data parsing and scalability for easy subsequent retrieval and analysis.

[0103] Among them, "When to Receive Strategy" identifies the starting time of the strategy's execution process, reflecting the timeliness of the strategy's response; "Market Data Snapshot" records the market state at the moment the decision occurs, including key information such as price and order volume, thus restoring the external environment at that time; "Decision Judgment Basis" records the conditions or logical basis on which the declaration behavior is based, making each operation interpretable; "Generated Declaration Instruction" records specific transaction parameters, such as the target day, time period, electricity volume, and electricity price; "Risk Control Check Result" indicates whether the operation has passed various risk verification conditions; "Submission Time" records the time when the declaration instruction is sent to the trading system; and "Return Status" indicates the transaction execution result, including success, failure, or specific error code.

[0104] Each log entry includes a timestamp accurate to milliseconds and an operator identifier (human or agent). The timestamp ensures accurate recording of the order of events, allowing for a clear reconstruction of the temporal relationships between different operations. The operator identifier distinguishes whether the operation was triggered manually or by automated execution logic, thus enabling differentiated recording of human and machine behavior. This fine-grained identification method allows subsequent analysis to accurately identify the impact of different executing entities on the results.

[0105] In terms of data organization, a unified structured format is used for storage, enabling different types of data to be linked within the same system. For example, a complete operation record can contain data from multiple dimensions, such as strategy information, market status, decision-making logic, and execution results. These are linked through a unified identifier, allowing any transaction to be fully reconstructed. This structured recording method not only improves data management efficiency but also provides a solid data foundation for subsequent analysis.

[0106] In terms of post-mortem analysis, precise post-mortem analysis can be conducted based on actual operation logs and the final clearing results. The post-mortem process involves retrospectively analyzing log data, comparing the strategy generation, execution process, and final results to evaluate the performance of each stage. For example, by comparing the theoretically optimal return (the baseline return assuming perfectly accurate load forecasting) with the actual return, the contribution of forecasting error and execution efficiency to the final profit and loss can be quantified.

[0107] The theoretical optimal return represents the achievable return level under ideal conditions, i.e., perfectly accurate load forecasting and unbiased execution; this value serves as the evaluation benchmark. Actual return represents the return achieved during actual transactions. By comparing the two, sources of return deviation can be identified, such as deviations caused by forecasting errors or efficiency losses during execution. Furthermore, by analyzing the decision-making basis and execution path in the operation logs, specific influencing factors can be pinpointed, thus providing direction for subsequent strategy optimization.

[0108] Retrospective analysis can be conducted from multiple dimensions, including time, strategy, and execution. The time dimension allows for analysis of differences in strategy performance across different time periods; the strategy dimension assesses the impact of different strategy parameter settings on returns; and the execution dimension analyzes the influence of factors such as application timing and execution speed on the results. Through multi-dimensional analysis, a more comprehensive evaluation can be achieved.

[0109] Furthermore, the accumulated log data can support trend analysis and experience accumulation. For example, by statistically analyzing the performance of strategies under different market conditions, a better range of parameter configurations can be identified; by analyzing the reasons for frequent failures, risk control rule settings can be optimized, thereby improving overall execution stability.

[0110] During the overall operation, it also plays a role in data support and feedback. Through continuous recording and analysis, the execution results are fed back to the strategy optimization stage, enabling the system to have self-improvement capabilities. Although it does not directly participate in transaction execution, it plays a fundamental role in improving the overall performance of the system.

[0111] Through the above structural design, the entire transaction process can be fully recorded and analyzed, making each operation traceable, interpretable, and optimizable, thus providing reliable data support for subsequent strategy iterations and performance improvements.

[0112] This invention significantly improves the efficiency and accuracy of transaction execution. By constructing an integrated workflow system that connects data acquisition, strategy generation, and automated execution, it eliminates inefficient steps such as repeated data switching and manual data transfer between different systems, enabling continuous and automated operation of the transaction process. Traders no longer need to perform tedious manual data entry, allowing them to focus their main efforts on higher-value aspects such as strategy monitoring, parameter optimization, and anomaly handling. Simultaneously, the automated data entry mechanism, combined with multi-level verification logic, rigorously checks key parameters such as electricity volume, price, and frequency before submission, preventing errors caused by human intervention and significantly improving the stability and accuracy of transaction execution.

[0113] This invention realizes a scientific transformation of trading strategies from experience-driven to data-driven. By constructing a quantitative decision-making model and a full-link log recording system, the strategy formulation process becomes calculable, verifiable, and optimizable. Each strategy can be quantitatively simulated before execution and analyzed based on real data after execution, thus forming a complete evaluation loop. By continuously accumulating historical execution data and analysis results, strategy parameters can be systematically optimized, gradually transforming strategies from artistic judgments relying on personal experience to scientific decisions based on data feedback, effectively improving the stability, replicability, and long-term evolution capabilities of the strategies.

[0114] This invention constructs a proactive, dynamic, and reliable real-time risk control system, embedding risk control mechanisms into the entire transaction execution process, rather than relying on post-trade verification or static threshold limits. During execution, by continuously monitoring transaction progress and key state variables, the operable range is dynamically calculated and execution strategies are adjusted in real time, ensuring that risk control and trading behavior are synchronized. Especially in continuous matching trading scenarios, the remaining operable space can be continuously updated based on real-time transaction data, effectively avoiding the triggering of medium- to long-term position ratio constraints or buyback mechanisms due to the accumulation of multiple transactions, thereby keeping trading behavior within safe boundaries in a complex and rapidly changing market environment.

[0115] This invention enhances the system's adaptability and scalability to heterogeneous trading environments. By designing execution capabilities as independent layers, it allows for flexible adjustments to implementation methods based on the interface conditions of different trading platforms. When the trading environment provides a standard interface, direct system integration enables automated execution; when only a graphical interface is available, operations can be completed through simulated interaction, ensuring consistent execution results under various technical conditions. This layered structure makes strategy generation and execution processes independent, allowing upper-layer strategy logic to be quickly migrated to different markets or platforms, reducing system modification costs while improving overall scalability and application flexibility.

[0116] This invention provides, for example Figure 2 The system shown is a human-machine collaborative intelligent agent decision-making system for daily rolling matching transactions of electricity, including a data acquisition and processing module, a prediction and strategy generation module, an automatic monitoring and reporting module, and a data and log module. Data acquisition and processing module: used to collect historical data of long-term rolling matching market transactions in the spot electricity market, historical load data of agent users, historical meteorological data, new energy output data, historical data of thermal power bidding space and historical operation record data, and to perform unified processing on the collected data to form a basic historical data set containing market status information, load change information, new energy output information and operation behavior information. The prediction and strategy generation module is used to replace manual multi-tool analysis and provide data-driven decision-making. It performs decoupling of new energy output, calculation of photovoltaic proportional coefficient and curve fitting processing, generates estimated wind power output curves and photovoltaic output curves, calculates the target day bidding space and makes time-segmented adjustments, further performs quantitative prediction of electricity price based on bidding space and new energy output, and constructs strategy decision-making logic based on recycling rules and price difference signals, outputting standardized strategy instructions including target day, time period, trading direction, target price and matched volume. The automatic market monitoring and reporting module displays market price information, electricity price forecasts, and changes in open positions on a visual monitoring screen, and provides strategy configuration and monitoring parameter settings. It executes standardized strategy commands and includes a monitoring module, a decision-making module, and a reporting module. The monitoring module acquires market data and tracks the total matching target volume and the volume already traded to calculate the open electricity volume. The decision-making module generates transaction reporting decisions based on real-time prices using unidirectional or bidirectional monitoring logic. The reporting module completes transaction reporting and performs verification of the reported volume, price, frequency, and market volume. It also supports semi-automatic mode, fully automatic mode, and one-click pause. The data and log module is used to record strategy reception, market data, decision-making process, reporting instructions and execution results, and to perform transaction review and profit attribution analysis based on the recorded data.

[0117] The present invention provides a human-machine collaborative intelligent agent decision-making method for daily rolling matching of electricity transactions, which is implemented through the aforementioned human-machine collaborative intelligent agent decision-making system for daily rolling matching of electricity transactions. For details of the specific method and process of the human-machine collaborative intelligent agent decision-making system for daily rolling matching of electricity transactions, please refer to the embodiment of the aforementioned human-machine collaborative intelligent agent decision-making method for daily rolling matching of electricity transactions, which will not be repeated here.

[0118] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A human-machine collaborative intelligent agent decision-making method for daily rolling matching transactions of electricity, characterized in that, Includes the following steps: Collect historical data of medium- and long-term rolling matching transactions of electricity, historical load data of agent users, historical meteorological data, historical data of new energy output and thermal power bidding space, and uniformly store and process the collected data to form a historical data set that includes market status information, load change information and user private output data. The system performs load forecasting, bidding space forecasting, and electricity price forecasting on historical data sets, and generates standardized trading strategy instructions based on the forecast results, including the trading target day, time period, trading direction, target price, and matched volume. For standardized trading strategy instructions, the system obtains real-time rolling matching trading price and transaction information, calculates the difference between the total matching target volume and the already traded volume to obtain the current remaining operating power, and generates corresponding trading declaration decisions based on changes in the trading price. During the execution of transaction declarations, the transaction declaration decisions are subject to constraints on declaration volume, price range, declaration frequency, and market validity. Based on the verification results, automatic declarations or manual confirmations are completed. At the same time, all operational information during the transaction execution process is recorded for subsequent review and analysis.

2. The human-machine collaborative intelligent agent decision-making method for daily rolling matching transactions of electricity as described in claim 1, characterized in that, Regarding the handling of proportional relationships in the process of decomposing new energy output, stable load forecasting results are formed by analyzing and combining different types of output data. The steps are as follows: The system receives total new energy data and wind power day-ahead output data, performs timestamp matching on the two types of data, divides the data into 96 continuous time series data points according to 15-minute intervals, performs interpolation to complete the remaining time points, and then calculates the difference between the two types of data to form an estimated photovoltaic output data sequence. Read the day-ahead photovoltaic power output data, calculate the ratio of it to the estimated photovoltaic power output data hour by hour, generate a photovoltaic proportional coefficient sequence, and perform screening processing on abnormal proportional values, including setting upper and lower limits of the proportional and removing data points that exceed the range. The proportional coefficient series is grouped according to the date dimension, and the mean is calculated, the discrete points are smoothed and the continuity is corrected for each time segment to form stable proportional curve data. Perform trend scanning processing on the proportional curve to extract the starting point, peak point and ending point of change, construct the corresponding time index set, and output the load forecast data sequence.

3. The human-machine collaborative intelligent agent decision-making method for daily rolling matching transactions of electricity as described in claim 1, characterized in that, By combining the time series variation characteristics, and through key node extraction and data fitting processing, a continuous wind power output variation trend curve is constructed. The steps are as follows: Perform a full traversal analysis on the scale curve, identify the start point, end point and peak point by detecting changes in the curve slope, and record the corresponding time index and scale value; Based on the identification results, read the new energy output data and wind power ratio data at the corresponding time points, and map the two types of data in a unified manner to construct a key node data set; A weighted combination calculation is performed on the set of key nodes, the wind power output value of each node is derived according to the proportional relationship, and a set of discrete wind power output data points is formed by inter-node interpolation. Curve fitting is performed on discrete data points to generate a continuous wind power output trend curve, and the complete output sequence is output in chronological order.

4. The human-machine collaborative intelligent agent decision-making method for daily rolling matching transactions of electricity as described in claim 1, characterized in that, Based on the correlation between changes in new energy output and historical bidding space data, the future bidding space is derived and calculated. The steps are as follows: Extract the target date bidding space data and read the corresponding date new energy output data. Perform unified time granularity processing on the two types of data and construct a data alignment sequence. Differential calculations are performed on the continuous date renewable energy output data to obtain a time-by-time output change sequence, and the changes are marked as positive or negative. The output change is mapped to the bidding space data, and the original bidding space data is corrected time by time through addition and subtraction operations to form the bidding space calculation result for the target day. The calculation results are subjected to continuity verification processing, including outlier correction and sequence smoothing, and a complete bidding space prediction curve is generated.

5. The human-machine collaborative intelligent agent decision-making method for daily rolling matching transactions of electricity as described in claim 1, characterized in that, Based on the price differences between different markets, trading signals are generated by comparing multiple price data. The steps are as follows: Read medium- and long-term price data and day-ahead clearing forecast prices, and perform time alignment and format standardization processing on the two types of data; Perform a comparison operation at each time point, extract the minimum value from the medium- and long-term prices and the day-ahead clearing forecast prices, and generate a reference price series. Acquire real-time price data from the matching market, and calculate the difference between it and the reference price every hour to form initial price difference data; Perform sign recognition and numerical output processing on the price spread data to generate a standardized price spread signal sequence.

6. The human-machine collaborative intelligent agent decision-making method for daily rolling matching transactions of electricity as described in claim 1, characterized in that, Based on the changes in electricity volume during the transaction execution process, an operable electricity control mechanism is calculated using the difference between the target electricity volume and the transaction volume. The steps are as follows: Extract the total matching target from the strategy instructions and divide it into multiple time-segment target sequence according to the time dimension; Read the current real-time market transaction data, perform cumulative calculations on the declared and traded electricity volumes, and form a sequence of currently traded electricity volumes; The difference between the total matching target and the already traded volume is calculated at each time step to obtain the sequence of remaining eligible volumes. The remaining declared electricity is mapped to the current trading cycle to limit the range of declared electricity and update the execution status data.

7. The human-machine collaborative intelligent agent decision-making method for daily rolling matching transactions of electricity as described in claim 6, characterized in that, Based on the dynamic changes in the remaining available electricity volume sequence, the target volume sequence for each time period is synchronously corrected, and the correction result is matched and verified with the current traded electricity volume sequence to generate an updated operable electricity volume allocation result. At the same time, the allocation result is linked to the transaction declaration process to form continuous execution control data.

8. The human-machine collaborative intelligent agent decision-making method for daily rolling matching transactions of electricity as described in claim 1, characterized in that, Based on the price fluctuation characteristics and monitoring parameter configuration, a corresponding trading strategy is formed by determining the price range. The steps are as follows: Receive monitoring price parameters and simultaneously read real-time buyer and seller price data that meet the reporting rules to construct a set of order book price data; Compare the monitored price with the buyer's price, and output the result of the corresponding relationship judgment. The position of the monitored price within the range between the buyer's price and the seller's price is determined, forming a range judgment result; The comparison results and the interval judgment results are combined to generate the declared price data and the transaction direction data.

9. The human-machine collaborative intelligent agent decision-making method for daily rolling matching transactions of electricity as described in claim 1, characterized in that, Based on the multi-dimensional data conditions in the transaction declaration process, consistency checks are performed on the declaration behavior using multiple verification rules. The steps are as follows: The scope of the declared electricity volume is tested to determine whether it meets the minimum and maximum declared volume requirements, and the test results are marked. Perform range matching on the declared prices to determine whether they are within the preset price range and record the verification status; Perform statistical calculations on the number of applications within a unit of time, compare the statistical results with a preset threshold, and output the frequency detection results; The order book volume is filtered to identify orders below a preset threshold, and the corresponding price signals are filtered to form the final verification result.

10. A human-machine collaborative intelligent agent decision-making system for daily rolling matching of electricity transactions, used to implement the human-machine collaborative intelligent agent decision-making method for daily rolling matching of electricity transactions as described in any one of claims 1-9, characterized in that, It includes modules for data acquisition and processing, prediction and strategy generation, automated market monitoring and reporting, and data and logs. Data acquisition and processing module: used to collect historical data of long-term rolling matching market transactions in the spot electricity market, historical load data of agent users, historical meteorological data, new energy output data, historical data of thermal power bidding space and historical operation record data, and to perform unified processing on the collected data to form a basic historical data set containing market status information, load change information, new energy output information and operation behavior information. The prediction and strategy generation module is used to replace manual multi-tool analysis and provide data-driven decision-making. It performs decoupling of new energy output, calculation of photovoltaic proportional coefficient and curve fitting processing, generates estimated wind power output curves and photovoltaic output curves, calculates the target day bidding space and makes time-segmented adjustments, further performs quantitative prediction of electricity price based on bidding space and new energy output, and constructs strategy decision-making logic based on recycling rules and price difference signals, outputting standardized strategy instructions including target day, time period, trading direction, target price and matched volume. The automatic monitoring and reporting module displays market price information, electricity price forecast results, and changes in open positions through a visual monitoring screen. It also provides strategy configuration and monitoring parameter setting functions, and executes standardized strategy instructions. The module includes a monitoring module, a decision-making module, and a reporting module. The monitoring module is used to acquire market data and track the total matching target volume and the volume of transactions to calculate the open electricity volume. The decision-making module generates transaction reporting decisions based on real-time prices and executes one-way or two-way monitoring logic. The reporting module is used to complete transaction reporting and perform verification of reporting volume, price, frequency, and market volume. It also supports semi-automatic mode, fully automatic mode, and one-click pause. The data and log module is used to record strategy reception, market data, decision-making process, reporting instructions and execution results, and to perform transaction review and profit attribution analysis based on the recorded data.

Images